Physics of Three-Dimensional Bosonic Topological Insulators: Surface-Deconfined Criticality and Quantized Magnetoelectric Effect

Physics of Three-Dimensional Bosonic Topological Insulators: Surface-Deconned Criticality
and Quantized Magnetoelectric Effect

Ashvin Vishwanath
Department of Physics, University of California, Berkeley, California 94720, USA
Materials Science Division, Lawrence Berkeley National Laboratories, Berkeley, California 94720, USA
T. Senthil
Department of Physics, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA
(Received 4 October 2012; revised manuscript received 31 December 2012; published 28 February 2013)
We discuss physical properties of integer topological phases of bosons in D 3 1 dimensions,
protected by internal symmetries like time reversal and/or charge conservation. These phases invoke
interactions in a fundamental way but do not possess topological order; they are bosonic analogs of free-
fermion topological insulators and superconductors. While a formal cohomology-based classication of
such states was recently discovered, their physical properties remain mysterious. Here, we develop a
eld-theoretic description of several of these states and show that they possess unusual surface states,
which, if gapped, must either break the underlying symmetry or develop topological order. In the latter
case, symmetries are implemented in a way that is forbidden in a strictly two-dimensional theory. While
these phases are the usual fate of the surface states, exotic gapless states can also be realized. For example,
tuning parameters can naturally lead to a deconned quantum critical point or, in other situations, to a
fully symmetric vortex metal phase. We discuss cases where the topological phases are characterized by a
quantized magnetoelectric response 0, which, somewhat surprisingly, is an odd multiple of 2r. Two
different surface theories are shown to capture these phenomena: The rst is a nonlinear sigma model with
a topological term. The second invokes vortices on the surface that transform under a projective
representation of the symmetry group. We identify a bulk-eld theory consistent with these properties,
which is a multicomponent background-eld theory supplemented, crucially, with a topological term. We
also provide bulk sigma-model eld theories of these phases and discuss a possible topological phase
characterized by the thermal analog of the magnetoelectric effect.
DOI: 10.1103/PhysRevX.3.011016 Subject Areas: Condensed Matter Physics, Strongly Correlated Materials,
Topological Insulators
I. INTRODUCTION
Following the discovery of topological insulators (TIs)
[1], intense theoretical efforts have resulted in a good
understanding of the topological phases of free fermions,
including a complete classication of such phases that are
stable to disorder [2]. In these phases, the bulk appears to
local probes as a rather conventional gapped state. The
surface is gapless, however, unless one of the symmetries
protecting the phase is broken.
In contrast, our understanding of the topological phases
of interacting particles is much less complete. The frac-
tional quantum Hall effect has inspired much work on
phases with topological order [3]. These phases have the
remarkable property that the degeneracy of their ground
states depends on the topology of the space on which they
are dened. These phases are often associated with edge
states, but, additionally, excitations with exotic statistics
also occur in the bulk. Moreover, these phases of interact-
ing particles are characterized by a topological entangle-
ment entropy, implying long-range quantum entanglement
in the ground state. It seems appropriate to exclude such
phases from a minimal generalization of topological insu-
lators to interacting systems.
We dene a short-range entangled (SRE) state [4] as a
gapped state with a unique ground state on all closed
manifolds, i.e., one that has no topological order. In the
presence of interactions, do new SRE phases appear that
share the same symmetry but differ at the level of topol-
ogy? A possible distinction, for example, is the presence of
protected states at the boundaries.
In this paper, we study the physics of such SRE phases in
systems of interacting bosons. For bosons, the noninteract-
ing limit is a simple condensate so that interactions are
necessary to stabilize gapped phases. Thus, we necessarily
need to free ourselves from the crutch of free-fermion
Hamiltonians and band topology on which most current
discussions of topological insulators are based. Studying
bosonic generalizations of topological insulators is poten-
tially a useful step toward solving the harder problem of
interaction-dominated fermionic topological insulators.
Published by the American Physical Society under the terms of
the Creative Commons Attribution 3.0 License. Further distri-
bution of this work must maintain attribution to the author(s) and
the published articles title, journal citation, and DOI.
PHYSICAL REVIEW X 3, 011016 (2013)
2160-3308,13,3(1),011016(28) 011016-1 Published by the American Physical Society
Several examples of interacting-boson systems exist. A
very natural realization of a strongly interacting boson
system is a quantum magnet made up of localized quantum
spins on a lattice. In that context, the phases we are inter-
ested in may be dubbed topological paramagnets. They
have a bulk gap and no bulk topological order or fractional
quantumnumbers, but theyhave protected surface states, all
features in close analogy to electronic topological insula-
tors. The topological paramagnet should be distinguished
from a more familiar (and more exotic) paramagnetthe
quantum spin-liquid statewhich has been extensively
discussed in the literature. Quantum spin-liquid states
have either bulk topological order or gapless excitations.
Other realizations of strongly interacting bosons are, of
course. provided by ultracold atoms in optical lattices.
A famous example of a topological paramagnet already
exists: the spin-1 Haldane (or AKLT) chain, which has a
bulk gap and no fractionalization but has dangling spin-1,2
moments at the edge that are protected by symmetry [6].
Using a powerful matrix-product representation of gapped
states [710], topological phases in one dimension are
completely classied by the second group cohomology of
symmetry group G. In 2D and 3D, such rigorous results are
not available. However, Kitaev pointed out that 2D SRE
phases of bosons with chiral edge states are nevertheless
possible [11]. Recently, Chen et al. [5] have proposed that
SRE topological phases of bosons protected by symmetry
are captured by the higher-dimensional cohomology
groups. While this insight provides an efcient mathemati-
cal scheme to enumerate phases, unfortunately, their prop-
erties are not transparently obtained.
Progress on clarifying the physics of these states has
subsequently been made by several authors. Levin and Gu
[12] explicitly studied a specic example in 2D protected
by Z
2
symmetry. 2D SRE topological phases were de-
scribed using a simpler Chern-Simons approach [13],
which provided a eld theory and explicit edge theories
of these states. It was found that bosons with a conserved
global U1 but no other symmetries can display an
integer quantum Hall state with quantized Hall conduc-
tance predicted to be even-integer multiples of q
2
,h, i.e.,
a
xy
2nq
2
,h, where q is the elementary charge of the
bosons [13]. A simple physical realization of such a phase
of two-component bosons in the lowest Landau level has
recently been provided [14]. Other physical interpretations
have also been described [15,16]. Thus, some dening
properties of several symmetry-protected topological
(SPT) phases in 1D and 2D are now understood.
However in 3D, although the cohomology classication
predicts various SPT phases, including bosonic general-
izations of topological insulators, the physical properties of
SPT phases have not thus far been elucidated. This eluci-
dation is our primary task in this paper. Potentially, our
approach could also be used to classify 3D bosonic SPT
phases, which we leave to future work. We note, however,
that, for all symmetry classes that we considered both with
time-reversal (Z
T
2
, U1 Z
T
2
, U1 2 Z
T
2
) and without
(U1 2 Z
2
), we identify candidate phases that exhaust
those predicted by the cohomology classication [5].
Further, we identify a possible topological phase that ap-
pears to be outside the classication of Ref. [5].
Much of our discussion focuses on a theory of the novel
surface states of these 3D bosonic topological insulators.
We construct effective Landau-Ginzburg eld theories of
these surface states. We also identify bulk eld theories that
correctly yield the proposed surface theory. Akey feature of
this surface theory is that it does not admit a trivial gapped
symmetry-preserving surface phase. The surface either
spontaneously breaks symmetry, or, if gapped, develops
surface topological order (even though there is no bulk
topological order). Other, more exotic symmetry-
preserving states with gapless excitations are also possible.
In all these cases, the dening global symmetries are im-
plemented in a way not allowed in strictly 2D systems. As a
specic example, consider insulating bosonic phases
with the symmetries of a topological insulator: charge
conservation and time-reversal symmetry [formally de-
noted as U1 2 Z
T
2
]. For these phases, the surface can break
a symmetry, e.g., by forming a surface superuid.
Alternatively, the surface can remain insulating, while
breaking time-reversal symmetry. The bulk is assumed
to retain the symmetry. Then these surface-ordered phases
can reect their special origin by exhibiting features that are
forbidden in purely 2D phases. For example, we show that
the vortices of the surface superuid mentioned above are,
in a precise sense, fermionic. Furthermore, the surface
insulator with broken time-reversal symmetry is shown to
have a Hall conductance of a
xy
1 (in units of q
2
,h), in
contrast to the 2D integer quantum Hall phases of bosons,
which are allowed only to even-integer Hall conductance.
Equivalently, the surface quantum Hall effect may be
considered a quantized 3D response: the magnetoelectric
polarizability 0. Recall, in the context of free-fermion
topological insulators with broken time-reversal on the
surface that gaps the surface states, that a quantized mag-
netoelectric effect appears [17] that is captured by the
topological theta term:
L
0

0
4r
2
E

B, (1)
where h c e 1 and

E and

B are applied electric and
magnetic elds. For free-fermion topological insulators,
0 rmod 2r, corresponding to a half-integer Hall ef-
fect on the surface. The 2r ambiguity in 0 corresponds to
the fact that one may deposit a fermionic integer quantum
Hall layer on the surface [18].
Here, for bosonic topological insulators, 0 is only de-
ned modulo 4r, and the topological phase corresponds to
0 2r. This theta value implies, for example, that the
domain wall between opposite time-reversal-symmetry-
breaking regions on the surface induces a protected
ASHVIN VISHWANATH AND T. SENTHIL PHYS. REV. X 3, 011016 (2013)
011016-2
mode, which is the edge state corresponding to the a
xy
2
quantized Hall effect of bosons. The 4r ambiguity in 0
corresponds to the fact that one may deposit an integer
quantum Hall layer of bosons on the surface, which must
have an even-integer Hall conductance.
A symmetry-preserving surface state can be accessed
from the superuid by condensing vortices that transform
trivially under the symmetry. However, as the vortex of the
surface superuid has fermionic statistics (in the sense
discussed in Sec. IVD, it cannot condense. This unusual
nature of the vortices precludes the possibility of a trivial
gapped surface insulator. The fermionic vortices can, of
course, pair and condense. However, as is well known
[19], paired-vortex condensation leads to a 2D state with
topological order (described in the present context by a
deconned Z
2
gauge theory). We show that this surface
topological order realizes symmetry in a manner not al-
lowed in strictly 2D systems.
Exotic gapless surface states that preserve all symme-
tries are also conceivable. For example, the gapless surface
state may intuitively be viewed as a quantum Hall state that
uctuates between a
xy
1. The theory of such a state is
constructed using a network model that captures the
quantum-phase transition between distinct integer quan-
tum Hall states of bosons. The same approach, when
applied to fermionic topological insulators, correctly
yields the single-Dirac-cone surface state. The eld theory
thus obtained of the bosonic model poised at the transition
naturally leads to the required surface theory, which is
closely related to the deconned-quantum-critical theory
[20] previously proposed in the context of frustrated quan-
tum magnets.
3D eld theory.The general arguments above are
shown to be consistent with the following J 3 eld
theory:
2rL
3D

I
e
:a
B
I
:
o
I
a

I,1
K
I1
4r
e
:a
o
I
:
o
1
a
, (2)
where the index I refers to boson species; bosons four
currents are represented by ]
I
1
2r
e
:a
:
B
I
a
; and the
curl of o
I
represents the vortex lines. The rst background-
eld (BF) term [21] just represents the 2r phase factor of
taking a particle around its vortex. The key topological
properties, however, are determined by the second BF
term, which attaches quantumnumbers to vortices. To avoid
topological order at the surface and to ensure bosonic ex-
citations, detK 1 and diagonal entries are even integers.
In most cases, we take two species with K a
x
. Here, !
2r is assumed to lead to an equivalent theory that
differs only in details of surface termination. Furthermore,
time-reversal symmetry constrains 0, r, the latter
being the topological phase. Note that for the internal
gauge elds o
I
is distinguished from 0 for the external
electromagnetic eld discussed above. Coupling to an
external electromagnetic eld allows one to obtain the quan-
tized magnetoelectric effect discussed above. Related theo-
ries have appeared in the context of 3D topologically
ordered phases and superconductors [22] where only the
rst BF term in Eq. (2) appears with a different coefcient.
On the other hand, the eld theory discussed in Ref. [23]
retains only the second term, which leads to gapless excita-
tions in the bulk, as noted in Ref. [24], which in turn differ
fromthe gapped phases of interest here. Thus, it is important
to combine both the rst and the second terms.
Recently, it was proposed that free fermion topological
insulators are captured by similar theories [25], with a
single component eld and the rst term of Eq. (2) along
with coupling to the external eld. The surface states in
such theories were argued to be bosonized Dirac fermions
[25,26]. While this is an intriguing idea, we point out (see
Appendix A) certain problems with the identication of a
metallic surface in Ref. [25]. Moreover, identication of a
bulk fermionic operator is also problematic in this theory
[25,26]. One approach, taken in a recent paper [27], is to
regard this as a partial theory that provides a purely hydro-
dynamic description that excludes fermionic excitations
altogether. Thus, nding a complete, effective eld-theory
description for 3D fermionic topological insulators re-
mains an open problem.
We also display continuum eld-theoretic models in
D 3 1 dimension that realize some of the topological
phases we describe. These models are obtained as pertur-
bations of nonlinear sigma models in the presence of a
topological theta term. The theta term has the effect of
endowing topological-defect congurations of the contin-
uum elds with nontrivial global quantum numbers. We
showthe connection to the topological BF theory and to the
theory of protected surface states.
Possible phases with half-quantized surface thermal Hall
effect.Interestingly, our general approach and the theory
in Eq. (2) both predict a new 3D SPT phase, protected by
time-reversal symmetry, that is not found within the coho-
mology classication of Chen et al. [5]. Just as the quan-
tized Hall effect of J 2 SRE bosons immediately
constrained the physics in 3D, an analogous argument for
thermal Hall conductance can be made. In J 2, this is
quantized to 8 times the quantum of thermal conductance:
k
xy
,T 8n
r
2
k
2
B
3h
for a 2D SRE phase of bosons. A realiza-
tion of n 1 is the Kitaev E
8
state [11], with 8 chiral bosons
at the edge. Therefore, a 3D phase protected by time-
reversal symmetry can be conceived of, on the surface of
which a domain wall between opposite Tbreaking regions
hosts 8 chiral boson modes and which is described by
Eq. (2), with the 8D K
E
8
matrix discussed in Ref. [13].
This putative phase lies outside the cohomology classica-
tion of Chen et al. [5], which reports a single nontrivial
topological phase with this symmetry, for which a different
candidate, with K a
x
, is identied in Sec. VII.
PHYSICS OF THREE-DIMENSIONAL BOSONIC . . . PHYS. REV. X 3, 011016 (2013)
011016-3
II. OVERVIEW
As this paper is long and discusses bosonic SPT phases
from several points of view, it is helpful to provide an
overview. We seek an effective Landau-Ginzburg descrip-
tion of the surface of a 3D SPT phase. We employ various
approaches that, satisfyingly, lead to consistent conclu-
sions. In this section, we motivate these different ap-
proaches as preparation for the rest of the paper.
A key physical requirement of the surface theory is that
there be no trivial gapped-symmetry-preserving surface
phase. This feature is reminiscent of the Lieb-Schultz-
Mattis (LSM) theorem [6] and its generalization [28,29]
to states of bosonic systems at a fractional lling on clean
2D lattices. Indeed, in both cases, either a symmetry must
be broken or there is topological order, or there is more
exotic long-range entanglement, possibly with protected
gapless excitations. There is one important difference,
however. For the surface states discussed in this paper,
the trivial insulating phase does not exist even in the
presence of disorder that breaks translation symmetry.
They are protected by internal symmetries rather than by
lattice-translation symmetry (as is the case for the LSM
theorem). Nevertheless, we exploit insights from existing
effective eld theories of clean lattice bosons that build in
the LSM restrictions in order to construct the desired
surface theory of the bosonic topological insulator.
We begin with a physical discussion of the constraints on
quantized electrical and thermal responses in 3D SRE bo-
sonic insulators, imposed by our knowledge of 2D SRE
phases (Sec. III). In particular, we argue that the quantized
magnetoelectric coupling for bosonic topological insulators
(0 2r) is doublethat of the fermionic case (0 r), which
parallels the doubling of Hall conductance [13,14,16] for 2D
integer states of bosons compared to fermions.
Next, we borrow an approach that is useful in classifying
SPT phases of 2D bosons. There, the edge states are
typically described by a 1D Luttinger liquid theory. What
makes them special is that symmetry acts on these edge
states in a way that is impossible to realize in a purely 1D
system [1316]. This property ensures that all perturba-
tions that lead to a gapped phase also break symmetry. By
analogy, to discuss a 3D SPT phase, we model the 2D
surface by a conventional 2D theory of bosons (e.g., an XY
model). However, we identify symmetry operations that
are forbidden in a conventional 2D system by demanding
that the 2D system can never enter a trivial gapped phase.
As a useful device, consider an enlarged U1 U1
global symmetry, along with time-reversal symmetry T
(technically, U1 U1 2 Z
T
2
), and eventually break it
down to the symmetry of interest. Physically, this enlarged
symmetry corresponds to separate conservation of two spe-
cies of bosons: species 1 and 2. Consider the surface of a 3D
SPT phase with a broken symmetry, e.g., a condensate
of species-1 bosons that breaks the rst U1 symmetry.
This superuid must also be unusual in that it cannot be
connected to a fully symmetric insulator. Guidance from
effective eld theories of clean 2D lattice bosons at frac-
tional lling suggests thinking in terms of the vortices of the
superuid. Since the insulating state is obtained by condens-
ing vortices, we can ask, What vortex properties provide
the required obstruction? Unlike a particle, a vortex is a
nonlocal object and can transform projectively under the
remaining U1 and T symmetry. Vortices with projectively
realized symmetries provide an obstruction to realizing a
trivial insulator and can describe the surface of an SPT
phase. In this example, the projective transformation re-
quires that the two species of vortices (c
) carry half-
charge of species 2, and these two species are exchanged
by time-reversal symmetry. A minimal theory of these vor-
tices is obtained by representing the density of the con-
densed species 1 by the curl of a vector potential
N
1

x
n
y

y
n
x
,2r, and this vector potential couples
minimally to the vortices:
L
edge2D
|n
|a
2
A
2
_
c
a
e
:
2r
A
1
:
n
n
:
:
n
2
, (3)
where we have inserted external electromagnetic elds A
1
,
A
2
that couple to the two conserved currents. To obtain an
insulating surface, we must condense the vortices, but this
action inevitably breaks symmetry. Since the vortices trans-
form into one another under time reversal, one cannot con-
dense one species and not the other. Condensing them both,
however, implies breaking the other U1 symmetry, since
FIG. 1. Schematic depiction of a 3D symmetry-protected
topological phase, with two conserved species of bosons
[U1 U1, blue and red] and time-reversal symmetry (Z
T
2
).
The bulk is insulating and corresponds to a condensate of
vortices of both species (shown as black and white loops). In
the topological phase, the vortex line of one species that ends on
a surface carries half charge of the other species. Such a surface,
it may be argued, does not have a trivial gapped phase, where the
symmetries are preserved. Pictorially, the vortex lines may be
viewed in 1D Haldane phases, with half-charged end states.
011016-4
the vortices carry the charge of species 2. Athird option is to
condense pairs of vortices, but this results in Z
2
topological
order [19]. This discussion points to a schematic picture of
the SPT phase as shown in Fig. 1. A bulk insulator may be
regarded as a condensate of vortex loops. However, the
special feature here is that the condensed vortex lines carry
a half charge at their ends when they intersect the surface.
The vortex lines, viewed as 1D objects, are in a topological
state analogous to the Haldane phase, and thus carry edge
states.
Note that, if a single bosonic vortex species was present at
low energies, we could perform a duality back to the usual
boson phase variables. However, having multiple vortex
elds demands a dual description like the one above. In
fact, an identical theory appeared in the discussion of
deconned quantum critical points [20,30,31]; however,
there the symmetries included both internal and spatial
symmetries. Here all symmetries are internal and hence
are restricted to the surface of a 3D topological phase.
Let us consider in more detail the insulating surface
obtained on breaking time-reversal symmetry. This break-
ing is obtained by condensing just one species of vortex,
say, c
, in Eq. (3), which forces n

1
2
A
2
. When substi-
tuted into the action above, this calculation yields the
electromagnetic response
L
em

1
4r
A
2
e
:
:
A
1
. (4)
If the separate species are now identied with a single
conserved charge, we can set A
1
A
2
A, which yields
a
xy
1 on the surface, indicating a magnetoelectric re-
sponse 0 2r, as predicted. In the absence of time-
reversal breaking, the surface may be assumed to uctuate
between a
xy
1. We provide an alternate derivation of
the same surface theory [Eq. (3)] by modeling it using a
network model, poised at the transition point between two
bosonic integer quantum Hall states. This model is analo-
gous to obtaining the surface state of the 3D fermionic
topological insulators, the single Dirac cone, by a network
model at the critical point between integer quantum Hall
plateaus [32] in a clean system. In both cases, the time-
reversal symmetry automatically tunes the system to criti-
cality. For the bosonic SPT surface, the resulting theory is
an O4 nonlinear sigma model with a topological term, as
in the Euclidean Lagrangian
L
0
edge2D
1
2k
Tr
g
y
g
|
r
24r
2
e
:
Trg
y
gg
y
:
gg
y
g, (5)
where the O4 vector has been written in terms of an
SU2 matrix g. Reassuringly, this model has been argued
in Ref. [33] to be equivalent to Eq. (3) once appropriate
anisotropies are introduced. These two descriptions of
surface properties of SPT phases are discussed in Sec. IV
along with the connection between them.
We also emphasize the properties of a gapped symmetric
surface state obtained by condensing paired vortices. This
state has surface topological order described by a decon-
ned Z
2
gauge theory (even though the bulk has no such
order). This state provides a particularly simple perspective
on why a trivial gapped symmetric phase is forbidden at the
surface. Indeed, we show that bosonic topological quasi-
particles of this state carry fractional quantum numbers.
Destroying the topological order by condensing one of
these quasiparticles necessarily breaks a symmetry. Not
surprisingly, we show that the implementation of symme-
try in this surface topological ordered state is distinct from
what is allowed in strict 2D systems.
In Sec. V, we discuss bulk theories that are consistent
with the surface descriptions above. We describe SPT
phases in quantum magnetsthe ones we dubbed topo-
logical paramagnetsprotected by other symmetries such
as time-reversal (Z
T
2
), or time-reversal along with one
component of spin rotation [U1 Z
T
2
] in Sec. VI.
(Appendix C contains an example without time-reversal
symmetry [U1 2 Z
2
]. Section VII closes the paper with
the discussion of a new topological phase that is predicted
by this approach, a 3D extension of Kitaevs E
8
state, along
with some additional comments.
III. TRANSPORT PROPERTIES OF 3D BOSONIC
TOPOLOGICAL INSULATORS:
GENERAL CONSTRAINTS
We begin by considering a system of interacting
bosons in J 3 space dimensions in the presence of
time-reversal and particle-number-conservation symme-
tries. Specically, let us consider the situation in which
the boson eld l carries charge 1 under a global U1
symmetry and transforms as l ! l under time reversal.
The corresponding symmetry group is U1 2 Z
T
2
. Assume
the system is in a gapped insulating phase (at least in the
absence of any boundaries) and that there is a unique
ground state on topologically nontrivial manifolds. For
any such insulator in 3D, the effective Lagrangian for an
external electromagnetic (EM) eld obtained by integrat-
ing out all the matter elds takes the form
L
eff
L
Max
L
0
. (6)
The rst term is the usual Maxwell term and the second is
the theta term in Eq. (1).
Several properties of the theta term are well known.
First, under time reversal, 0 ! 0. Next, on closed mani-
folds, the integral of
1
4r
2

E.

Bis quantized to be an integer so
that the quantum theory is periodic under 0 ! 0 2r.
These two facts together imply that time-reversal-
symmetric insulators have 0 nr, with n an integer.
Trivial time-reversal-symmetric insulators have 0 0,
while free-fermion topological insulators have 0 r.
011016-5
If we allow for a boundary to the vacuum and further
assume that the boundary is gapped (if necessary by break-
ing time-reversal symmetry), then the 0 term leads to a
surface Hall conductivity of
0
2r
. To see this, assume a
boundary (at, say, z 0), then 0 0z is zero for z <0
and constant 0 for z >0. The action associated with the 0
term is
S
0

1
8r
2
_
J
3
xJi0z
(7)

1
8r
2
_
J
3
xJi
J0
Jz
K
z
(8)
0
8r
2
_
B
J
2
xJie
z:k
A
:
A
k
, (9)
where A is the external electromagnetic potential and
K
e
:k
A
:
A
k
. Equation (9) is a surface Chern-
Simons term and leads to a Hall conductivity of 0,2r.
For fermion topological insulators, 0 r so that the
surface a
xy

1
2
. If we shift 0 ! r 2nr, then the sur-
face a
xy
n
1
2
. This shift corresponds to simply de-
positing an ordinary integer quantum Hall state of fermions
at the surface of this insulator. Hence, this shift should not
be regarded as a distinct bulk state. The only nontrivial
possibility is 0 r.
Now let us consider bosonic insulators. Again, T rever-
sal and periodicity imply that 0 nr and that a surface
a
xy
n,2. A crucial observation is that now0 2rmust
be regarded as distinct from0 0. At 0 2r, the surface
a
xy
1. However, this Hall conductivity cannot be ob-
tained from the surface of the 0 0 insulator by deposit-
ing any 2D integer quantum Hall effect (IQHE) state of
bosons. Recent work [13,14] has shown that 2D IQHE
states of bosons necessarily have a
xy
even. (See
Ref. [14] for a simple argument.) Thus, the surface state
of the 0 2r boson insulator is not a trivial 2D state but
rather requires the presence of the 3D bulk.
Therefore, 0 2r necessarily corresponds to a non-
trivial 3D bosonic TI. 0 4r is trivial, however, as then
the surface state can be regarded as a 2D bosonic IQHE
state. One may still obtain a 3D topological phase, but the
topology is not manifest in the electromagnetic response.
We can sharpen and generalize this result. Under T
reversal, as 0 ! 0, nr ! nr. As the bulk state is T
reversal invariant, we require that the surface state at
0 nr be obtainable from the surface state at
0 nr by depositing a 2D IQHE boson state. Let us
characterize the surface state by both its electrical and its
thermal Hall conductivities a
xy
, k
xy
. Under T reversal,
both Hall conductivities change sign. The requirement
described above then means that 2a
xy
, 2k
xy
must corre-
spond to the allowed electrical and thermal Hall conduc-
tivities of a 2D boson IQHE state.
For 0 2r, it follows that a
xy
1, k
xy
0. It is thus
half of the elementary 2D boson IQHE state.
For 0 r, 2a
xy
1, which is not allowed for the 2D
bosonic IQHE. It follows, therefore, that 3D bosonic TIs
that have no intrinsic topological order cannot have
0 r. It is, of course, very easy to construct such states
[3436] (or other states with fractional 0) if we allow for
fractionalization of the boson, but that violates our original
assumption.
A 2D IQHE state with 2a
xy
8, 2k
xy
8 is allowed; it
is discussed by Kitaev [11]. Thus, a 3D boson TI with
surface a
xy
4, k
xy
4 is allowed. Combining these two
types of fundamental states generates the allowed thermal
and electrical Hall responses on the surface.
Later in this paper, we discuss how these results t with
the formal classication of SPT and other short-ranged
entangled phases in 3D. For now, we reiterate the crucial
observation of this section: A state with EM response of
0 2r necessarily describes a topological insulator of
T-reversal-symmetric bosons while 0 r requires the
presence of intrinsic topological order. In the next
section, we study the properties of this 0 2r boson
topological insulator in detail.
IV. SURFACE THEORY OF 3D BOSONIC
SPT PHASES
In this section, we derive the nontrivial surface theory of
one example of a 3D bosonic SPT phase. We soon special-
ize to the symmetries of the topological insulator: charge
conservation and time-reversal symmetry [U1 2 Z
T
2
] and
exhibit a nontrivial topological phase in three dimensions,
built purely of bosons. To begin, we assume that there are
two species of bosons whose numbers are separately con-
served and that there is enlarged U1 U1 2 Z
T
2
. Later,
we break this enlarged symmetry to just U1 2 Z
T
2
sym-
metry by including interspecies boson mixing terms in the
Hamiltonian. A similar construction [13] has proven to be
very powerful in J 2. We consider two approaches.
Projective vortices.We exploit the fact that a bosonic
SPT phase in J dimensions has surface states that corre-
spond to a conventional theory of bosons in J 1 dimen-
sions except in the way that symmetries are implemented.
For example, the edges of SPT phases of bosons in
D 2 1 dimensions correspond to conventional 1D
Luttinger liquids except for their unusual symmetry trans-
formations [1315]. We therefore consider a 2D bosonic
state to model the surface and assume that the surface is a
superuid breaking one of the U1 symmetries. Then,
vortices of this condensate may transform under a projec-
tive representation of the remaining symmetry group. In a
projective representation, even the identity element of the
symmetry group induces a phase rotation. Hence, local
operators, which can be physically measured, must remain
unchanged under the identity operation of the symmetry
group, since this operation corresponds to doing
011016-6
nothing. However, vortices, which are nonlocal objects,
can transform projectively.
One may attempt to restore the U1 symmetry by con-
densing vortices. However, the projective transformation
ensures that, when vortices condense, they necessarily
break another symmetry. In this way, both the boson and
vortex condensates lead to symmetry breaking, in line with
our general expectation for the surface of a 3D SPT phase.
It is important that vortices transform projectively so that
they cannot be screened by bosons to obtain a trivial
representation of the symmetry group. The projective rep-
resentation is a generalization of the idea of quantum-
number fractionalization. For example, a particle with
half charge changes sign under the 2r-phase rotation of
bosons, implying a projective representation. Clearly, a
half charge cannot be screened by any nite number of
bosons. Projective representations were also recently used
to classify SPT phases in D 1 1, where they corre-
spond to the ends of gapped 1D topological phases. For
example, the half-integer spin edge states of spin-1
Haldane chains furnish a projective representation of the
rotation group. The reasoning above suggests a physical
picture of a 3D SPT phase in which the vortex line in the
bulk is similar to a Haldane-chain-type gapped phase,
which requires low-energy states on the surface where
the vortex ends. In this section, we specialize to the sym-
metries of the topological insulator. Then this procedure
explicitly produces a topological phase characterized by
the quantized magnetoelectric effect 0 2r.
Network model.We directly implement the property
discussed in Sec. III that, if the surface breaks T reversal
and is gapped, then it has quantized Hall transport. If T
reversal is not broken, a powerful approach to obtaining the
surface theory is to start with the theory of the quantum
phase-transition point between the two bosonic quantum
Hall phases that correspond to the two T-broken surfaces.
In the case of free-fermion topological insulators, a similar
reasoning leads to the single Dirac-cone-surface state that
describes the transition between the a
xy

1
2
states on
the surface. For free fermions, the transition between these
integer quantum Hall states is described by a Chalker-
Coddington network model [37]. For the bosonic problem
of interest here, we construct an analogous network model
and show that it leads to a sigma model with a topological
term.
The results of these two approaches are readily seen to
be connected. In both cases, the eld theories we obtain for
the surface have been discussed previously in the context
of deconned quantum criticality. We here discuss the
phase diagram of the surface states described by these eld
theories. When interspecies tunneling is included, the vor-
tices of the two species of bosons are conned to each
other. The resulting single vortex no longer transforms
projectively under the physical symmetries. However, we
argue that it is most conveniently viewed as a fermion. This
unusual nature of the vortices precludes the possibility of
obtaining a trivial insulating phase at the surface by con-
densing vortices.
A. Surface states and projective vortices
Consider a boson eld at the surface with phase degree
of freedom
1
, l
y
1
e
|
1
. We assume that the bulk is
insulating and that the surface is in the x-y plane.
The surface theory could spontaneously break a global
U1 symmetry of boson number conservation (a surface
superuid) or it could stay insulating. More precisely, as
the bulk is always assumed to be insulating, the vortex line
loops have proliferated in the bulk. These vortex lines
penetrate the surface at points that may be viewed as point
vortices of the 2D surface theory since there is no vortex
line tension in the insulating bulk. These point vortices are
gapped when the surface is a superuid. If, instead, they
are condensed, the surface is insulating. To describe vorti-
ces, we move to a dual description [38,39] in which we
write the density and currents of the boson l
1
on the
surface in terms of the eld strengths of a gauge eld
]
1
e
:
:
n
2
,2r. In particular, the density of bosons
is n
1

x
n
2y

y
n
2x
,2r. (The reason for the subscript
2 on n will soon be apparent.) The boson insertion opera-
tors e
|m
1
correspond to monopole insertion operators
since they insert 2rm magnetic ux. Now, the vortices
2
are particles that couple minimally to the gauge eld
n
2
. In general, multiple vortex species transform into each
other under the symmetry operation. Let us label them by |,
so
2
|
c
2|
. All these elds couple minimally to the
gauge eld.
Thus, we have for the dual surface theory
L
surf

|,
j
1
|n
2
c
2|
j
2
V
2|

1
2k
]
2
2:
. . . ,
(10)
where ]
2:

n
2:

:
n
2
.
As argued above, one route to obtaining topological
surface states is if the surface vortices transform under a
projective representation of the remaining symmetry.
Vortices that transform projectively under a global sym-
metry are actually quite familiar. They describe the generic
situation of 2D bosons on a lattice, say, at some commen-
surate lling [30,40]. These projective vortices play a
crucial role in the theory of deconned quantum criticality.
We need consider only two component vortex elds
2
c
2
, c
2
for the cases considered in this paper.
The gauge-invariant combination
y
2
a
2
c
2
c
2
e
|
2
l
y
2
then denes another bosonic eld. Note that
Eq. (10) closely resembles the action for a deconned
quantum critical point (noncompact CP
1
theory with
easy-plane anisotropy) [20,41]. We demonstrate how
Eq. (10) emerges as the theory for the surface states, and
011016-7
also, in the next section, describe a 3D bulk theory that
leads to this edge theory.
Surface states of a bosonic topological insulator:
Symmetry U1 2 Z
T
2
The topological insulator has these symmetries: a
conserved U1 charge and Z
T
2
time-reversal symmetry.
The semidirect product appears so that the charge-insertion
operator e
|
is invariant under time reversal, which
involves both ! and | ! |. Here, we construct
the surface theory of a 3D topological phase that has these
symmetries.
Let us begin with an enlarged symmetry, two species of
bosons that are separately conserved. Consider a conden-
sate of one species l
1
. Vortices in this condensate are
created by the eld
2
. We need to specify the projective
representation for the vortices
2
and the transformation of
the bosons l
y
1
. The remaining symmetry group U1 2 Z
T
2
has a single projective representation (P1), which acts as
follows: Under a U1 rotation by angle e, the elds c
2
!
e
|e,2
c
2
and, under time-reversal Z
T
2
, the elds c
2
!
c
2
and c
2
! c
2
. Or, more compactly,
2
! e
|e,2a
z
2
: U1,
2
! a
x
2
: Z
T
2
. (11)
Here, the a are the Pauli matrices in the standard represen-
tation. Thus, thevortices carry charge 1,2 of bosons of the
other species. The time-reversal symmetry that inter-
changes the two vortex elds ensures that the vortex charge
is xed exactly at half. It is impossible to screen this
charge with regular integer charged bosons.
These transformation laws, of course, determine how the
boson operator l
y
2
e
|
2
and how their density n
2
trans-
forms. We also need to specify how the bosons l
y
1
e
|
1
and how density n
1
transforms. The symmetry transforma-
tions are
1,2
!
1,2
efor U1,
1,2
!
1,2
for Z
T
2
. (12)
The conjugate boson numbers therefore transform as
n
1,2
! n
1,2
for U1, n
1,2
! n
1,2
for Z
T
2
. (13)
A necessary compatibility check is that Eq. (11) is
invariant under the symmetry operation, which can be
veried for these transformations. For example, time-
reversal symmetry is implemented via Eq. (10) on the
vortex elds, which is compatible with n
2
(and hence n
2
)
remaining invariant while | ! | under time reversal.
Moreover, since the bosons carry charge, the monopole
insertion operators are forbidden.
Condensing single vortices then breaks symmetry, as
described below. This situation is best analyzed by assum-
ing separate number conservation of each species of boson,
n
1
, n
2
, in which case they can be coupled to an external
gauge potential A
1
, A
2
. Then the effective Lagrangian
equation (10) reads
L
1

|n
2
|
sA
2
2
_
c
2s
2
. . .
1
2k
1
e
:
:
n
2
1
2r
A
1
e
:
:
n
2
. (14)
The c
2
are vortex elds of l
1
, which carry half charge of
boson species 2. The ux of the gauge eld n
1
is precisely
the conserved density of species 1, hence, the last term in
the above action where the external probe gauge eld A
1
couples to this current. This action needs to be modied by
including all symmetry-allowed perturbations. We do this
and analyze the possible phases in the rest of this section,
but, as a preview, consider the effect of breaking time-
reversal symmetry by condensing just one species of vor-
tex, say, c
2
. Anticipating a single charge, consider
an external eld that couples equally to the two charge
densities A
1
A
2
A. The vortex condensate forces
n
2

1
2
A, which, when substituted into the action above,
yields the electromagnetic response
1
4r
A
e
:
:
A
,
which yields a
xy
1 on the surface, indicating a magneto-
electric response 0 2r, as advertised. We now change
track and obtain the same surface theory from a very
different point of view.
B. Construction of the network model
The general considerations of Sec. III show that a time-
reversal-symmetric boson insulator with electromagnetic
response characterized by 0 2r is in a topological-
insulator phase. This key result relies on the observation
that, if T is broken at the surface to gap it out, then such
a state has a quantized electrical Hall conductivity a
xy

1, and a thermal Hall conductivity k
xy
0. What if T
reversal is not explicitly broken at the surface? The surface
can then potentially be gapless. What then is the nature of
the resultant theory? To construct this theory, it is extremely
instructive to learn from the example of the free-fermion T-
reversal-symmetric topological insulator. In that case, if T
is explicitly broken to gap out the surface, then we obtain
a
xy

1
2
. When T is unbroken, it is possible to get a
single massless Dirac cone, which is exactly the low-energy
theory of the transition between two integer quantum Hall
plateaus of fermions in J 2. Generically, we can tune the
chemical potential to move away fromthe Dirac point to get
a Fermi surface that encloses the Dirac point.
Note that the quantized Hall conductance jumps by 1
across the integer quantum Hall plateau transition. When
applied to the surface of the 3D fermionic topological
insulator, the transition connects the two possible
T-breaking surface states that go into each other under
time reversal. Thus, each such surface state must be as-
signed a
xy

1
2
and the critical theory itself is time-
reversal invariant. In contrast, when applied to strictly 2D
systems, the network model describes a transition between
plateaus with a
xy
0 and a
xy
1.
011016-8
The familiar free-fermion example gives us a crucial
clue for constructing the theory of the T-reversal symmet-
ric surface state of the boson topological insulator. First,
construct the low-energy theory of the J 2 integer quan-
tum Hall state of bosons as a potential candidate for the
gapless surface state of the 3D topological insulator. Then
add perturbations allowed by symmetry to obtain the ge-
neric surface theory. Across the boson integer quantum
Hall transition, a
xy
jumps by 2. As for the fermionic
example, when this transition is realized at the surface of
the bosonic topological insulator, the two plateau states on
ether side of the transition are related by T reversal and
must be assigned a
xy
1, consistent with our earlier
arguments.
With this motivation, we now study the IQHE plateau
transition of bosons in J 2.
IQHE quantum phase transition of bosons in J 2.
We now study the phase transition using a network
model construction. We start with the theory of the edge
state and couple together opposite edges. Let us warm up
with the familiar example of the IQHE transition of fermi-
ons from a state with a
xy
1 to one with a
xy
0. The
model, dened by Fig. 2, is described by the Euclidean
action S
_
JxJiL,
L
|
" c
|
r
|s
|
x
c
|

i
|
" c
|1
c
|
" c
|
c
|1
, (15)
with
i
| even
i
e
, (16)
i
| odd
i
o
, (17)
s
|
1
|
. (18)
The rst term is the sum of the actions of a single chiral
edge mode of the a
xy
1 fermion IQHE state taken to
propagate in opposite directions for adjacent |. The second
term describes electron hopping between opposite-moving
edge channels.
When i
e
<i
o
, all chiral edge channels are paired with
partners and a
xy
0. Conversely, if i
e
>i
o
, then all edge
channels get paired except the two end ones and we get the
fermion IQHE state with a
xy
1. The transition occurs at
i
e
i
o
and is readily seen to yield a single massless Dirac
fermion if the continuum limit is taken in the | direction.
Let us now repeat this construction for the bosonic
IQHE transition. The edge theory for the boson IQHE state
with a
xy
2, k
xy
0 has a pair of counterpropagating
edge modes: One carries the charge and the other is neutral
[1315]. It is convenient to write the effective action of the
edge as an SU2
1
Wess-Zumino-Witten (WZW) theory:
S
eff

_
JxJr
1
2
tr
g
y
g |S
WZW
g. (19)
Here, g is a 2 2 matrix with entries
g
l
1
l
2
l
2
l
1
_ _
.
The l
1
, l
2
are the two physical boson elds that form the
IQHE quantum Hall state.
A network model capable of describing the boson IQHE
state may now be written; it is dened by Fig. 3. Again, we
have an array of opposite edge channels which are coupled
together by boson hopping
o1,2
l
y
|o
l
|1,o
H.c. /
trg
y
|
g
|1
H.c. The full effective action is then
FIG. 2. Network model for the fermion IQHE transition. On
the top, all chiral edge modes are paired to yield an ordinary
insulator. On the bottom, there is an unpaired edge mode to yield
an integer quantum Hall insulator.
FIG. 3. Network model for the boson IQHE transition. Each
edge channel now has a charged chiral mode and a counter-
propagating neutral mode. The rest is the same as for fermions in
Fig. 2.
011016-9
S S
0
S
W
S
i
, (20)
S
0

_
JxJr
1
2
|
tr
g
y
|

g
|
, (21)
S
W
|
|
s
|
S
WZW
g
|
, (22)
S
i

|
i
|
trg
y
|
g
|1
H.c., (23)
with s
|
and i
|
as before. If i
o
) i
e
, we obtain the trivial
insulator, while, if i
e
) i
o
, we get the boson IQHE state.
The transition occurs at i
e
i
o
. Alow-energy theory of the
transition is obtained by taking the continuum limit in the |
direction. As the opposite-moving edge channels have
opposite WZW terms, they nearly cancel, and it is neces-
sary to carefully sum them. Fortunately, precisely this sum
was performed in Ref. [33] where the same model arose in
a different context. The result is the effective D 2 1
dimensional action
S
eff

_
J
3
x
1
2k
tr
g
y
g |rL
0
g. (24)
The second term is a 0 term for the SU2-matrix-valued
eld g in 21 dimensions corresponding to
3
SU2Z.
In the present context, our calculation has yielded this term
at 0 r.
We do not, of course, have full SU2 symmetry rotating
between l
1
and l
2
in the microscopic system. For the time
being, let us assume that we have U1 U1 symmetry
corresponding to separate conservation of the l
1
, l
2
bo-
sons. Furthermore, let us also assume that there is a Z
2
symmetry interchanging l
1
and l
2
. (Later, we relax all
these assumptions.) Here, the results of Ref. [33] show that
the eld theory described above at 0 r maps onto the
self-dual easy-plane noncompact CP
1
(NCCP
1
) model.
Equivalently, it also maps onto a model of two species of
spacetime loops with a phase r associated with each link-
ing of the two loop species.
The 0 r SU2-matrix eld theory [with the
U1U1 anisotropy] or the equivalent easy-plane
NCCP
1
model arises also in the theory of deconned quan-
tumcriticality in two space dimensions. Remarkably, we see
that the eld theories describing the boson IQHE plateau
transition (and hence the surface states of the 3D boson
topological insulator) are closely related to the theory of
deconned quantum criticality. In the previous subsection,
we obtainedthis connection froma different point of view. In
the rest of the paper, we explore this connection in more
detail and generality. For now, we merely point out that the
results of Ref. [33] (see also Ref. [42]) show that the 0 r
SU2 matrix eld theory in two space dimensions does not
have a trivial gapped disordered phase. Its phases either
break symmetry, are gapless, or have topological order.
This phase structure is a hallmark of the surface state of a
symmetry-protected topological phaseThere is no trivial
gapped phase that preserves all the symmetries.
Surface of the bosonic TIs: Field theories.We now
exploit our intuition about deconned criticality to obtain
the theory of possible surface states of the 3D boson TI,
starting with the 0 r SU2-matrix eld theory. We rst
describe a number of equivalent eld-theoretic descrip-
tions of the surface state, paying particular attention to
the realization of the physical U1 2 Z
T
2
symmetry. First,
we note that the SU2 matrix g is related to the physical
boson elds l
1,2
through
g
l
1
l
2
l
2
l
1
_ _
. (25)
Under the global U1 symmetry, both bosons transform
with charge 1, i.e.,
l
1,2
! l
1,2
e
|
. (26)
We implement time reversal by simply requiring that
l
1,2
! l
1,2
. (27)
In terms of the phases of the bosons, dened through l
1,2
$
e
|
1,2
, and the conjugate bosons densities n
1,2
, the symme-
try transformations are the same as in Eqs. (12) and (13) so
that we are indeed describing the same symmetry class in
the two approaches. We remind the reader that the total
number n
1
n
2
of the two boson species is conserved due
to the global U1 symmetry, but the relative number
n
1
n
2
is, in general, not conserved. As promised, we
rst analyze the theory in a limit where this relative num-
ber is also conserved (so that there is U1 U1 sym-
metry), and then we include interspecies tunneling terms to
recover the generic case.
As argued in Ref. [33], the 0 term of the SU2 matrix
eld theory with U1 U1 anisotropy has a simple
interpretation: It is the phase that is picked up when the
vortex of the boson l
1
is taken around the vortex of the
boson l
2
. At 0 r, the two vortices are mutual semions.
We may thus readily write a dual eld theory in terms of
the vortices
1,2
of the two bosons l
1,2
, respectively.
This dual eld theory has the structure
L L
1
L
2
L
0
L
A
, (28)
L
|
j
|o
|

|
|
j
2
, (29)
L
0

|
r
1
e
:
2
, (30)
L
A
|A
|
]
|
. (31)
Here, , :, , . . . . represent spacetime indices in 2 1
dimensions. The o
|
, | 1.2 are the usual dual-gauge elds
of the vortex theory. The physical current ]
|
of the bosons
l
1,2
is given as usual by
011016-10
]
|

1
2r
e
:
:
o
|
. (32)
We have included external probe gauge elds A
|
that
couple to these currents. The
|
are statistical gauge
elds that serve to impose the mutual statistics of the two
vortex species through the mutual Chern-Simons term in
L
0
. We have also tuned away a chemical potential that
couples to the total boson number so that the effective
action is relativistic. We relax that assumption shortly.
In passing, we note that recently related models of two
species of bosons with mutual r statistics have been
studied numerically through Monte Carlo simulations
[31]. The relevance of these models to the surface of the
3D boson topological insulator (and the related 2D boson
integer quantum Hall transition) should give further impe-
tus for such studies.
C. Synthesis of the two approaches
We now provide a synthesis of the results of the two
approaches taken in Sec. IV. We rely closely on the results
of Ref. [33] to provide two alternate eld-theoretic repre-
sentations of the theory described by Eq. (28). Rather than
repeat the derivation from Ref. [33], we provide a physical
description. The r phase picked up when the vortex
1
goes around the vortex
2
suggests that
1
carries 1,2
charge under the global U1 symmetry associated with
species 2 and vice versa [i.e., that
2
carries 1,2 charge
under the global U1 of species 1]. However the rphase is
obtained for both charge 1,2 and charge 1,2. We thus
should expect that the vortex of either species carries frac-
tional charge 1,2 of the global U1 quantum number of
the other species. This expectation is formalized by the
derivation [33]. First, by performing a duality on one
species (say, 1), we explicitly map to an easy-plane non-
compact CP
1
model with the action given by Eq. (14).
If instead we had performed a duality transformation on
species 2, we would have obtained an equivalent action in
terms of the fractionalized elds c
1
related to l
1
through
l
y
1
c
y
1
c
1
. (33)
This action takes the form
L
2

|n
1
|
sA
1
2
_
c
1s
1
2k
1
e
:
:
n
1
|
2r
A
2
e
:
:
n
1
, (34)
with the physical U1 current of the l
2
bosons given by
]
2

1
2r
e
:
:
n
1
. Note the obvious similarity between
Eq. (34) and Eq. (14) after interchange of the 1 and 2
labels. This similarity is a reection of the self-duality of
the easy-plane NCCP
1
model rst pointed out in Ref. [41].
This self-duality is obvious when both theories are
obtained starting with the sigma model or the equivalent
dual-vortex theory [Eq. (28)].
D. Analysis of the surface eld theory: Phase diagram,
deconned criticality, and fermionic vortices
Having obtained a eld-theoretic description of the sur-
face states, we now analyze the phase diagram. The sym-
metry transformations summarized in Table I enable us to
deduce the allowed perturbations to the actions above. A
crucial allowed perturbation is chemical-potential terms
that couple to the boson number
1
n
1

2
n
2
. Another
crucial allowed perturbation is an interspecies boson-
tunneling term l
y
1
l
2
H.c.. Let us rst discuss the
phase diagram when these terms are tuned to zero.
Depending on the question being asked, we nd it useful
to use one or the other of the formulations provided above.
For clarity of presentation, however, we use the theory in
Eq. (14) when it is convenient.
T-breaking states and the quantum Hall effect.
Consider condensing just one of the vortex elds
in Eq. (14):
hc
2
i 0, hc
2
i 0. (35)
Such a phase clearly breaks T-reversal symmetry, which
interchanges the vortices. However, the phase is an insulator
because the gauge-invariant combination hc
2
c
2
i
he
|
2
i 0. The transport properties of this phase are readily
obtained by noticing that the combination n
1

A
2
2
is gapped
by the Higgs condensate. Therefore, at long wavelengths,
we may set
n
2
%
A
2
2
. (36)
Furthermore, we may integrate out the eld c
2
in the
Lagrangian equation (14). The effective long-wavelength
Lagrangian for the external probe gauge elds then becomes
L
eff

|
4r
A
1
e
:
:
A
2
. (37)
Dening the charge andpseudospin probe gauge elds
A
c

A
1
A
2
2
, A
s

A
1
A
2
2
, we obtain
TABLE I. Symmetry properties of the c
1
, c
2
elds. q
1,2
are the charges under the two U1 symmetries associated with
l
1,2
, respectively. n
1,2
are the vorticities in the phase of l
1,2
.
They can also be viewed as the gauge charge for the coupling to
the corresponding U1 gauge elds. The last column gives the
transformation under time reversal.
Field q
1
q
2
n
1
n
2
Z
T
2
c
y
1
1
2
0 0 1 c
1
c
y
1

1
2
0 0 1 c
1
c
y
2
0
1
2
1 0 c
2
c
y
2
0
1
2
1 0 c
2
011016-11
L
eff

|
4r
A
c
e
:
:
A
c
A
s
e
:
:
A
s
, (38)
which implies that the charge Hall conductivity a
xy
1
while the pseudospin Hall conductivity a
s
xy
1. Taken
together, the thermal Hall conductivity k
xy
0. If, on the
other hand, we had condensed c
2
without condensing
c
1
, we would have found the time-reversed partner with
a
xy
1 and a
s
xy
1.
Consider now adding symmetry-allowed perturbations to
the action. The surface state described above is gapped and
hence is unaffected by the chemical potential terms if they
are weak. The interspecies tunneling term destroys conser-
vation of pseudospin ( n
1
n
2
) and hence a
s
xy
is no
longer quantized. However, the electrical and thermal
Hall conductivities continue to be well dened and have
quantized values a
xy
1, k
xy
0. This result is exactly
what we expect based on the general considerations of
Sec. III.
Pictorially, these edge states may be understood by con-
sidering the bulk system on a solid sphere and assuming
that c
1
is condensed on the top hemispherical surface
while c
1
is condensed on the bottom hemispherical sur-
face. Then, along the equator, there is a domain wall
between the two kinds of surface quantum Hall states. At
this domain wall, there are gapless 1D states identical to the
edge of the 2D boson IQHE state. Specically, there is one
charged chiral mode corresponding to the jump a
xy
2
across the domain wall and a counterpropagating neutral
mode that carries the pseudospin. When interspecies tun-
neling is added, the quantization of the pseudospin Hall
conductivity is not guaranteed, but the neutral edge mode is
protected so long as charge is still conserved.
Superuid state. Let us now consider T-reversal-
symmetric phases. A simple option is
hc
2s
i c
0
, (39)
independent of s. This state has hl
2
i 0. The gauge eld
n
2
is gapped by the Higgs c
2
condensate. Consider for a
moment the situation in which the boson number is inde-
pendently conserved for each species. Then this state
breaks the global U1 symmetry associated with l
2
but
preserves the other global U1 associated with l
1
. We
refer to it as SF
1
. If, on the other hand, both c
2
are
gapped, then they may be integrated out to leave a
Maxwell action for n
1
. Integrating out n
1
then gives a
Higgs mass for A
1
so that the global symmetry associated
with l
1
is now broken. We call this SF
2
. These two phases
are separated by a phase transition that is described by the
putative critical point of the easy-plane NCCP
1
eld the-
ory. In general, a chemical potential term can also be added
that tunes the system away from the NCCP
1
critical point.
Apart from SF
1
and SF
2
, we have the possibility of a phase
with the coexistence of the two superuid orders.
Inclusion of interspecies tunneling has a more dramatic
effect. First, there is now no real distinction between the
SF
1
and SF
2
phases so that the phase boundary between
them disappears. Second, and more important, as the rela-
tive phase of l
y
1
l
2
can no longer wind, vortices in l
1
are
bound to vortices in l
2
. Note that c
1
are vortices in l
2
and c
2
are vortices in l
1
. When we bind vortices in l
1
to
vortices in l
2
, the resulting vortices are created by elds,
V
y
ss
0
c
y
1s
c
y
2s
0
, (40)
with s, s
0
. Note that, for s s
0
, V
ss
0 carries charge 0
under the single remaining global U1 while, for s s
0
, it
carries charge 1. Thus, the vortices no longer carry frac-
tional charge. V
, V
can be obtained as a composite of

the boson creation operator and the vortex V
, V
so
that only the latter are elementary. Furthermore, V
can mix with V
due to the interspecies tunneling term. It

follows that there is a unique elementary vortex V $V
that carries charge 0. Also, under time reversal,

V ! V
y
. (41)
Thus, in the presence of interspecies mixing, there is a
unique vortex that does not transform projectively under
the global symmetries. Does this fact invalidate our earlier
analysis? In particular, can we now get a trivial insulator by
condensing this vortex? The answer is No. The point
that we demonstrate below is that the effective action for
the vortex V in the superuid phase is not the usual one;
rather, it contains an extra Chern-Simons term. The pres-
ence of this Chern-Simons term has a convenient rough
interpretation: It changes the statistics of the vortex to a
fermion. A simple way to picture it is to describe it in terms
of the vortex elds
1,2
in Eq. (28). The 0 term in the
sigma-model description means that the two vortices are
mutual semions, and it follows that their bound state is a
fermion.
To add some meat to this bare-bones picture, we start
with Eq. (28). In the presence of interspecies tunneling, the
individual vortex elds
1
,
2
are conned, but a bound
combination
c

1
2
survives. It is therefore nec-
essary to reformulate the action in terms of
c
. We nd it
convenient to do so rst even in the presence of the
enlarged U1 U1 symmetry and later to include the
interspecies tunneling. Now we introduce another eld
s

1
2
. The resulting Lagrangian takes the form
L L
c
, o
L
s
, o
|
4r
e
:

|
4r
e
:
.
(42)
Here, o
o
1
o
2
,

1

2
. o
are the usual

dual-gauge elds whose curl gives the charge and pseudo-
spin current, respectively. The most interesting terms are
the coupling to the gauge elds
, which have
011016-12
selfChern-Simons interactions. Including interspecies
tunneling leads to linear connement of
s
. The effective
dual Landau-Ginzburg theory of the superuid then has the
usual form but with the additional Chern-Simons term, as
promised.
Surface topological order.A wide variety of other
phases is possible depending on the details of the surface
interactions. For instance, a gapped topologically ordered
Z
2
liquid is possible, accessed within the present formula-
tion by condensing the paired vortex (c
2
c
2
H.c.)
without condensing any other elds. In this situation, the
full 3D system when placed on a solid torus has a ground-
state degeneracy of 4 coming from the surface topological
order. It is interesting to consider a bit further the properties
of this state and its relationship with the superuid state. In
the Z
2
topologically ordered insulator, there is an unpaired
vortex c
2
$ c
2
that survives as a gapped excitation and
that carries physical boson charge 1,2 of the U1
2
global
symmetry associated with the boson l
2
. We refer to it as a 2
chargon. Following standard reasoning, this phase may
equivalently also be understood as a paired condensate of
c
1
c
1
H.c.. Thus, there is another gapped excitation
corresponding to the eld c
1
$ c
1
which, in the
present context also carries charge 1,2 of the physical
boson l
1
. We refer to it as the 1 chargon. These two
chargons are mutual semions, as expected for Z
2
topologi-
cal order. Note that they have bosonic self-statistics. In the
presence of interspecies tunneling, a pair of 1 chargons can
mix with a pair of 2 chargons. Both species of chargons
continue to exist as independent excitations, but now they
carry charge-1,2 of the remaining global U1. Finally, the
bound state of these two kinds of chargons is a fermion that
does not carry fractional charge. We can take it to be charge
neutral. It is convenient to regard this neutral fermion as
the vison, and the two kinds of bosonic chargons as the
other two nontrivial quasiparticles expected for a Z
2
topo-
logical ordered state. These transformation laws are sum-
marized in Table II.
Now let us consider the relationship of this topologically
ordered state to the superuid state discussed above, to
clarify the nature of the vortices of the superuid state.
Coming from the superuid side, the Z
2
topological state is
obtained by condensing paired vortices. In the presence of
interspecies tunneling, we have argued above that there is a
unique vortex V. The unpaired vortex survives as a nite-
energy vison in this vortex-pair condensate. That this vison
is a fermion ties in nicely with the observation that the
superuid vortex V is conveniently regarded as a fermion.
Thus, as the transition to this Z
2
insulator is approached,
the vortex statistics becomes well dened and fermionic.
The topologically ordered phase provides a particularly
simple perspective on why a trivial gapped paramagnet is
not allowed. Generally, to go from a topologically ordered
insulator to a trivial insulator, we must conne the topo-
logical quasiparticles. For a Z
2
gauge theory, this is done
by condensing one of the three nontrivial kinds of quasi-
particles (usually dubbed the electric, the magnetic, and
their composite). For the Z
2
topological state that can
appear at the surface of the SPT phase we are discussing,
the electric and magnetic particles are both (half)-charged
under the global U1 symmetry, and their condensation
breaks this symmetry. On the other hand, the neutral topo-
logical quasiparticle is a fermion and hence it cannot
condense. At the same time, time reversal prevents one
from altering the Chern number associated with this
gapped fermion. Thus, we see clearly that a trivial gapped
state obtained by connement from the Z
2
topological state
is not possible at the surface.
An interesting property of this Z
2
topological ordered
state is that it realizes symmetry differently fromstrictly 2D
systems. Such a strictly 2D gapped Abelian insulator may
be described within the usual K matrix formulation. For Z
2
topological order, K 2a
x
. If both bosonic chargons carry
charge 1,2 as we argued, then the charge vector r 1, 1.
It is then easy to see that the resulting topological phase has
nonzero electrical Hall conductivity and therefore must
break time-reversal invariance. However, when realized at
the surface of the 3D insulator, a time-reversal-symmetric
Z
2
topological phase where both bosonic quasiparticles
carry charge-1,2 is allowed. In Appendix D, we collect
the properties of the Z
2
topological ordered state for the
various SPT phases discussed here and show using the
results of Ref. [43] that they all realize symmetry differ-
ently from what is allowed in strictly 2D systems.
While such interesting topologically ordered (or other
even more exotic) states are allowed, they are not required:
The surface could be in a superuid or T-broken insulating
quantum Hall state with no ground-state degeneracy. The
most important conclusion, however, is that a trivial
gapped insulating state that preserves all the symmetries
and has no topological order is not possible on the surface.
This realization is a key property of a symmetry-protected
topological phase and is satised by our example.
Some remarks are now in order. We have constructed
one topological phase, which, when the surface spontane-
ously breaks time-reversal symmetry, leads to a quantized
magnetoelectric effect. The cohomology classication of
Chen et al. [5] gives Z
2
2
so that there are three nontrivial
TABLE II. Symmetry properties of the topological excitations
of a Z
2
gauge theory, realized on the surface of a 3D SPT phase
with charge conservation and time-reversal symmetry
[U1 2 Z
T
2
], as for the topological insulator. The bosonic quasi-
particles carry half charge but transform linearly under time
reversal. Further details are in Appendix D.
Excitation Charge T
2
Boson 1 (e) q 1,2 1
Boson 2 (m) q 1,2 1
Fermion (]) q 0 1
011016-13
states. Of the these three, one must have vanishing mag-
netoelectric effect since that is an additive quantity. We
expect that this phase is simply the SPT phase associated
with just the Z
T
2
symmetry discussed in Sec. VI A. The
remaining phase is then obtained by combining the
other two phases. In the particular example discussed in
this section, the fact that there is a single U1 charge-
conservation symmetry introduces additional terms that
imply a deformation of the deconned-criticality action.
Despite this deformation, the degrees of freedom of this
action provide the useful elds in terms of which the
effective theory of the surface state may be described.
We discuss other examples with different symmetries in
subsequent sections.
V. 3D TOPOLOGICAL FIELD THEORIES
To write a 3D theory of particles and their vortices,
we choose to represent particle currents by ]
e
:a
:
B
a
,2r. The vortex lines, being loops in 3D
space, sweep out a surface in spacetime dened by the
two form
:
. Relating this two form to a vector potential
o, whose curl is the location of the vortex loop, we dene
:
e
:a
o
a
,2r. The quantization of boson particle
number to integers implies that the charges that couple to
this vector potential o are quantized. (Equivalently, the
dual vector potential is compact, i.e., dened only modulo
2r.) Clearly, gauge transformations o
! o
and
B
:
! B
:

n
:

:
n
do not change physical

variables. Taking a particle around a vortex leads to a phase
of 2r, which is captured by the minimal coupling L
o
, which may be rewritten as

L
1
2r
o
e
:a
:
B
a
. (43)
The right-hand side is often written as eBo,2r and
called the BF action. The unit coefcient in the action
ensures that there is no topological order, i.e., that it is a
unique ground state in the absence of surfaces [44]
appropriate to the current discussion. Then, the theory
above only states the obvious: that particles and their
vortices have a mutual phase factor of 2r.
Let us briey describe how this term arises from a
microscopic theory of bosons. A lattice-regularized theory
of bosons can be captured by a loop model of integer-
valued closed loops with Euclidean Lagrangian L
l

1
2
]
. The integer constraint is implemented by sum-

ming over the auxiliary vector eld o
that is an integer
multiple of 2r. Now, the current ]
takes real (rather than

integer) values, and its divergence-free condition can be
implemented by writing ]
e
:a
:
B
a
,2r, where the
two-form B is also a real eld. This gives
L
E
l

1
8r
2
eB
2
|
2r
eoB, (44)
where the second term is the desired statistical interaction.
(The factor of | appears because of the Euclidean formu-
lation.) However, at this point, o is an integer (times 2r)
eld. One can softly introduce this constraint by assuming
o to be real but adding the cosine term L
cos
, where we have utilized the fact that

the longitudinal component can always be added to the
gauge eld [39]. The phase is actually the phase of the
original bosons, and, when the bulk is insulating, the cosine
is irrelevant since the bosons are gapped. Therefore, in the
insulating phase, we may use Eq. (44) in which both B, o
elds are taken as real, with the caveat that charges are
ultimately quantized. Further discussion of basic issues
related to this theory is at the end of Appendix A.
As shown in Appendix A, surface states dened based
on the BF-theory Eq. (44) are usual 2D bosonic modes that
are not topologically protected. To encode a SRE topologi-
cal phase, an additional term must be added, as shown
below.
Based on the discussion on surface states, where a
pair of bosonic elds were invoked, we consider two
species of bosons to write a topological term. This step
also follows the two-component U1 U1-symmetric
Chern-Simons approach for the 2D systems, which was
found to be successful in describing 2D SRE topological
phases [13]. Therefore, we introduce two B elds that
represent their conserved currents, and two o elds that
are vortices in these elds. Let us begin with the general
case of N species of bosons, with
L
BF

1
2r
N
I1
eB
I
o
I
, (45)
where e is the antisymmetric symbol, I labels the distinct
boson species, and indices have been suppressed.
Note that the apparently more general version is L
B

Q
I1
2r
eB
I
o
1
where Qis a unimodular, i.e., detQ 1, integer
matrix, which ensures absence of topological order.
However, this can be brought into the canonical form of
Eq. (45) by redening B
I
Q
1
KI
B
0
K
. The transforma-
tion matrix Q
1
is also an integer matrix, since detQ 1
and the minors of an integer matrix are also integers.
Now, we can add an additional topological term to the
action:
L
3D
L
BF
L
FF
, L
FF

8r
2
K
I1
eo
I
o
1
. (46)
The action must be invariant under ! 2r to allow
for addition of 2D layers at the surface. Therefore, the
action dened on a closed 3D space should be invariant
under the shift ! 2r. We show in Appendix B that
this condition xes the entries K
I1
to be integers.
A stronger condition on K can be applied as follows:
Values of that differ by 2r simply correspond to differ-
ent ways of terminating the surface. Hence, at a domain
011016-14
wall where changes by 2r at the surface, we demand
that all excitations present are bosonic, which is the same
as the requirement placed on K matrices describing 2D
SPT phases, i.e., that Det K 1 and that all diagonal
entries are even integers.
The simplest choice of K matrix with these properties is
K
0 1
1 0
_ _
. (47)
We nd that, for most of the 3D bosonic SPT phases that
we are interested in, it sufces to consider this matrix. This
choice is similar to the 2D situation [13], where the above
K matrix describes a large set of SPT phases that differ in
the way symmetry is implemented.
A. Two-component BF theory of bosonic
topological insulator
Let us specialize to the symmetries of the topological
insulator U1 2 Z
T
2
and consider a two-component theory
with the simplest allowed K matrix given by Eq. (47). Then
we can write
L
tot
L
BF
L
FF
L
em
, (48)
L
BF

1
2r
eB
1
o
1
B
2
o
2
, (49)
L
FF

4r
2
eo
1
o
2
. (50)
(We discuss coupling to the external electromagnetic eld
L
em
later in the paper.) Under time-reversal symmetry,
we have
B
I,0|
! B
I,0|
, B
I,|]
! B
I|]
, (51)
o
I,0
! o
I,0
, o
I,]
! o
I,]
, (52)
where indices |, ] refer to spatial coordinates. The trans-
formation of B elds is obtained by relating them to the
boson densities and currents, while the o elds are chosen
to transform such that the BF term is left invariant. Since
both species o
I
transform in the same way under time
reversal, we may conclude that ! under Z
T
2
. A
time-reversal-invariant bulk action can then be constructed
for 0, r (given the ambiguity in modulo 2r). Of
course, we pick r in the topological phase. While we
have not derived this action, we have written the simplest
possible topological theory that meets the general con-
straints required of SPT phases. We now proceed to show
that this topological theory produces a surface with the
same physical properties as predicted in the previous sec-
tion. We study three different situations: First, we study the
surface superuid and determine the quantum numbers of
vortices. Second, we investigate the electromagnetic re-
sponse, particularly the magnetoelectric polarizability.
Third, we analyze the case in which time reversal is broken
at the surface, in opposite ways, leading to a domain wall.
(i) Fractionally charged vortices.Consider a surface
of the topological phase at z 0 with r0 for
z >0z <0. Then the effective action at the surface
arising from L
FF
is
S
edge

1
4r
_
JiJxJye
zn
o
1n
o
2
, (53)
where indices n, , run over i, x, y, and the elds are
evaluated at z 0 [45]. At the surface, we may replace
o
I|

|
I
(see Appendix A), where | x, y. Consider
now a surface superuid of component I 2, with a vortex
at the origin x y 0. This conguration implies that
2
winds around the origin or that
x
y

y
2

2rn
2
uxuy, where we have allowed for a vortex of
strength n
2
. Substituting this for
2
in Eq. (53), we have
S
o

1
2
_
Jin
1
. (54)
Similarly, a vortex in the eld
1
couples to the phase of
2
. Given the conjugate relation between number and
phase, this implies that a vortex of strength n
2
in compo-
nent I 2 carries charge n
2
,2 of component I 1. Thus
we see that unit vortices in one bosonic eld carry a half
charge of the other eld.
A different perspective on this result is obtained by
thinking about the fate of external monopoles of the
gauge elds o
1
, o
2
. These monopoles are sources for vortex
lines of the two boson elds l
1
and l
2
, respectively. Now,
the well-known Witten effect implied by the term tells us
that a 2rmonopole in o
1
carries gauge charge

2r

1
2
which
couples to o
2
, i.e., it carries charge 1,2 of the global U1
associated with l
2
. Similarly, a 2rmonopole in o
2
carries a
charge 1,2 of the global U1 associated with l
1
. At the
surface, this monopole creates a 2r vortex of the corre-
sponding boson, which is then seen to carry a half charge of
the other boson, thereby recovering the result in the pre-
vious paragraph.
(ii) Electromagnetic response.Since we have U1
symmetry, we can couple to an external electromagnetic
eld and write the following terms:
L
em

e
2r
q
1
B
1
A
1
q
2
B
2
A
2
, (55)
where q
I
is the charge on the Ith boson, given that the
current of bosons eB
I
] couples minimally to A
I
.
If a single U1 charge is present, then we identify A
1

A
2
A. We see that this minimal coupling, along with
Eq. (46), ensures that the charged vortex ends are also
coupled to the external electromagnetic eld.
To nd the electromagnetic response, we integrate out
the Bs, and then the o. The rst step gives o
I
q
I
A
I
.
Substituting this, we obtain
011016-15
L
em-response

4r
2
q
1
q
2
eA
1
A
2
. (56)
Setting r, q
1
q
2
1,
L
em-response

1
4r
eA
1
A
2
, (57)
which indicates that ends of vortices in the condensate of
the rst species of bosons (which are induced by inserting
2r ux in A
1
) carry a half charge of the second species.
Finally, to obtain the response to the external electro-
magnetic eld, we identify A
1
A
2
A, and then L
em

2r
4r
2

E

B, which corresponds to a magnetoelectric polar-
izability of 0 2r, i.e., it is an odd multiple of 2r, as
expected.
The theory above can also be derived using the hydro-
dynamic approach of [27]. Note that we could absorb the
FF term in Eq. (48) [27] by redening B
I
!
B
I

4r
2
K
I1
eo
1
. However, this redenition leads to an
additional electromagnetic coupling:

4r
2
q
I
K
I1
eA
I
o
1
.
We still prefer to work with the original form since it is a
eld theory written in terms of internal elds, that is, it is
well dened even in the absence of a conserved charge and
external electromagnetic couplings.
(iii) Surface domain wall.Finally, we consider an in-
sulating surface in the z 0 plane, on which time reversal
is broken in opposite ways for y >0, y <0. This leads to
edge modes along the x direction, localized near z y
0. This conguration is modeled with a spatially varying
eld, where z >0 0, while z <0, y rsigny.
Introducing this prole in L
tot
, with a gapped B eld on the
surface, the edge theory is readily shown to be
S
domain-wall

1
2r
_
JiJx
x
2
(58)
A
0
x
q
2
1
q
1
2
, (59)
where the rst term denes the commutation relations of a
regular Luttinger liquid, the second term identies the
coupling to the external eld (assuming a gauge where
A
x
0), and the dots refer to nonuniversal potential terms
for the edge elds. This situation is identical to the edge
state of the integer quantum Hall effect of bosons [13,14]
on setting q
1
q
2
1, which is also constant with the
magnetoelectric polarizability of 0 r in this phase.
The general problem of deriving the above 3D eld
theory from microscopic models is left for future work.
Next, we describe a bulk nonlinear sigma model, which in
some ways may be considered a microscopic theory since it
assumes additional ingredients, over and above the purely
symmetry-group-based approach of the cohomology theory
[5]. Therefore, instead of writing sigma models with topo-
logical terms, where the target manifold is the symmetry
group, we allow the target manifold to be the four-sphere
S
4
, which assumes a particular microscopic representation.
However, since we are not concerned with classifying
phases, but rather with providing physical examples, this
additional assumption is convenient. This procedure is quite
analogous to the common practice of considering crystal-
line band structures for free-fermion topological insulators,
although they can (and, strictly speaking, should) be de-
ned in the absence of translation symmetry [2].
B. Bulk sigma model theory
The bulk eld theory discussed in the previous section is
topological and has no bulk dynamical degrees of freedom.
In this section, we describe a different bulk theory, one
with dynamical boson elds, that gives rise to the topo-
logical effective eld theory of the previous section in a
disordered phase. This theory may thus be viewed as a
eld-theory realization of a model with a bosonic SPT
phase in three space dimensions. This eld theory takes
the form of a 3 1-dimensional nonlinear sigma model
supplemented with a topological 0 term. This form
generalizes to three space dimensions the continuum
eld-theory model that realizes the 2D integer quantum
Hall state of bosons.
Following the 2D example and the discussion in pre-
vious sections of this paper, we enlarge the symmetry of
the boson system from U1 to a larger symmetry group
and then add perturbations to reduce to the symmetry of
interest. For the construction of this section, it is extremely
convenient to consider a generalization in which we rst
embed the U1 symmetry into an SO5 group. Consider
therefore a ve-component unit vector eld ^ n. Later, we
describe exactly how the physical symmetry [U1 Z
T
2
or
U1 2 Z
T
2
, etc.] are realized by the components of this
eld. For now, we write a continuum eld theory for ^ n.
On a closed spacetime manifold, say, the 4-sphere, the
Lagrangian takes the form
L L
0
^ n L
0
^ n, (60)
L
0
^ n
1
2
^ n
2
, (61)
L
0
^ n |0Q, (62)
Q
1
4
_
J
3
xJre
olcJe
n
o
x
n
l
y
n
c
z
n
J
r
n
e
(63)
3
8r
2
_
J
3
xJr det ^ n
x
^ n
y
^ n
z
^ n
r
^ n, (64)
where
4

8r
2
3
is the volume of the unit 4D sphere. Q is
the integer invariant corresponding to
4
S
4
Z and
counts the number of times that spacetime congurations
of the ^ n eld wrap around unit 4-sphere. Clearly, the theta
term does not affect the bulk physics if 0 2nr with
integer n. (We note that, throughout this section, 0 denotes
011016-16
the theta angle for the bulk sigma model; it should not be
confused with the same symbol used previously for the
electromagnetic response.) We are interested in disordered
phases of this eld theory where ^ n is gapped in the bulk.
Consider a spatial domain wall between a state where
0 2r and one where 0 0. If this domain wall has a
nontrivial surface state, then the 0 2r theory describes
an SPT phase in the bulk. Such a domain wall corresponds
to a situation where 0 varies spatially. To handle this
situation, it is convenient to elevate 0 to be a new dynami-
cal eld and to dene a sigma model for a new six-
component unit vector eld
^
, dened by
^

cosn
^ n sinn
_ _
. (65)
The eld denes a map from spacetime (taken to be the
4D sphere S
4
) to the 5D unit sphere S
5
. Consider a eld
theory for
^
that includes (apart from the usual gradient
terms) a WZW term (dened as usual as 2r times the
fraction of the volume of S
5
) that is bounded by the hyper-
surface traced out by
^
. Formally, let
^
x, u be a smooth
extension of
^
x such that
^
x, 0
^
0
,
^
x, 1
^
.
Then
S
WZW
^

2
r
2
_
x,r
_
1
0
Judet
^
x
^
y
^
z
^
r
^
u
^
.
(66)
Consider an extension where
^

cosnu
^ nsinnu
_ _
, (67)
with n0 0, n1 n, nu is independent of x, r), and
^ n is independent of u. The determinant in S
WZW
is readily
calculated and reduces to the theta term L
0
^ n for the ^ n
eld, with
0
16
3
_
n
0
Jn
0
sin
4
n
0
. (68)
In particular, when n 0, we get 0 0, and when n r,
we get 0 2r. Thus the WZW model for
^
with constant
n r describes the 3 1-dimensional nonlinear sigma
model for ^ n at 0 2r.
To study a domain wall between 0 2r and 0 0
along, say, z 0, let n nz such that
nz ! 1 0, (69)
nz ! 1 r. (70)
We further assume that
Jn
Jz
is localized to within a short
distance of z 0. To evaluate the WZW term in this
conguration, it is convenient to use a different extension
of the
^
eld. Specically, let
^

cosnz
^ n x, r, u sinnz
_ _
, (71)
with n now independent of u, x, y, r, and ^ nx, 0 ^ n
0
,
^ n
x,1
^ nx. The determinant in S
WZW
is again readily
evaluated and becomes
S
WZW

2
r
2
_
1
1
Jz
Jn
Jz
_
x,u
det ^ n
x
^ n
y
^ n
z
^ n
r
^ n
u
^ n.
(72)
As
Jn
Jz
is localized at the domain wall at z 0, we can
replace ^ n in the integral by its conguration at z 0. The z
integral can now be performed and leads to
L
WZW

3
4r
_
x,u
det ^ n
x
^ n
y
^ n
z
^ n
r
^ n
u
^ n. (73)
This is exactly the WZW term (at level 1) for the ^ n eld at
the boundary. Thus the domain wall in question is de-
scribed by a 2 1-dimensional SO5 nonlinear sigma
model with a WZW term.
This eld-theoretic result is very useful for constructing
a bulk-sigma-model description of the SPT phases dis-
cussed in this paper. The simplest application is to bosons
with symmetry U1 Z
T
2
, discussed in detail in the next
section. To illustrate this application, let us rst introduce a
U1 SO3 anisotropy and write ^ n Rec, Imc,

N,
where, under the global U1 symmetry, we choose c !
e
|e
c, but

N !

N. Under time-reversal, we let c !
c

N !

N. Finally, under the global SO3 symmetry,
N transforms as a vector while c is invariant.

The level-1 WZW term plays the following crucial role
[33] in this eld theory: It implies that the vortex of the c
eld transforms as spin-1,2 under the SO3 rotation.
Indeed, the ve-component sigma model with global
SO3 U1 and Z
T
2
symmetries implemented in this
way, and supplemented with a level-1 WZW term, pre-
cisely arises also as the theory of the deconned critical
point between Neel and VBS states in 2D. There, the
spin-1,2 attached to the vortex captures the physical pic-
ture that a VBS vortex is a spinon.
For our present purposes, we need to further explicitly
break the SO3 symmetry while preserving time reversal.
Then the

N eld is no longer a freely uctuating variable.
However, the crucial point is that, as the vortices of c form
a spinor, their Kramers degeneracy is preserved so long as
Z
T
2
is preserved even if the full SO3 is not present. This
Kramers degeneracy is exactly the dening property of the
surface theory of one of the SPT phases for bosons with
U1 Z
T
2
symmetry described in the next section. We
have thus obtained this surface theory from a bulk sigma
model.
(i) Meaning of the theta term of the bulk sigma model.
The sigma model description is useful because it suggests a
route to obtaining a physical realization of this SPT phase.
011016-17
First, let us understand the meaning of the bulk 0 termin this
sigma model. As a topological term, it depends on the global
conguration of the ^ n eld. In general, for a theory of

N, c
with SO3 U1 symmetry, if there are no topological
defects ineither the

Nor the c eld, it is easytosee that the 0
term vanishes. In 3D, the

N eld admits point hedgehog
defects, while the c eld admits vortex loops. The 0 term
implies that, during a process in which a hedgehog is taken
around a vortex line, a phase e
|0
accumulates. For 0 2r, it
follows that the hedgehog of

N has charge 1 under the global
U1 symmetry associated with c. (This charge ensures that
the hedgehog acquires phase 2r when it moves around a
vortex line.) Thus this kind of SPT phase may potentially be
engineered by constructing a physical situation in which the
hedgehog defect of a three-component order parameter is
charged under the global U1 symmetry of the bosons of
interest.
(ii) Other SPT phases.To describe the other SPT
phases discussed above with the same symmetries, we
must implement the symmetry differently. First, we con-
sider a different situation, where the SO3 vector has only
U1 Z
T
2
symmetry. Then we write ^ n Rec
1
, Imc
1
,
Rec
2
, Imc
2
, N
z
. Following the reasoning of the previous
sections, we take both c
1
and c
2
to be charged under the
global U1 symmetry. For U1 Z
T
2
, we take under time
reversal c
1,2
! c
1,2
and N
z
! N
z
. If time reversal is
preserved, then hN
z
i 0. In the presence of anisotropy
that favors the ^ n to have zero component of N
z
, we may
drop N
z
to obtain an effective eld theory for the surface.
Then the level-1 WZW term for ^ n becomes the familiar 0
term at 0 r for the remaining four-components in the
2 1-dimensional surface theory. To understand the bulk,
note that, when there is U1 Z
T
2
anisotropy on an SO3
vector, the vortex lines of the U1 eld c
2
come in two
kinds, which are distinguished by the sign of N
z
in the core.
The hedgehogs of the original

N eld are then domain walls
within the cores of these vortices where the N
z
changes sign
[46]. Thus the hedgehog must be regarded as a composite
of two kinds of monopole sources for the two kinds of
vortex lines. Formally, we may write the hedgehog creation
operator h
y
as
h
y
m
y
2
m
2
, (74)
where m
2
create the two kinds of monopole sources. Now
the charge 1 of the hedgehog implied by the bulk 0 term
implies that these monopole sources are charged. Further, as
Z
T
2
changes the sign of N
z
, it interchanges the two monopole
sources. It follows that the monopoles m
y
each carry charge

1,2 of the c
1
eld. This is precisely what is implied by
the Witten effect as applied to the two-component BF FF
topological eld theory of the previous section.
Exactly the same description can be provided for the
boson topological insulator with symmetry U1 2 Z
T
2
.
Then we consider anisotropy similar to that above with
c
1,2
charged under the global U1 symmetry, but we let
c
1,2
! c
1,2
, N
z
! N
z
under Z
T
2
. The rest of the discus-
sion is identical to the one above.
This analysis establishes the connection between the bulk
sigma model andtopological eld theorydescriptions. Apart
from giving an alternate perspective, we hope that the ideas
of this section provide insights into physical realization of
these SPT phases, a task we leave for the future.
VI. OTHER SYMMETRIES: TOPOLOGICAL
PARAMAGNETS
Now let us study various other symmetries that are par-
ticularly appropriate to quantum spin systems. By analogy
with electronic topological insulators, SPT phases in quan-
tummagnets may be christened topological paramagnets.
In particular, we highlight two cases: (i) Z
T
2
time-reversal
symmetry. This case is the simplest symmetry that produces
a topological phase, and we construct a nontrivial phase,
thus indicating a Z
2
class. In the absence of a conserved
charge, there is no quantized magnetoelectric effect. In the
next section, we construct a separate topological phase with
this symmetry but with chiral modes on a domain wall.
(ii) U1 Z
T
2
. This case corresponds physically to a time-
reversal invariant spin system in which the z component of
spin is conserved. Two nontrivial phases are constructed:
The rst has a quantized 0 2rbut the symmetry prohibits
background charge, which allows us to sharply dene sta-
tistics of vortices. The possibility of an exotic type of Bose
liquid, the vortex metal, as a generic surface state is dis-
cussed. The second nontrivial phase has 0 0. However, in
this case, we show that a deconned quantumcritical action
could emerge on tuning just a few parameters. We relegate
to Appendix C a third symmetry, which is also readily
analyzed: U1 2 Z
2
, for which we obtain Z
2
topological
phases. This symmetry is of interest since it does not involve
time-reversal symmetry.
A. Symmetry Z
T
2
We now consider the case of only time-reversal symme-
try, both by analyzing the projective representations of
surface vortices and by constructing bulk eld theories.
Surface theory.As usual, it is convenient to assume a
slightly bigger symmetry to identify the relevant physics
and then to break it down to the physical symmetry. Here it
is sufcient to enlarge the symmetry to U1 Z
T
2
so that
we may discuss vortices in a boson eld l
y
1
e
|
1
.
Let us rst discuss the transformation of the
1
eld
under time reversal. If this condition were like charge-
phase, then, under time reversal,
1
!
1
,r,
where , 0, 1. However, in such case, we would be able
to pin
1
for either value of ,by adding either a cos
1
or a
sin
1
term. So this transformation does not correspond to
an SPT phase boundary state, since the surface can be
gapped in a trivial fashion without breaking symmetry.
The other option is that
1
transforms like the XY spin,
011016-18
i.e.,
1
!
1
,r. In this case, for , 1, the term that
can be added to the Lagrangian is cos2
1
c, and the
1
eld cannot be gapped without breaking symmetry. Nowwe
need to consider the transformation of the vortices because
that can lead to a gapped state even if
1
itself cannot be
condensed. First, though, we note that the eld n
1
conjugate
to the phase
1
transforms as n
1
! n
1
under time reversal
(i.e., like S
z
spin) to preserve the commutation relations.
Let us now discuss the transformation properties of the
vortices under the remaining symmetry Z
T
2
. What are the
projective representations of the symmetry group Z
T
2
?
These are essentially the end states of a 1D topological
phase with this symmetry. It is well known that there is a Z
2
classication of such phases and that the nontrivial phase is
just the Haldane phase with gapless edge states that are
spin 1,2 objects. Therefore, the projective representation
of Z
T
2
that we need is the following transformation of two
vortex elds (just like spin 1,2) under time reversal, so
c
2
! c
2
, c
2
! c
2
or, more compactly,
2
! |a
y
2
. (75)
Note that, if the vortex elds condense, they can condense
either individually or simultaneously. In the individual
case, the gauge-invariant operator N
z
jc
2
j
2
jc
2
j
2
takes on a nonzero expectation value. This object, under
time reversal, transforms as N
z
! N
z
, so it breaks time
reversal. If both elds condense simultaneously, then the
gauge-invariant eld c
2
c
2
e
|
2
acquires an expec-
tation value. However, under time reversal, e
|
2
transforms
nontrivially as
2
!
2
r (note
1
and
2
transform
the same way) and cannot take on an expectation value
without breaking time-reversal symmetry. (Both of these
vortex elds are essentially spin operators.) A third option
is that c
2s
do not condense individually, but rather a pair
condenses, which leads to a Z
2
topologically ordered state
that does not break symmetry, making it also consistent
with a topological surface state.
The effective theory for bosons at the edge consistent
with this symmetry is
L
e
s
j
|n
2
c
2s
j
2
1
2k
]
2
2:
m
V
2m
c.c.,
(76)
where V
2m
e
|2m
1
is the 2mmonopole insertion operator,
which is allowed once we break the U1 symmetry to leave
just the time-reversal invariance allowing even numbers of
monopoles. Note that the background magnetic eld
x
n
2y
y
n
2x
2rn
1
is odd under time-reversal sym-
metry and is not allowed. Also note, however, that the boson
mixing terms e
|
1
are allowed by symmetry. Other

symmetry-allowed terms are the same as with U1 Z
T
2
symmetry, which is discussed following Eq. (80).
Surface Z
2
topological order and symmetry.Just as we
did in Sec. IVD, we nd it extremely instructive to con-
sider the question of why a trivial paramagnetic state is not
allowed at the surface from the point of view of the Z
2
topological surface state. Here, of the three nontrivial
topological quasiparticles, the two bosonic ones each
transform as Kramers doublets under Z
T
2
, i.e., T
2
1.
They are simply the unpaired vortex of either l
1
or l
2
.
They have mutual semionic statistics. Their bound state is a
Z
T
2
singlet (T
2
1), but it is a fermion. To destroy the
topological order, we must condense one of these non-
trivial quasiparticles. However, when either of the two
bosonic excitations condense, Z
T
2
is spontaneously broken.
The fermion cannot condense, and time-reversal symmetry
prohibits a nontrivial Chern number for fermions. Thus
there is no possibility of a trivial paramagnet.
The cohomology classication [5] also produces one
nontrivial SPT phase with this symmetry. Our analysis
gives a direct understanding of the allowed surface struc-
ture of this phase. However, later we point out another
distinct nontrivial SPT phase with Z
T
2
symmetry that ap-
pears to be beyond the cohomology classication.
3D bulk Lagrangian.Consider the bulk Lagrangian
L
BF

e
2r
B
1
o
1
B
2
o
2

e
4r
2
o
1
o
2
. (77)
The rst term is invariant under time reversal if we assume
B
0|
I
! B
0|
I
and o
|
I
! o
|
I
while B
|]
I
! B
|]
I
and o
0
I
! o
0
I
under time reversal. The transformation law for B is ob-
tained by assuming that it is connected to a conserved
current that transforms like spin current. Thus, unlike
Chern-Simons in 2D, the bulk action is naturally invariant
under T. However, the second term changes sign if both o
I
transform in the same way. This observation xes to one
of the two values 0 and r, yielding at least two phases.
Since there is no conserved charge, there is no coupling to
an external eld. However, the fractionalized degrees
of freedom at the ends of vortices are captured in this
formalism.
It is relevant to note that phenomena previously used to
distinguish topological phases do not apply for this sym-
metry. For example, the absence of a conserved charge
does not allow us to dene the magnetoelectric polariz-
ability. Also, it turns out that the domain wall between
opposite surfaces that break time-reversal symmetry does
not carry gapless modes. Recall that, in free-fermion 3D
topological insulators and class DIII topological supercon-
ductors, there are chiral edge modes on T-breaking domain
walls. Also, for bosonic topological insulators, there is a
nonchiral but protected domain wall mode. Here, however,
the domain wall is nonchiral [as can be seen from the
K a
x
matrix that enters in the second term of Eq. (77)],
and, in the absence of a conserved charge, the oppositely
propagating modes can acquire a gap. Nevertheless, the
surface states are still special, and they either are gapless,
break symmetry, or develop topological order.
011016-19
B. Symmetry U1 Z
T
2
Here the U1 can be interpreted as spin-rotation sym-
metry about z axis. However, we often continue referring to
the conserved quantity as charge. We construct two differ-
ent topological phases (and the composition of these
phases denes a third nontrivial phase), each of which is
interesting for different reasons.
Surface theory of phase 1 and the deconned criticality
action.Here we do not enlarge the symmetry. Let bosons
l
y
1
e
|
1
be charged under the U1 symmetry so that
1
!
1
e: U1, n
1
! n
1
: U1,
1
!
1
r: Z
T
2
, n
1
! n
1
: Z
T
2
.
(78)
Next, we consider the vortices
2
of the eld e
|
1
and
specify their transform under the remaining time-reversal
symmetry, which has a single projective representation:
2
! |a
y
2
: Z
T
2
. (79)
Now, since c
2
c
2
$e
|
2
, we have
2
!
2
r under
time reversal. The effective eld theory is written as
L
e
a
jf
|n
2
c
2a
j
2
1
2k
]
2
2:
c
2
c
2
2
H.c. Vj
2
j
2
g. (80)
The second to the last term is cos2
2
, which preserves
time-reversal symmetry. The ux
x
n
2y

y
n
2x
,2r
n
1
vanishes on average since the density n
1
changes sign
under time reversal. No monopole insertion operators are
allowed since changing the ux corresponds to inserting
conserved U1 charge. We note that this action is very
similar to the easy-plane noncompact CP
1
(NCCP
1
)
action, proposed [20] as the critical theory between a
spin 1,2 easy-plane Neel antiferromagnet and valence-
bond-solid (VBS) order. The ux here is just the spin
density, while the vortex bilinear e
|
2
correspond to the
VBS order. In contrast to the square lattice with fourfold
rotation symmetry, here the square of the VBS order
parameter is allowed, as on a rectangular lattice. An
important distinction from previously discussed decon-
ned criticality is that here translation symmetry is not
invoked.
Symmetry actually permits other terms in this action, for
example, the linear derivative terms (
i
n
2|

|
n
20
),
which correspond to electric elds (spin currents) in the
ground state. Similarly, nite gauge charge is also allowed
in the ground state, corresponding to nite vortex density,
since vortices here do not break time-reversal symmetry.
This nite vortex density introduces linear time-derivative
terms in the action above. However, if we expand the
symmetry to include a Z
2
that reverses the orientation of
the U1 rotation axis, i.e., that sends n
1
! n
1
(which is a
rotation by r around the S
x
axis in spin notation), then the
additional terms discussed here are prohibited, since both
electric eld and gauge charge are odd under this Z
2
.
Thus, for the topological surface state with symmetry
U1 2 Z
2
Z
T
2
, the eld theory is given by Eq. (80).
Parenthetically, we note that precisely this internal sym-
metry was also assumed in the original discussion of
deconned criticality in 2D quantum magnets with easy-
plane anisotropy [20]. Generically, either the bosons or the
vortices are condensed, which implies that either U1
symmetry or time-reversal symmetry is broken. However,
if the critical point separating these states is stable to
uctuations, then one could tune a single parameter and
access a deconned critical point on the surface. It is at
present unclear if this holds true for the theory in Eq. (80),
which is an easy-plane NCCP
1
with a twofold anisotropy
term. There is mounting evidence that the SU2 symmet-
ric NCCP
1
model supports a stable quantum critical point.
While initial studies were divided between continuous
[41,47] and weak rst order [48,49], recent studies of
quantum models seem to favor continuous transition [50
52]. However, the situation is less clear with easy-plane
anisotropy [41,53] and the anisotropy term above. The
connection to SPT surface states should provide additional
motivation for further study.
Thus far, we have assumed translation invariance on the
surface, but in fact only internal symmetries are required to
dene the phase. The presence of surface randomness that
respects internal symmetries provides random variations in
the local critical coupling. This random energy density
term is known to be typically relevant at a quantum critical
point [54], since it requires a rather stringent condition to
be met, : >1 for irrelevance in a clean critical point in
J 2. Here we emphasize a crucial difference from the
realization of the deconned critical theory in 2D quantum
magnets. In that case, the presence of disorder leads to a
random eld that couples linearly to the VBS order pa-
rameter. In the spinon representation, this is a random
monopole insertion term. Alternatively, in the dual vortex
representation, this is a random term that couples to
c
2
c
2
. This coupling is expected to be relevant at the
clean deconned critical point and might potentially lead
to connement at the resulting disordered xed point. Thus
it is not clear if the NCCP
1
description is a useful one in the
presence of disorder. In the present problem, however, a
linear coupling to c
2
c
2
remains forbidden even in the
presence of disorder. The random energy terms, though
relevant, are still not expected to lead to connement by
themselves. More dangerous potentially are random varia-
tions in the coupling . The fate of the disordered NCCP
1
model in the presence of this particular kind of randomness
remains to be investigated. In this context, it may be
relevant to note that even the fate of the 3D fermionic-
topological-insulator surface states in the presence of dis-
order and interactions is also not a settled issue. It is
currently unclear if one of the symmetries is spontaneously
broken in the low-energy limit. If the symmetries are
011016-20
preserved, a critical metal with universal conductance is
predicted [55].
As in the other examples, a surface state with Z
2
topo-
logical order that preserves all symmetries is allowed and
is readily accessed by condensing paired vortices. For both
phase 1 and phase 2 discussed in this subsection with
symmetry U1 Z
T
2
, the symmetry properties of the cor-
responding surface topological order is summarized in
Appendix D.
Surface theory of phase 2 and vortex spin metal.A
different topological phase is accessed by enlarging the
symmetry momentarily to U1 U1 Z
T
2
. Now we
can assume that a boson of species 1 l
y
1
e
|
1
be charged
under the rst U1 symmetry and that it transforms exactly
as in Eq. (78). Vortices in this boson eld transform under
the remaining U1 and time-reversal symmetry as
2
! e
|e
0
a
z
,2
2
: U1, (81)
2
! |a
y
2
: Z
T
2
. (82)
Now the effective theory at the surface is
L
e

a
j
|n
2
c
2a
j
2
1
2k
]
2
2:
Vj
2
j
2
,
which has neither monopole insertion nor anisotropy terms
due to the presence of conservation of separate boson
species. There is no background ux due to time-reversal
symmetry. However, by breaking down the symmetry to
the single U1 symmetry, one is allowed the following
term cos
1

2
, since both bosonic elds transform the
same way under symmetry. Now V
1
e
|
1
corresponds to
a monopole insertion operator. Thus this is a composite
operator, which, in the variables above, may be written as
c
2
c
2
V
1
H.c.. It breaks down the U1 U1 sym-
metry to a single U1 and leads to a binding of their
vortices. If only the time-reversal symmetry is broken,
then a quantized Hall effect results from the conserved
spin. The discussion closely parallels that in Sec. .
However, if U1 symmetry is broken, the surface is an
XY-ordered state of the spin system, and the vortices can be
shown to be fermionic as in Sec. IVD. An advantage in the
topological paramagnet compared to the bosonic topologi-
cal insulator surface is the absence of background spin
density that implies that the fermionic vortices move in
zero background eld. This constraint allows for a sharp
denition of their statistics in terms of the Berry phase under
exchange. Moreover, since the vortex density does not break
time-reversal symmetry, generically, a nite vortex density
is present in the ground state. In the XY ordered state, these
vortices form a vortex solid, and their statistics is not very
important. It is, however, extremely interesting to ask about
the result of destroying the XY order by melting the vortex
solid and proliferating the vortices. With Fermi statistics, the
vortices form a Fermi surface that is coupled to the non-
compact U1 gauge eld. The resulting state is a vortex
spin metala compressible metallic phase of spins with
many interesting properties. It is a gapless spin liquid with a
vortex Fermi surface and is distinct from the more familiar
2D quantum spin liquids with a spinon Fermi surface.
Ref. [56] proposed a very analogous vortex metal phase as
an exotic possibility for a magnetic-eld-driven quantum
vortex liquid state in two space dimensions. There the
magnetic eld explicitly breaks time-reversal invariance.
In contrast, the vortex spin metal obtained at the surface of
the 3D topological paramagnet is a phase that preserves the
dening U1 Z
T
2
symmetry. As with the other examples
discussed in this paper, such a time-reversal-invariant vortex
spin metal is presumably forbidden in strict 2Dspin systems.
3D bulk theory.For both phase 1 and phase 2, the 3D
topological theories are identical; they differ only in the
coupling of the conserved charge to the external eld:
L
tot
L
topo
L
em
, (83)
L
topo

1
2r
eB
I
o
I

e
4r
2
o
1
o
2
. (84)
Under the Z
T
2
symmetry, o
1|
! o
1
, and o
2|
! o
2|
while
their 0 components change sign. Thus the axion eld
must be odd under Z
2
, so the action as a whole is invariant.
This property allows us to x 0, r, and, of course, we
pick the latter value in the topological phase. In general, in
a 3D topological phase protected by time reversal, both
elds should transform in the same way.
Now phase 1 has a single charged boson,
1
:
L
phase 1
em

1
2r
eB
1
A. (85)
The bulk theory predicts that 0 0, i.e., no magnetoelec-
tric effect for this phase.
However, for phase 2, both bosons are charged, so
L
phase 2
em

1
2r
eB
1
B
2
A, (86)
and the bulk theory predicts 0 2rmagnetoelectric effect
for this phase.
VII. 3D E
8
PHASE WITH HALF-QUANTIZED
SURFACE THERMAL HALL EFFECT AND
MISCELLANEOUS COMMENTS
Thus far, we have based our discussion of novel 3D SPT
phases on the 2D K a
x
matrix. When a conserved
charge is present, these phases often lead to a quantized
magnetoelectric effect, or, equivalently, to a half-quantized
surface Hall effect. On general grounds, one may expect
additional phases based on the fact that thermal transport
can also be quantized. In these phases, chiral modes are
expected at the domain walls between opposite-symmetry-
breaking regions, which lead to the quantized thermal
Hall conductance. In this section, we provide a possible
eld-theoretic description of such a phase.
011016-21
Recall that, in a 2D system, the combination k
xy
,T
:
T
r
2
k
2
B
3h
is quantized, where k
xy
is the thermal Hall con-
ductance and T is temperature, in the limit of T ! 0. Here,
:
T
counts the number of chiral boson modes at the edge.
For bosons with SRE in J 2, it is known that the quan-
tization takes values :
T
8n that are multiples of 8 times
quantum of thermal conductance. Anything else leads to
topological order. These states are based on the K matrix of
the Kitaev E
8
state [13]:
K
E
8
2 1 0 0 0 0 0 0
1 2 1 0 0 0 0 0
0 1 2 1 0 0 0 1
0 0 1 2 1 0 0 0
0 0 0 1 2 1 0 0
0 0 0 0 1 2 1 0
0 0 0 0 0 1 2 0
0 0 1 0 0 0 0 2
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
. (87)
One may utilize this fact to construct the following 3D
SPT phase. Assume time-reversal symmetry Z
T
2
is present.
Consider the 3D theory given by
L
1
2r
8
I1
eB
I
o
I

I,1
K
E
8
I1
8r
2
eo
I
o
1
. (88)
As long as all the elds o
I
transform the same way under
time-reversal symmetry, the coefcient is quantized to
0, r. The latter leads to a topological phase. If time-
reversal symmetry is preserved in the bulk but broken on
the surface, it is readily seen that each domain has thermal
Hall conductivity :
T
4, and a domain wall between
opposite domains has the eight chiral edge states of the 2D
theory specied by the K matrix in Eq. (87).
It is likely that such a state lies beyond cohomology
classication since we have already identied a phase
based on K a
x
in this symmetry class that exhausts
the set of states predicted by cohomology theory [5]. A
question that is relevant in this context is whether the eld
theory above can be realized within a lattice model. One
difference from the other topological phases we have de-
scribed based on the K a
x
FF term is that, in those cases,
a lattice regularization of the eld theory can be readily
envisaged since it involves a Berry phase for the product of
electric and magnetic elds representing two different
species of vortices eE
1
E
2
! E
1
_
B
2

E
2

B
1
. This term
is naturally discretized by assuming that the corresponding
vector potentials live on the links of the direct lattice and
the dual lattice. However, the diagonal entries of the K
E
8
matrix above lead to terms that are not obviously compat-
ible with a lattice. Whether this fact imposes an additional
constraint on possible phases is an important open ques-
tion. If, indeed, the phase described above is physically
admissible, then it remains to be claried if the additional
states lead to a Z
2
or a Z extension (assuming just time-
reversal symmetry). We leave these questions to future
study. In this context, it may be relevant to note that the
analogous free-fermion phases are topological supercon-
ductors in 3D protected by time-reversal symmetry (Class
DIII), which are classied by integers. On the other hand,
one may ask, What are the properties of the topologically
ordered surface state that is fully symmetric? We conjec-
ture that a candidate state is a Z
2
topological ordered state
in which all three nontrivial excitations are fermionic and
have r mutual statistics. Such a state when realized in 2D
is given by the K matrix:
K
SO8

2 1 1 1
1 2 0 0
1 0 2 0
1 0 0 2
_
_
_
_
_
_
_
_
_
_
_
_
_
_
, (89)
which is the Cartan matrix of SO8. This state, when
realized in 2D, has four chiral edge states and hence must
break time-reversal symmetry. However, it may appear on
the surface of a 3D SPT phase, protected by time-reversal
symmetry.
VIII. CONCLUSIONS
In summary, we highlight the remarkable similarities
between free fermion topological insulators and the bosonic
interacting topological phases described here. In the former
case, the surface is gapped only on breaking one of the
dening symmetries of charge conservation or time-
reversal symmetry. Then, the resulting ordered phase also
possesses unusual properties. For example, when charge
conservation is destroyed by a superconducting surface, the
vortices carry a Majorana zero mode. Similarly, for the
bosonic topological insulator with the same symmetries,
breaking charge conservation at the surface leads to fermi-
onic vortices (albeit without an attached Majorana zero
mode). On the other hand, breaking just time-reversal sym-
metry leads for the fermionic case to a quantized magneto-
electric effect of 0 r, whereas, for bosonic TIs in the
same situation, the same response is quantized but at
0 2r. The fully symmetric surface of the fermionic TI,
from which these conclusions can be readily derived, is a
Dirac dispersion of free fermions. We propose that the
analog for bosons is the deconned quantum critical action,
which describes a putative gapless state from which, on
being subjected to various perturbations, realizes different
ground states of the surface. It is also relevant to note that
bosonic analogs of topological superconductors exist where
domain walls between regions of opposite time-reversal-
symmetry breaking carry gapless chiral modes.
It is interesting to further highlight the particular case of
spin systems. The spin analogs of topological insulators
the topological paramagnetsmay potentially be the most
011016-22
important realization in solid-state systems of the class of
phases we have described. The surface of the topological
paramagnet either spontaneously breaks symmetry or is in
a quantum spin liquid state that is not allowed to exist in
strict two dimensions with the same symmetry. We have
discussed examples of such quantum spin liquids with
surface topological order or with exotic gapless excita-
tions. Our work raises fascinating questions on what kinds
of spin liquids with symmetry are actually allowed to exist
in strictly 2D systems that should be of direct importance
to studies of 2D quantum magnetism.
There are several open questions for future work.
Clearly, a central question is whether there are microscopic
models, or perhaps even experimentally relevant systems,
that could realize these phases. The bulk-sigma-model
eld theories may provide useful guidance in searching
for such realizations. One route to accessing SPT phases in
two dimensions is to start with a fractionalized phase and
conne the fractionalized excitations. Our analysis sug-
gests that a similar route may also be possible in three
dimensions by starting with a fractionalized phase with
emergent deconned U1 U1 gauge elds if these are
conned by condensation [57] of mutual dyons [where a
monopole of one U1 gauge eld is bound to particles that
carry gauge charge of the other gauge eld]. Exploring this
possibility might also suggest physical realizations of the
3D SPT phases. A more formal question is whether one can
push the eld-theoretic descriptions of this paper to obtain
all possible SPT phases in 3D, which could shed light on
the way in which the chiral phases augment the cohomol-
ogy characterization. The 3D BF FF theories seem a
convenient tool for capturing bosonic SPT phases.
However, general constraints on the form of such theories
are at present unclear.
ACKNOWLEDGMENTS
A. V. thanks Ari Turner and especially Yuan-Ming Lu
for stimulating discussions and collaborations on related
topics, and acknowledges support from NSF Contract
No. DMR-1206728. T. S. thanks Liang Fu, Michael
Levin, Chong Wang, Z. Gu, and Xiao-Gang Wen. T. S.
was supported by NSF Contract No. DMR-1005434. We
both thank Matthew Fisher for scintillating discussions and
encouragement and for stimulating discussions on dyon
condensation in 3D gauge theories as a route to describing
exotic phases. This material is based on work supported in
part by the National Science Foundation under Grant
No. PHYS-1066293 and the hospitality of the Aspen
Center for Physics. We also thank the Perimeter Institute
for Theoretical Physics, and the Kavli Institute for
Theoretical Physics where parts of this work were under-
taken. This work was partially supported by Simons
Foundation Nos. 229736 (T. S.) and 231377 (A. V.). On
completing this work, we became aware of other studies
that have some overlap with the present work [58,59].
APPENDIX A: 3D BF THEORYSURFACE
STATES AND EM RESPONSE
As a warmup, let us recall the derivation of the edge
states of a Chern-Simons theory in 2D [3]. We specialize to
the K a
x
Chern-Simons theory:
L
CS

e
:
2r
o
1
:
o
2
. (A1)
Note that gauge invariance at the surface can be ensured by
working in the gauge o
0
0. This condition implies
Jo
I
0 so o
I
|

|
I
and gives the edge Lagrangian
L
1
2r
2
, (A2)
leading to the usual Kac-Moody commutation relations.
The edge dynamics originates from other terms. For
example, we can add a Maxwell term to the original action
o
I
:

:
o
I
2
. The only low-derivative term that ap-
pears at the edge is from
y
o
I
x
. Substituting the edge eld
o
I
x

x
I
and noticing that the derivative perpendicular
to the edge (i.e., along y) picks up only the conning wave
function of the edge states, we are led to
y
o
I
x
/
x
I
. This
reasoning gives potential terms
L
1

x
2

x
2
.
Note that the pair of elds
1
,
2
, which are canonically
conjugate Eq. (A2), is like any regular 1D Luttinger liquid.
The special physics of SPT phases arises from the fact that
the elds can transform under the symmetry in ways that a
1D system cannot. For example, in the U1-protected
integer quantum Hall phase of bosons, the transformation
law of the rst nontrivial phase is
|
!
|
e. In other
words, both elds transform under the charge rotation,
which leads to protected edge states. By analogy, it appears
that we should nd that the surface of a 3 1D SPT phase
of bosons is a regular 2D bosonic system, apart from
application of symmetries. Indeed, we show below that
this is the surface state of the 3 1D BF theory.
Let us begin with the following 3D Lagrangian:
L
BF

e
:a
2r
B
:
o
a
. (A3)
Here, a bosonic current has been written as ]
e
:a
:
B
a
,2r, and ^ o represents the vortex loops.
To derive surface properties, again we see that the non-
dynamical parts of the Lagrangian implement the con-
straint: e
|]
|
o
]
0 and e
|]k
|
B
]k
0. One can solve this
expression to obtain o
|

|
and B
|]
e
|]
|
n
]
. We take
the gauge B
0|
o
0
0. The topological part of the edge
Lagrangian is computed next. [The edge is taken to be
perpendicular to z, and we use the indices o, l x, y.]
L
e
ol
2r
o
n
l
r
. (A4)
011016-23
Now, for the dynamics, once again we introduce
Maxwell terms in the bulk: Jo
2
and JB
2
. Again, the
ones that survive with low derivatives have
z
acting on
them. The terms that appear are
L
1
2

2
x
n
y
y
n
x
2
. (A5)
Thus our boundary Lagrangian is (given S
_
JxJyJrL
e
)
L
e
e
ol
2r
o
n
l
x
n
y
y
n
x
2
. (A6)
One interpretation of this Lagrangian is that of a photon
on a 2D surface where the Gauss law constraint has been
solved, i.e., 2 1D; the Gauss law
x
E
x

y
E
y
0 can
be solved by writing E
o
e
ol
l
,r. Then the term that
leads to canonical quantization of electric elds E
o
r
n
o
is
now replaced by
1
r
x
n
y

y
n
x
, the rst term in L
e
.
The Hamiltonian of the Maxwell theory
1
E
2
x
E
2
y

x
n
y

y
n
x
2
is the term written above in the
Lagrangian.
Although Refs. [25,26] follow a rather similar deriva-
tion, they interpret the theory above as the bosonized
description of a 2 1D Dirac fermion. This interesting
speculation does not appear to be compatible with the well-
known fact that the theory described by Eq. (A6) is dual to
a 2 1D theory of bosons. Explicitly, this can be seen as
follows: Since e
ol
o
n
l
,2ris conjugate to the phase , we
denote it by
e
ol
o
n
l
,2r and use this eld to write
the Hamiltonian of the surface theory as
H 4r
2

1
r
2
. (A7)
This Hamiltonian is just the theory of a boson in the two
spatial dimensions of the surface (as expected, since we
began with a bosonic theory).
It is useful to catalog the connection between the dual
descriptions. The monopole insertion operator in the sur-
face electrodynamics is e
|
and actually corresponds to the
insertion of particles. The other excitations, gauge
charges of the gauge theory, are the ends of vortices of
the 3Dbulk and are point particles on the surface. Here they
behave like charges in the 2D electrodynamics, since vor-
tices of the eld are equivalent to violating Gauss law for
the electric eld. Also, since Curlo ! Curlr, they corre-
spond to the ends of the 3D vortices. The vortex insertion
operator is of course a nonlocal object, which reects the
fact that one cannot insert a gauge-charged particle in the
bulk without changing the gauge elds everywhere. The 2D
surface is gapped either by monopoles e
|
or by vortex
condensation (Higgs mechanism). However, symmetry
may forbid these, leading to SPT phases.
Let us briey review some questions that arise in the
context of the Lagrangian (A3). One can add terms such as
L
1
$ eB
2
and L
2
$
o
:

o
:
2
, which are local
and respect symmetries. Integrating out B now appears to
give the Higgs term o
2
. However, it is readily seen that

this still describes an insulator by coupling the charge to an
external electromagnetic potential A via eAB,2r. Now
integrating the elds B essentially enforces o $A, which
when substituted into L
2
simply produces a Maxwell action
for the external eld: approximately
:
A
A
:
2
, as
expected for an insulator. A 3D topological EM response
appears in other cases where an FF term is present, by the
same substitution.
APPENDIX B: PERIODICITY IN
MULTICOMPONENT BF THEORY
In this appendix, we prove the 2r periodicity of for
the multicomponent BF theory. The Lagrangian is
L
1
2r
o
I
e
:k
:
B
I
k

8r
2
K
I1
e
:k
o
I
:
o
1
k
. (B1)
Summation is implicit over repeated component indices
I, 1.. The crucial second term, when expressed in terms of
the electric elds e
I
and the magnetic elds

l
I
, takes the
form
8r
2
I
K
II
_
2 e
I

l
I

1>I
K
I1
2 e
I

l
1
e
1

l
I
_
.
(B2)
Consider the theory on a closed three manifold such as a
3-torus of size L L L. Through one cycle, say, the xy
cycle, slowly insert 2rn
I
magnetic ux of species I at a
rate
J
I
Ji
. This operation leads to a bulk electric eld along
the z direction:
e
I
z

1
L
J
I
Ji
. (B3)
Next, slowly turn on 2rm
I
ux of l
I
z
in the bulk so that
l
I
z

2rm
I
L
2
. (B4)
The quantum amplitude for these processes is given by the
term in the action and takes the form
e
|,4r
2
_
JiL
3
2r,L
2
f
I
K
II
m
I
J
I
,Ji
1>I
K
I1
m
1
J
I
,Jig
e
|
I
K
II
n
I
m
I
I1
K
I1
n
I
m
1
. (B5)
For some particular pair I, 1, choose n
I
1, m
1
1, and
all other n
I
0 m
1
0 0. Then the amplitude simply be-
comes e
|K
I1
. If all the elements K
I1
are integers, it follows
that is periodic under a 2r shift.
APPENDIX C: OTHER SYMMETRIESU1 2 Z
2
Here the U1 can be interpreted as spin-rotation sym-
metry about the z axis, while the Z
2
is spin rotation by 180
degrees about the x axis.
011016-24
Surface theory.Let bosons l
y
1
e
|
1
be charged under
the U1 symmetry. Then the phase
1
and conjugate
number n
1
transform as
1
!
1
e: U1, n
1
! n
1
: U1,
1
!
1
: Z
2
, n
1
! n
1
: Z
2
.
(C1)
Now we would like to understand how a vortex in a
superuid surface state of this boson eld transforms.
The remaining Z
2
symmetry acts on the vortices; however,
it is readily seen that this symmetry switches vortices to
antivortices. More formally, the vortex elds
2
are
coupled minimally to the gauge eld n
2
, whose ux is
the number density n
1

x
n
2y

y
n
2x
,2r. Now since
the number density changes sign under Z
2
, so does the
gauge eld n
2
! n
2
. This relationship implies that, for
the minimal coupling to remain invariant, we need
2
!
2
. In fact, the desired transformation is
2
! |a
y
2
: Z
2
, (C2)
which may be viewed as the single projective representa-
tion of U1 2 Z
2
, where the U1 may be viewed as the
gauge U1 that changes sign under Z
2
. It is readily veried
that the gauge-invariant combinations jc
2
j
2
jc
2
j
2
and c
2
c
2
e
|
2
both transform nontrivially under
Z
2
, verifying that, if vortices condense, they always break
the symmetry. A second species of bosons is dened by
l
y
2
e
|
2
, which transforms as
2
!
2
r: Z
2
, n
2
! n
2
: Z
2
(C3)
and is neutral under the global U1. This transformation
satises the intuitive requirements of a topological surface
state, and therefore we conclude that a U1 2 Z
2
symmetry
group also leads to Z
2
topological phases. Note, however, if
the two symmetries were in direct product, there would be
no topological phases.
Let us now write the eld theory for the surface in terms
of vortices of
2
. They are minimally coupled to a vector
potential n
2
whose ux is the boson density r n
1
n
1
.
L j
|n
2
2
j
2
1
2k
]
2
2:
Vj
2
j
2
.
Since the eld l
1
is charged, monopole insertion operators
are forbidden, but various anisotropy terms involving four
vortex elds are allowed. These and other allowed pertur-
bations are readily identied given the symmetry trans-
formations above.
A dual description of the same theory is obtained by
fractionalizing the boson eld l
y
1
c
1
c
1
, where
1
c
1
, c
1
may be viewed either as a Schwinger
boson representation of l
1
[20] or as vortices of l
2
. Now
these transform under a projective representation of the
global symmetry U1 2 Z
2
:
1
! e
|ea
z
,2
1
: U1, (C4)
1
! a
x
1
: Z
2
, (C5)
which is compatible with the transformations in Eq. (C1).
We see that this implies that vortices in l
2
carry half unit of
global charge at the surface. This information helps us to
x the bulk eld theory.
3D bulk theory.Given the characterization of the sur-
face states above, we can write a bulk 3D theory that
reproduces these features. We write the following theory
based on K a
x
, where the conserved charge is coupled
to an external electromagnetic eld A, and justify it later:
L
tot
L
topo
L
em
, (C6)
L
topo

1
2r
eB
1
o
1
B
2
o
2

e
4r
2
o
1
o
2
,
L
em

1
2r
eB
1
A.
(C7)
Under the Z
2
symmetry, B
1
! B
1
, o
1
! o
1
, but
B
2
! B
2
, o
2
! o
2
. Thus the axion eld is odd under
Z
2
, so the action as a whole is invariant. This observation
allows us to x 0, r; the latter value yields the
topological phase. Also, only one of the boson species
carries global U1 charge, which implies that there is no
topological contribution to the magnetoelectric polariz-
ability, i.e., 0 0.
APPENDIX D: SYMMETRY TRANSFORMATION
OF SURFACE STATES WITH
TOPOLOGICAL ORDER
For convenience, we accumulate in this appendix the
properties of the surface state with Z
2
topological order of
the SPT phases with various symmetries. As described in
the main paper, such a surface topologically ordered phase
provides a particularly simple perspective on why a trivial
gapped symmetry-preserving surface is not allowed. The
Z
2
topological order has four distinct quasiparticles, which
we denote as 1, e, m, ]. The trivial quasiparticle sector is
described by 1 and consists of all local operators. We take e
(for electric) and m(for magnetic) to be bosons and ]
to be a fermion. e, m, and ] are all mutual semions. Below
TABLE III. U1 2 Z
T
2
.
Field q T
2
e
1
2
1
m
1
2
1
TABLE IV. Z
T
2
.
Field T
2
e 1
m 1
011016-25
we summarize how the physical symmetry is realized for
each of the three nontrivial quasiparticles for the various
phases. In what follows, q denotes the charge under the
global U1 symmetry. In all cases, other possible quasi-
particles in the same sector are obtained by adding trivial
quasiparticles. The symmetry properties of ] follow from
those of e and m as ] is a bound state of e and m.
(1) Symmetry U1 2 Z
T
2
.Here both e and m carry
charge 1,2 and are time-reversal invariant. It is
important only to ask about the presence or absence
of Kramers degeneracy under Z
T
2
corresponding to
T
2
1. See Table III.
(2) Symmetry Z
T
2
.Here there are a pair of e particles
(denoted e
n
e
"
, e
#
, a pair of m particles (m
n

m
"
, m
#
), and a single ] particle. We note that, in the
absence of other symmetries, e
"
is able to mix with
e
#
so that we may regard them as a single particle
and its antiparticle, which together form a Kramers
doublet. The same reasoning also applies to the m
particles. Therefore, we work with just a single e
and a single m particle. See Table IV.
(3) Symmetry U1 Z
T
2
.Here we discuss two
phases, phase 1 and phase 2. Their symmetries are
described in Tables Vand VI.
As we emphasized in the main paper, the realization of
symmetry is such that these Z
2
topologically ordered
phases cannot arise in strict 2D models with local action
of the symmetry group, as is readily seen from the general
K-matrix classication of 2D time-reversal-invariant
gapped Abelian phases in Ref. [43]. Here we present an
elementary analysis that is sufcient for the purposes of
this paper. Strictly 2D systems admit an edge to the vac-
uum (or equivalently a trivial gapped insulator) without
changing the symmetry. This is a key difference from the
surface topological order of interest to us here, where a
domain wall with a trivial gapped insulator is not possible
without breaking symmetry. The analysis below, following
the reasoning of Ref. [43], relies crucially on analyzing the
symmetries of the edge Lagrangian that describes a strictly
2D system. Hence it distinguishes between topological
order that is allowed in strict 2D and orders that require a
3D bulk. For the case of Z
2
topological order, we may, as
usual, take
K
0 2
2 0
_ _
. (D1)
The corresponding Chern-Simons Lagrangian is simply
L
1
r
o
e
Jo
m

1
2r
Ar
e
Jo
e
r
m
Jo
m
. (D2)
Here, o
1,2
are internal gauge elds and A is an external
probe gauge eld. The charge vector r
e
, r
m
has integer
components. Physically, Jo
e,m
are 2r]
e,m
of the e and m
particles, respectively. If both e and m carry global U1
charge 1,2, then r
e
r
m
1. As we already noted, such a
charge assignment implies a nonzero Hall conductivity and
hence cannot describe a strict 2Dsystemwith time-reversal
invariance. Thus the surface topological order described in
Tables III and VI cannot occur in strict 2D systems. It
remains for us to discuss the other two surface topological
orders. In both (as well as in the case in Table VI) both e
and m are Kramers doublets. We now show that this is not
possible in any time-reversal-invariant strict 2D Z
2
topo-
logically ordered state.
The 1 1 dimensional edge theory corresponding to the
Chern-Simons Lagrangian above is
L
edge

1
r
m
, (D3)
with o
|1,2

|
1,2
. Demanding time-reversal invariance
of the edge Lagrangian, we see immediately that the edge
densities
x
e
,
x
m
must transform with opposite signs.
However, if e
|
e
creates one member of a Kramers doublet,
it must transform as
e
|
e
! |e
|
e
(D4)
under time reversal. Equivalently,
e
!
e

r
2
so that the
edge density
x
e
!
x
e
. Thus if both e and m are
Kramers doublets, then both corresponding edge densities
must be even under time reversal. But this result is incon-
sistent with our deduction above from demanding time-
reversal invariance of the edge Lagrangian. We thus con-
clude that, in strict 2D systems, both e and m cannot be
Kramers pairs in a time-reversal-invariant system. It is
allowed to happen, however, at the surface of the 3D
SPT phases described in this paper. In passing, we note
that this analysis precludes the possibility that strictly 2D
spin models have gapped Z
2
topological phases where both
nontrivial bosonic quasiparticles carry spin-1,2 (i.e., are
spinons) while the fermionic quasiparticle carries no spin.
[1] M. Z. Hasan and C. L. Kane, Colloquium: Topological
Insulators, Rev. Mod. Phys. 82, 3045 (2010); X.-L.
Qi and S.-C. Zhang, Topological Insulators and
TABLE V. U1 Z
T
2
: Phase 1.
Field q T
2
e
1
2
1
m 0 1
TABLE VI. U1 Z
T
2
: Phase 2.
Field q T
2
e
1
2
1
m
1
2
1
011016-26
Superconductors, Rev. Mod. Phys. 83, 1057 (2011); M. Z.
Hasan and J. E. Moore, Three-Dimensional Topological
Insulators, Annu. Rev. Condens. Matter Phys. 2, 55
(2011).
[2] Andreas P. Schnyder, Shinsei Ryu, Akira Furusaki,
Andreas W. W. Ludwig, Classication of Topological
Insulators and Superconductors in Three Spatial
Dimensions, Phys. Rev. B 78, 195125 (2008); Alexei
Kitaev, Periodic Table for Topological Insulators and
Superconductors, arXiv:0901.2686.
[3] X. G. Wen, Quantum Field Theory of Many-Body Systems:
From the Origin of Sound to an Origin of Light and
Electrons (Oxford University Press, Oxford, England,
2004).
[4] This terminology differs slightly from that of Chen et al.
[5], who require that, for a state to be short-range en-
tangled, it must also be nonchiral.
[5] Xie Chen, Zheng-Cheng Gu, Zheng-Xin Liu, and Xiao-
Gang Wen, Symmetry Protected Topological Orders and
the Group Cohomology of Their Symmetry Group,
arXiv:1106.4772.
[6] A. Auerbach, Interacting Electrons and Quantum
Magnetism, Graduate Texts in Contemporary Physics
(Springer-Verlag, New York, 1994).
[7] L. Fidkowski and A. Kitaev, Topological Phases of
Fermions in One Dimension, Phys. Rev. B 83, 075103
(2011).
[8] A. M. Turner, F. Pollmann, and E. Berg, Topological
Phases of One-Dimensional Fermions: An Entanglement
Point of View, Phys. Rev. B 83, 075102 (2011).
[9] X. Chen, Z.-C. Gu, and X.-G. Wen, Classication of
Gapped Symmetric Phases in One-Dimensional Spin
Systems, Phys. Rev. B 83, 035107 (2011).
[10] N. Schuch, D. Perez-Garcia, and I. Cirac, Classifying
Quantum Phases Using Matrix Product States and
Projected Entangled Pair States, Phys. Rev. B 84,
165139 (2011).
[11] Alexei Kitaev, Anyons in an Exactly Solved Model and
Beyond, Ann. Phys. (Amsterdam) 321, 2 (2006); Alexei
Kitaev, Toward Topological Classication of Phases with
Short-Range Entanglement (unpublished). See http://online
.kitp.ucsb.edu/online/topomat11/kitaev/.
[12] M. Levin and Z. Gu, Braiding Statistics Approach to
Symmetry-Protected Topological Phases, Phys. Rev. B
86, 115109 (2012).
[13] Yuan-Ming Lu and Ashvin Vishwanath, Theory and
Classication of Interacting Integer Topological
Phases in Two Dimensions: A Chern-Simons Approach,
Phys. Rev. B 86, 125119 (2012).
[14] T. Senthil and Michael Levin, Integer Quantum Hall
Effect for Bosons: A Physical Realization, Phys. Rev.
Lett. 110, 046801 (2013).
[15] Zheng-Xin Liu and Xiao-Gang Wen, Symmetry-Protected
Quantum Spin Hall Phases in Two Dimensions, Phys. Rev.
Lett. 110, 067205 (2013).
[16] Xie Chen and Xiao-Gang Wen, Chiral Symmetry on the
Edge of Two-Dimensional Symmetry Protected
Topological Phases, Phys. Rev. B 86, 235135 (2012).
[17] Xiao-Liang Qi, Taylor L. Hughes, and Shou-Cheng
Zhang, Topological Field Theory of Time-Reversal
Invariant Insulators, Phys. Rev. B 78, 195424 (2008).
[18] Andrew M. Essin, Joel E. Moore, and David Vanderbilt,
Magnetoelectric Polarizability and Axion
Electrodynamics in Crystalline Insulators, Phys. Rev.
Lett. 102, 146805 (2009).
[19] T. Senthil and M. P. A. Fisher, Z
2
Gauge Theory of
Electron Fractionalization in Strongly Correlated
Systems, Phys. Rev. B 62, 7850 (2000).
[20] T. Senthil, A. Vishwanath, Leon Balents, Subir Sachdev,
and M. P. A. Fisher, Deconned Quantum Critical Points,
Science 303, 1490 (2004); T. Senthil, Leon Balents, Subir
Sachdev, Ashvin Vishwanath, and Matthew P. A. Fisher,
Quantum Criticality beyond the Landau-Ginzburg-Wilson
Paradigm, Phys. Rev. B 70, 144407 (2004).
[21] D. Birmingham, M. Blau, M. Rakowski, and G.
Thompson, Topological Field Theory, Phys. Rep. 209,
129 (1991).
[22] T. H. Hansson, V. Oganesyan, and S. Sondhi,
Superconductors Are Topologically Ordered, Ann. Phys.
(Amsterdam) 313, 497 (2004).
[23] Kevin Walker and Zhenghan Wang, (3 1)-TQFTs and
Topological Insulators, arXiv:1104.2632.
[24] C. W. von Keyserlingk, F. J. Burnell, and Steven H. Simon,
Three-Dimensional Topological Lattice Models with
Surface Anyons, Phys. Rev. B 87, 045107 (2013).
[25] G. Y. Cho and J. E. Moore, Topological BF Field Theory
Description of Topological Insulators, Ann. Phys.
(Amsterdam) 326, 1515 (2011).
[26] H. Aratyn, Fermions from Bosons in 2 1 Dimensions,
Phys. Rev. D 28, 2016 (1983).
[27] AtMa Chan, T. Hughes, S. Ryu, and E. Fradkin, Effective
Field Theories for Topological Insulators by Functional
Bosonization, arXiv:1210.4305.
[28] Masaki Oshikawa, Commensurability, Excitation Gap,
and Topology in Quantum Many-Particle Systems on a
Periodic Lattice, Phys. Rev. Lett. 84, 1535 (2000).
[29] M. B. Hastings, Lieb-Schultz-Mattis in Higher
Dimensions, Phys. Rev. B 69, 104431 (2004).
[30] C. Lannert, Matthew P. A. Fisher, and T. Senthil, Quantum
Connement Transition in a J-Wave Superconductor,
Phys. Rev. B 63, 134510 (2001).
[31] Scott D. Geraedts and Olexei I. Motrunich, Monte Carlo
Study of a U1 U1 System with r-Statistical
Interaction, Phys. Rev. B 85, 045114 (2012).
[32] Andreas W. W. Ludwig, Matthew P. A. Fisher, R. Shankar,
and G. Grinstein, Integer Quantum Hall Transition: An
Alternative Approach and Exact Results, Phys. Rev. B 50,
7526 (1994).
[33] T. Senthil and Matthew P. A. Fisher, Competing Orders,
Nonlinear Sigma Models, and Topological Terms in
Quantum Magnets, Phys. Rev. B 74, 064405 (2006).
[34] B. Swingle, M. Barkeshli, J. McGreevy, and T. Senthil,
Correlated Topological Insulators and the Fractional
Magnetoelectric Effect, Phys. Rev. B 83, 195139
(2011).
[35] Michael Levin, F. J. Burnell, Maciej Koch-Janusz, and
Ady Stern, Exactly Soluble Models for Fractional
Topological Insulators in Two and Three Dimensions,
Phys. Rev. B 84, 235145 (2011).
[36] Joseph Maciejko, Xiao-Liang Qi, Andreas Karch, and
Shou-Cheng Zhang, Fractional Topological Insulators in
Three Dimensions, Phys. Rev. Lett. 105, 246809 (2010).
011016-27
[37] J. T. Chalker and P. D. Coddington, Percolation, Quantum
Tunnelling and the Integer Hall Effect, J. Phys. C 21, 2665
(1988).
[38] C. Dasgupta and B. I. Halperin, Phase Transition in a
Lattice Model of Superconductivity, Phys. Rev. Lett. 47,
1556 (1981).
[39] M. P. A. Fisher and D. H. Lee, Correspondence between
Two-Dimensional Bosons and a Bulk Superconductor in a
Magnetic Field, Phys. Rev. B 39, 2756 (1989).
[40] Leon Balents, Lorenz Bartosch, Anton Burkov, Subir
Sachdev, and Krishnendu Sengupta, Putting Competing
Orders in Their Place near the Mott Transition, Phys. Rev.
B 71, 144508 (2005).
[41] Olexei I. Motrunich and Ashvin Vishwanath, Emergent
Photons and Transitions in the O(3) Sigma Model with
Hedgehog Suppression, Phys. Rev. B 70, 075104 (2004).
[42] Cenke Xu and A. W. W. Ludwig, Nonperturbative Effects
of Topological Theta-Term, arXiv:1112.5303.
[43] Michael Levin and Ady Stern, Classication and Analysis
of Two-Dimensional Abelian Fractional Topological
Insulators, Phys. Rev. B 86, 115131 (2012).
[44] With an integer coefcient Q, L
E
l

Q
2r
eBo describes a
topologically ordered phase with ground-state degeneracy
of Q
3
on a three torus.
[45] This is obtained by applying Gausss law
_
JV
JS
, where ]
e
:a
o
1:
o
2a
.
[46] O. I. Motrunich and T. Senthil, Origin of Articial
Electrodynamics in Three-Dimensional Bosonic Models,
Phys. Rev. B 71, 125102 (2005).
[47] O. I. Motrunich and A. Vishwanath, Comparative Study of
Higgs Transition in One-Component and Two-Component
Lattice Superconductor Models, arXiv:0805.1494.
[48] A. B. Kuklov, M. Matsumoto, N. V. Prokofev, B. V.
Svistunov, and M. Troyer, Deconned Criticality:
Generic First-Order Transition in the SU2 Symmetry
Case, Phys. Rev. Lett. 101, 050405 (2008).
[49] F. Jiang, M. Nyfeler, S. Chandrasekharan, and U. Wiese,
From an Antiferromagnet to a Valence Bond Solid:
Evidence for a First-Order Phase Transition, J. Stat.
Mech. (2008) P02009.
[50] A. W. Sandvik, Evidence for Deconned Quantum
Criticality in a Two-Dimensional Heisenberg Model with
Four-Spin Interactions, Phys. Rev. Lett. 98, 227202
(2007); Anders W. Sandvik, Continuous Quantum Phase
Transition between an Antiferromagnet and a Valence-
Bond Solid in Two Dimensions: Evidence for Logarithmic
Corrections to Scaling, Phys. Rev. Lett. 104, 177201
(2010).
[51] A. Banerjee, K. Damle, and F. Alet, Impurity Spin Texture
at a Deconned Quantum Critical Point, Phys. Rev. B 82,
155139 (2010).
[52] R. G. Melko and R. K. Kaul, Scaling in the Fan of an
Unconventional Quantum Critical Point, Phys. Rev. Lett.
100, 017203 (2008); Ribhu K. Kaul and Anders W.
Sandvik, Lattice Model for the SUN Neel to Valence-
Bond Solid Quantum Phase Transition at Large N, Phys.
Rev. Lett. 108, 137201 (2012).
[53] Anatoly Kuklov, Nikolay Prokofev, and Boris
Svistunov, Weak First-Order Superuid-Solid Quantum
Phase Transitions, Phys. Rev. Lett. 93, 230402
(2004).
[54] J. T. Chayes, L. Chayes, Daniel S. Fisher, and T. Spencer,
Finite-Size Scaling and Correlation Lengths for
Disordered Systems, Phys. Rev. Lett. 57, 2999
(1986).
[55] P. M. Ostrovsky, I. V. Gornyi, and A. D. Mirlin,
Interaction-Induced Criticality in Z
2
Topological
Insulators, Phys. Rev. Lett. 105, 036803 (2010).
[56] Victor M. Galitski, G. Refael, Matthew P. A. Fisher, and T.
Senthil, Vortices and Quasiparticles near the
Superconductor-Insulator Transition in Thin Films, Phys.
Rev. Lett. 95, 077002 (2005).
[57] Matthew Fisher (personal communication).
[58] B. Swingle, Interplay between Short and Long-Range
Entanglement in Symmetry Protected Phases,
arXiv:1209.0776.
[59] Cenke Xu, Three Dimensional Symmetry Protected
Topological Phase Close to Antiferromagnetic Neel
Order, arXiv:1209.4399.
011016-28

Physics of Three-Dimensional Bosonic Topological Insulators: Surface-Deconfined Criticality and Quantized Magnetoelectric Effect

Uploaded by

Copyright:

Available Formats

Physics of Three-Dimensional Bosonic Topological Insulators: Surface-Deconfined Criticality and Quantized Magnetoelectric Effect

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Physics of Three-Dimensional Bosonic Topological Insulators: Surface-Deconfined Criticality and Quantized Magnetoelectric Effect

Uploaded by

Copyright:

Available Formats

Physics of Three-Dimensional Bosonic Topological Insulators: Surface-Deconned Criticality

and Quantized Magnetoelectric Effect

, in Eq. (3), which forces n

can be obtained as a composite of

can mix with V

due to the interspecies tunneling term. It

that carries charge 0. Also, under time reversal,

are the usual

do not change physical

, which may be rewritten as

. The integer constraint is implemented by sum-

takes real (rather than

, where we have utilized the fact that

N transforms as a vector while c is invariant.

each carry charge

are allowed by symmetry. Other

. However, it is readily seen that

You might also like