Full Text

Turn on search term navigation

ARTICLE

Received 30 Aug 2013 | Accepted 25 Feb 2014 | Published 26 Mar 2014

symmetry-protected topological phases, write down ground state wave function and discover topological properties of symmetry defects both in the bulk and on the edge of the system. We consider symmetries that include a Z2 subgroup, which allows us to dene domain walls.

While the usual disordered phase is obtained by proliferating domain walls, we show that symmetry-protected topological phases are realized when these domain walls are decorated, that is, are themselves symmetry-protected topological phases in one lower dimension. This construction works both for unitary Z2 and anti-unitary time reversal symmetry.

DOI: 10.1038/ncomms4507

Symmetry-protected topological phases from decorated domain walls

Xie Chen1, Yuan-Ming Lu1,2 & Ashvin Vishwanath1,2

1 Department of Physics, University of California, Berkeley, California 94720, USA. 2 Materials Science Division, Lawrence Berkeley National Laboratories, Berkeley, California 94720, USA. Correspondence and requests for materials should be addressed to A.V. (email: mailto:[email protected]

Web End [email protected] ).

NATURE COMMUNICATIONS | 5:3507 | DOI: 10.1038/ncomms4507 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 1

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms4507

Symmetry-protected topological (SPT) phases are gapped quantum phases with topological properties protected by symmetry1. The ground states of SPT phases contain only

short-range entanglement and can be smoothly deformed into a totally trivial product state if the symmetry requirement is not enforced in the system. However, with symmetry, the nontrivial SPT order is manifested in the existence of gapless edge states on the boundary of the system that cannot be removed as long as symmetry is not broken. Many SPT phases have been discovered over the past decades222 and a general structure of the theory of SPT phases is emerging. In 1d, SPT phases have been completely classied for general interacting bosonic/fermionic systems that carry nontrivial projective representations of the symmetry in their degenerate edge states2325. In fact, the Haldane/AKLT phase of spin-one quantum antiferromagnet is a physical realization of a one-dimensional (1D) SPT phase, which is protected by spin rotation or time reversal symmetries. Two-dimensional (2D) SPT phases were rst discovered in topological insulators and superconductors, which have subsequently been generalized to three dimensions2628. Such free fermion SPT phases have been realized experimentally2932 and also classied completely for systems with internal symmetries3335. Classication of free fermion SPT phases with spatial symmetries is not complete yet but a lot of progress has been made, see for example, refs 36 and 37. Recently, it was realized that two- and higher-dimensional SPT phases exist not only in free fermion systems but also in strongly interacting boson systems, and a systematic construction is given based on the group cohomology of the symmetry1. While the construction provides a xed point description in the bulk, it is hard to access edge dynamics and hence how the system responds to physical perturbations. Also it is not clear in what physically realistic systems can these bosonic SPT phases be realized.

Much progress has been achieved recently in understanding the low energy physics and nding physical realizations for some of the bosonic SPT phases. In 2D, a general understanding of SPT phases with Abelian (and time reversal) symmetry has been given in terms of ChernSimons K-matrix theory6, which also provides a eld theory of the protected edge states. A physical realization of the U(1) SPT phases (which are expected to have even integer-quantized quantum Hall conductance6) has been proposed in bosonic cold atom systems with articial gauge elds8. The 2D SPT phases with nonabelian SO(3) and SU(2) symmetry was studied with nonlinear sigma model with quantized topological y terms and are found to have quantized spin transport13. More recently, some three-dimensional (3D) SPT phases have been understood within a eld theoretic approach, which predicts surface vortices with projective representations and quantized magnetoelectric responses9. Recently, 3D SPT phases have been discussed from number of different theoretical perspectives, including twisted vortex condensates18 and the statistical magneto-electric effect20. A useful perspective on SPT phases appears on gauging the symmetry discussed by Levin and Gu5 in 2D, and recently extended in other studies20,21,3840 to 3D phases. Finally, we note a special feature of 3D SPT phases is that their surface states could be gapped and fully symmetric if they develop topological order just at the surface9,2022.

In this paper, we focus on SPT phases with symmetry group Z2 G and present a simple construction of d dimensions SPT

phases by decorating the domain walls of Z2 congurations in the bulk with d 1 dimensional SPT states with G symmetry. Such a

construction naturally reveals some of the special topological features of such phases. For example, it is easy to see that when a domain wall is cut open at the boundary of the system, the end points/loops of the domain wall carry gapless edge states of the d 1 dimensional SPT state with G symmetry. The same is true

on the ux point/loops in the system when the Z2 part of the symmetry is gauged. In particular, in the Result section we are going to present the construction of a 2D SPT phase with Z2 ZT2

symmetry by decorating the 1D Z2 domain walls with Haldane chains and a 3D SPT phase with Z2 Z2 symmetry by decorating

the 2D Z2 domain walls with the nontrivial 2D SPT phase with Z2 symmetry. Supporting numerical evidence and generalizations of our results are presented in the Methods section, which include numerical analysis of symmetry on the edge of the 2D SPT phase with Z2 ZT2 symmetry, discussion of gauging the Z2 symmetry

and the construction of a 1D SPT phase with Z2 Z2 symmetry

and 3D SPT phases with ZT2, ZT2 U1 and ZT2 SO3

symmetries. These results are summarized in Figs 13. The relation between the domain wall construction and the Knneth formula for group cohomology is also discussed in the methods section. Brief reviews of the group cohomoloy description of SPT phases, the eld theory description of 2D and 3D SPT phases are given in the Supplementary Material.

Results2D SPT phase with Z2 ZT2 symmetry. Lets start with the

simplest example in this construction: a 2D SPT phase with Z2 ZT2 symmetry where ZT2 represents time reversal symmetry.

Intuitively, the state is constructed by attaching 1D Haldane chains with time-reversal symmetry to the domain walls between Z2 congurations on the plane and then allow all kinds of uctuations in the domain wall congurations. With such a structure, it is then easy to see that when the system has a boundary where domain walls can end, the end point would carry a spin 1/2 degree of freedom (edge state of Haldane chain) that transforms projectively under time reversal.

First, we describe a 2D lattice wave function of the ground state that is gapped and does not break any symmetry. Then, we discuss what happens on the boundary of the system. By identifying the relation between symmetry action on the edge and the nontrivial cocycle of Z2 ZT2 symmetry group, we establish the

existence of nontrivial SPT order in this model. It is known that

Figure 1 | SPT phase in 2D with Z2 and time-reversal symmetry (Z2 ZT2).

A snapshot of the ground state wave function: blue and grey are oppositely directed domains of the Z2 symmetry. The ground state preserves the

Z2 symmetry since it is a superposition of domain congurations. The domain walls themselves (black lines) are in a d 1 SPT phase (the

Haldane/AKLT phase) protected by time reversal symmetry. When they end at the edge of the system they create Kramers doublets, leading to a gapless edge state.

2 NATURE COMMUNICATIONS | 5:3507 | DOI: 10.1038/ncomms4507 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms4507 ARTICLE

Figure 4 | Two-dimensional SPT model with Z2 ZT2 symmetry. Lattice

model: (a) each plaquette hosts a Z2 variable (big black dot) (b) each vertex hosts four spin s (small blue dots) (c) two spin s on the same link form a singlet if they are on a Z2 domain wall, other spin s form singlets within each vertex.

Figure 2 | SPT phase in 3D with Z2

2 symmetry. Red (blue) surfaces represent domain walls of the Z2 (2) symmetry. They intersect along curves (black lines). The ground state wave function is a superposition of all domain wall congurations that differ by a sign depending on whether there is an even or odd number of intersection curves. This automatically implies a protected edge state when a domain wall intersects the surface of the sample (grey dashed line).

[afii9848]

[afii9846]

Figure 3 | An SPT phase in 1D protected by Z2 Z2 symmetry. This phase

involves two sets of spins s, t and emerges from the ordered phase of the s spins by condensing domain walls attached to spin ip excitations of the t spins.

Figure 5 | One term in

PiPsixTiP. Two spin s connected by dotted lines are in the singlet state j00i j11i=

p , four spin s connected by dotted lines are in the symmetric total singlet state j0000i j1111i=

p , arrows

denote the state of unpaired spin s, which are preserved in the ipping.

this SPT phase with Z2 ZT2 symmetry can be described with

U(1) U(1) ChernSimons theory with a nontrivial symmetry

action. We review this description and demonstrate the nontrivial topological feature of domain walls in the eld theory language.

Bulk wave function on 2D lattice. Consider a honeycomb lattice as shown in Fig. 4a where each plaquette hosts a Z2 variable (big black dot) in state |0S or |1S. The Z2 part of the symmetry ips |0S into |1S and |1S into |0S. At each vertex, there are four spin 1/2s (one at the center and three on links as shown in Fig. 4b). On each spin 1/2 time-reversal symmetry acts as T isyK and

satises T 2 1. On the total Hilbert space of each vertex, time

reversal still satises T 2 1 and forms a linear representation of

ZT2. Now consider a Hamiltonian term

Pk Vk that enforces that in the ground state two spin 1/2s on the same link form a time-reversal singlet pair j00i j11i=

i ips the Haldane chain conguration around it. P projects onto the low energy sector of

Pk Vk. Figure 5 illustrates one term in Pi PtixTiP. Note that when ipping a segment of the Haldane chain from one side of the plaquette to another side, the state of the edge spin 1/2s are preserved.

The Vk terms and the PtixTiP terms are local and commute with each other. The ground state of the total Hamiltonian

H XkVk

p if the link is on a Z2 domain wall while the remaining spin 1/2s that are not on a domain wall form singlets within each vertex (there is always an even number of these at each vertex). For the vertex that is not on a domain wall, we can put the four spins into the symmetric total singlet j0000i j1111i=

p state. The effect of this term can be

thought of as attaching Haldane chains to all the Z2 domain walls. A possible conguration in the ground state is shown in Fig. 4c where the dotted lines represent singlet pairing. Note that while the Haldane chain is usually dened in spin-1 systems, here for simplicity of discussion we use the xed point form of the wave function where each spin-1 is replaced with two spin-1/2s. A tunneling term between the congurations

Pi PtixTiP is then added to the Hamiltonian that in the low energy sector of

Pk Vk ips the Z2 variables together with the related singlet congurations. tix ips the Z2 spin in a plaquette and T

XiPtixTiP 1

is exactly solvable and is an equal-weight superposition of all possible consistent congurations of Z2 variables and singlets.

The ground state is unique, gapped and preserves Z2 ZT2

symmetry.

To see this, note that to each domain wall conguration there is a unique state of the spins that live at the vertices. This is enforced by the rst term in the Hamiltonian in equation (1) that projects into the manifold of states that follow this association. In the absence of any decoration, the equal superposition of domains is

NATURE COMMUNICATIONS | 5:3507 | DOI: 10.1038/ncomms4507 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 3

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms4507

[afii9846]2

[afii9846]3

[afii9846]4

[afii9848]1

[afii9848]2

[afii9848]3 [afii9848]4

[afii9846]5

[afii9848]5

Figure 6 | Two equivalent descriptions of the boundary of 2D SPT model with Z2 ZT2 symmetry. (a) thick bonds represent Z2 variable in state j1i

and thin bonds j0i, spin 1/2 exists on their domain wall. (b) One Z2 variable

per bond and one spin variable per vertex. Time reversal acts on spins in a way dependent of neighboring Z2 congurations. Dotted boxes represent local degrees of freedom on the 1D boundary labelled by group elements.

Pi txi. Now, in order that the decorated spin congurations are properly generated, one induces the same matrix elements between domain wall congurations as txi, but where each domain wall conguration appears along with its unique spin state attached. Let us denote these operators as ~txi.

Since these only affect the spins near the ipped bit, they are local operators and take the form of the second term in the Hamiltonian in equation (1). Moreover, they are readily seen to commute with one another and the constraint, when acting within the low energy manifold, since they essentially mimic the action of the txi, with the decorating spin conguration simply coming for the ride.

Edge state. Interesting things happen when the system has a boundary. When the system is cut open, the Z2 domain walls will have end points on the boundary. As the spin singlets are tied to domain walls in the bulk, on the boundary there will be isolated spin 1/2s on the end points of Z2 domain walls. Imagine that we cut the system through the middle of plaquettes (In order not to cut through degrees of freedom, we can rst split the Z2 variable into two, one on each side of the cut and add a Hamiltonian term for them to be equal. This does not violate the symmetry of the system and is therefore allowed). The 1D boundary degrees of freedom then contain two parts: the bond Z2 variables that come from plaquettes on the cut and the vertex spin variables when the neighboring bonds contain different Z2 congurations, as shown in Fig. 6a. Symmetry acts by ipping the Z2 variable and time-reversing the spins.

To study the boundary as an effective 1D, we will use a slightly different description of the boundary degrees of freedom. In the previous description, the existence of vertex degrees of freedom are dependent on the bond degrees of freedom. Equivalently, we can also think of the boundary as a local 1D system with independent degrees of freedom on the bonds and the vertices, as shown in Fig. 6b. That is, we can imagine that there exist pseudo-spins (empty circles) on the vertices when neighboring Z2 variables are the same. On the pseudo-spins time reversal acts simply as T sxK and satises T 2 1. The pseudo-spin can be

thought of as the direct sum of two spin-0 dimensions |0S |1S

and i(|0S |1S). The total symmetry action is thenZ2 : tx on each t spin

ZT2 : isyK if s is on t domain wall

sxK otherwise

It is easy to see that such a description of the boundary can be reduced to the previous one by polarizing each pseudo-spin in the z direction. As T 2 1 on each pseudo-spin, polarizing them does

not break time-reversal symmetry. Using this description, we can see explicitly how this model is related to the nontrivial third cocycle of the symmetry group and therefore contains nontrivial SPT order. We can also write down effective Hamiltonians for the boundary and solve for the low-energy dynamics, as we demonstrate in the following.

In terms of the local degrees of freedom on the 1D edge state, s and t, the Z2 symmetry action on the edge reads

Z2 :

Yitix 2

and time reversal acts as

ZT2 :

Yi;i 1I tizti1z2 si1x I tizti1z2 isi1yK 3

We can write down an effective Hamiltonian satisfying the symmetry and solve for the dynamics on the 1D edge. Some simple interaction terms that satisfy both the Z2 symmetry (equation 2) and the time-reversal symmetry (equation 3) include tix siztixsi1z and six ti 1zsixtiz. Therefore, a possible form of

the dynamics of the edge is given by Hamiltonian

Xitix siztixsi1z six ti 1zsixtiz 4

This Hamiltonian can be mapped exactly to an XY model by applying unitary transformations U

Qn i tzn txn

the ground state of H0

Qn 11 tnz1 snz=4

h i

to each pair of ti and si variables. The Hamiltonian that we get after these transformations is

e UHeU 1 Xitixsi1x tizsi1z sixtix siztiz 5

and the symmetry transformations are mapped to

2 :

Yitizsiz 6

and

T2 :

Yitix ti 1x2 six tix ti 1x2 isiyK 7

In order to see how the symmetry acts on the low-energy effective theory, we diagonalize the XY Hamiltoniane, identify the free boson modes in the low energy eigenstates and calculate the action of the symmetry on the low-energy states. The low-energy states are labelled by quantum numbers nkAZ and

nk 2 Z,

k 0, 1, 2y, where n0 and

n0 label the total angular momentum and the winding number of the boson eld respectively and nk and

nk, k40, label the left/right moving boson modes. The Z2 symmetry acts by mapping state n0;

n0; nk;

h i

f g

j i to

Pk40 nk nkj n0; n0; fnk; nkgi and time-reversal symmetry acts by mapping state jn0; n0; fnk; nkgi to

Pk 0nk nkjn0; n0; f nk; nkgi, which is consistent with the low-energy description given in the next section where Z2 symmetry acts as f1- f1, f2- f2 and time-reversal acts as

f1- f1 p, f2-f2 p on the chiral boson elds {f1,2}. Effective eld theory description. As discussed in ref. 6, SPT phases in 2D may be characterized by the unusual transformation properties of their edge states, under the action of symmetry. The simplest situation, which captures a large fraction of SPT phases, is a Luttinger liquid edge with a single gapless bosonic mode. The edge elds are characterized by compact conjugate bosonic elds

4 NATURE COMMUNICATIONS | 5:3507 | DOI: 10.1038/ncomms4507 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms4507 ARTICLE

f1, f2 such that

f1x;

@x0 f2x0 2p

idx x0 8

Physically, eif1x inserts a boson at point x along the edge, while eif2 inserts a phase slip. We see that the 2D SPT phase of interest implies the following transformation law:

Z2 : f1 ! f1 p f2 ! f2

ZT2 : f1 ! f1 f2 ! f2 p

This is different from the form we discussed in the last section. However, as we show in the Supplementary Note 3 and 4, these two forms are actually equivalent to each other.

Note, this edge cannot be gapped out without breaking symmetry. Terms that gap the edge such as cosf1,2 are forbidden

by symmetry. If either one of the symmetries is broken, then a trivial edge is possible. Finally, let us verify that a domain wall of the Z2 order, terminating at the edge carries a Kramers pair, as the pictorial description implies.

The operator that creates a domain wall, rotates f1 by p everywhere (say) to the right of the point x. By the commutation relation 8, this is identied with the operator

Dx eif2x=2 10

Now, let us consider how the domain wall insertion operator transforms under time reversal T . In particular, we would like to

calculate the action of acting twice with time reversal and see

whether T 2 1. Now Dx !

T e if2 p=2 iD x. Apply-

1 o3a 1iai

1; a 1i1a 1; a 13 Now we can show that the symmetry action on the boundary of the Z2 ZT2 model we constructed is of exactly this form. We

can consider the pair of Z2 variables ai (ti, si) (dotted box in

Fig. 6b) as labelling group elements in the group Z2 ZT2. That is,

we consider the Z2 state |0S/|1S of t as labelling the trivial/ nontrivial element in group Z2 and the spin state |mS/|kS of s as labelling the trivial/nontrivial element in group ZT2. Then, the Z2 part of the symmetry is the following mapping

jtii !j tii 14 and the time-reversal part of the symmetry (given by either

T sxK on the pseudo-spin or T isyK on the spin) also

involves the mapping

jsii !j

i 15 Therefore, a general symmetry operation labelled by a (t, s),

tAZ2, s 2 ZT2, will rst change group elements labels in each box

jai ti; sii!jaai tti; ssii: 16 Moreover, the time-reversal symmetry also adds a ( 1) phase

factor when the neighboring Z2 variables are different and when the vertex variable was originally in state |mS. Such a phase factor on spin si can be written as

si1z

1 t

ing this again we nd:

Dx !

izti 1

Here sz 1 in |mS, sz 1 in |kS and tz 1 in |mS,

Dx 11 thus, we nd that for this operator T 2 1. Hence, it must

carry a Kramers degeneracy as expected from the physical picture.

Connection to group cohomology. Now let us establish the nontrivial SPT order in this state by identifying the connection of the symmetry actions on the boundary with nontrivial third cocycles. When we think of the boundary as a local 1D system with independent bond variables and vertex variables (that is, treating spins and pseudo-spins as equivalent variables), symmetry no longer acts on the degrees of freedom independently. The Z2 part of the symmetry Z2 :

Qi tix still acts on each Z2 variable independently. However, time-reversal symmetry on the vertices depends on the Z2 congurations on the bonds, as shown in equation (3). This is exactly the signature of SPT phases.

Indeed, as was shown in ref. 1, the boundary of dD STP phases can be thought of as a d 1D local system where the symmetry

acts in a non-onsite way. Without symmetry, the boundary can be easily gapped out. If the symmetry acts in an on-site way, the boundary can be simply gapped out by satisfying the symmetry on each site. However, with non-onsite symmetry related to nontrivial group cocycles, the boundary must remain gapless as long as symmetry is not broken.

In particular, in the group cohomology construction as reviewed in the Supplementary Note 2, the boundary local degrees of freedom are labelled by group elements ai of the symmetry group and the action of the symmetry operator involves two parts: rst, changing the local states |aiS to |aaiS;

second, multiplying a phase factor given by nontrivial cocycles to each pair of ai and ai 1.

More specically, on the 1D boundary of 2D SPT phases, the symmetry acts as

Oaja1; ::::; aNi Yif aai; ai1jaa1; :::; aaNi 12

where f a(ai, ai 1) is a phase factor given by the nontrivial third

cocycle o3

f aai; ai

T 2

1 in |kS. Therefore, the symmetry action on the boundary

can be put exactly into the form of equation (12) with

f aai; ai

izti 1

2 ; 18 when the symmetry action a involves time reversal andf aai; ai

1 si1z

1 t

1 1; 19 when a does not involve time reversal. We can reorganize the variables and write the phase factor as a function of a 1iai 1, a 1i1a 1 o3

a 1iai

1; a 1i1a 1; a f aai; ai

1 20

It can be checked that o3a 1iai

1; a 1i1a 1; a is a nontrivial

third cocycle of group Z2 ZT2, using the cocycle conditions

introduced in the Supplementary Note 1. Therefore, we can show using methods in ref. 3 that the boundary must be either gapless or symmetry breaking and the 2D bulk is in a nontrivial SPT phase.

3D SPT phase with Z2 Z2 symmetry. Now we go one

dimension higher and consider the Z2 Z2 group. To differ

entiate the two Z2s, we write them as Z2 and2. We can start with similar constructions in the bulk where the nontrivial2 SPT phase in two dimension is attached to the domain wall of the Z2 conguration in the cubes. A simple understanding of the ground state wave function exists starting from Levin and Gus5 construction of 2D SPT phase with Z2 symmetry. It was shown that in the 2D SPT model the ground state wave function takes the simple form of

c2DC 1NC 21

NATURE COMMUNICATIONS | 5:3507 | DOI: 10.1038/ncomms4507 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 5

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms4507

where NC is the number of domain walls in conguration C. Such

a wave function has gapless edge states when the system has a

boundary. Now consider the symmetry Z2

2. This has elements {1, g1, g2, g3}. We will pick two of the three nontrivial elements say g1, g2. Choose any one of these two generators (say g1) and consider domain walls in 3D between regions that break this symmetry in opposite ways. This denes closed 2D manifolds. Now, on these closed manifolds the domain walls of the second generator (g2) are examined and number of closed loops counted (essentially these loops are intersections of domain walls of g1 and g2). Now, one uses this set of closed loops and denes a wave function as in equation (21).

c3DC 1N

g1g2 C

2 symmetric state where Ng1g2C is the number of closed loops formed by the intersection of

the g1 and g2 domain walls in conguration C (see Fig. 2). What is

the physical consequence of this wave function? Consider breaking one of the Z2 symmetries and forming a domain wall (of say g1). Now, the domain wall is the edge state of the 2D SPT phase protected by the unbroken Z2 symmetry. This is also evident from the wave function. Owing to this nontrivial topological features, the wave function describes a nontrivial SPT phase.

As there are three different ways to pick two generators out of the three nontrivial elements in Z2

This is the wave function of a Z2

2 symmetry.

Connecting the edge state to group cohomology. Imagine that we cut the system open and expose the boundary. Note that when doing the cut, we need to be careful and not break the symmetry of the system (in particular, what we can do is double the Z2 and2 spins along the cut and cut through each pair). Then on the boundary, each plaquette hosts a Z2 variable ti and each vertex hosts a2 variable si. It is then easy to see that on the 1D domain wall of Z2 variables on the boundary is attached the 1D edge state of the2 SPT phase. The Z2 symmetry acts by ipping the plaquettes

X jtii j tii 24 and the2 symmetry acts on the s variables living on the t domain walls as

~X js1; :::sNi

YNi1f si; si1j

s1; :::

2, we can construct three different SPT wave function in this way. Therefore, this construction allows us to access all possible nontrivial phases classied using group cohomology theory. In the following sections we will present a more detailed study of this construction. Starting from a bulk Hamiltonian, we analyse its edge state and demonstrate its relation to nontrivial group cocycles. Finally, we present the eld theory description of these phases9.

Ground state wave-function on a 3D lattice. Consider a 3D cubic lattice where each cube in the bulk hosts a Z2 variable t (big black dot in Fig. 7) and the Z2 symmetry ips |0S to |1S and |1S to |0S. Each vertex in the 3D bulk hosts a2 variable s (small green dot in Fig. 7) and the2 symmetry ips j

~0i to j

~1i and j

~1i to j

~0i. To

dene the Hamiltonian of the system, we rst x a Z2 conguration in the cubes and dene the interaction between the2 vertices. Start with a magnetic eld in the x direction on all the2 spins. If a vertex is on the domain wall between Z2 congurations, modify the Hamiltonian term sx by an extra

factor of

ndwp 23 where ndwp is the number of2 domain wall pairs along the loop of all nearest neighbor2 spins on the same Z2 domain wall as the original2 spin. If two or more Z2 domains walls touch at the2 vertex, then one factor is added for each domain wall (we can pick a particular way to separate the domain walls at each vertex). Now it is easy to see that the2 spins form 2D nontrivial2 SPT states on the Z2 domain wall. Away from the domain wall, they are polarized in the x direction. Now we include tunneling between the Z2 congurations tx, which in the low-energy sector of the previous Hamiltonian term not only ips the Z2 spins but also changes the corresponding2 interaction pattern in its neighborhood. All these terms are local. The ground state is then unique and gapped and is the equal-weight superposition of all Z2 congurations together with the2 SPT state on its domain wall and polarized2 spin away from its domain walls. The ground state is symmetric under the Z2

i 25

where f(si, si 1) adds a phase factor of 1 if and only if both si

and si 1 are in the state j

~1i. f si; si

1 1 s

iz 1 si1z=4 26

where sz 1 for j

~0i and sz 1 for j

~1i. Here we use the

symmetry convention as inref. 3, which is equivalent to the convention used in ref. 5.

[afii9848]j

[afii9846]k

[afii9846]j

[afii9848]i

[afii9848]

[afii9846]i

[afii9846]

2 symmetry. Lattice model: each cube hosts a Z2 variable t (big black dot) and each vertex hosts a2 variable s (small green dot) (a) a cube and the vertices around it (b) a vertex and the cubes around it. Shaded surfaces represent Z2 domain walls.

[afii9848]k

2 symmetry: Each plaquette hosts a Z2 variable t and each vertex hosts a2 variable s.

Degrees of freedom within a circle are labelled by group elements of Z2

Figure 7 | 3D SPT model with Z2

Figure 8 | Boundary of 3D SPT phase with Z2

6 NATURE COMMUNICATIONS | 5:3507 | DOI: 10.1038/ncomms4507 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms4507 ARTICLE

Equivalently, just like in the previous case, we can think of the 2D boundary as a system of independent degrees of freedom with one Z2 variable t per plaquette and one2 variable s per vertex, as shown in Fig. 8. The t domain walls lives on the solid lines. Then, the Z2 part of the symmetry simply acts by ipping the Z2 variables ti

Z2 : jtii !j tii: 27 The2 part of the symmetry action acts by ipping the2 variables si

2 : jsii !j

1 that is

charged only under the U(1) symmetry. A possible SPT surface state is to spontaneously break this U(1) symmetry by condensing this boson. Restoring the symmetry requires condensing vortices c of this bosonic eld. However, for this to be the surface of a topological phase, the vortex transformation law must forbid a vortex condensate that preserves the remaining Z2 Z2 symme

try. This can be achieved if the transformation law is projective, which implies at least a twofold degeneracy of the vortex elds hence c (c , c ). Intuitively, one may consider the elements

of the group Z2 Z2 {1, gX, gY, gZ} as representing 180

rotations about the x, y, z axes. The projective representation is then just the spin 1/2 doublet, with the nontrivial generators represented by the Pauli matrices ga isa, thus c !

i: 28 Moreover, the2 symmetry also involves a phase factor of 1

whenever the t domain wall connects two s spins in the j~1i state.

For example, at the shaded intersection shown in Fig. 8, the phase factor is given by

1 s

ga isac.

Physical operators, which can be directly measured, do not realize symmetry in a projective fashion; however, the vortex is a nonlocal object and can hence transform projectively. Thus, the effective Lagrangian for the vortices is:

Lsurface

iz1 s

i 1

1 t

izt

Such a symmetry action can be shown to be related to cocycles in a similar way as in the 2D example. In particular, in the group cohomology construction as reviewed in Supplementary Note 2, the boundary local degrees of freedom are labelled by group elements ai of the symmetry group and the action of the symmetry operator involves two parts: rst, changing the local states |aiS to |aaiS; second, multiplying a phase factor given by nontrivial cocycles to each triangle of ai, aj, ak.

More specically, on the 2D boundary of 3D SPT phases, the symmetry acts as

Oaja1; ::::; aNi

Yf aai; aj; akjaa1; :::; aaNi 30 where f a(ai, aj, ak) is a phase factor given by the nontrivial fourth cocycle o4

f aai; aj; ak o4a 1iaj; a 1jak; a 1ka 1; a 31 To put the above symmetry action of Z2

2 into this form, we can relate each plaquette with one of its vertices (the lower right corner, for example) and think of the Z2 and2 variables in this pair as labelling group element ai (ti, si) in the symmetry

group, as indicated by circles in Fig. 8. Then, the symmetry action labelled by a t, s ips the variables as

jai ti; sii !jaai tti; ssii 32 Moreover, the phase factor is related to the triangle of ai (ti,

si), aj (tj, sj) and ak (tk, sk) (although it depends trivially on

tk and si). Including the dependence of the phase factor on the symmetry action applied, we have

f aai; aj; ak 1 if s 2 a is trivial f aai; aj; ak

1 s

Xsj@m iamcs j2 K@man @nam2 33

where, by the usual duality41,42, the vortex elds are coupled to a vector potential a whose ux is the number density of b1 bosons.

Now, a vortex condensate will necessarily break symmetry since the gauge invariant combination Na cwsac will acquire an

expectation value. However, this transforms like a vector under rotations and will necessarily break the Z2 Z2 symmetry. Thus,

the theory passes the basic test for the surface state of a topological phase. This continues to be true when we break down the U(1) symmetry to Z2 and identify it with one of the existing

Z2 symmetries. That is, the boson b1 now transforms under the ga. This gives us three choices, which can be labelled by a X, Y, Z,

which identies the generator that leaves b1 invariant. Thus, the phase labelled X has b1- b1 under the generators gY, gZ but is

left invariant under gX. Even under this reduced symmetry the surface states cannot acquire a trivial gap since the condensate of b1 continues to break some of the symmetries. Now, there are three distinct nontrivial phases implied by this construction labelled a X, Y, Z, which, combined with the trivial phase,

conrms the Z2 Z2 classication of topological phases with this

symmetry1. With this reduced symmetry, additional terms can be added to the surface Lagrangian including DLX (cwsxc

cosf1) We would like to now verify that domain walls at the surface carry protected modes along their length.

Now, consider breaking the symmetry at the surface down to a single Z2. When the remaining Z2 generator is different from the one used to label the phase, we will see a protected mode is present along the surface domain walls. For example, in the setup here where the phase is labelled by X, consider breaking the symmetry down to just the Z2 generated by gZ. This is realized by condensing just c or c vortices. Consider a domain wall in

the xy plane along x 0, where for x40 (xo0) we have c

(c ) condensed. The domain wall is the region of overlap of these condensates where we have cy c eif

iz1 s

i 1

1 t

izt

if s 2 a is nontrivialNow we can reorganize the variables and check that o4

a 1iaj; a 1jak; a 1ka 1; a f aai; aj; ak is indeed a nontri

vial four cocycle using the condition given in Supplementary Note 1.

Therefore, as discussed in Supplementary Note 2, the state constructed in this way corresponds to a nontrivial SPT phase with Z2

2 symmetry and the physical features include gapless states on Z2 domain walls on the boundary.

Effective eld theory description. We access this 3 1D topological

phase by directly constructing the 2 1 dimensional boundary.

The surface is a conventional 2 1D bosonic system except in

the way symmetries are implemented, which prohibit a fully symmetric, gapped and non-fractionalized surface. We expand the physical symmetry Z2 Z2 to (Z2 Z2) U(1). Eventually we

will break the U(1) to Z2 and identify it with one of the Z2 generators.

We begin with a bosonic eld on the surface: b1 eif

2 , which represents a vortex tunneling operator across the domain wall. The phase f2 is therefore conjugate to the boson phase f1 as in a Luttinger liquid. Note however the transformation law under the remaining

Z2 symmetry is: f1;2 !

gZ f1;2 p, which is the transformation law

of the protected edge of the 2D Z2 SPT phase5,6. A similar conclusion can be drawn for surface domain walls, when the remaining Z2 symmetry is gY. On the other hand, in this X phase, preserving the gX symmetry does not lead to protected modes

NATURE COMMUNICATIONS | 5:3507 | DOI: 10.1038/ncomms4507 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 7

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms4507

since f1 is invariant under this generator and can be locked at a particular value without breaking the symmetry.

Now, let us write down a 3 1D topological eld theory that

describes this phase. We will restrict to an abelian theory, which is not ideal given that the symmetry acts like spin rotation along orthogonal directions. Nevertheless the abelian theory gives us valuable insights. Consider the two species of bosons eif1;2

introduced earlier. In the phase labelled X above, the number density n1 conjugate to f1 is invariant under all transformations, while the number density n2 changes sign under two of them. We can always pick these to be group elements gX, gY as above. Then we write down the following BF FF theory9:

2pL EB1@a1 B2@a2

Y2p E@a1@a2 34

where E is short for Emvls and the indices are suppressed. Here, 2pn1 E0ijkqiB1jk is the number density of boson b1, and

F1ij qia1j qja1i represents the vortex lines of boson b1. Thus

the vector potentials a1 are invariant under the transformations, but a2 changes sign under gX, gY. Therefore, the coefcient of the second term in the equation above must be quantized Y 0, p by

the usual arguments. The latter case represents the topological phase. The unusual properties of domain walls when the symmetry is broken down from Z2 Z2-Z2 at the surface are

readily deduced from this theory. A domain wall that breaks gX,

gY occurs when Y p- p on the surface leads to the surface

Luttinger liquid action:

SLL

K 38

In this section, we show how to extract the low-energy effective action of the symmetry on the edge state from exact diagonalization.

The Hamiltonian is an XY model (written in xz plane here) on a spin 1/2 chain and the low-energy effective theory is known to be described by the compactied free boson theory with Lagrangian

2pLedge @tf1@xf2 u

@xf1

@xf2

" #

The low-energy states are labelled by quantum numbers nk 2 Z and

nk 2 Z,

k 0, 1, 2y, where n0 and

n0 label the total angular momentum and the winding number of the boson eld respectively and nk and

nk, k40, label the left/right moving boson modes. If we normalize the ground state energy to be 0 and the rst excited-state energy to be 1/4, then the energy of each low-energy state is given by

n20

@xf1@tf2 dx dt 35

which describes a domain wall located along z 0, y 0, where

we have replaced a1,2i qif1,2. These are just the phase elds we

discussed above, and in particular, under the remaining gZ transformation they are both shifted by f1;2 !

gZ f1;2 p, which is the transformation property of the nontrivial edge of a Z2 SPT phase in 2D.

DiscussionIn conclusion, we have presented the construction of a 2D SPT phase with Z2 ZT2 symmetry by attaching Haldane chains, a 1D

SPT phase with ZT2 symmetry, to the Z2 domain walls in the 2D bulk and also a 3D SPT phase with Z2

Xk40knk nk 40

and the lattice momentum p ( p/aoprp/a) of each state is given by

pa n

" #

Xk402kL nk nk

where L is the total system size. n0 is given by the conserved U (1) quantum number

Xisiy tiy 42

The boson eld f f f2 can be thought of as describing the direction the

spin 1/2 is pointing to in the xz plane. Therefore, it corresponds to the spin state cos f j"i sin f j#i. From this, it is easy to see that the Z2 symmetry maps f to f. However, it is not easy to see the action of the time-reversal symmetry on the

low-energy state because of its complicated form. In order to obtain this information, we perform exact diagonalization of the Hamiltonian, identify the nk,

nk quantum numbers of the eigenstates from quantities like E, p, Sy and nd out how time-reversal symmetry acts on these states.

Energy levels in the spectrum can be degenerate and p and Sy allow us to partially split the degeneracy. However, states with fn0 0;

n0; fnk;

2 symmetry by attaching the 2D SPT state with2 symmetry to the domain wall of the Z2 variables in the 3D bulk. Such a construction leads directly to the special topological feature of Z2 domain walls on the boundary of the system: the Z2 domain walls on the boundary carry gapless edge states of the other symmetry (ZT2 and2). We established the nontrivial SPT order in the system by relating the symmetry action on the boundary of the system to nontrivial cocycles and also demonstrated how the SPT order can be properly described using eld theories.

This domain wall construction also applies to 3D SPT phases with ZT2, ZT2 U1 and ZT2 SO3 symmetry, including the time-

reversal symmetric topological superconductor phase, which features chiral E8 edge modes along surface domain walls that break time-reversal symmetry. This state lies beyond the group cohomology classication. The domain wall construction of attaching d 1 dimensional SPT states of G2 symmetry to d 1

dimensional defects in G1 congurations was shown to be related to one term in the Knneth formula for the group cohomology of groups of the form G1 G2. Other terms in the formula may be

related to attaching lower-dimensional SPT states with G2 symmetry to lower-dimensional defects in G1 congurations, exploring which is left to future work. A 1D SPT phase with Z2 Z2 symmetry, when approached in this manner, readily

suggests a parent Hamiltonian. It would be interesting to nd

physically well-motivated Hamiltonians that realize the higher-dimensional topological phases as well. The physical viewpoint on SPT phases described in this work may help guide such a search.

Methods

Symmetry on the edge of 2D SPT with Z2 ZT2 symmetry. For the 2D SPT phase

with Z2 ZT2 symmetry, we studied one particular realization of the edge dynamics.

The Hamiltonian governing the edge dynamics is given by

Xitixsi1x tizsi1z sixtix siztiz 36

and the Z2 symmetry acts on the edge as

~X Yitizsiz 37

while the ZT2 symmetry acts as

~T Yitix ti 1x2 six tix ti 1x2 isiy

nkgg and

fn0 0; n0; fnk; nkgg have the same E, p and Sy. In order to tell them apart, we

need to use our knowledge of the action of the Z2 symmetry. Using the decomposition of the f eld

fx; t f0 p utL p xL

Pk401

4pk

k e ipxut

h i

where p0 measures total angular momentum (with quantum number n0),

bke ipx ut by

k eipx ut

bkeipx ut

measures winding number (with quantum number

n0) and bk, bk are the annihilation operators for left/right moving boson modes (with occupation number

labelled by nk and

nk). As the Z2 symmetry maps f to f, it maps p0,

p0, bkby

all to minus themselves. Therefore, under Z2 symmetry action,

jn0;

n0; nk;

nki goes to

j n0;

n0; fnk;

nkgi. Therefore,

using the action of the Z2 symmetry, we can further distinguish states with

fn0 0; n0; fnk; nkgg and x the relative phase between them. If we want to x

the global phase factor of these two states, we can calculate the action of complex conjugation on them, which is expected to act as jn0;

n0; fnk;

nkgi to

j n0;

n0; fnk;

nkgi.

Now we are ready to look at each degenerate sector labelled by E, p, Sy and see how symmetry acts on them. We discuss the rst few sectors as an illustration of method. The spectrum is obtained by diagonalizing a system with 16 spin s.

8 NATURE COMMUNICATIONS | 5:3507 | DOI: 10.1038/ncomms4507 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms4507 ARTICLE

First, the ground state with E 0, p 0, Sy 0 is nondegenerate. If the phase

factor is xed such that the state is unchanged under complex conjugation, then it is invariant under both Z2 and time-reversal symmetry.

The rst excited states with E 1/4 and p 0 are twofold degenerate, one with

Sy 1 and one with Sy 1. The two states are j1; 0; f0; 0gi and j 1; 0; f0; 0gi. If we x gauge such that complex conjugation acts as 0 1

1 0

in this subspace

and Z2 symmetry acts as 0 1

1 0

in this subspace, then time reversal acts as

1:0000 0:0065i 0:0000 0:0000i

0:0000 0:0000i 1:0000 0:0095i

The third energy level at P 0 is also twofold degenerate with statesj2; 0; f0; 0gi and j 2; 0; f0; 0gi. If we x gauge such that complex conjugation acts as 0 1

1 0

in this subspace and Z2 symmetry acts as 0 1

1 0

in this

subspace, then time reversal acts as

0:9619 0:0004i 0:0381 0:0099i

0:0381 0:0099i 0:9619 0:0004i

The third energy level also has twofold degeneracy at P 1 with statesj0; 1; f0; 0gi and j0; 1; f0; 0gi. If we x gauge such that complex conjugation acts as 1 0

0 1

in this subspace and Z2 symmetry acts as 0 1

1 0

2 symmetry. Consider gauging g1 the rst Z2 symmetry, but leaving the remaining2 engaged so it continues to act as a global symmetry. Now consider and inserting a gauge ux along a curve. It is readily seen that this ux curve will have an SPT edge state along it, specically, the edge state of the d 2 SPT phase protected by

in this

subspace, then time reversal acts as

0:0380 0:6342i 0:9621 0:0251i 0:9621 0:0251i 0:0380 0:6342i

2 symmetry, as shown in Fig. 9. It can therefore be distinguished from a trivial phase, where gauging one of the Z2 degrees of freedom does not lead to protected excitations along Z2 ux lines. To see this, note that gauging the symmetry just means that the domain walls of g1 ends along the ux line. However, this domain wall is now in a2 SPT phase. This follows from the wave function in equation (22), since the loops on the surface bounded by this curve, formed by domains of g1, are weighed to give a 2D

LevinGu wave function5.

An interesting question is the result of gauging entirely the Z2

The third energy level also has twofold degeneracy at p

1L=2 with states

2 symmetry.

If we begin in the SPT phase this should lead to a Z2 Z2 topological order that is

distinct from the conventional one, which is described by the deconned phase of a Z2 Z2 gauge theory. Characterizing these subtle differences in topological order is

left to future work.

1D SPT phase with Z2 Z2 symmetry. We discuss how the well-known Haldane

phase in 1D can be understood within the framework of decorated domain walls. An advantage of this picture is that it leads directly to a model Hamiltonian that realizes this phase, and provides a simple rationale for the string-order parameter of this state. Consider breaking down the full SO(3) spin rotation symmetry to just Z2 Z2 symmetry, which is sufcient to dene this topological phase. The two Z2s

can be modeled by a pair of Ising models s,t, and the ordered phases are given by

hszi 6 0 or htzi 6 0 or hsztzi 6 0. Consider beginning in the partially ordered

phase hszi 6 0 but htzi 0. Now, the domain walls of s and the Z2 quanta of

t are both gapped. If we condense the former, we enter the completely disordered phase. Condensing the latter leads to the completely ordered phase. However, one can consider the following scenario. What happens when we condense the bound state of s domain wall and the t Z2 quanta (see Fig. 3)?

It is readily seen that this describes the SPT phase. First, since we condense domain walls of s, the Z2 symmetry is restored. Note, although t quanta are condensed, they are condensed along with the domain walls, so this is not a local operator, which implies that Z2 is also unbroken. So we have the full Z2 Z2

symmetry. The easiest way to see this is the SPT phase is to consider the order parameter for this phase, which is just the condensate r ztz, wherez is the

disorder parameter that creates/destroys a domain wall in the s elds. So we expect long range order in:

hrirji htzizizjtji

htzi

Yiokoj sxk

j0; 0; fn1 1; others are 0gi and j0; 0; f

n1 1; others are 0gi. If we x gauge such that complex conjugation acts as 0 1

1 0

in this subspace and Z2 symmetry acts

as 1 0

0 1

in this subspace, then time reversal acts as 0 1 1 0

Therefore, up to nite size inaccuracy, the action of time reversal on the low energy state is consistent with jn0;

n0; fnk;

nkgi to

jn0;

n0; fnk;

nkgi.

Gauging the Z2 symmetry. In ref. 5, Levin and Gu discussed the topological phase in d 2 protected by just Z2 symmetry. In that case, since there is no additional

symmetry, we cannot interpret the domain walls with SPT phases in d 1.

However, there is a different mechanism whereby an SPT phase is generated. The domain wall ends, which are now particles, carry fractional statistics (semionic statistics in this case). In fact this is readily seen from the symmetry transformation law for the edge state of this phase f1,2-f1,2 p. Now, to generate a domain wall,

both elds must be rotated, which is achieved by the domain wall operator Dx eif x f x=2. It is readily seen that D(x)D(x0) iSign(x x0)D(x0)D(x),

which is the hallmark of semionic statistics in 1D. One can now gauge the Z2 symmetry to obtain a topologically ordered state that is distinct from the regular Z2 gauge theory38,39,4345. In fact, the particle excitations in this theory are semions and anti-semions, and this is termed the double-semion theory. Indeed, the act of gauging the Z2 symmetry simply implies the possibility of domain walls ending within the 2D sample. The location of these ends are just the gauge charges and uxes. Given our experience with domain walls ending at the edge of the sample, where they have been shown to be simians, these end points are then expected to be semions.

We now apply this intuition to the cased discussed here. First, consider thed 2 SPT phase with Z2 ZT2 symmetry. If the Z2 part of the symmetry is gauged,

domain walls could end within the sample. We now expect these ends, which are Z2 uxes, to carry a spin 1/2 degrees of freedom, as shown in Fig. 9. The gauged version of this model is similar to the one discussed in ref. 46. From the cohomology classication we know that 2D phases with Z2 ZT2 symmetry form a

Z2 Z2 group. Therefore, with this construction together with the nontrivial SPT

phase that appears with just the Z2 symmetry, we are able to account for all SPT phases possible in this system.

Now consider the d 3 SPT phase with Z2

Before gauging

After gauging

5tzji

However, this is just the string order parameter for the Z2 Z2 SPT phase47.

The following spacetime picture motivates why it has edge states. Consider the path integral representation in imaginary time, with a spatial boundary. Then spacetime is a cylinder, the periodic direction being time. The condensate of Z2 charged domain walls leads to world lines that sometimes intersect the boundary. When they do, since they carry Z2 quantum number, they ip the spin at the edgeso there must be gapless edge degrees of freedom that uctuate in time.

This picture also motivates the following Hamiltonian. First, consider a pair of decoupled Ising Models, s and t as shown in Fig. 3. To enforce the binding of charge to domain walls we add the following Hamiltonian:

H1 l

Xisziszi1txi1=2 tzi 1=2tzi1=2sxi

Figure 9 | Result of gauging the Z2 symmetry: The gauge ux now allows the domain walls to terminate on them. They are point particles carrying a Kramers doublet in the d 2 system, and ux lines that carry a 1D edge

state protected by the remaining2 global symmetry in the d 3 system.

The Hamiltonian HCluster1 (refs 48,49) is just a set of commuting projectors, and has a unique ground state on a system with periodic boundary conditions.

However, gapless edge states appear when the system is terminated at a boundary.

NATURE COMMUNICATIONS | 5:3507 | DOI: 10.1038/ncomms4507 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 9

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms4507

3D time-reversal symmetric SPT phases. The decorating domain wall picture allows us to construct nontrivial SPT phases with time-reversal symmetry also. In this section we are going to discuss 3D phases with ZT2 symmetry only, ZT2 U1

symmetry and ZT2 SO3 symmetry. In particular, the phase we construct with ZT2

symmetry only is beyond the cohomology classication.

Consider a 3D lattice with a spin 1/2 sitting in each cube. Time-reversal symmetry maps between spin up j"i to spin down j#i states together with the

complex conjugation operation in this basis. Now we can decorate the domain wall in this spin conguration with some 2D states. First, we need to specify the orientation of the domain walls as pointing from j#i to j"i. Then we can put chiral

states on the domain wall whose chirality matches the orientation of the domain walls. In particular, we put the Kitaevs E8 state50 on the domain wall, which is a bosonic 2D state with no fractional excitations in the bulk and a c 8 edge state.

After this decoration, we sum over all possible spin congurations, together with the E8 decoration. This wave function is invariant under time-reversal symmetry, which can be seen as follows. Time reversal maps between j#i and j"i changing the

orientation not the potision of the domain walls. At the same time, time reversal maps the E8 state to E8 state with inverse chirality. Therefore, the chirality of the

E8 is always consistent with the orientation of the domain wall and the total wave function is time reversal invariant.

Such a construction gives rise to a 3D SPT state with only time-reversal symmetry. Following similar arguments as in previous sections, we see that on the boundary of the system if we break time-reversal symmetry in opposite ways on neighboring regions, then the domain wall between these regions carry a chiral edge state with c 8. This is half of what one can get in a pure 2D system without

fractionalization. This is consistent with the eld theory in ref. 9. On the other hand, if time reversal symmetry is not broken, then the surface is gapless or has topological order. In fact, the surface topological state is that with three fermions21,22, which cannot be realized in pure 2D system with time-reversal symmetry.

Similar construction applies to 3D SPT phases with ZT2 U1 and ZT2 SO3

symmetry. In 2D with U (1) and SO (3) symmetry, there is an integer class of SPT phases with even-integer-quantized charge or spin Hall conductance8,13,14,45. Similar to the construction above, we can put the rst nontrivial phase in this class to the domain wall of time reversal. The Hall conductance chirality should match the domain wall orientation. By summing over all spin congurations, we obtain a state with both time reversal and U (1) or SO (3) symmetry. One signature of the state would be even-integer-quantized Hall conductance on the time reversal domain wall on the 2D surface of the system.

Relation to Knneth formula for group G G1 G2. If a group G is the direct

product of two subgroups G G1 G2, then the Knneth formula for the group

cohomology of G can be written as38

Hd 1G; U1

Xd 1k0 HkG1; Hd 1 kG2; U1 47

which says that cohomology groups of G in d 1 dimension can be obtained from

cohomology groups of G1 and G2 in lower dimensions. Owing to the relation between cohomology groups and SPT phases, this implies that SPT phases with G symmetry in d dimension can be constructed from SPT phases with G1 and G2 symmetry in lower dimensions.

When k 0, the term in the formula is

H0G1; Hd 1G2; U1 Hd 1G2; U1 48 which means that some SPT phases with symmetry G in d dimension are just SPT phases with symmetry G2 in d dimension.

When k d 1, the term in the formula is

Hd 1G1; H0G2; U1 Hd 1G1; U1 49 which means that some SPT phases with symmetry G in d dimension are just SPT phases with symmetry G1 in d dimension.

Our domain wall construction correspond to the term with k 1

H1G1; HdG2; U1 50

Hd(G2, U (1)) labels SPT phases with G2 symmetry in d 1 dimensions. If Hd(G2,

U (1)) M (M Zn, Z for example), then the term becomes H1(G1, M), which

labels 1D representations of G1 using M coefcient. Therefore, this term says that some d dimensional SPT phase with G symmetry can be obtained from d 1

dimensional SPT phases with G2 symmetry and 1D representation of G1 with M coefcient.

Our domain wall construction provides a physical interpretation of the above statement. With discrete G1 group, as in the cases we are interested in, consider a d dimensional conguration with group elements in G1. The domain walls are then also labelled by group elements g1 2 G1 given by the difference of the group

elements on the two sides of the wall. Then on the d 1 dimensional domain wall

labelled by g1, we can choose to put d 1 dimensional SPT phases from the class

M. Our choice should be consistent with the fusion rules of the domain walls. That is, the phases ma and mb 2 M that we choose to put onto domain walls ga1 and gb1

should combine into the phase mab, which we put onto the domain wall labelled by

ga1gb1. In other words, this correspond to a 1D representation of group G1 in coefcient M.

Therefore, the term with k 1 in the formula 47 corresponds exactly to our

domain wall construction. Note that this formula also gives the condition of when such domain wall construction can be consistent. In particular, when putting the d 1 dimensional SPTs onto the domain walls, we need to put them according to

the corresponding 1D representation of G1 in M. Suppose we have M Z3, then

putting the rst nontrivial one onto a Z2 domain wall is not allowed because two Z2 domain walls can merge to trivial and so should the corresponding SPTs.

Terms with higher ks would correspond to constructing d dimensional SPT phases with G symmetry by putting d k dimensional SPT phases with G2

symmetry onto d k dimensional defects in G1 congurations.

References

1. Chen, X., Gu, Z.-C., Liu, Z.-X. & Wen, X.-G. Symmetry-protected topological orders in interacting bosonic systems. Science 338, 16041606 (2012).

2. Gu, Z.-C. & Wen, X.-G. Tensor-entanglement-ltering renormalization approach and symmetry-protected topological order. Phys. Rev. B 80, 155131 (2009).

3. Chen, X., Liu, Z.-X. & Wen, X.-G. Two-dimensional symmetry-protected topological orders and their protected gapless edge excitations. Phys. Rev. B 84, 235141 (2011).

4. Gu, Z. & Wen, X. G. Symmetry-protected topological orders for interacting fermionsfermionic topological non-linear sigma-models and a group super-cohomology theory. Preprint at http://arXiv.org/abs/1201.2648

Web End =http://arXiv.org/abs/1201.2648 (2012).

5. Levin, M. & Gu, Z.-C. Braiding statistics approach to symmetry-protected topological phases. Phys. Rev. B 86, 115109 (2012).

6. Lu, Y.-M. & Vishwanath, A. Theory and classication of interacting integer topological phases in two dimensions: a Chern-Simons approach. Phys. Rev. B 86, 125119 (2012).

7. Levin, M. & Stern, A. Classication and analysis of two-dimensional abelian fractional topological insulators. Phys. Rev. B 86, 115131 (2012).

8. Senthil, T. & Levin, M. Integer quantum hall effect for bosons. Phys. Rev. Lett. 110, 046801 (2013).

9. Vishwanath, A. & Senthil, T. Physics of three-dimensional bosonic topological insulators: surface-deconned criticality and quantized magnetoelectric effect. Phys. Rev. X 3, 011016 (2013).

10. Lu, Y.-M. & Lee, D.-H. Quantum phase transitions between bosonic symmetry protected topological phases in two dimensions: emergent $qed_3$ and anyon superuid. Preprint at http://arXiv.org/abs/1210.0909

Web End =http://arXiv.org/abs/1210.0909 (2012).

11. Grover, T. & Vishwanath, A. Quantum phase transition between integer quantum hall states of bosons. Phys. Rev. B 87, 045129 (2013).

12. Xu, C. Three-dimensional symmetry-protected topological phase close to antiferromagnetic nel order. Phys. Rev. B 87, 144421 (2013).

13. Liu, Z.-X. & Wen, X.-G. Symmetry-protected quantum spin hall phases in two dimensions. Phys. Rev. Lett. 110, 067205 (2013).

14. Chen, X. & Wen, X.-G. Chiral symmetry on the edge of two-dimensional symmetry protected topological phases. Phys. Rev. B 86, 235135 (2012).

15. Lu, Y.-M. & Lee, D.-H. Spin quantum hall effects in featureless nonfractionalized spin-1 magnets. Preprint at http://arXiv.org/abs/1212.0863

Web End =http://arXiv.org/abs/1212.0863 (2012).

16. Oon, J., Cho, G. Y. & Xu, C. Two dimensional Symmetry Protected Topological Phases with PSU(N) and time reversal symmetry. Preprint at http://arXiv.org/abs/1212.1726

Web End =http://arXiv.org/ http://arXiv.org/abs/1212.1726

Web End =abs/1212.1726 (2012).

17. Ye, P. & Wen, X.-G. Projective construction of two-dimensional symmetry-protected topological phases with u(1), so(3), or su(2) symmetries. Phys. Rev. B 87, 195128 (2013).

18. Xu, C. & Senthil, T. Wave functions of bosonic symmetry protected topological phases. Phys. Rev. B 87, 174412 (2013).

19. Geraedts, S. D. & Motrunich, O. I. Exact realization of integer and fractional quantum hall phases in models in. Ann. Phys. 334, 288315 (2013).

20. Metlitski, M. A., Kane, C. L. & Fisher, M. P. A. Bosonic topological insulator in three dimensions and the statistical witten effect. Phys. Rev. B 88, 035131 (2013).

21. Wang, C. & Senthil, T. Boson topological insulators: a window into highly entangled quantum phases. Phys. Rev. B 87, 235122 (2013).

22. Burnell, F. J., Chen, X., Fidkowski, L. & Vishwanath, A. Exactly Soluble Model of a 3D Symmetry Protected Topological Phase of Bosons with Surface Topological Order. Preprint at http://arXiv.org/abs/1302.7072

Web End =http://arXiv.org/abs/1302.7072 (2013).

23. Chen, X., Gu, Z.-C. & Wen, X.-G. Classication of gapped symmetric phases in one-dimensional spin systems. Phys. Rev. B 83, 035107 (2011).

24. Fidkowski, L. & Kitaev, A. Topological phases of fermions in one dimension. Phys. Rev. B 83, 075103 (2011).

25. Turner, A. M., Pollmann, F. & Berg, E. Topological phases of one-dimensional fermions: an entanglement point of view. Phys. Rev. B 83, 075102 (2011).26. Hasan, M. Z. & Kane, C. L. Colloquium: topological insulators. Rev. Mod. Phys. 82, 30453067 (2010).

10 NATURE COMMUNICATIONS | 5:3507 | DOI: 10.1038/ncomms4507 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms4507 ARTICLE

27. Qi, X.-L. & Zhang, S.-C. Topological insulators and superconductors. Rev. Mod. Phys. 83, 10571110 (2011).

28. Hasan, M. Z. & Moore, J. E. Three-dimensional topological insulators. Annu. Rev. Condens. Matter Phys. 2, 5578 (2011).

29. Konig, M. et al. Quantum spin hall insulator state in hgte quantum wells. Science 318, 766770 (2007).

30. Hsieh, D. et al. A topological dirac insulator in a quantum spin hall phase. Nature 452, 970974 (2008).

31. Hsieh, D. et al. Observation of time-reversal-protected single-dirac-cone topological-insulator states in Bi2Te3 and Sb2Te3. Phys. Rev. Lett. 103, 146401 (2009).

32. Hsieh, D. et al. Observation of unconventional quantum spin textures in topological insulators. Science 323, 919922 (2009).

33. Schnyder, A. P., Ryu, S., Furusaki, A. & Ludwig, A. W. W. Classication of topological insulators and superconductors in three spatial dimensions. Phys. Rev. B 78, 195125 (2008).

34. Kitaev, A. Periodic table for topological insulators and superconductors. AIP Conf. Proc. 1134, 2230 (2009).

35. Wen, X.-G. Symmetry-protected topological phases in noninteracting fermion systems. Phys. Rev. B 85, 085103 (2012).

36. Fu, L. Topological crystalline insulators. Phys. Rev. Lett. 106, 106802 (2011).37. Teo, Jeffrey C. Y. & Taylor, L. Hughes existence of Majorana-fermion bound states on disclinations and the classication of topological crystalline superconductors in two dimensions. Phys. Rev. Lett. 111, 047006 (2013).

38. Hung, L.-Y. & Wen, X.-G. Quantized topological terms in weak-coupling gauge theories with a global symmetry and their connection to symmetry-enriched topological phases. Phys. Rev. B 87, 165107 (2013).

39. Hung, L.-Y. & Wan, Y. k matrix construction of symmetry-enriched phases of matter. Phys. Rev. B 87, 195103 (2013).

40. Wen, X.-G. Topological invariants of symmetry-protected and symmetry-enriched topological phases of interacting bosons or fermions. Preprint at http://arXiv.org/abs/1301.7675

Web End =http://arXiv.org/abs/1301.7675 (2013).

41. Dasgupta, C. & Halperin, B. I. Phase transition in a lattice model of superconductivity. Phys. Rev. Lett. 47, 15561560 (1981).

42. Fisher, M. P. A. & Lee, D. H. Correspondence between two-dimensional bosons and a bulk superconductor in a magnetic eld. Phys. Rev. B 39, 2756 (1989).

43. Mesaros, A. & Ran, Y. Classication of symmetry enriched topological phases with exactly solvable models. Phys. Rev. B 87, 155115 (2013).

44. Essin, A. M. & Hermele, M. Classifying fractionalization: symmetry classication of gapped z2 spin liquids in two dimensions. Phys. Rev. B 87, 104406 (2013).

45. Lu, Y.-M. & Vishwanath, A. Classication and Properties of Symmetry Enriched Topological Phases: A Chern-Simons approach with applications to Z2 spin liquids. Preprint at http://arXiv.org/abs/1302.2634

Web End =http://arXiv.org/abs/1302.2634 (2013).

46. Yao, H., Fu, L. & Qi, X.-L. Symmetry fractional quantization in two dimensions. Preprint at http://arXiv.org/abs/1012.4470

Web End =http://arXiv.org/abs/1012.4470 (2010).

47. Bahri, Y. & Vishwanath, A. Detecting Majorana fermions in quasi-1D topological phases using non-local order parameters. Preprint at http://arXiv.org/abs/1303.2600

Web End =http:// http://arXiv.org/abs/1303.2600

Web End =arXiv.org/abs/1303.2600 (2013).

48. Nielsen, M. A. Cluster-state quantum computation. Rep. Math. Phys. 57, 147161 (2006).

49. Son, W., Amico, L. & Vedral, V. Topological order in 1d cluster state protected by symmetry. Quant. Inform. Process. 11, 19611968 (2012).

50. Kitaev, A. Anyons in an exactly solved model and beyond. Ann. Phys. 321, 2111 (2006).

Acknowledgements

X.C. would like to thank Xiao-Gang Wen for pointing out the relation between the domain wall construction and the Knneth formula and to thank Ying Ran for pointing out an error in four cocycles. X.C. is supported by the Miller Institute for Basic Research in Science at UC Berkeley. A.V. thanks T. Senthil, Ying Ran, Ehud Altman, Yasaman Barhi, Lukasz Fidkowski and Michael Levin for insightful discussions, and is supported by NSF- DMR 0645691. Y.-M.L. is supported by the Ofce of BES, Materials Sciences Division of the US DOE under contract No. DE-AC02-05CH11231.

Author contributions

X.C. and A.V. came up with the idea of using decorated domain wall to construct SPT phases. X.C. studied the connection of this construction to group cohomology. Y.-M.L. and A.V. studied the eld theoretical description of the phases. X.C., Y.-M.L. and A.V. all contributed to the writing of the manuscript.

Additional information

Supplementary Information accompanies this paper at http://www.nature.com/naturecommunications

Web End =http://www.nature.com/ http://www.nature.com/naturecommunications

Web End =naturecommunications

Competing nancial interests: The authors declare no competing nancial interests.

Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions/

Web End =http://npg.nature.com/ http://npg.nature.com/reprintsandpermissions/

Web End =reprintsandpermissions/

How to cite this article: Chen, X. et al. Symmetry-protected topological phases from decorated domain walls. Nat. Commun. 5:3507 doi: 10.1038/ncomms4507 (2014).

NATURE COMMUNICATIONS | 5:3507 | DOI: 10.1038/ncomms4507 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 11

Word count: 11745

Show less

Abstract

Translate

Symmetry-protected topological phases generalize the notion of topological insulators to strongly interacting systems of bosons or fermions. A sophisticated group cohomology approach has been used to classify bosonic symmetry-protected topological phases, which however does not transparently predict their properties. Here we provide a physical picture that leads to an intuitive understanding of a large class of symmetry-protected topological phases in d=1,2,3 dimensions. Such a picture allows us to construct explicit models for the symmetry-protected topological phases, write down ground state wave function and discover topological properties of symmetry defects both in the bulk and on the edge of the system. We consider symmetries that include a Z₂ subgroup, which allows us to define domain walls. While the usual disordered phase is obtained by proliferating domain walls, we show that symmetry-protected topological phases are realized when these domain walls are decorated, that is, are themselves symmetry-protected topological phases in one lower dimension. This construction works both for unitary Z₂ and anti-unitary time reversal symmetry.

Details

Title

Symmetry-protected topological phases from decorated domain walls

Author

Chen, Xie; Lu, Yuan-ming; Vishwanath, Ashvin

Pages

3507

Publication year

2014

Publication date

Mar 2014

Publisher

Nature Publishing Group

e-ISSN

20411723

Source type

Scholarly Journal

Language of publication

English

DOI

https://doi.org/10.1038/ncomms4507

ProQuest document ID

1510129787

Symmetry-protected topological phases from decorated domain walls

Jump to:

Full Text

Abstract

Details

Suggested sources