From the adiabatic theorem of quantum mechanics to topological states of matter

Jan Carl Budich

1,2
and Bjorn Trauzettel
2
1
Department of Physics, Stockholm University, Se-106 91 Stockholm, Sweden;
2
Institute for Theoretical Physics and Astrophysics,
University of W urzburg, 97074 W urzburg, Germany
(Dated: February 14, 2013)
Owing to the enormous interest the rapidly growing eld of topological states of matter (TSM)
has attracted in recent years, the main focus of this review is on the theoretical foundations of TSM.
Starting from the adiabatic theorem of quantum mechanics which we present from a geometrical
perspective, the concept of TSM is introduced to distinguish gapped many body ground states that
have representatives within the class of non-interacting systems and mean eld superconductors,
respectively, regarding their global geometrical features. These classifying features are topological
invariants dened in terms of the adiabatic curvature of these bulk insulating systems. We review the
general classication of TSM in all symmetry classes in the framework of K-Theory. Furthermore,
we outline how interactions and disorder can be included into the theoretical framework of TSM
by reformulating the relevant topological invariants in terms of the single particle Greens function
and by introducing twisted boundary conditions, respectively. We nally integrate the eld of TSM
into a broader context by distinguishing TSM from the concept of topological order which has been
introduced to study fractional quantum Hall systems.
I. INTRODUCTION
Stimulated by the theoretical prediction
13
and
experimental discovery
4
of the quantum spin Hall
(QSH) state, tremendous interest has been recently
attracted by the study of topologocical properties of
non-interacting band structures
57
. A topological state
of matter (TSM) can be understood as a non-interacting
band insulator which is topologically distinct from a
conventional insulator. This means that a TSM cannot
be adiabatically, i.e., while maintaining a nite bulk
gap, deformed into a conventional insulator without
breaking its fundamental symmetries. This classica-
tion approach provides a new paradigm in condensed
matter physics which goes beyond the mechanism of
local order parameters associated with spontaneous
symmetry breaking
8
: Dierent TSM dier in the value
of their dening topological invariant but concur in all
conventional symmetries. Interestingly, these global
topological features are not always immediately visible
in the microscopic equations of motion. However, the
bulk topology leads to unique nite size eects at the
boundary of a nite sample which has been coined bulk
boundary correspondence
7,9,10
. This general mechanism
gives rise to peculiar holographic transport properties
of topologically non-trivial systems. Predicting and
probing the rich phenomenology of these topological
boundary eects in mesoscopic samples has become one
of the most rapidly growing elds in condensed matter
physics in recent years.
The focus of this review is threefold. First, we view the
topological invariants characterizing TSM as the global
analogues of geometric phases associated with the adi-
abatic evolution of a physical system
1113
. Geometric
phases are well known to reect the local inner-geometric
properties of the Hilbert space of a system with a param-
eter dependent Hamiltonian
14,15
. The topological invari-
ants represent global features of a Bloch Hamiltonian,
i.e., of a system whose parameter space is its k-space.
This approach integrates TSM and physical phenomena
related to geometric phases into a common theoretical
context. Second, we present the topological classica-
tion of TSM in the framework of K-Theory
16
in a more
accessible way than in the pioneering work by Kitaev
17
.
Third, we discuss very recent developments concerning
the generalization of the relevant topological invariants
to interacting and disordered systems several of which
were published after existing reviews on TSM.
II. THE ADIABATIC THEOREM
The gist of the adiabatic theorem can be understood
at a very intuitive level: Once prepared in an instanta-
neous eigenstate with an eigenvalue which is separated
from the neighboring states by a nite energy gap , the
system can only leave this state via a transition which
costs a nite excitation energy . A simple way to esti-
mate whether such a transition is possible is to look at
the Fourier transform

1() of the time dependent Hamil-
tonian 1(t). If the time dependence of 1 is made su-
ciently slow,

1() will only have nite matrix elements
for . In this regime the system will stick to the
same instantaneous eigenstate. This behavior is known
as the adiabatic assumption.
A. Proof due to Born and Fock
The latter rather intuitive argument is at the heart of
the adiabatic theorem of quantum mechanics which has
been rst proven by Born and Fock in 1928
18
for non-
degenerate systems. Let [n(t)
n
be an orthonormal
a
r
X
i
v
:
1
2
1
0
.
6
6
7
2
v
2


[
c
o
n
d
-
m
a
t
.
m
e
s
-
h
a
l
l
]


1
3

F
e
b

2
0
1
3
2
set of instantaneous eigenstates of 1(t) with eigenvalues
E
n
(t)
n
. The exact solution of the Schrodinger equation
can be generally expressed as
[(t) =

n
c
n
(t)[n(t)e
i
n
D
(t)
, (1)
where the dynamical phase
n
D
(t) =
_
t
t0
E
n
()d has
been separated from the coecients c
n
(t) for later conve-
nience. Plugging Eq. (1) into the Schrodinger equation
yields
c
n
= c
n
n[
d
dt
[n

m=n
c
m
n[
_
d
dt
1
_
[m
E
m
E
n
e
i
nm
D
(t)
(2)
with
nm
D
(t) =
n
D
(t)
m
D
(t). The salient consequence
of the adiabatic theorem is that the last term in Eq.
(2) can be neglected in the adiabatic limit since its de-
nominator [E
n
E
m
[ is nite whereas the ma-
trix elements of
d
dt
1 become arbitrarily small. More
precisely, if we represent the physical time as t = Ts,
where s is of order 1 for a change in the Hamiltonian
of order and T is the large adiabatic timescale, then
d
dt
=
1
T
d
ds
. Now,
d
ds
1(t(s)) is by construction of order
. The entire last term in Eq. (2) is thus of order
1
T
.
Under these conditions
19
, Born and Fock
18
showed that
the contribution of this second term vanishes in the adi-
abatic limit T . Note that this is not a trivial result
since the dierential equation (2) is supposed to be in-
tegrated from t = 0 to t T, so that one could naively
expect a contribution of order 1 from a coecient that
scales like 1/T. The coecient of c
n
in the rst term on
the right hand side of Eq. (2) is purely imaginary since
0 =
d
dt
n[n = (
d
dt
n[)[n + n[
d
dt
[n and hence does not
change the modulus of c
n
when the dierential equation
c
n
= c
n
n[
d
dt
[n is solved as
c
n
(t) = c
n
(t
0
) e

t
t
0
n|
d
d
|nd
. (3)
Born and Fock
18
argue that n[
d
dt
[n = 0
t
amounts to
a choice of phase for the eigenstates and therefore ne-
glect also the rst term on the right hand side of Eq. (2).
This review article is mainly concerned with physical
phenomena associated with corrections to this in general
unjustied assumption.
B. Notion of the geometric phase
By neglecting the rst term on the right hand side
of Eq. (2), Ref. 18 overlooks the potentially nontrivial
adiabatic evolution, known as Berrys phase
12
, associ-
ated with a cyclic time dependence of 1. After a period
[0, T] of such a cyclic evolution, Eq. (3) yields
c
n
(T) = c
n
(0)e

T
0
n|
d
d
|nd
. (4)
To understand why the phase factor e

T
0
n|
d
d
|nd
can
in general not be gauged away, we remember that
the Hamiltonian depends on time via the time depen-
dence R(t) of some external control parameters. Hence,
n[
d
dt
[n = n[

[n

R

, where

=

R

. To reveal the
mathematical structure of the latter expression, we dene
/
B
_
d
dt
_
= /
B

= in[

[n

R

, (5)
where /
B
= A
B

dR

is called Berrys connection.

/
B
clearly has the structure of a gauge eld: Under the
local gauge transformation [n e
i
[n with a smooth
function R (R), Berrys connection transforms like
/
B
/
B
+d.
Furthermore, the cyclic evolution denes a loop : t
R(t), t [0, T] , R(0) = R(T) in the parameter manifold
!. If can be expressed as the boundary of some piece
of surface, then, using the theorem of Stokes, we can
calculate
i
_
T
0
n[
d
d
[nd =
_

/
B
=
_
S
d/
B
=
_
S
T
B
, (6)
where in the last step Berrys curvature T
B
= T
B

dR

dR

is dened as
T
B

= i (

n[

n[

n) = 2Im

n[

n
with the shorthand notation [

n =

[n. Note that

T
B
is a gauge invariant quantity that is analogous to the
eld strength tensor in electrodynamics. Dening the
Berry phase associated with the loop as
B

=
_

/
B
=
_
S
T
B
we can rewrite Eq. (4) as
c
n
(T) = c
n
(0)e
i
B

. (7)
The manifestly gauge invariant Berry phase
B

can have
observable consequences due to interference eects be-
tween coherent superpositions that undergo dierent adi-
abatic evolutions. The analogue of this phenomenology
due to an ordinary electromagnetic vector potential is
known as the Aharonov-Bohm eect
20
. The geometri-
cal reason why Berrys connection /
B
cannot be gauged
away all the way along a cyclic adiabatic evolution is the
same as why a vector potential cannot be gauged away
along a closed path that encloses magnetic ux, namely
the notion of holonomy on a curved manifold. We will
come back to the concept of holonomy shortly from a
more mathematical point of view. For now we only com-
ment that the Berry phase
B

is a purely geometrical
quantity which only depends on the inner-geometrical re-
lation of the family of states [n(R) along the loop and
reects an abstract notion of curvature in Hilbert space
which has been dened as Berrys curvature T
B
.
3
C. Proof due to Kato
For a degenerate eigenvalue, Berrys phase is promoted
to a unitary matrix acting on the corresponding degener-
ate eigenspace
13
. The rst proof of the adiabatic theorem
of quantum mechanics that overcomes both the limita-
tion to non-degenerate Hamiltonians and the assumption
of an explicit phase gauge for the instantaneous eigen-
states was reported in the seminal work by Tosio Kato
11
in 1950. We will review Katos results briey for the
readers convenience and use his ideas to illustrate the
geometrical origin of the adiabatic phase. The explicit
proofs are presented at a very elementary and self con-
tained level in Ref. 11. Our notation follows Ref. 21
which is convenient to relate the physical quantities to
elementary concepts of dierential geometry.
Let us assume without loss of generality that the sys-
tem is at time t
0
= 0 in its instantaneous ground state
[
0
(0) or, more generally, since the ground state might
be degenerate, in a state [ satisfying
P(0)[ = [, (8)
where P(t) is the projector onto the eigenspace associated
with the instantaneous ground state energy E
0
(t) which
is dened as
P(t) =
1
2i
_
c
dz
z 1(t)
,
where the complex contour c encloses E
0
(t) which is again
assumed to be separated from the spectrum of excita-
tions by a nite energy gap > 0. To understand the
adiabatic evolution, we are not interested in the dynam-
ical phase
D
(t) =
_
t
0
E
0
()d. We thus dene a new
time evolution operator

|(t, 0) = e
i
D
(t)
|(t, 0). Clearly,

| represents the exact time evolution operator of a sys-

tem which has the same eigenstates as the original system
but has been subjected to a time dependent energy shift
that transforms E
0
(t)

E
0
(t) = 0
t
. Kato proved the
adiabatic theorem in a very constructive way by writing
down explicitly the generator / of the adiabatic evolu-
tion:
/
_
d
dt
_
=
_

P, P
_
. (9)
In the adiabatic limit,

|(t, 0)P(0) was shown
11
to con-
verge against the adiabatic Kato propagator /, i.e.,

|(t, 0)P(0)
adiabatic limit
/(t, 0) = T e

t
0
A(
d
d
)d
. (10)
The adiabatic assumption is now a direct corollary from
Eq. (10) and can be elegantly expressed as
21
P(t)/(t, 0) = /(t, 0)P(0), (11)
implying that a system, which is prepared in an instan-
taneous ground state at t
0
= 0, will be propagated to a
state in the subspace of instantaneous ground states at
t by virtue of Katos propagator /. Note that / is a
completely gauge invariant quantity, i.e., independent of
the choice of basis in the possibly degenerate subspace of
ground states. The Kato propagator /(T, 0) associated
with a cyclic evolution in parameter space thus yields the
Berry phase
12
and its non-Abelian generalization
13
, re-
spectively. We will call this general adiabatic phase the
geometric phase (GP) in the following. The GP /

rep-
resenting the adiabatic evolution along a loop in pa-
rameter space can be expressed in a manifestly gauge
invariant way as
/

= T e

A
. (12)
Katos propagator is the solution of an adiabatic ana-
logue of the Schrodinger equation, an adiabatic equation
of motion that can be written as
_
d
dt
+/
_
d
dt
__
[(t) = 0, (13)
for states satisfying P(t)[(t) = [(t), i.e., states
in the subspace of instantaneous groundstates. Before
closing the section, we give a general and at least numer-
ically always viable recipe to calculate the Kato propaga-
tor /(t, 0). We rst discretize the time interval [0, t] into
n steps by dening t
i
= i
t
n
. The discrete version of Eq.
(13) for the Kato propagator reads (see Eq. (9))
/(t
i
, 0) /(t
i1
, 0) = (P(t
i
) P(t
i1
)) P(t
i1
)
P(t
i
) (P(t
i
) P(t
i1
)) /(t
i1
, 0). (14)
Using P(t
i1
)/(t
i1
, 0) = /(t
i1
, 0) and P
2
= P, Eq.
(14) can be simplied to
/(t
i
, 0) = P(t
i
)/(t
i1
, 0),
which is readily solved by K(t
i
, 0) =

i
j=0
P(t
j
). Taking
the continuum limit yields
13,14,21
/(t, 0) = lim
n
n

i=0
P(t
i
), (15)
which is a valuable formula for the practical calculation
of the Kato propagator.
III. GEOMETRIC INTERPRETATION OF
ADIABATIC PHASES
We now view the adiabatic time evolution as an ab-
stract notion of parallel transport in Hilbert space and re-
veal the GP associated with a cyclic evolution as the phe-
nomenon of holonomy due to the presence of curvature
in the vector bundle of ground state subspaces over the
manifold ! of control parameters. Interestingly, Katos
approach to the problem provides a gauge invariant, i.e.,
a global denition of the geometrical entities connection
4
and curvature, whereas standard gauge theories are de-
ned in terms of a complete set of local gauge elds along
with their transition functions dened in the overlap of
their domains. This dierence has an interesting physi-
cal ramication: Quantities that are gauge dependent in
an ordinary gauge theory like quantum chromodynamics
(QCD) are physical observables in the theory of adia-
batic time evolution. To name a concrete example, only
gauge invariant quantities like the trace of the holonomy,
also known as the Wilson loop, are observable in QCD
whereas the holonomy itself, in other words the GP de-
ned in Eq.(12), is a physical observable in Katos theory.
This subtle dierence has been overlooked in standard
literature on this subject
15,22
which we interpreted as an
incentive to clarify this point below in greater detail.
A. Adiabatic time evolution and parallel transport
To get accustomed to parallel transport, we rst
explain the general concept with the help of a very
elementary example, namely a smooth piece of two
dimensional surface embedded in R
3
. If the surface is
at, there is a trivial notion of parallel transport of
tangent vectors, namely shifting the same vector in the
embedding space from one point to another. However,
on a curved surface, this program is ill-dened, since
a tangent vector at one point might be the normal
vector at another point of the surface. Put shortly,
a tangent vector can only be transported as parallel
as the curvature of the surface admits. On a curved
surface, parallel transport along a curve is thus dened
as a vanishing in-plane component of the directional
derivative, i.e., a vanishing covariant derivative of a
vector eld along a curve. The normal component of
the directional derivative reects the rotation of the
entire tangent plane in the embedding space and is
not an inner-geometric quantity of the surface as a two
dimensional manifold.
The analogue of the curved surface in the context of
adiabatic time evolution is the manifold of control pa-
rameters !, parameterizing for example external mag-
netic and electric elds. The analogue of the tangent
plane at each point of the surface is the subspace of de-
generate ground states of the Hamiltonian 1(R) at each
point R in parameter space. An adiabatic time depen-
dence of 1 amounts to traversing a curve t R(t) in
! at adiabatically slow velocity. A cyclic evolution is
uniquely associated with a loop in !. We will now ex-
plicitly show that the adiabatic equation of motion (13)
denes a notion of parallel transport in the ber bundle of
ground state subspaces over ! in a completely analogous
way as the ordinary covariant derivative on a smooth
surface denes parallel transport in the tangent bundle of
the smooth surface. We rst note that
d
dt
=

R

is re-
ferring to a particular direction

R

in parameter space,
which depends on the choice of the adiabatic time de-
pendence of 1. We can get rid of this dependence by
rephrasing Eq. (13) as
D[ = (d +/) [ = 0, (16)
where the adiabatic derivative D = d + / has been
dened, / = [(dP), P] and here as in the following
P[ = [. The R-dependence has been dropped for
notational convenience. The adiabatic derivative D takes
a tangent vector, e.g.,
d
dt
, as an argument to boil down to
the directional adiabatic derivative
d
dt
+ /
_
d
dt
_
appear-
ing in Eq. (13). For the following analysis the identities
P
2
= P and P[ = [ are of key importance. It is now
elementary algebra to show
P(dP)P = 0. (17)
Eq. (17) has a simple analogue in elementary geometry:
Consider the family of unit vectors n(t)
t
where t pa-
rameterizes a curve on a smooth surface. Then, since
1 = n[n, we get 0 =
d
dt
n[n = 2n[ n, i.e., the change
of a unit vector is perpendicular to the unit vector itself.
Using Eq. (17), we immediately derive P/[ = 0 and
with that
D[ = 0 Pd[ = 0. (18)
This makes the analogy of our adiabatic derivative D to
the ordinary notion of parallel transport manifest: [ is
parallel-transported if the in-plane component of its
derivative vanishes.
Curvature and holonomy
Let us again start with a very simple example of a
curved manifold, a two dimensional sphere S
2
, which
has constant Gaussian curvature. Parallel-transporting
a tangent vector around a geodesic triangle, say the
boundary of an octant of the sphere gives a defect angle
which is proportional to the area of the triangle or, more
precisely, the integral of the Gaussian curvature over the
enclosed area. This defect angle is called the holonomy
of the traversed closed path. This elementary example
suggests that the presence of curvature is in some sense
probed by the concept of holonomy. This intuition is
absolutely right. As a matter of fact, the generalized
curvature at a given point x of the base manifold of
a ber bundle is dened as the holonomy associated
with an innitesimal loop at x. More concretely, the
curvature is usually dened as

= [

] which
represents an innitesimal parallel transport around a
parallelogram in the -plane.
In total analogy, we dene
T

[ = [D

, D

] [ = P [P

, P

] P[, (19)
with the shorthand notation P

P. Restricting the
domain of T to states which are in the projection P, we
5
can rewrite Eq. (19) as the operator identity
T = T

dR

dR

= P [(dP) (dP)] P. (20)

In the general case of a non-Abelian adiabatic connec-
tion, i.e., if the dimension of P is larger than 1, we cannot
simply use Stokes theorem to reduce the evaluation of Eq.
(12) to a surface integral of T over the surface bounded
by , as has been done in the case of the Abelian Berry
curvature in Eq. (6). However, the global one to one cor-
respondence between curvature and holonomy still exists
and is the subject of the Ambrose-Singer theorem
23
.
Relation between Katos and Berrys language
In order to make contact to the more standard lan-
guage of gauge theory, we will now express Katos mani-
festly gauge invariant formulation
11
in local coordinates
thereby recovering Berrys connection /
B12,14
and its
non-Abelian generalization
13
, respectively. For this pur-
pose, let us x a concrete basis [(R)

, R O ! in
an open subset O of the parameter manifold. We assume
the loop to lie inside of O. Otherwise we would have
to switch the gauge while traversing the loop. We will
drop the R-dependence of [ right away for notational
convenience. The projector P can then be represented as
P =

[[. Let us start the cyclic evolution without

loss of generality with [(0) = [(0). From Eq. (11) we
know that the solution [(t) = /(t, 0)[(0) of Eq. (13)
satises P(t)[(t) = [(t) at every point in time during
the cyclic evolution. Hence, we can represent [(t) in
our gauge as
[ =

[[ = U
B

[, (21)
where the t-dependence has been dropped for brevity.
From Eq. (18), we know that P
d
dt
[ = 0 which implies
[
d
dt
[ = 0. Plugging this into Eq. (21) yields
d
dt
U
B

[
d
dt
[U
B

. (22)
Redening /
B
for the non-Abelian case as a matrix val-
ued gauge eld through /
B

= i[

[dR

, Eq. (22)
is readily solved as
U
B
(t) = T e
i

t
0
A
B
(
d
d
)d
.
The representation matrix of the GP associated with the
loop then reads
U
B

= T e
i

A
B
. (23)
By construction, U
B

is the representation matrix of the

GP /

, i.e.,
_
U
B

_
,
= (0)[/

[(0),
or, more general, for any point in time along the path
_
U
B
(t)
_
,
= (t)[/(t, 0)[(0). (24)
Eq. (24) makes the relation between Katos formulation
of adiabatic time evolution and the non-Abelian Berry
phase manifest. In contrast to the gauge independence
of Katos global connection /, /
B
behaves like a local
connection (see Ref. 24 for rigorous mathematical def-
initions) and depends on the gauge, i.e., on our choice
of the family [(R)

of basis states. Under a smooth

family of basis transformations U(R)
R
acting on the
local coordinates /
B
transforms like
23,24
/
B

/
B
= U
1
/
B
U +U
1
dU (25)
resulting in the following gauge dependence of Eq. (23),
U
B

U
B

= U
1
U
B

U, (26)
which only depends on the basis choice U = U (R(0)) at
the starting point of the loop .
Inserting our representation P =

[[ into the
gauge independent form of the curvature, Eq.(20), we
readily derive
T
B,

= [ [P

, P

] [ = (d/
B
)

+ (/
B
A
B
)

,
which denes T
B
as the usual curvature of a non-Abelian
gauge eld
23
, i.e.,
T
B
= d/
B
+/
B
/
B
, (27)
which transforms under a local gauge transformation
U like
T
B
U
1
T
B
U.
B. Gauge dependence and physical observability
The gauge dependence of the non-Abelian Berry
phase U
B

(see Eq. (26)) has led several authors

15,22
to
the conclusion that only gauge independent features like
the trace and the determinant of U
B

can have physical

meaning. However, working with Katos manifestly
gauge invariant formulation, it is understood that the
entire GP /

is experimentally observable. In the

remainder of this section we will try to shed some light
on this ostensible controversy.
In gauge theory, it goes without saying that explicitly
gauge dependent phenomena are not immediately physi-
cally observable and that only the gauge invariant infor-
mation resulting from a calculation performed in a spe-
cial gauge can be of physical signicance. At a formal
level this is a direct consequence of the fact that the La-
grangian of a gauge theory is constructed in a manifestly
6
gauge invariant way by tracing over the gauge space in-
dices. The physical reason for this is quite simple: A
concrete gauge amounts to a local choice of the coordi-
nate system in the gauge space. Under a local change
of basis, a non-abelian gauge eld A transforms like (see
also Eq. (25))
A

A = U
1
AU +U
1
dU,
where U(x) is a smooth family of basis transformations,
with x labeling points in the base space of the theory,
e.g., in Minkowski space. Now, since the gauge space is
an internal degree of freedom, the basis vectors in this
space are not associated with physical observables. This
situation is fundamentally changed in Katos adiabatic
analogue of a gauge theory. Here, the non-Abelian struc-
ture is associated with a degeneracy of the Hamiltonian,
e.g., Kramers degeneracy in the presence of time rever-
sal symmetry (TRS). For a system in which spin is a
good quantum number, Kramers degeneracy is just spin
degeneracy, which makes the spin the analogue of the
gauge degree of freedom in an ordinary gauge theory.
However, the magnetic moment associated with a spin is
a physical observable which can be measured. The basis
vectors, e.g., [, [ have an objective meaning for the ex-
perimentalist (a magnetic moment that points from the
lab-oor to the sky which we call z- direction). For con-
creteness, let us assume that we have calculated a GP
/

= [[ +[[. The representation matrix of /

in
this basis of S
z
eigenstates is clearly the Pauli matrix

x
. Choosing a dierent gauge, i.e., a dierent basis for
the gauge degree of freedom at the starting point of the
cyclic adiabatic evolution, we of course would have ob-
tained a dierent representation matrix U
B

for /

, e.g.,

z
, had we chosen the basis as eigenstates of S
x
(see Eq.
(26)). However, the fact that /

rotates a spin which is

initially pointing to the lab-ceiling upside down is gauge
independent physical reality.
IV. FROM GEOMETRY TO TOPOLOGY
In Section III, we worked out the relation between
the GP and the notion of curvature as a local geometric
quantity. The topological invariants characterizing TSM
are in some sense global GPs. They measure global
properties which cannot be altered by virtue of local
continuous changes of the physical system. Continuous is
at this stage of the analysis synonymous with adiabatic,
i.e., happening at energies below the bulk gap. Later
on, we will additionally require local continuous changes
to respect the fundamental symmetries of the physical
system, e.g., particle hole symmetry (PHS) or time
reversal symmetry (TRS).
Let us illustrate the correspondence between local cur-
vature and global topology of a manifold with the help
of the simplest possible example. We consider a two di-
mensional sphere S
2
with radius r. This manifold has a
constant Gaussian curvature of =
1
r
2
. The integral of
over the entire sphere obviously gives 4, independent
of r. The Gauss-Bonnet theorem in its classical form
(see, e.g., Ref. 25) relates precisely this integral of the
Gaussian curvature of a closed smooth two dimensional
manifold / to its Euler characteristic in the following
way:
1
2
_
M
= (/) (28)
Note that is a purely algebraic quantity which is de-
ned as the number of vertices minus the number of edges
plus the number of faces of a triangulation
26
of /. is
by construction of simplicial homology
26
a topological in-
variant which can only be changed by poking holes into
/ and gluing the resulting boundaries together so as
to create closed manifolds with dierent genus. Hence,
Eq. (28) nicely demonstrates how the integral of the local
inner-geometric quantity over the entire manifold yields
a topological invariant. Concretely, for our example S
2
, a
triangulation is provided by continuously deforming the
sphere into a tetrahedron. Simple counting of vertices,
edges, and faces yields (S
2
) = 46+4 = 2, in agreement
with Eq. (28). More generally speaking, e =

2
is our
rst encounter with a characteristic class
27
, the so called
Euler class of /, which upon integration over / yields
the topological invariant . Similar mathematical struc-
tures will be ubiquitous when it comes to the classica-
tion of TSM. The simplest example in this context is the
quantum anomalous Hall (QAH) state
2830
, a 2D insu-
lating state in symmetry class A (see Section VA) which
is characterized by its rst Chern number (see Section
VI A for a detailed discussion)
(
1
=
_
BZ
iTr [T]
2
. (29)
The formal analogy between Eq. (28) and Eq. (29) is
striking. The parameter space over which the adiabatic
curvature T is integrated in Eq. (29) is the k-space of
the physical system, i.e., the Brillouin zone (BZ) for a
periodic system, which has the topology of a torus. In
Section II, we showed that the GP can be viewed as the
ux through the surface in parameter space which is
bounded by the corresponding cyclic adiabatic evolution.
Along similar lines, the rst Chern number is analogous
to a monopole charge enclosed by the entire BZ of the
QAH insulator.
The close correspondence between adiabatic evolution
and the rst Chern number has also been discussed in
Refs.
3133
: For a 2D system in cylinder geometry, a cir-
cumferential electric eld can be modeled by a time de-
pendent magnetic ux threading the cylinder in axial di-
rection. In this scenario, the rst Chern number can
be viewed as the shift of the charge polarization in ax-
ial direction associated with the adiabatic threading of
one ux quantum. Along these lines, Laughlin
31
had in-
terpreted the quantum Hall eect as an adiabatic charge
7
pumping process, even before the formal relation between
the Hall conductivity and the rst Chern number was es-
tablished in Refs. 3436.
V. BULK CLASSIFICATION OF ALL
NON-INTERACTING TSM
Very generally speaking, the understanding of TSM
can be divided into two subproblems. First, nding
the group that represents the topological invariant for
a class of systems characterized by their fundamental
symmetries and spatial dimension. Second, assigning
the value of the topological invariant to a representative
of such a symmetry class, i.e., measuring to which
topological equivalence class a given system belongs. We
address the rst problem in this section and the second
problem in VI.
The general idea that yields the entire table of TSM
is quite simple: In addition to requiring a bulk insulat-
ing gap, physical systems of a given spatial dimension
are divided into 10 symmetry classes distinguished by
their fundamental symmetries, i.e., TRS, PHS, and chiral
symmetry (CS)
37
. The topological properties of the cor-
responding Cartan symmetric spaces of quadratic candi-
date Hamiltonians determine the group of possible topo-
logically inequivalent systems. We outline the mathemat-
ical structure behind this general classication scheme in
some detail. First, we briey review the construction of
the ten universality classes
37
. Then, we present the asso-
ciated topological invariants for non-interacting systems
of arbitrary spatial dimension giving a complete list of all
TSM
6
that can be distinguished by virtue of this frame-
work. Finally, we discuss in some detail the origin of
characteristic patterns appearing in this table using the
framework of K-Theory along the lines of the pioneering
work by Kitaev
17
.
A. Cartan-Altland-Zirnbauer symmetry classes
A physical system can have dierent types of symme-
tries. An ordinary symmetry
6
is characterized by a set of
unitary operators representing the symmetry operations
that commute with the Hamiltonian. The inuence of
such a symmetry on the topological classication can
be eliminated by transforming the Hamiltonian into a
block-diagonal form with symmetry-less blocks. The
total system then consists of several uncoupled copies
of symmetry-irreducible subsystems which can be clas-
sied individually. In contrast, the extremely generic
symmetries
6
follow from the anti-unitary operations of
TRS and PHS. Involving complex conjugation according
to Wigner, they impose certain reality conditions on
the system Hamiltonian. In total, the behavior of the
system under these operations, and their combination,
the CS operation, denes ten universality classes which
we call the Cartan-Altland-Zirnbauer (CAZ) classes.
For disordered systems, these classes correspond to
ten distinct renormalization group (RG) low energy
xed points in random matrix theory
37
. The spaces of
candidate Hamiltonians within these symmetry classes
correspond to the ten symmetric spaces introduced by
Cartan in 1926
38
dened in terms of quotients of Lie
groups represented in the Hilbert space of the system.
For translation-invariant systems, the imposed reality
conditions are inherited by the Bloch Hamiltonian
h(k) (see Eq. (33) below).
In the following, the anti-unitary TRS operation will
be denoted by T and the anti-unitary PHS will be de-
noted by (. The Hamiltonian 1 of a physical system
satises these symmetries if
T 1T
1
= 1, (30)
and
(1(
1
= 1, (31)
respectively. According to Wigners theorem, these anti-
unitary symmetries can be represented as a unitary op-
eration times the complex conjugation K. We dene
T = TK, ( = CK. Using the unitarity of T, C along
with 1 = 1

we can rephrase Eqs. (30-31) as

T1
T
T

= 1,
C1
T
C

= 1. (32)
There are two inequivalent realizations of these anti-
unitary operations distinguished by their square which
can be plus identity or minus identity. For example,
T
2
= 1 for the unfolding of a particle with integer/half-
integer spin, respectively. Clearly, T
2
= 1 TT

=
1 and (
2
= 1 CC

= 1. In total, there are thus

nine possible ways for a system to behave under the two
anti-unitary symmetries: each symmetry can be absent,
or present with square plus or minus identity. For eight
of these nine combinations, the behavior under the com-
bination T ( is xed. The only exception is the so called
unitary class which breaks both PHS and TRS and can
either obey or break their combination, the CS. This class
hence splits into two universality classes which add up to
a grand total of ten classes shown in Tab. I. For a pe-
riodic system, symmetry constraints similar to Eq. (32)
hold for the Bloch Hamiltonian h(k), namely
Th
T
(k)T

= h(k),
Ch
T
(k)C

= h(k), (33)
where T, C now denote the representation of the unitary
part of the anti-unitary operations in band space.
For a continuum model, the real space Hamiltonian
H(x) is dened through
1 =
_
d
d
x

(x)H(x)(x),
8
Class TRS PHS CS
A (Unitary) 0 0 0
AI (Orthogonal) +1 0 0
AII (Symplectic) -1 0 0
AIII (Chiral Unitary) 0 0 1
BDI (Chiral Orthogonal) +1 +1 1
CII (Chiral Symplectic) -1 -1 1
D 0 +1 0
C 0 -1 0
DIII -1 +1 1
CI +1 -1 1
TABLE I. Table of the CAZ universality classes. 0 denotes
the absence of a symmetry. For PHS and TRS, 1 denotes the
square of a present symmetry, the presence of CS is denoted
by 1. The last four classes are Bogoliubov de Gennes classes
of mean eld superconductors where the superconducting gap
plays the role of the insulating gap.
where is a vector/spinor comprising all internal degrees
like spin, particle species, etc.. The k-space on which the
Fourier transform

H(k) of H(x) is dened does not have
the topology of a torus like the BZ of a periodic system.
However, the continuum models one is concerned with in
condensed matter physics are eective low energy/large
distance theories. For large k,

H(k) will thus generically
have a trivial structure so that the k-space can be en-
dowed with the topology of the sphere S
d
by a one point
compactication which maps k to a single point.
The symmetry constraints on

H(k) have the same form
as those on the Bloch Hamiltonian h(k) shown in Eq.
(33). By abuse of notation, we will denote both

H(k) and
h(k) by h(k). Nevertheless, we will point out several dif-
ferences between periodic systems and continuum models
along the way.
B. Denition of the classication problem for
continuum models and periodic systems
For translation-invariant insulating systems with n oc-
cupied and m empty bands and continuum models with
n occupied and m empty fermion species, respectively,
the projection P(k) =

n
=1
[u

(k)u

(k)[ onto the oc-

cupied states is the relevant quantity for the topological
classication. The spectrum of the system is not of in-
terest for adiabatic quantities as long as a bulk gap be-
tween the empty and the occupied states is maintained.
We thus deform the system adiabatically into a at band
insulator, i.e., a system with eigenenergy

= 1 for all
occupied states and eigenenergy
+
= +1 for all empty
states. The eigenstates are not changed during this defor-
mation. The Hamiltonian of this at band system then
reads
39,40
Q(k) = (+1) (1 P(k)) + (1)P(k) = 1 2P(k)
Obviously, Q
2
= 1, Tr [Q] = m n. Without further
symmetry constraints, Q is an arbitrary U(n+m) matrix
which is dened up to a U(n) U(m) gauge degree of
freedom corresponding to basis transformations within
the subspaces of empty and occupied states, respectively.
Thus, Q is in the symmetric space
G
n+m,m
(C) = G
n+m,n
(C) = U(n +m)/(U(n) U(m)).
Geometrically, the complex Grassmannian G
k,l
(C) is a
generalization of the complex projective plane and is de-
ned as the set of l-dimensional planes through the ori-
gin of C
k
. The set of topologically dierent translation-
invariant insulators is then given by the group g of homo-
topically inequivalent maps k Q(k) from the BZ T
d
of
a system of spatial dimension d to the space G
n+m,m
(C)
of possible Bloch Hamiltonians. For continuum models
T
d
is replaced by S
d
and g is by denition given by
g =
d
(G
n+m,m
(C)) , (34)
where the n-th homotopy group
n
of a space is by
denition the group of homotopically inequivalent
maps from S
d
to this space. For translation-invariant
systems dened on a BZ, the classication can be
more complicated than Eq. (34) if the lower homo-
topy groups
s
, s = 1, . . . , d 1 are nontrivial. For
the quantum anomalous Hall (QAH) insulator
28
, a
2D translation-invariant state which does not obey
any fundamental symmetries, we can infer from

2
(G
n+m,m
(C)) = Z,
1
(G
n+m,m
(C)) = 0 that an inte-
ger topological invariant must distinguish possible states
of matter in this symmetry class, i.e., possible maps
T
2
G
n+m,m
(C). The condition
1
(G
n+m,m
(C)) = 0 is
necessary because the
2
classies maps from S
2
, which
is only equivalent to the classication of physical maps
from T
2
if the fundamental group
1
of the target space
is trivial. The dierence between the base space of a
periodic systems which is a torus and of continuum
models which has the topology of a sphere has interesting
physical ramications: The so called weak topological
insulators are only topologically distinct over a torus
but not over a sphere. Physically, this is visible in the
lacking robustness of these TSM which break down with
the breaking of translation symmetry.
Requiring further symmetries as appropriate for the
other nine CAZ universality classes is tantamount to
imposing symmetry constraints on the allowed maps
T
d
G
n+m,m
(C), k Q(k) for translation-invariant
systems and S
d
G
n+m,m
(C), k Q(k) for continuum
models, respectively. The set of topologically distinct
physical systems is then still given by the set of homo-
topically inequivalent maps within this restricted space,
i.e., the space of maps which cannot be continuously de-
formed into each other without breaking a symmetry con-
straint. For example, for the chiral classes characterized
9
by CS = 1, Q can be brought into the o diagonal form
40
Q =
_
0 q
q

0
_
with qq

= 1, which reduces the corresponding target

space to U(n). For the chiral unitary class AIII with-
out further symmetry constraints, the calculation of
g amounts to calculating g =
d
(U(n)) = Z for odd d and
g =
d
(U(n)) = 0 for even d, respectively, provided
n (d + 1)/2. Additional symmetries will again impose
additional constraints on the map T
d
U(n), k q(k).
This procedure rigorously denes the group g of topo-
logical equivalence classes for non-interacting translation-
invariant insulators in arbitrary spatial dimension and
CAZ universality class. However, the practical calcula-
tion of g can be highly non-trivial and has been achieved
for continuum models via various subtle detours, for ex-
ample the investigation of surface nonlinear -models, in
Refs. 6 and 41. In the following, we outline a mathemat-
ical brute force solution to the classication problem in
terms of K-Theory which has been originally introduced
in the seminal work by Kitaev in 2009
17
. This method
will naturally explain the emergence of weak topological
insulators. We will not assume any prior knowledge on
K-Theory. In Tab. II, we summarize the resulting groups
of topological sectors g for all possible systems
41
. We no-
tice an interesting diagonal pattern relating subsequent
symmetry classes to neighboring spatial dimensions. Fur-
thermore, the pattern of the two unitary classes shows a
periodicity of two in the spatial dimension. If we had
shown the classication for higher spatial dimensions, we
would have observed a periodicity of eight in the spatial
dimension for the eight real classes. The following dis-
cussion is dedicated to provide a deeper understanding
of these fundamental patterns as pioneered in Ref. 17.
The mentioned periodicities have rst been pointed out
in Refs. 39 and 41.
C. Topological classication of unitary vector
bundles
In order to prepare the reader for the application of K-
Theory and to motivate its usefulness, we rst formulate
the classication problem in the language of ber bun-
dles. The mathematical structure of a non-interacting
insulator of spatial dimension d is that of a vector bundle
E

/ (see Ref. 42 for a pedagogical review on vector
bundles in physics). Roughly speaking, a vector bundle
consists of a copy of a vector space V which is called the
typical ber over each point of its so called base manifold
/. The projection of the bundle projects the entire
ber V
p
V over each point p / to p. Conversely,
V
p
can be viewed as the inverse image of p under , i.e. ,
V
p
=
1
[p]. The base manifold / of E is the d dimen-
sional k-space of the system, and the ber over a point
k given by the (projective) space of occupied states P(k).
The gauge group of the bundle is U(n), where n is the di-
mension of P. Locally, say in a neighborhood | / of
a given point p /, any vector bundle looks like the
trivial product | V . Globally however, this product
can be twisted which gives rise to topologically distinct
bundles. If no further symmetry conditions are imposed
(see class A in Tab. I), the question of how many topo-
logically distinct insulators in a given dimension exist is
tantamount to asking how many homotopically dierent
U(n) vector bundles can be constructed over /. This
question can be formally answered for arbitrary smooth
manifolds / as we will outline now. The general idea
is the following. There is a universal bundle

A into
which every bundle E can be embedded through a bundle
map
23
f : E such that
f

= E, (35)
where f : /A is the map between the base manifolds
associated with the bundle map

f. That is to say every
bundle can be represented as a pullback bundle f

23
of
the universal bundle by virtue of a suitable bundle map

f. The key point is now that homotopically dierent bun-

dles E are distinguished by homotopically distinct maps
f. Thus, the set of dierent TSM is the set [/, A] of
homotopy classes of maps from the k-space /to the base
manifold A of the universal bundle. A is also called the
classifying space of U(n) and is given by the Grassmanian
G
N,n
(C) = U(N)/(U(n)U(Nn)) for suciently large
N, i.e., N > ,
d
2
+ n|. To be generic in the dimension of
the system d, we take the inductive limit A = G
n
((

) =
lim
N
G
N,n
(C) = lim
N
U(N)/(U(n) U(N n)).
We thus found for the set Vect
n
(/, C) of inequivalent
U(n) bundles over / the expression
Vect
n
(/, C) = [/, G
n
((

)] ,
which is known for some rather simple base manifolds.
In particular for spheres S
d
, there is a trick to calcu-
late Vect
n
(S
d
, C): S
d
can always be decomposed into
two hemispheres which are individually trivial. The ho-
motopy of a bundle over S
d
is thus determined by the
clutching function f
c
dened in the overlap S
d1
of the
two hemispheres, i.e., along the equator of S
d
. Phys-
ically, f
c
translates a local gauge choice on the upper
hemisphere into a local gauge choice on the lower hemi-
sphere and is thus a function f
c
: S
d1
U(n). The
group of homotopy classes of such functions is by deni-
tion
d1
(U(n)). Interestingly, for n >
d1
2
, these groups
are given by
Vect
n
(S
d
, C) =
d1
(U(n)) =
_
Z, d 1 odd
0 , d 1 even
(36)
This periodicity of two in (d 1) is known as the
complex Bott periodicity. The physical meaning of Eq.
(36) is the following: In the unitary universality class A,
10
Class constraint d = 1 d = 2 d = 3 d = 4
A none 0 Z 0 Z
AIII none on q Z 0 Z 0
AI Q
T
(k) = Q(k) 0 0 0 Z
BDI q

(k) = q(k) Z 0 0 0
D xQ
T
(k)x = Q(k), m = n Z2 Z 0 0
DIII q(k)
T
= q(k), m = n even Z2 Z2 Z 0
AII iyQ
T
(k)(iy) = Q(k), m, n even 0 Z2 Z2 Z
CII iyq

(k)(iy) = q(k), m = n even Z 0 Z2 Z2

C yQ
T
(k)y = Q(k), m = n 0 Z 0 Z2
CI q(k)
T
= q(k), m = n 0 0 Z 0
TABLE II. Table of all groups g of topological equivalence classes. The rst column denotes the CAZ symmetry class, divided
into two unitary classes without anti-unitary symmetry (top) and eight real classes with at least one anti-unitary symmetry
(bottom). The second column shows the symmetry constraints on the at band maps, where we have chosen the representation
T = K for T
2
= 1, T = iyK for T
2
= 1, as well as C = xK for C
2
= 1, C = yK for C
2
= 1. Here, y denotes the
Pauli matrix in spin space, x, y denote Pauli matrices in the particle hole pseudo spin space of Bogoliubov de Gennes Hilbert
spaces. In the last four columns, g is listed for d = 1, . . . , 4.
there is an integer topological invariant in even spatial
dimension (e.g., QAH in d = 2) and no TSM in odd
spatial dimension.
This classication has two shortcomings. First, it can-
not be readily generalized to other CAZ classes at this
simple level. Second, only systems with the same num-
ber of occupied bands n can be compared. However,
adding some topologically trivial bands to the system
should yield a system in the same equivalence class, if
those bands can be considered as inert, i.e., if they do not
change the low energy physics close to the Fermi surface.
Both shortcomings can be overcome in the framework of
K-Theory
16,43
.
D. K-Theory approach to a complete classication
K-Theory
16,43
is concerned with vector bundles which
have a suciently large ber dimension. This means,
that topological defects which can be unwound by just
increasing the ber dimension are not visible in the
resulting classication scheme. This is physically reason-
able, as trivial occupied bands from inner localized shells
for example increase the number of bands as compared
to the eective low energy models under investigation.
Models of dierent number of such trivial bands should
be comparable in a robust classication scheme. In Ref.
44, K-Theory was used to discuss analogies between the
emergence of D-branes in superstring theory and the
stability of Fermi-surfaces in non-relativistic systems.
The use of K-Theory for the classication of TSM has
been pioneered in Refs. 1 and 2 for the QSH state and
more systematically been discussed for general TSM in
Ref. 17. This analysis in turn can be understood as a
special case of a twisted equivariant K-Theory as has
been reported very recently
45
.
Crash-course in K-Theory
The direct sum of two vector bundles EF is the direct
sum of their bers over each point. This addition has only
a semi-group structure, since E G = F G E F.
A minimal counterexample is given by E = TS
2
, F =
S
2
R
2
. F is clearly trivial, whereas TS
2
, the tangent
bundle of S
2
, i.e., the disjoint union of all tangent planes
of S
2
, is well known to be non-trivial. However, adding
NS
2
, the bundle of normal vectors to S
2
, to both bundles
E, F we obtain the same trivial bundle S
2
R
3
. This
motivates the concept of stable equivalence
E
s
F E Z
m
F Z
n
, (37)
where Z
n
= /K
n
, K = R, C is the trivial bundle over
the xed base manifold /, which plays the role of an ad-
ditive zero as far as stable equivalence is concerned. We
denote the set of K-vector bundles over /by 1
K
(/) in
the following. Note that stably equivalent bundles can
have dierent ber dimension, as m ,= n in general in
Eq. (37). The benet of this construction is:
E G = F G E
s
F (38)
This is because for vector bundles on a smooth manifold
every bundle can be augmented to a trivial bundle, i.e.,

H,l
GH = Z
l
. (39)
Eq. (38) naturally leads to the notion of a subtraction
on 1
K
(/) by virtue of the Grothendieck construction:
Consider the pairs (E
1
, E
2
) 1
K
(/) 1
K
(/) and de-
ne the equivalence relation
(E
1
, E
2
) (F
1
, F
2
) H F
1
E
2
H E
1
F
2
H.
(40)
11
Looking at Eq. (40), we can intuitively think of the
equivalence class (E
1
, E
2
)

as the formal dierence E

1

E
2
. We now dene the K-group as the quotient
K(/) = (1
K
(/) 1
K
(/)) / , (41)
which identies all formal dierences that are equiv-
alent in the sense of Eq. (40). Due to Eq. (39),
every group element in K(/) can be represented in
the form (E, Z
n
). However, (E, Z
n
) (E, Z
m
) for
n ,= m. We dene the virtual dimension of (E, F) as
d
v
= rk(E) rk(F), where rk denotes the rank, i.e.,
the ber dimension of a vector bundle. By restricting
K(/) to elements with d
v
= 0, we obtain the restricted
K-group

K(/) = g K(/)[d
v
(g) = 0.

K(/) is
isomorphic to the set of stable equivalence classes
of 1
K
(/). Up to now, the construction has been
independent of the eld over which the vector spaces
are dened. In the following, we will distinguish the
real and complex K-groups K
R
(/), K
C
(/). Physically,
K
C
will be employed to characterize systems without
anti-unitary symmetries, whereas K
R
is relevant for
systems in which at least one anti-unitary symmetry
imposes a reality constraint on the k-space.
A crucial notion in K-Theory which is also our main
physical motivation to study it is that of the stable range.
The idea is that at suciently large ber dimension n
no new bundles can be discovered by looking at even
larger ber dimension. Suciently large in terms of the
dimension d of / means n n
C
= d/2 + 1 for the
complex case and n n
R
= d + 1 for the real case,
respectively. More formally, every bundle E with n >
n
K
can be expressed as a sum
E F Z
nn
K
(42)
of a bundle F with ber dimension n
K
and a trivial bun-
dle for K = R, C. Since clearly E
s
F (see Eq. (38)),
this means that all stable equivalence classes have rep-
resentatives in ber dimension n n
K
. Furthermore, a
situation like our counterexample above where we aug-
mented two non-isomorphic bundles by the same trivial
bundle NS
2
to obtain the same trivial bundle cannot oc-
cur in the stable range. That is to say F as appearing
in Eq. (42) is uniquely dened up to isomorphisms. The
stable range hence justies the approach of K-theory of
ignoring ber dimension when dening the stable equiva-
lence. The key result in the stable range which connects
K-Theory to our goal of classifying all inequivalent vector
bundles with suciently large but arbitrary ber dimen-
sion on equal footing reads
43

K
K
(/) = Vect
n
(/, K) = [/, G
n
(K

)]
nn
K
.
(43)
The complex Bott periodicity Eq. (36) with period p
C
=
2 has a real analogue concerning the homotopy groups of
O(n) with period p
R
= 8. This immediately implies in
the language of K-Theory

K
K
(S
d+p
K
) =

K
K
(S
d
), K = R, C. (44)
We dene

K
d
K
(/) =

K
K
(S
d
/), (45)
where S is the reduced suspension (see Refs. 26 and 43
for a detailed discussion) which for a sphere S
k
indeed
satises SS
k
= S
k+1
. The stronger version of the Bott
periodicity in K-Theory now reads
43

K
dp
K
K
(/) =

K
d
K
(/), K = R, C, (46)
which only for / = S
l
trivially follows from Eq. (44).
Using this periodicity, the denition of K
d
K
in Eq. (45)
can be formally extended to d Z.
Class Classifying Space
A C0 = U(n +m)/ (U(n) U(m))
AIII C1 = U(n)
AI R0 = O(n +m)/ (O(n) O(m))
BDI R1 = O(n)
D R2 = U(2n)/U(n)
DIII R3 = U(2n)/Sp(2n)
AII R4 = Sp(n +m)/ (Sp(n) Sp(m))
CII R5 = Sp(n)
C R6 = Sp(2n)/U(n)
CI R7 = U(n)/O(n)
TABLE III. Table of all classifying spaces Cq, Rq of complex
and real K-Theory, respectively. The rst column denotes
the CAZ symmetry class. From top to bottom, the next com-
plex/real classifying space is the loop space of its predecessor,
i.e., Cq+1 = Cq(mod 2), Rq+1 = Rq(mod 8)
The Bott clock
From the very basic construction of homotopy groups
the following identities for the homotopy of a topological
space X are evident:

d
(X) =
_
S
d
, X

=
_
SS
d1
, X

=
_
S
d1
, X

,
where X denotes the loop space
43
of X, i.e., the space
of maps from S
1
to X. Iterating this identity gives

d
(X) =
0
(
n
X) using the complex Bott periodicity
(36), we immediately see that counting the connected
components
0
(U(n)) ,
1
(U(n)) =
0
(U(n)) of the
unitary group and its rst loop space, we can classify
all U(n) vector bundles over S
d
in the stable range, i.e.,
with n >
d
2
. The real analogue of the Bott periodic-
ity with period p
R
= 8 leads to analogous statements
for O(n) bundles over S
d
which depend only on the con-
nected components of O(n) and its rst seven loop spaces
12

i
O(n), i = 1, . . . , 7 (see Tab. III). This denes a Bott
clock with two ticks for the complex case and eight ticks
for the real case, respectively. Interestingly, these ten
spaces, for the complex and real case together, are pre-
cisely the ten Cartan symmetric spaces in which the time
evolution operators associated with Hamiltonians in the
ten CAZ classes lie. After this observation, only two
points are missing until a complete classication of all
TSM of continuum models can be achieved. The rst
point is a subtlety related to the interdependence of the
two wave vectors k and k as shown in Eq. (33), which
makes the real Bott clock tick counter clockwise. The sec-
ond point is the inclusion of symmetry constraints into
the scheme which leads to the clockwise ticking Cliord
clock (see Eq. (50) below). The combination of both im-
plies that the topological invariant of a continuum model
of dimension d in the CAZ class q only depends on the
dierence q d (mod 8) for the eight real classes and on
q d (mod 2) for the two complex classes A and AIII,
respectively.
Reality and k-space topology
For systems which obey anti-unitary symmetries the
real structure of the Hamiltonian H(x) is most con-
veniently accounted for in its Majorana representation.
H(x) can in this representation be expressed in terms of
a real antisymmetric 2n 2n-matrix B,

(x)H(x)(x) =
i
4
B
ij
c
x,i
c
x,j
, (47)
where c
x,i
, i = 1, . . . , 2n are the Majorana operators rep-
resenting the n fermion species at x. On Fourier trans-
form, 1 =
_

H can be written as
17
1 =
i
4
_
d
d
kA
ij
(k)c
k,i
c
k,j
, (48)
where A is skew hermitian and satises
A

(k) = A(k). (49)

Eq. (49) naturally leads to a real vector bundle structure
as dened in Ref. 46 for the bundle of eligible A-matrices
over the k-space (R
d
, ), where the involution (see Ref.
46) is given by k k. On one-point compactication,
this real k-space becomes a sphere

S
d
= (S
d
, ) with the
same involution
17
. Whereas the ordinary sphere S
d
can
be viewed as a reduced suspension S of S
d1
over the
real axis,

S
d
can be understood as the reduced suspen-
sion

S of

S
d1
over the imaginary axis. This picture is
algebraically motivated by comparing the involution to
the ordinary complex conjugation which, restricted to the
imaginary axis of the complex plane is of the same form.
Interestingly, in the language of denition (45),

S plays
the role of an inverse to S
17,46
, i.e.,

K
R
(/) =

K
1
R
(

S/) =

K
R
(S

S/).
This means that the Bott clock over

S
d
is reversed as
compared to its analogue over S
d
.
Real K-theory and the Cliord clock
The main reason for the real construction of Eq. (48)
is that the anti-unitary symmetry constraints yielding
the eight real CAZ classes (all except A and AIII) can
be distinguished in terms of anti-commutation relations
of the A-matrix with real Cliord generators
16,17
. At
a purely algebraic level, these constraints can be trans-
formed so as to be expressed only in terms of positive
Cliord generators
17
, i.e., generators that square to plus
identity. We call the restricted K-group of a vector bun-
dle of A-matrices over /that anti-commute with q pos-
itive Cliord generators

K
q
R
(/). Interestingly
16,17
,

K
q
R
(/)

K
q
R
(/). (50)
Eq. (50) denes a Cliord clock that runs in the oppo-
site direction as the

S
d
Bott clock. This algebraic phe-
nomenon explains the full periodic structure of the table
of TSM of continuum models (see Tab. II). The classify-
ing spaces of A-matrices for systems that anti-commute
with q Cliord generators are shown in Tab. III.
Periodic systems
The classication of periodic systems is much more
complicated from a mathematical point of view. Their
base space is the real Brillouin zone

T
d
= (T
d
, ), where
the involution giving rise to the real structure is again
given by k k. For

T
d
the reduced suspension does
not provide a trivial relation between the K-Theory of
dierent spatial dimension like

S

S
d
=

S
d+1
for the base
space of continuum models. The general calculation of
all relevant K-groups over

T
d
has been reported in Ref.
17. Interestingly, the resulting groups always contain the
respective classication of continuum models in the same
symmetry class as an additive component. Additionally,
the topological invariants of weak TSM, i.e., TSM which
are only present in translation-invariant systems, can be
inferred. The Cliord clock dened in Eq. (50) is in-
dependent of the base space and hence still applicable.
Here, we only review the general result calculated in Ref.
17
K
q
R
(

T
d
) K
q
R
(

S
d
)
_
d1

s=0
_
d
s
_
K
q
R
(

S
s
)
_
. (51)
The second term on the right hand side of Eq. (51) en-
tails the notion of so called weak topological insulators
which are obviously due to TSM in lower dimensions.
To name the most prominent example, the Z
2
invariant
characterizing the QSH insulator in d = 2 in the presence
of TRS, CAZ class AII, yields a 3Z
2
topological invari-
ant characterizing the weak topological insulators with
the same symmetry in d = 3.
13
Lattice systems with disorder
In a continuum model, disorder that is not too short
ranged so as to keep the k-space compactication for
large k valid, can be included into the model system with-
out changing the classication scheme. However, per-
turbing a translation-invariant lattice system with disor-
der also gives its k-space (now dened in terms of a dis-
crete Fourier transform) a discrete lattice structure which
is not directly amenable to investigation in the framework
of K-Theory which we only dened over smooth base-
manifolds. Ref. 17 shows that a Hamiltonian featuring
localized states in the energy gap can be transformed
into a gapped Hamiltonian upon renormalization of pa-
rameters. The physical consequence of this statement is
that only a mobility gap is needed for the classication
of a TSM and no energy gap in the density of states.
Furthermore, Ref. 17 argues without explicit proof that
the classication problem of gapped lattice systems with-
out translation invariance is equivalent to the classi-
cation problem of continuum models. This statement
agrees with the physical intuition that the breaking of
translation-invariance must remove the additional struc-
ture of weak TSM as described for periodic systems by
Eq. (51).
VI. CALCULATION OF TOPOLOGICAL
INVARIANTS OF INDIVIDUAL SYSTEMS
In Section V, we have shown how many dierent TSM
can be expected in a given spatial dimension and CAZ
class. Now, we outline how insulating systems within
the same CAZ class and dimension can be assigned a
topological equivalence class in terms of their adiabatic
connection dened in Eq. (9) and their adiabatic curva-
ture dened in Eq. (20), respectively. A complete case
by case study in terms of Dirac Hamiltonian representa-
tives of all universality classes of this problem has been
reported in Ref. 6. We outline the general patterns relat-
ing the classication of neighboring (see Tab. II) univer-
sality classes following the analysis in Refs. 6, 39, and 40.
Interestingly, all topological invariants can be calculated
using only complex invariants, namely Chern numbers
and chiral unitary winding numbers. The anti-unitary
symmetries are accounted for by the construction of a
dimensional hierarchy in Section VI B starting from a so
called parent state in each symmetry class for which the
complex classication concurs with the real classication.
In Section VI C, we show how the topological invariants
can be dened for disordered systems with the help of
twisted boundary conditions. Furthermore, we discuss a
generalization of the non-interacting topological invari-
ants to interacting systems in Section VI D.
A. Systems without anti-unitary symmetries
Chern numbers of unitary vector bundles
Eq. (35) shows that every U(n) bundle E / can
be represented as a pullback from the universal bun-
dle G
n
(C

) by some bundle map

f. Chern
classes are de Rham cohomology classes, i.e., topologi-
cal invariants
47
that are dened as the pullback of cer-
tain cohomology classes of the classifying space G
n
(C

).
The cohomology ring H

(G
n
(C

)) consists only of even

classes and is generated by the single generator c
j

H
2j
(G
n
(C

)) , j = 1, . . . , n for every even cohomology

group
48
. The Chern classes c
i
of E are dened as the
pullback c
i
= f

c
i
from the classifying space by the map
f : / G
n
(C

) associated with the bundle map

f.
Due to the Chern-Weyl theorem
23
, Chern classes can be
expressed in terms of the curvature, i.e., in our case, the
adiabatic curvature T dened in Eq. (20) of E. Explic-
itly, the total Chern class c can be expressed as
23
c = det
_
1 +
iT
2
_
= 1 +c
1
(T) +c
2
(T) . . . . (52)
The determinant is evaluated in gauge space and prod-
ucts of T are understood to be wedge products. c
j
is
the monomial of order j in T. Obviously, c
j
is a 2j-form
and can only be non-vanishing for 2j d, where d is
the dimension of the base manifold /, i.e., the spatial
dimension of the physical system. Another characteris-
tic class which generates all Chern classes is the Chern
character
23
ch = Tr
_
e
iF
2
_
= 1 + ch
1
(T) + ch
2
(T) +. . . . (53)
Due to their importance for later calculations, we ex-
plicitly spell out the rst two Chern characters ch
1
=
Tr
_
iF
2

, ch
2
=
1
8
2
Tr [T T]. Importantly, for even
d = 2p, the integral
(
p
=
_
M
ch
p
yields an integer, the so called p-th Chern number
24
.
These Chern numbers characterize systems in the uni-
tary symmetry class A which can only be non-trivial in
even spatial dimension (see Tab. II).
Winding numbers of chiral unitary vector bundles
In Section V, we have shown that the classifying space
for a chiral unitary (AIII) system is given by U(n) and
that the topological sectors are dened by homotopically
distinct maps k q(k) U(n). Now, we discuss how
to assign an equivalence class to a given map q by cal-
culating its winding number
4951
following Ref. 6. From
Tab. II it is clear that only in odd spatial dimension
14
d = 2j 1 there can be a non-trivial winding number.
We dene
w
q
2j1
=
((j 1)!)
(2j 1)!(2i)
j
Tr
_
(q
1
dq)
2j1

, (54)
which has been dubbed winding number density
6
. Inte-
grating this density over the odd-dimensional base man-
ifold / representing the k-space of the physical system,
we get the integral winding number
2j1

2j1
=
_
M
w
q
2j1
, (55)
which is well known to measure the homotopy of the map
k q(k).
Relation between chiral winding number and Chern Simons
form
So far, the relation between the adiabatic connection of
a chiral system and its topological invariant has not been
made explicit. Since characteristic classes like Chern
characters are closed 2j-forms, they can locally be ex-
pressed as exterior derivatives of (2j 1)-forms. These
odd forms are called the Chern Simons forms associated
with the even characteristic class
23,52
. For the j-th Chern
character ch
j
, which is a 2j form, the associated Chern
Simons form Q
2j1
reads
23
Q
2j1
(/, T
t
) =
1
(j 1)!
_
i
2
_
j
_
1
0
dt STr
_
/, T
j1
t
_
,
(56)
where T
t
= tT + (t
2
t)/ / is the curvature of the
interpolation t/ between the zero connection and / and
STr denotes the symmetrized trace. Explicitly, we have
Q
1
=
i
2
Tr [/] , Q
3
=
1
8
2
Tr
_
/d/+
2
3
/
3

.
It is straightforward to show
6
, that in a suitable gauge,
the Berry connection of a chiral bundle yields /
B
=
1
2
qdq

, where q U(n) is again the chiral map char-

acterizing the system. This is not a pure gauge due to
the factor
1
2
which entails that the associated curvature
T
B
does not vanish. Plugging /
B
and T
B
into Eq. (56)
immediately yields
6
Q
2j1
(/
B
, T
B
t
) =
1
2
w
q
2j1
. (57)
Eq. (57) directly relates the winding number density to
the Chern Simons form. We dene the Chern Simons
invariant of an odd dimensional system as
CS
2j1
=
_
M
Q
2j1
(mod 1),
where (mod 1) accounts for the fact that
_
M
Q
2j1
has
an integer gauge dependence due to
2j1
(U(n)) = Z for
n > j. Looking back at Eq. (55), we immediately get

2j1
(mod 2) = 2CS
2j1
(mod 2).
FIG. 1. Illustration of the WZW dimensional extension. The
circle at v = 0 represents the physical system. The poles
at v = represent the trivial reference system without k-
dependence. The two interpolations are conjugated by an
anti-unitary symmetry, here exemplary denoted by TRS.
We note that the (mod 2) can be dropped if we x the
gauge as described above to /
B
=
1
2
qdq
1
. This estab-
lishes the desired relation between the winding number
of a chiral unitary system and its adiabatic curvature.
B. Dimensional hierarchy and real symmetry
classes
Until now, we have only discussed how to calculate
topological invariants of systems in the complex sym-
metry classes A and AIII. Interestingly, for some real
universality classes, the classication in the presence of
anti-unitary symmetries concurs with the unitary classi-
cation (see Tab. IV). The rst known example of this
type is in the symplectic class AII in d = 4 which is char-
acterized by the second Chern number of the correspond-
ing complex bundle
53,54
. Another example of this kind is
the p+ip superconductor in d = 2 and symmetry class D
which is characterized by its rst Chern number, i.e., in
the same way as the QAH eect in class A. In odd dimen-
sions similar examples exist for real chiral classes, e.g., for
DIII in d = 3, where the winding number is calculated
using Eq. (54) in the same way as for the chiral unitary
class AIII in the same dimension. All the topological in-
variants just mentioned are integer invariants. In some
universality classes, these integers can only assume even
values (see Tab. IV). For physically relevant dimensions,
i.e., d = 1, 2, 3, these exceptions are CII in d = 1, C in
d = 2, and CI in d = 3. All other states where the com-
plex and the real classication concur, can be viewed as
parent states of a dimensional hierarchy within the same
symmetry class from which all Z
2
invariants appearing
in Tab. IV can be obtained by dimensional reduction.
This approach was pioneered in the seminal work by Qi,
Hughes, and Zhang
39
.
The general idea is more intuitive if we consider
the parent state as a Wess-Zumino-Witten (WZW)
dimensional extension
55,56
of the lower dimensional
descendants instead of thinking of a dimensional re-
15
Class d = 1 d = 2 d = 3 d = 4 d = 5 d = 6 d = 7 d = 8
A 0 Z 0 Z 0 Z 0 Z
AIII Z 0 Z 0 Z 0 Z 0
AI 0 0 0 2Z 0 Z2 Z2 Z
BDI Z 0 0 0 2Z 0 Z2 Z2
D Z2 Z 0 0 0 2Z 0 Z2
DIII Z2 Z2 Z 0 0 0 2Z 0
AII 0 Z2 Z2 Z 0 0 0 2Z
CII 2Z 0 Z2 Z2 Z 0 0 0
C 0 2Z 0 Z2 Z2 Z 0 0
CI 0 0 2Z 0 Z2 Z2 Z 0
TABLE IV. Table of all groups of topological equivalence classes. The rst column denotes the symmetry class, divided into
two complex classes without any anti-unitary symmetry (top) and eight real classes with at least one anti-unitary symmetry
(bottom). Chiral classes are denoted by bold letters. The parent states of dimensional hierarchies are boxed. For all non-chiral
boxed states, the classication concurs with that of class A in the same dimension. For all chiral boxed states, the classication
concurs with that of class AIII in the same dimension. 2Z indicates that the topological integer can only assume even values
in some cases. Such states are never parent states.
duction from the d-dimensional parent state to its
descendants. This works as follows: We x a localized
(d 1)-dimensional insulator without any hopping
that satises the required anti-unitary symmetries
as a trivial reference state. This reference state is
described by the k-independent Bloch Hamiltonian h
0
.
The (d 1)-dimensional physical system of interest is
characterized by the Bloch Hamiltonian h(k). Then,
we interpolate by varying the parameter v between
the (d 1)-dimensional physical system (v = 0) and
the trivial state (v = ) without closing the insulating
gap. However, the intermediate (d 1)-dimensional
system at xed v ,= 0, might well break the required
anti-unitary symmetries. The crucial step is now to
do the interpolation for v [0, ] and v [, 0] in a
symmetry conjugated way (see Fig. 1). That is to say,
we require our (d 1)-dimensional system of interest
and the resulting d-dimensional extended system to be
in the same CAZ class. This d-dimensional system is
characterized by the Bloch Hamiltonian h(k, v). The
v [, 0] and the v [0, ] half of the extended
k-space then are not independent of each other but give
equal contributions to the integer topological invariant
of the d-dimensional extended system
6,39
. One might
now ask to which extend the resulting integer invariant
of the extended system depends on our choice of the
interpolation h(k, v) between h(k) = h(k, v = 0) and
h
0
= h(k, v = ). To answer this question, one
considers two interpolations h(k, v),

h(k, v). It is then

elementary to show
39
that the dierence between the
integer invariants of these two d-dimensional systems
is an even integer. This implies that a Z
2
information,
namely the parity of the integer invariant associated
with the extended system, is well dened only in terms
of the physical system with spatial dimension (d 1).
A similar procedure can be repeated a second time to
obtain a Z
2
invariant for a (d 2) dimensional second
descendant
39
. From the procedure just sketched, it is
obvious why the exceptional phases which are character-
ized by an even integer are not parent states of such a
dimensional hierarchy. The generic constructions for all
possible classes can be found in Refs. 6 and 39. With
that, we are provided with a general and fairly explicit
recipe for the practical calculation of the topological in-
variants for all possible CAZ classes in all spatial dimen-
sions.
C. Bulk invariants of disordered systems and
twisted boundary conditions
Our practical calculation of topological invariants
so far has been focused on periodic systems with a
BZ, i.e., / = T
d
and continuum models where the
k-space can be compactied to a sphere, i.e., / = S
d
.
As already pointed out in Section V, the situation
is more complicated for disordered lattice models.
Seminal progress along these lines was reported for the
quantum Hall state by Niu et al. in 1985
57
. These
authors use twisted boundary conditions (TBC) to
dene the quantized Hall conductivity
xy
for a 2D
system as a topological invariant only requiring a
bulk mobility gap. We briey review their analysis
and propose the framework of TBC as a general recipe
to calculate topological invariants for disordered systems.
The Hall conductivity
xy
resulting from a linear re-
sponse calculation at zero temperature yields

xy
=
2
A
Im

n=0
0[1
x
[nn[1
y
[0
(E
n
E
0
)
2
,
where [0 is the many body ground state, A is the area
of the system, and 1
i
=
H
k
i
. In the presence of a mag-
16
netic eld, translation-invariance is dened in terms of
the magnetic translation operator T
B
36
which concurs
with the ordinary translation operator T(a) = e
ia

k
in
the absence of a magnetic eld. TBC now simply mean
that a (magnetic) translation by the system length L
j
in
j-direction gives an additional phase factor e
ij
.
j
is
called the twisting angle in j-direction. Gauging away
this additional phase to obtain a wave function with pe-
riodic boundaries amounts to a gauge transformation of
the Hamiltonian which shifts the momentum operator
like
i
j
i
j
+
j
(58)
Using 1
i
= 1
i
= [
i
, 1] the Hall conductivity can
be expressed as the sensitivity of the ground state wave
function to TBC.

xy
=
2
A
Im
x

0
[
y

0
,
where [
0
is the many body ground state after the men-
tioned gauge transformation which depends on the twist-
ing angles. Dening = L
x

x
, = L
y

y
,

xy
= 2Im

0
[

0
= iT

. (59)
The main merit of Ref. 57 is to show that this expression
actually does not depend on the value of (, ) as long
as the single particle Greens function of the system is
exponentially decaying in real space. This condition is
met if the Fermi energy lies in a mobility gap. Hence, a
trivial integration can be introduced as follows:

xy
=
i
4
2
_
2
0
d
_
2
0
dT

= G
0
_
T
2
iT
2
= G
0
(
1
,
(60)
where G
0
=
e
2
h
=
1
2
and the integer (
1
is by denition
the rst Chern number of the ground state line bundle
over the torus of twisting angles. This construction
makes the topological quantization of the Hall conduc-
tivity manifest.
Eq. (58) shows the close relation between momentum
and twisting angles. One is thus tempted to just replace
the BZ of each periodic system by the torus of twisting
angles for the corresponding disordered system which is
topologically equivalent to a ctitious periodic system
with the physical system as single lattice site
58,59
. We
will proceed along these lines below but would like to
comment briey on the special role played by the quan-
tum Hall phase rst. Eq. (59) represents the physical
observable
xy
in terms of the twisting angles. Niu et al.
argued rigorously
57
that
xy
of a bulk insulating system
can actually not depend on the value of these twisting
angles which allows them to express
xy
as a manifestly
quantized topological invariant in Eq. (60). For a
generic TSM, the topological invariant of the clean sys-
tem does in general not represent a physical observable.
Furthermore, the integration over the twisting angles
will not be trivial, i.e., the function to be integrated will
actually depend on the twisting angles. Employing the
picture of a periodic system with the physical system
as a single site is problematic inasmuch as the bulk
boundary correspondence at the boundary of a single
site is hard to dene mathematically rigorously. In the
quantum Hall regime for example, it is well known that
in disordered systems a complicated landscape of local-
ized states and current carrying regions produces the
unchanged topologically quantized Hall conductivity
9
.
However, the edge states of the disordered quantum Hall
state are in general not strictly localized at the boundary.
Replacing the BZ of a translation-invariant system by
the torus of twisting angles in the disordered case yields
a well dened topological invariant which adiabatically
connects to the topological invariant of the clean system
where the relation

k
j =
j
(61)
follows from Eq. (58). This is because from a purely
mathematical perspective it cannot matter which torus
we consider as the base space of our system. In this
sense, the framework of TBC is as good as it gets concern-
ing the denition of topological invariants for disordered
systems. When calculating the topological invariant of a
symmetry protected TSM through dimensional extension
(see Section VI B), a hybrid approach between TBC in
the physical dimensions and periodic boundaries in the
extra dimension can be employed to dene a topological
invariant for the disordered system which can be calcu-
lated more eciently
60
. The fact that in some symme-
try protected topological phases the topological invari-
ants are not directly representing physical observables
is a not a problem of the approach of TBC but is a re-
markable dierence between these TSM and the quantum
Hall state at a more fundamental level. Recently, an S-
matrix approach to calculating topological invariants of
non-interacting disordered TSM has been reported
61
D. Taking into account interactions
Up to now, our discussion has only been concerned
with non-interacting systems. As a matter of fact,
the entire classication scheme discussed in Sec. V
massively relies on the prerequisite that the Hamiltonian
is a quadratic form in the eld operators. The violation
of this classication scheme for systems with two particle
interactions has been explicitly demonstrated in Ref. 62.
As we are not able to give a general classication of
TSM for interacting systems, we search for an adiabatic
continuation of the non-interacting topological invariants
to interacting systems. This procedure does from its
outset impose certain adiabaticity constraints on the
17
interactions that can be taken into account. The topo-
logical invariants for non-interacting systems are dened
in terms of the projection P on the occupied single
particle states dening the ground state of the system.
The main assumption is thus that the gapped ground
state of the non-interacting system is adiabatically
connected to the gapped ground state of the interacting
system. A counter-example of this phenomenology are
fractional quantum Hall states
63,64
, where a gap due to
non-adiabatic interactions emerges in a system which is
gapless without interactions. However, it is clear that
the phase space for low energy interactions will be much
larger in a gapless than in a gapped non-interacting
system. We thus generically expect the classication
scheme at hand to be robust against moderate inter-
actions. However, beyond mean eld interactions, the
Hamiltonian cannot be expressed as an eective single
particle operator. Hence, we need to nd a formulation
of the topological invariants that adiabatically connects
to the non-interacting language and is well dened
for general gapped interacting systems. The key to
achieving this goal is to look at the single particle
Greens function G instead of the Hamiltonian. This
approach has been pioneered in the eld of TSM by Qi,
Hughes, and Zhang
39
who formulated a topological eld
theory for TSM in the CAZ class AII.
Chern numbers and Greens function winding numbers
The role model for this construction is again the Hall
conductivity
xy
of a gapped 2D system. In Ref. 49,

xy
has been expressed in terms of G by perturbative ex-
pansion of the eective action of a gauge eld A that is
coupled to the gapped fermionic system in the framework
of quantum electrodynamics in (2+1)D. The leading con-
tribution stemming from a vacuum polarization diagram
yields the Chern Simons action
S
CS
=

xy
2
_
d
2
xdt

=

xy
2
_
A dA.
The prefactor
xy
in units of the quantum of conductance
assumes the form

xy
=
1
24
2
_
d
2
kdTr
_
(GdG
1
)
3

, (62)
where d now denotes the exterior derivative in combined
frequency-momentum space and G is the time ordered
Greens function, or, equivalently as far as the calcula-
tion of topological invariants is concerned, the contin-
uous imaginary frequency Greens function as used in
zero temperature perturbation theory. Eq. (62) has
rst been identied as a topological invariant and been
proven in a non-relativistic condensed matter context in
Refs. 6567. An analogous expression has been derived
by Volovik using a semi-classical gradient expansion
50
.
The similarity between Eq. (54) and the integrand of
Eq. (62) is striking. Obviously, Eq. (62) represents

xy
as a winding number in 3D frequency-momentum
space. If this construction makes sense, we should by
integration of Eq. (62) over recover the representa-
tion of
xy
as the rst Chern number in the 2D BZ for
the special case of the non-interacting Greens function
G
0
(, k) = (i h(k))
1
. This straightforward calcula-
tion relies on the residue theorem and has been explicitly
presented in Ref. 39. Its result can be readily general-
ized to higher even spatial dimensions and higher Chern
numbers, respectively. In Ref. 51, a perturbative expan-
sion similar to Ref. 49 has been presented for fermions
coupled to a gauge eld in arbitrary even spatial dimen-
sion 2n. The resulting analogue of the Hall conductivity,
i.e., the prefactor of the Chern Simons form in (2n+1)D
(see Eq. (56)) can be expressed as
10,39,51
N
2r+1
[G] = A(r)
_
BZR
Tr
_
_
GdG
1
_
2r+1
_
, (63)
A(r) =
r!
(2r + 1)!(2i)
r+1
.
Performing again the integration over the frequency ana-
lytically for the noninteracting Greens function G
0
yields
N
2r+1
[G
0
] = (
r
. (64)
Eq. (64) makes manifest that N
2r+1
[G], which
can be formulated for an interacting system, repro-
duces the non-interacting classication for the free
Greens function G
0
of the non-interacting system.
The topological invariance of N
2r+1
[G] is clear by
analogy with Eq. (55): Whereas the winding num-
ber
2j1
measures the homotopy of the chiral map
k q(k) U(n) which, properly normalized, yields
an integer due to
2j1
(U(n)) = Z, n > j, Eq. (63)
measures the homotopy of G GL(n + m, C) in the
(2r + 1)D frequency-momentum space which is also
integer due to
2r+1
(GL(n +m, C)) = Z, n +m > r.
The dimensional hierarchy for symmetry protected de-
scendants of a parent state which is characterized by a
Chern number (see Section VI B) can be constructed in
a completely analogous way for the interacting general-
ization N
2r+1
of the Chern number (
r
56
. The resulting
topological invariants for the descendant states have been
coined topological order parameters in Ref. 56. Disor-
der can again be accounted for by imposing TBC and
replacing the k-space of the system by the torus of twist-
ing angles (see Section VI C). Our discussion is limited
to insulating systems. A detailed complementary anal-
ysis of the topological properties of dierent quantum
vacua can be found in Ref. 68.
Interacting chiral systems
The integer invariant of chiral unitary systems (class
AIII) in odd spatial dimension 2r 1 is not a Chern
18
number but a winding number (see Section VI A). For all
these systems and dimensional hierarchies with a chiral
parent state, i.e., all chiral TSM (see Tab. IV), a similar
interacting extension of the denition of the invariants in
terms of G(i, k) has been reported in Ref. 69:
I
2r
[G] = n(r)
_
BZR
Tr
_
Q(dQ)
2r
_
, (65)
where n(r) is a normalization constant, and Q(i, k) =
G
1
(i)U
CH
G(i, k), with the unitary representation
matrix U
CH
of the chiral symmetry operation. In the
non-interacting limit, I
2r
reduces to
2r1
as dened in
Eq. (55)
69
.
Fluctuation driven topological transitions
Thus far, we have shown that for a non-interacting
system the integration over reproduces the band struc-
ture classication scheme formulated in terms of the adi-
abatic curvature. However, the additional frequency de-
pendence of the single particle Greens function can cause
phenomena without non-interacting counterpart. To see
this, we represent the single particle Greens function of
an interacting system as
G(, k) = (i h(k) (, k))
1
,
where is the self-energy of the interacting system. In
Ref. 69, it has been pointed out, that the value of
N
2r+1
[G] cannot only change due to gap closings in the
energy spectrum as in the case of the Chern number (
r
.
This is due to the possibility of poles in the -dependence
of the self-energy which give rise to zeros of the Greens
function, whereas gap closings correspond to poles of G.
From the analytical form of N
1
it is immediately clear
that both poles and zeros of Gcan change the value of N
1
.
More generally, the G G
1
symmetry of N
2r+1
makes
clear that poles of G can be seen as zeros of G
1
and
vice versa on an equal footing. In Ref. 70, it has been
demonstrated that the -dependence of can change a
non-trivial winding number into a trivial one. The emer-
gence of a topologically nontrivial phase due to dynami-
cal uctuations which has no non-interacting counterpart
has been presented in Refs. 71 and 72.
Chern numbers of eective single particle Hamiltonians
Due to its additional -integration, the practical cal-
culation of N
2r+1
can be numerically very challenging
once the single particle Greens function of the interact-
ing system has been calculated. A major breakthrough
along these lines has been the observation that an eec-
tive single particle Hamiltonian dened in terms of the
inverse Greens function at zero frequency can be dened
to eectively reduce the topological classication to the
non-interacting case. This possibility has rst been men-
tioned in Refs. 73 and 74 and been generally proven in
Ref. 75. The authors of Ref. 75 show, using the spectral
representation of the Greens function, that one can al-
ways get rid of the -dependence of G. We only review
the physical results of this analysis. The both accessible
and explicit proof can be found in Ref. 75. The physical
conclusion is as elegant as simple: Instead of calculating
N
2r+1
we can just calculate the Chern number (
r
asso-
ciated with the ctitious Hamiltonian

h(k) = G
1
(0, k), (66)
the occupied states of which are just its eigenstates with
negative eigenvalues which have been dubbed R-zeros
75
since they are positive energy eigenstates of G
1
(0, k).
Obviously, G
1
0
(0, k) = h(k) for the non-interacting
Greens function. Hence,

h(k), which has recently been
coined topological Hamiltonian
76
, adiabatically connects
to h(k) in the non-interacting limit. Note that the pos-
sibility of eliminating the -dependence is not in con-
tradiction to the relevance of this -dependence for the
topology of the interacting system. All it shows is that
the relevant changes due to a dierent pole structure of
G as a function of can be inferred from its value at
= 0.
Topological Hamiltonian for chiral interacting systems
In principle, the construction of the topological Hamil-
tonian

h(k) can be readily generalized to chiral interact-
ing systems as has been mentioned in Ref. 77. To see
this, we note that the crucial argument for the construc-
tion of the topological Hamiltonian brought forward in
Ref. 75 is the following: The continuous interpolation
G(i, k, ) = (1 )G(i, k) +
_
i +G
1
(0, k)

1
does not contain any singularities or gap closings. Thus,
as long as the calculation of a topological invariant
in terms of G(i, k) is concerned, we can also use

G(i, k) = G(i, k, = 1) =
_
i +G
1
(0, k)

1
. Ob-
viously,

G(i, k) =
_
i

h(k)
_
1
is the Greens func-
tion of a ctitious non-interacting system which is gov-
erned by the topological Hamiltonian

h(k). The mere
existence of the topological invariant for chiral systems
in terms of G(i, k) as presented in Ref. 69 (see also Eq.
(65)) hence suces to argue that one can equally well
investigate the topology of

h(k) and its symmetry pro-
tected descendants (see Section VI B) instead of directly
evaluating Eq. (65). Since the single particle Greens
function inherits the fundamental symmetries from the
Hamiltonian
69
,

h(k) will also obey these symmetries. In
particular, for an interacting system with chiral symme-
try, the topological Hamiltonian can be brought into the
19
at band o-diagonal form

h(k)
_
0 q(k)
q

(k) 0
_
, (67)
where q(k) U(n) for the topologically equivalent
at-band system. This construction generically denes
a topological invariant for the chiral interacting system
which adiabatically concurs with the non-interacting
system: The winding number associated with the
ctitious Hamiltonian

h(k). A similar construction for a
chiral 1D system has been presented in Ref. 78.
Discussion of the topological Hamiltonian and practical
consequences
As already mentioned, the above construction cannot
be valid for arbitrary interacting systems. In particular
in 1D, the breakdown of the Z classication in the
presence of general interactions has been investigated
in Ref. 62. However, this problem does not pertain
to the concept of the topological Hamiltonian itself
but rather reects the limited validity of the adiabatic
continuation of the non-interacting invariants in terms
of the single particle Greens function, i.e., the limited
validity of Eq. (63) and Eq. (65). In the validity
regime of these equations, one can equivalently use the
topological Hamiltonian

h(k) = G
1
(0, k) to classify
an interacting system in any symmetry class. This is of
enormous practical usefulness for at least two reasons.
First, we get rid of the -integration appearing in Eq.
(63) and Eq. (65) which is cumbersome to evaluate.
Second, the method of dimensional extension, though
generally valid, is not always the most convenient way
to calculate the topological invariant of a symmetry
protected descendant state. Provided with the formal
equivalence between the non-interacting classication
problem of the topological Hamiltonian and the Greens
function topology, we can directly apply all simplied
schemes that have been introduced to directly calculate
non-interacting invariants of symmetry protected states
(see, e.g., Refs. 33, 58, 7981) to the topological Hamil-
tonian. The framework of dimensional extension and
Eq. (63) or Eq. (65) for the parent state are, with the
benet of hindsight, only needed to justify the validity
of the topological Hamiltonian.
Before closing the section, we would like to discuss the
role of the bulk boundary correspondence, in the pres-
ence of interactions. In general, interactions can spon-
taneously break the protecting symmetry of a symmetry
protected TSM locally at the boundary thus gapping out
the characteristic metallic surface states. Importantly,
this spontaneous symmetry breaking at the gapless sur-
face will typically happen at a lower critical interaction
strength than in the gapped bulk. This is because the
gapless surface modes oer more phase space for interac-
tions. A generally valid bulk boundary correspondence
is hence absent in the interacting case. Within the va-
lidity regime of Eq. (63) and Eq. (65) for chiral TSM,
respectively, an interacting analogue of the bulk bound-
ary correspondence has been reported in Refs. 69 and
82. The main dierence to the non- interacting case is
that boundary zero-modes, which represent poles of the
Greens function can be canceled by zeros of the Greens
function as far as the calculation of topological invari-
ants is concerned. Note that the Greens function of a
non-interacting system does not have zeros.
VII. LIMITATIONS OF THE FRAMEWORK OF
TSM
Finally, we would like to point out some limitations
of the concept of TSM. The two main aspects that one
could see critical in the eld of TSM are outlined in the
following.
First, whereas the topologically quantized Hall
conductivity in the integer quantum Hall state, the
historical role model of all TSM, is a physical observable,
the topological invariants of symmetry protected TSM
like the QSH state are not directly physically observable
without additional unitary symmetries. The quantum
Hall eect can be understood in terms of the spectral
ow associated with the threading of a ux tube
83
.
Along similar lines, the QSH eect can be understood
in terms of a spin charge separation associated with
the threading of a spin ux
84
. However, this spin ux,
as opposed to an ordinary magnetic ux tube, is not
immediately experimentally accessible and the general
observable consequences of the QSH state have been
shown to be much more subtle
85
. For several TSM,
the directly measurable consequences of the respective
topological invariants are still under debate or unknown.
Second, the entire construction and classication of
TSM is based on single particle Hamiltonians. In Sec-
tion VI D, we discussed how adiabatic interactions can be
taken into account and argued that interactions of mod-
erate strength are not likely to destroy the phenomenol-
ogy of TSM. In order to position the eld of TSM in a
broader context, we would like to point out that there are
also phenomena of topological origin which emerge only
due to the presence of interactions. The historically rst
phenomenon is the
1

FQH eect
63,64
which cannot be
adiabatically connected to an insulating non-interacting
state. The non-interacting state is in this case a partially
lled Landau level which provides an enormous density
of states at the Fermi energy. In a system with periodic
boundaries, the
1

FQH has a characteristic -fold ground

state degeneracy. Interestingly, this simplest FQH state
can still be analyzed in the framework of TBC
57
. For
more general FQH systems the concept of topological or-
20
der has been introduced by Wen
86
. A crucial notion in
this framework is the quantum dimension of the topolog-
ically ordered system which can be viewed as the ground
state degeneracy of the system on a torus, i.e., with pe-
riodic boundary conditions. Most TSM discussed in this
article have quantum dimension one, like a trivial insu-
lator. From the vantage point of topological order, these
states are thus trivial. Recently, the concepts of TSM
and FQH physics have been combined to the denition
of the fractional Chern insulator
8789
and the fractional
topological insulator
9093
the rst lattice realization of
which has been reported in Ref. 94. These states are
translation-invariant realizations of the FQH eect and
its time reversal symmetry protected analogues, respec-
tively. A general hierarchy of fractional topological insu-
lators has been reported in Ref. 95.
VIII. OUTLOOK
From a conceptual point of view, the entire zoo of
topological states of matter can be seen as conclusively
understood in the framework outlined in this review.
However, there are at least two general routes to be
considerably further explored by future research. First,
the precise experimental implications of many topolog-
ical states of matter have not been fully unraveled yet.
Whereas in the quantum Hall state, the topological in-
variant directly represents a physical observable, namely
the Hall conductivity of the sample, the observability of
the topological invariants of several topological states
of matter is unknown or still under active debate.
This issue is from our point of view closely related to
the rather limited number of promising proposals for
concrete technological applications based on these novel
states of matter. Obviously, successful research in this
direction will be of decisive importance for the long
term future of the entire eld of topological states of
matter. Second, the inuence of interactions and open
quantum system eects on topological states of matter
is by no means conclusively understood, let alone an
exhaustive topological classication of interacting or
dissipative systems. As a rst step along these lines,
a purely dissipation driven topological state has been
reported in Ref. 96 and 97.
ACKNOWLEDGEMENT
We acknowledge interesting discussions with Hans
Hansson, Patrik Recher, Dietrich Rothe, Ronny
Thomale, Grigori Volovik, and Shou-Cheng Zhang as
well as nancial support from the DFG-JST research unit
Topotronics (BT and JCB) and from the Swedish Sci-
ence Research Council (JCB).
1
C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 226801
(2005).
2
C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 146802
(2005).
3
B. A. Bernevig, T. L. Hughes, and S.-C. Zhang, Science
314, 1757 (2006).
4
M. K onig, S. Wiedmann, C. Br une, A. Roth, H. Buhmann,
L. Molenkamp, X.-L. Qi, and S.-C. Zhang, Science 318,
766 (2007).
5
M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045
(2010).
6
S. Ryu, A. P. Schnyder, A. Furusaki, and A. W. W. Lud-
wig, New Journal of Physics 12, 065010 (2010).
7
X.-L. Qi and S.-C. Zhang, Rev. Mod. Phys. 83, 1057
(2011).
8
P. W. Anderson, Basic Notions Of Condensed Matter
Physics (perseus, 1997).
9
B. I. Halperin, Phys. Rev. B 25, 2185 (1982).
10
G. E. Volovik, The Universe in a Helium Droplet (Claren-
don Press, 2003).
11
T. Kato, Journal of the Physical Society of Japan 5, 435
(1950).
12
M. V. Berry, Proc. R. Soc. Lond. A 392, 45 (1984).
13
F. Wilczek and A. Zee, Phys. Rev. Lett. 52, 2111 (1984).
14
B. Simon, Phys. Rev. Lett. 51, 2167 (1983).
15
A. Bohm, A. Mostafazadeh, H. Koizumi, Q. Niu, and
J. Zwanziger, The geometric phase in quantum systems
(Springer, 2003).
16
M. Karoubi, K-Theory: An Introduction (Springer, 1978).
17
A. Kitaev, AIP Conference Proceedings 1134, 22 (2009).
18
M. Born and V. Fock, Zeitschrift f ur Physik A Hadrons
and Nuclei 51, 165 (1928).
19
Note1, as a minor technical point, we note that the proof
by Born and Fock
18
also takes into account level crossings
at isolated points. These slightly more general conditions
are not of relevance for our purposes as we will only dis-
cuss fully gapped systems. More recent work by Avron
and coworkers
98
reported a proof of the adiabatic theo-
rem which, under certain conditions on the level spectrum,
works without any gap condition.
20
Y. Aharonov and D. Bohm, Phys. Rev. 123, 1511 (1961).
21
J. E. Avron, L. Sadun, J. Segert, and B. Simon, Commu-
nications in Mathematical Physics 124, 595 (1989).
22
A. Zee, Phys. Rev. A 38, 1 (1988).
23
M. Nakahara, Geometry, Topology and Physics (Taylor &
Francis, 2003).
24
Y. Choquet-Bruhat and C. deWitt Morette, Analysis Man-
ifolds and Physics (Elsevier Science, 1982).
25
W. Kuehnel, Dierential Geometry: Curves - Surfaces -
Manifolds (American Mathematical Society, 2005).
26
C. Nash and S. Sen, Topology and Geometry for Physicists
(Dover Publications, 2011).
27
J. W. Milnor and J. Stashe, Characteristic Classes
(Princeton University Press, 1974).
28
F. D. M. Haldane, Phys. Rev. Lett. 61, 2015 (1988).
29
X.-L. Qi, Y.-S. Wu, and S.-C. Zhang, Phys. Rev. B 74,
085308 (2006).
21
30
C.-X. Liu, X.-L. Qi, X. Dai, Z. Fang, and S.-C. Zhang,
Phys. Rev. Lett. 101, 146802 (2008).
31
R. B. Laughlin, Phys. Rev. B 23, 5632 (1981).
32
J. Zak, Phys. Rev. Lett. 62, 2747 (1989).
33
L. Fu and C. L. Kane, Phys. Rev. B 74, 195312 (2006).
34
D. J. Thouless, M. Kohmoto, M. P. Nightingale, and
M. den Nijs, Phys. Rev. Lett. 49, 405 (1982).
35
J. E. Avron, R. Seiler, and B. Simon, Phys. Rev. Lett. 51,
51 (1983).
36
M. Kohmoto, Annals of Physics 160, 343 (1985).
37
A. Altland and M. R. Zirnbauer, Phys. Rev. B 55, 1142
(1997).
38
E. Cartan, Bull. Soc. Math. France 54, 214264 (1926).
39
X.-L. Qi, T. L. Hughes, and S.-C. Zhang, Phys. Rev. B 78,
195424 (2008).
40
A. P. Schnyder, S. Ryu, A. Furusaki, and A. W. W. Lud-
wig, Phys. Rev. B 78, 195125 (2008).
41
A. P. Schnyder, S. Ryu, A. Furusaki, and A. W. W. Lud-
wig, AIP Conference Proceedings 1134, 10 (2009).
42
T. Eguchi, P. B. Gilkey, and A. J. Hanson, Physics Reports
66, 213 (1980).
43
C. Nash, Dierential Topology and Quantum Field Theory
(Academic Press, 1991).
44
P. Horava, Phys. Rev. Lett. 95, 016405 (2005).
45
D. S. Freed and G. W. Moore, arXiv:1208.5055 (2012).
46
M. F. Atiyah, The Quarterly Journal of Mathematics 17,
367 (1966).
47
R. Bott and L. W. Tu, Dierential Forms in Algebraic
Topology (Springer, 1982).
48
H. Sato, Algebraic Topology: An Intuitive Approach
(American Mathematical Society, 1996).
49
A. N. Redlich, Phys. Rev. D 29, 2366 (1984).
50
G. Volovik, JETP 67(9), 1804 (1988).
51
M. F. L. Golterman, K. Jansen, and D. B. Kaplan, Physics
Letters B 301, 219 (1993).
52
S.-S. Chern and J. Simons, The Annals of Mathematics
99, pp. 48 (1974).
53
J. E. Avron, L. Sadun, J. Segert, and B. Simon, Phys. Rev.
Lett. 61, 1329 (1988).
54
S.-C. Zhang and J. Hu, Science 294, 823 (2001).
55
E. Witten, Nuclear Physics B 223, 422 (1983).
56
Z. Wang, X.-L. Qi, and S.-C. Zhang, Phys. Rev. Lett. 105,
256803 (2010).
57
Q. Niu, D. J. Thouless, and Y.-S. Wu, Phys. Rev. B 31,
3372 (1985).
58
L. Fu and C. L. Kane, Phys. Rev. B 76, 045302 (2007).
59
B. Leung and E. Prodan, Phys. Rev. B 85, 205136 (2012).
60
J. C. Budich and B. Trauzettel, arXiv:1207.1104 (2012).
61
I. C. Fulga, F. Hassler, and A. R. Akhmerov, Phys. Rev.
B 85, 165409 (2012).
62
L. Fidkowski and A. Kitaev, Phys. Rev. B 81, 134509
(2010).
63
H. L. Stormer, A. Chang, D. C. Tsui, J. C. M. Hwang,
A. C. Gossard, and W. Wiegmann, Phys. Rev. Lett. 50,
1953 (1983).
64
R. B. Laughlin, Phys. Rev. Lett. 50, 1395 (1983).
65
H. So, Progress of Theoretical Physics 74, 585 (1985).
66
K. Ishikawa and T. Matsuyama, Zeitschrift f ur Physik C
Particles and Fields 33, 41 (1986).
67
K. Ishikawa and T. Matsuyama, Nuclear Physics B 280,
523 (1987).
68
G. E. Volovik, arXiv:1111.4627 (2011).
69
V. Gurarie, Phys. Rev. B 83, 085426 (2011).
70
L. Wang, X. Dai, and X. C. Xie, Phys. Rev. B 84, 205116
(2011).
71
J. C. Budich, R. Thomale, G. Li, M. Laubach, and S.-C.
Zhang, Phys. Rev. B 86, 201407 (2012).
72
J. C. Budich, B. Trauzettel, and G. Sangiovanni,
arXiv:1211.3059 (2012).
73
G. Volovik, JETP Letters 91, 55 (2010).
74
J. V ayrynen and G. Volovik, JETP Lett. 93, 344 (2011).
75
Z. Wang and S.-C. Zhang, Phys. Rev. X 2, 031008 (2012).
76
Z. Wang and B. Yan, arXiv:1207.7341 (2012).
77
Z. Wang and S.-C. Zhang, arXiv:1204.3149 (2012).
78
S. R. Manmana, A. M. Essin, R. M. Noack, and V. Gurarie,
arXiv:1205.5095 (2012).
79
A. Kitaev, Physics-Uspekhi 44, 131 (2001).
80
E. Prodan, Phys. Rev. B 83, 235115 (2011).
81
Z. Wang, X.-L. Qi, and S.-C. Zhang, Phys. Rev. B 85,
165126 (2012).
82
A. M. Essin and V. Gurarie, Phys. Rev. B 84, 125132
(2011).
83
J. E. Avron, R. Seiler, and B. Simon, Phys. Rev. Lett. 65,
2185 (1990).
84
X.-L. Qi and S.-C. Zhang, Phys. Rev. Lett. 101, 086802
(2008).
85
S.-C. Z. Xiao-Liang Qi, Taylor L. Hughes, Nature Physics
4, 273 (2008).
86
X.-G. Wen, Int. J. Mod. Phys. B 4, 239 (1990).
87
E. Tang, J.-W. Mei, and X.-G. Wen, Phys. Rev. Lett. 106,
236802 (2011).
88
K. Sun, Z. Gu, H. Katsura, and S. Das Sarma, Phys. Rev.
Lett. 106, 236803 (2011).
89
T. Neupert, L. Santos, C. Chamon, and C. Mudry, Phys.
Rev. Lett. 106, 236804 (2011).
90
B. A. Bernevig and S. C. Zhang, Phys. Rev. Lett. 96,
106802 (2006).
91
M. Levin and A. Stern, Phys. Rev. Lett. 103, 196803
(2009).
92
J. Maciejko, X.-L. Qi, A. Karch, and S.-C. Zhang, Phys.
Rev. Lett. 105, 246809 (2010).
93
B. Swingle, M. Barkeshli, J. McGreevy, and T. Senthil,
Phys. Rev. B 83, 195139 (2011).
94
T. Neupert, L. Santos, S. Ryu, C. Chamon, and C. Mudry,
Phys. Rev. B 84, 165107 (2011).
95
L. Santos, T. Neupert, S. Ryu, C. Chamon, and C. Mudry,
Phys. Rev. B 84, 165138 (2011).
96
S. Diehl, E. Rico, M. A. Baranov, and P. Zoller, Nature
Physics 7, 971 (2011).
97
C.-E. Bardyn, M. A. Baranov, E. Rico, A.

Imamo glu,
P. Zoller, and S. Diehl, Phys. Rev. Lett. 109, 130402
(2012).
98
J. E. Avron and A. Elgart, Communications in Mathemat-
ical Physics 203, 445 (1999).