Arxiv:1501.07285V3 [Cond-Mat.Str-El] 26 May 2015 Tecting SPT Phases

Topological Phases Protected By Reﬂection Symmetry and Cross-cap States

Gil Young Cho,1, 2 Chang-Tse Hsieh,1 Takahiro Morimoto,3 and Shinsei Ryu1 1Institute for Condensed Matter Theory and Department of Physics, University of Illinois at Urbana-Champaign, 1110 West Green St, Urbana IL 61801 2Department of Physics, Korea Advanced Institute of Science and Technology, Daejeon 305-701, Korea 3Condensed Matter Theory Laboratory, RIKEN, Wako, Saitama, 351-0198, Japan (ΩDated: May 28, 2015) Twisting symmetries provides an efficient method to diagnose symmetry-protected topological (SPT) phases. In this paper, edge theories of (2+1)-dimensional topological phases protected by reflection as well as other symmetries are studied by twisting reflection symmetry, which effectively puts the edge theories on an unoriented spacetime, such as the Klein bottle. A key technical step taken in this paper is the use of the so-called cross-cap states, which encode entirely the unoriented nature of spacetime, and can be obtained by rearranging the spacetime geometry and exchanging the role of space and time coordinates. When the system is in a non-trivial SPT phase, we find that the corresponding cross-cap state is non-invariant under the action of the symmetries of the SPT phase, but acquires an anomalous phase. This anomalous phase, with a proper definition of a reference state, on which symmetry acts trivially, reproduces the known classification of (2+1)-dimensional bosonic and fermionic SPT phases protected by reflection symmetry, including in particular the Z8 classification of topological crystalline superconductors protected by reflection and time-reversal symmetries.

I. INTRODUCTION other hand, gauging (non-spatial) symmetries effectively deconfines a set of quasiparticles (anyons). The fractional statistics of the braiding in the gauged theory can be used Symmetry-protected topological (SPT) phases are 24,27 gapped phases of matter which are not adiabatically to diagnose the original SPT phases. deformable, under a given set of symmetry conditions, Twisting non-spatial unitary symmetries of SPT to a topologically trivial phase. While SPT phases do phases is by now reasonably well-understood. Toward not have an intrinsic topological order (i.e., do not sup- further developments of methodologies to SPT phases, port (deconfined) fractional excitations), they are sharply it is necessary to extend the twisting or gauging pro- (topologically) distinct from topologically trivial states, cedure to spatial and/or antiunitary symmetries, such such as an atomic insulator, which respect the same set of as parity symmetry (P -symmetry) and time-reversal (T - symmetries. In other words, the distinction between SPT symmetry).21–23,26,28 The purpose of this paper is to and trivial phases cannot be made within Landau’s the- provide an efficient and intuitive method to diagnose ory; SPT phases are beyond the classification of phases SPT phases protected by a spatial symmetry such as of matter based on broken symmetries. A flurry of re- parity (reflection). We will study (1+1)-dimensional cent theoretical works includes, among others, the break- [(1+1)d] non-chiral gapless theories with spatial symme- down (or collapse) of the non-interacting classifications tries, which are the edge theories of the corresponding of fermionic SPT phases upon inclusion of interactions, (2+1)d bulk SPT phases protected by parity (P ) symme- non-trivial bosonic SPT phases and the possibility of try or parity combined with a unitary non-spatial sym- symmetry-respecting surface topological order, classifi- metry such as CP -symmetry (parity symmetry combined cation of interacting electronic topological insulators in with charge conjugation). In typical situations, in addi- three dimensions, and proposals for a possible complete tion to P - or CP -symmetry, there are other non-spatial classification of SPT phases. (For a partial and incom- symmetries such as U(1) symmetry or time-reversal (T ) plete list of recent works on SPT phases, see Refs. 1–23.) symmetry. One of the most efficient and powerful methods to Our starting point is Ref. 26, where twisting parity study SPT phases is to twist or gauge symmetries pro- within edge theories of SPT phases (protected by par- 9,24,25

arXiv:1501.07285v3 [cond-mat.str-el] 26 May 2015 tecting SPT phases. Quite generically, global sym- ity) has been used to diagnose the edge theories. This metries in quantum field theories can be twisted, i.e., twisting procedure leads to theories that are effectively can be used to define twisted boundary conditions. It defined on an unoriented manifold, e.g., the Klein bot- was proposed that the twisted theory can be used to di- tle, and thus suggests an interesting link between SPT agnose the original SPT phases, i.e., to judge whether phases and so-called orientifold field theories - type of or not the original theory is symmetry-protected, and theories discussed in unoriented superstring theory.29–38 distinct from topologically trivial phases. More specif- We make one further step in this paper by making use of ically, once twisted, the edge theory of an SPT phase the so-called cross-cap states in formulating orientifold suffers from various kinds of quantum anomalies, such field theories. Cross-cap states are quantum states in as a global anomaly under large U(1) gauge transfor- field theories (conformal field theories) obtained by twist- mations, or a global gravitational anomaly.9,25,26 On the ing parity symmetry and encode, quantum mechanically, 2 the unoriented topology; the fact that the theory is put anomalies. We also draw some parallel between twist- on an unoriented surface is entirely encoded in the cross- ing non-spatial symmetries (so-called orbifold procedure) cap states. They are akin to boundary states in boundary and twisting parity symmetries (orientifold). In Sec. III, conformal field theories, which are obtained by exchang- we apply our strategy to the bosonic SPT phases pro- ing the role of space and time coordinates. One promis- tected by P - or CP -symmetry together with other sym- ing aspect of cross-cap states is that they are formulated metries, and find that the non-invariance of the cross-cap and constructed fully in terms of many-body physics, and states reproduces all the non-trivial SPT phases found in hence are expected to capture the effects of interactions. Ref. 39. In Sec. IV, we discuss real fermionic SPT phases We will identify quantum anomalies in the edge theo- (topological superconductors) protected by parity sym- ries of SPT phases as the non-invariance of a cross-cap metry. By identifying quantum anomalies (anomalous state under the action of the other symmetry than par- phases) in the cross-cap states, the results consistent with ity. Correspondingly, a bulk theory which supports an known microscopic analysis of gapping potentials are ob- anomalous edge theory is diagnosed as a non-trivial SPT tained. phase. Our procedure to diagnose an edge theory with parity symmetry can be summarized as follows: II. CROSSCAP STATES AND SPT PHASES (i): The edge theory of a given bulk theory is put on an unoriented spacetime, the Klein bottle. In practice, this can be achieved by twisting boundary condi- In this section, some generalities of our approach to tions (orientifold procedure). SPT phases are presented. We start by briefly reviewing twisting and gauging non-spatial unitary symmetries of (ii): The edge theory on the Klein bottle is then quan- SPT phases, in particular within (1+1)d edge theories of tized. The cross-cap state corresponding to the (2+1)d SPT phases. When the edge theories are realized twisting is constructed. at the boundary of non-trivial SPT phases, this twisting procedure (orbifolding) reveals a conflict between differ- (iii): In the quantized theory, the effects of other sym- ent symmetries at the quantum level, even when they are metry, such as time-reversal, fermion number par- mutually consistent at the classical level. This is an ex- ity, etc. are studied. When these symmetries are ample of quantum anomalies, signaling the impossibility anomalous, i.e., when the cross-cap state is non- of realizing the edge theory on its own once symmetry invariant under these symmetries, the edge the- conditions are imposed. Hence, it is also indicative of ory cannot be gapped while preserving the symme- the presence of non-trivial bulk states. Twisting unitary tries. The corresponding bulk phase is symmetry- spatial symmetry, such as parity and CP symmetry, can protected and topologically distinct from a trivial be discussed in a similar way, resulting in an unoriented phase. spacetime manifold. Cross-cap states, which are quan- In the previous work,26 the edge theories of SPT phases tum states fully encoding the unoriented nature of the theory, are introduced. A potential conflict of other sym- with P o U(1)A symmetry and those with CP o U(1)V symmetry have been studied. The partition function metries with parity/CP symmetry can be studied in the on the Klein bottle was shown to be anomalous (non- language of cross-cap states. invariant) upon threading a unit flux of the U(1) symmetry. In fact, the partition function acquires the anoma- A. Edge theories of SPT phases lous (−1)-sign, implying the Z2 classification of these SPT phases. The proposed formulation in this paper in terms of cross-cap states reproduces these results for By definition, when going from a SPT phase to a triv- SPT phase with P oU(1)A and CP oU(1)V . In addition, ial phase in a phase diagram by changing parameters in this formulation in terms of cross-cap states offers a few the system’s Hamiltonian, one inevitably encounters a technical advantages. First, the new formulation allows quantum phase transition, if the symmetry conditions us to discuss SPT phases protected by a broader set of are strictly enforced. This in turn implies that if an symmetries than P o U(1)A or CP o U(1)V . That is, SPT phase is spatially proximate to a trivial phase, there we do not need a continuous U(1) symmetry and a large should be a gapless state localized at the boundary be- U(1) gauge transformation. Second, it is not necessary tween the two phases; this critical state can be thought of to compute the full (symmetry-twisted) partition func- as a phase transition occurring locally in space, instead tion; all information necessary to diagnose edge theories of the parameter space of the Hamiltonian. As implied are encoded in cross-cap states. by this construction, the edge state of a non-trivial SPT The rest of the paper is organized as follows. In Sec. phase should never be removable (completely gapped) II, some generalities on twisting/gauging parity symme- if the symmetries are strictly imposed. (It should be tries and finding cross-cap states are presented. We fur- noted, however, that there are other interesting pos- ther demonstrate that non-invariance of the cross-cap sibilities for symmetric surface states for (3+1)d SPT states under symmetry operations is related to the non- phases.15,16,40,41) Hence, this critical boundary state sig- invariance of the partition function, i.e., the quantum nals the topological distinction between the SPT and 3 trivial phases, and many properties of SPT phases can Observe that the boundary condition (??) is invariant be extracted from their boundary physics. For exam- under global U(1)c transformations ple, by inspecting under which symmetry conditions a 2πiα given edge theory is stable/unstable, one can predict un- 0 = ψs(x) − e ψs(x + `) der which symmetry conditions a given phase can be a = eiφQ ψ (x) − e2πiαψ (x + `) e−iφQ, (4) SPT phase. s s Although studying the boundary instead of the bulk where Q is the total charge operator associated to U(1)c reduces the dimensionality of the problem, it is still not symmetry, and φ is an arbitrary real parameter. Simi- straightforward to judge if a given state is topological larly, the boundary condition (??) is invariant under the or not. In principle, one could enumerate all possible iφSz global U(1)s transformation generated by e where symmetry-allowed perturbations within the edge theory, Sz is the total “charge” associated to U(1)s symmetry. which can potentially gap out the edge. Without any A crucial observation now is that, while the boundary guiding principle, however, such brute force approach is condition (??) is invariant under the global U(1)s, the quite cumbersome, and also, more fundamentally, does corresponding ground state may carry an “anomalous” not provide any intuition on the physics of SPT phases. Sz quantum number, Hence it is necessary to have an efficient and illuminating guiding principle for diagnosing topological properties of iQφ i×0×φ e |GSiα = e |GSiα = |GSiα, phases of matter with symmetries. iSz φ i×2α×φ 2iαφ e |GSiα = e |GSiα = e |GSiα. (5)

(The quantum numbers of the ground state can be com- B. Symmetry twist and conflicting symmetries puted explicitly, by using, for example, bosonization and identifying an operator corresponding to the ground Our approach to diagnose non-trivial SPT phases is state |GSiα. More precisely, the operator is given by to identify quantum anomalies within edge theories. ∼ exp(iα(ϕL + ϕR)), where the left- and right-moving We illustrate our strategy by considering a fermionic electrons are identified as ψL/R ∼ exp(±iϕL/R). For de- SPT phase protected by Uc(1) × Us(1) symmetry, where tails, see Ref. 18.) The anomalous phase is nothing but U (1) refers to U(1) symmetry associated with the con- c/s chiral anomaly; in the presence of a U(1)c magnetic flux servation of electromagnetic charge and z-component of twisting the boundary condition, the Sz quantum num- spin Sz, respectively. This phase is akin to (2+1)d time- ber, which is conserved at the classical level, is not con- reversal symmetric topological insulators (the quantum served at the quantum level. Hence, only way to reconcile spin Hall effect), although because of the imposed conser- the U(1)c × U(1)s symmetry and quantum mechanics is vation of Sz, states with Uc(1)×Us(1) symmetry are clas- to realize this (1+1)d theory as a boundary theory of sified by two integral topological invariants (i.e., “charge” a higher-dimensional system, a (2+1)d bulk SPT phase and “spin” Chern numbers). We will focus on the case of respecting the U(1)c × U(1)s symmetry. vanishing charge Chern number, and hence the allowed Let us now be more general. Consider an edge theory, phases are classified by the integer-valued spin Chern which is written in terms of a fundamental quantum field number. Φ(x) and is defined on a spatial circle x ∼ x+`. Suppose Consider the following Hamiltonian describing the there is a set of non-spatial unitary symmetry operators, edge of the Sz-conserving quantum spin Hall phase with G, which leave the edge theory invariant. We consider a U (1) × U (1) symmetry defined on a circle of circumfer- c s twisting boundary condition by a group element g1 ∈ G, ence ` Φ(x + `) = Φ(x) −1 = U · Φ(x), (6) Z ` G1 G1 g1 h † † i H = dx ψ↑(−vi∂x)ψ↑ + ψ↓(+vi∂x)ψ↓ , (1) 0 where G1 is the operator which implements a symmetry operation g1 in the Hilbert space of the edge theory, and where x ∈ [0, `] is the spatial coordinate of the edge, ψ↑/↓ Ug1 is a unitary matrix acting on the internal index of represents the fermion creation operator for up/down the field Φ(x), i.e., a unitary matrix representation of spin, and v is the Fermi velocity. Any global symme- the symmetry. (The field Φ(x) can carry a set of in- try in quantum field theories can be twisted (i.e., can dices representing internal degrees of freedom, which are be used to twist boundary conditions). We choose Uc(1) suppressed in the equation above and in the following.) symmetry to twist the boundary condition as With twisting, states in the Hilbert space, the ground

2πiα state |GSig1 in particular, obey ψs(x) = e ψs(x + `), s ∈ {↑, ↓}. (2)

[Φ(x + `) − Ug Φ(x)] |GSig = 0. (7) Let the ground state with the twisted boundary condition 1 1 be |GSiα, which satisﬁes As a next step, we consider the action of another symmetry g ∈ G.(g can be equal to g , which is the situation 2πiα 2 2 1 ψs(x) − e ψs(x + `) |GSiα = 0, s ∈ {↑, ↓}. (3) relevant to the quantum Hall eﬀect, but in our examples 4 below, g2 6= g1.) At the classical level, the g1-twisted C. Spatial symmetry boundary condition (??) may be invariant under g2 ∈ G, When considering SPT phases protected by spatial

0 = Φ(x + `) − Ug1 Φ(x) symmetries, one can consider twisting the spatial sym- −1 metries. Let us consider a parity symmetry: = G2 [Φ(x + `) − Ug1 Φ(x)] G2 . (8) −1 PΦ(x)P = UP Φ(` − x), (11) When this is the case, one may expect the twisted theory after quantization is invariant under g2 as well. In where the space is defined on a circle x ∼ x+`. Twisting particular, one expects the partition function and/or the by parity symmetry can be introduced in the following ground state of the twisted edge theory is invariant un- way.29,30,34,36–38 der g . When this expectation is betrayed, there is a 2 a. Loop channel We consider Euclidean spacetime quantum anomaly. [0, `] × [0, β] parameterized by (x1, x2), where x1 is the Before leaving this subsection, we comment on a con- spatial coordinate and x2 is the imaginary time coordi- nection between the twisting procedure and orbifold- nate. In the path integral picture, the fields obey the ing/gauging. Gauging and orbifolding have a similar following twisted boundary conditions (identical) effect in that we focus on a gauge singlet (G-invariant) sector of the theory [although the gaug- −1 Φ(x1, x2 + β) = PΦ(x1, x2)P = UP Φ(` − x1, x2), ing means in general imposing the singlet condition lo- −1 cally (e.g., at each site of a lattice), while the projection Φ(`, x2) = G Φ(0, x2)G , (12) in orbifolding is enforced only globally]. Let us again consider an edge theory with symmetry group G. By in which G may be implementing a non-spatial unitary 26 state-operator correspondence in quantum field theories, symmetry g ∈ G, e.g., the fermion number parity. In the operator language, twisting by parity symmetry can corresponding to the state |GSig in the twisted theory, be introduced as a projection operation. This amounts to there is an operator, the twist operator σg, which, when acting on the ground state of the untwisted theory, cre- inserting the parity operator into the partition function, 42–44 ates |GSig : K h −βH (`)i Z = Trg Pe loop , (13)

|GSig = σg(0)|0i. (9) where Hloop ≡ H is the Hamiltonian that generates time- translation in the x -direction. The trace is taken in the (The location of the insertion of the operator σg is taken 2 to be the origin in the radial quantization.) The twist Hilbert space defined for the quantum field obeying the −1 operator, when inserted in correlation functions, imple- boundary condition Φ(x1+`) = G Φ(x1)G , as indicated ments the symmetry twist by g, and satisfy the following by the subscript g. (For later use, we also introduce 0 0 −1 algebraic relation with Φ(x) on the complex plane: another parity operator, P , such that P · P = G .) This representation of the partition function with the choice of x as a direction of time-evolution will be called Φ(z, z¯)σ (w, w¯) = σ (w, w¯)U · Φ(z, z¯). (10) 2 g g g the ”loop channel” picture. b. Tree channel As we have seen in the case of non- Starting from the original, untwisted, theory, one can spatial symmetries, it is useful to discuss the twisted par- now consider including the ground states with twisted tition function in terms of the twist operator and the boundary conditions to define an extended theory. In the corresponding state. One may want to develop a similar extended theory, the ground states with twisted bound- alternative picture for the case of twisting by parity. To ary conditions (the ground states in the “twisted sec- this end, one needs to find a convenient and proper “time- tors”), and hence the corresponding twist operators, are slice” that allows us to define a quantum state (|GSig in considered as an excitation. This procedure to generate the notation of the previous section for the non-spatial a new theory from the untwisted theory is called orbifold. symmetry), which implements the twisting condition. Now, by further invoking bulk-boundary correspon- It is convenient to recall that any 2d compact unori- dence, there is a corresponding bulk excitation (anyon). ented surface without boundary can be generated from a Bulk statistical properties of the gauged theory can be sphere S2 by adding handles and cross-caps, which can read off from the operator product expansions and fu- be thought of as a real projective plane RP2. In partic- sion rules obeyed by the twist operator(s). Now a given ular, the Klein bottle can be generated by first cutting symmetry group G can be implemented in various differ- out two discs from S2, and then identifying antipodal ent ways in different SPT (and trivial) phases protected points of the resulting holes. In fact, one can rearrange by G leading to different choices of Ug, and to differ- the path integral on the Klein bottle such that the fields ent twist operators. By studying statistical (braiding) are now defined on [0, `/2] × [0, 2β]. The precise steps for properties of the twist operator(s), one can distinguish rearranging the spacetime is described in Fig.1. 24 different ungauged (original) theories. We use (σ1, σ2) to represent this rearranged spacetime. 5

(a) (b) x2 where Htree is the Hamiltonian that generates time- x2 A’ C’ translation in the tree channel picture. The loop channel- 2𝛽𝛽 tree channel duality (also known as open-closed duality)

D’ asserts that the partition functions computed in the loop B A’ B’ 𝛽𝛽 C B C and tree channels agree. 𝛽𝛽 c. Interplay with non-spatial symmetries By con-

D D structing the cross-cap states, we have “gauged” parity A B’ x1 (c) x1 A /2 /2 symmetry. As in the case of the SPT phases with non- x2 (d) spatial symmetries, we now consider the eﬀects of an- A’𝑙𝑙 C’𝑙𝑙 𝑙𝑙 C’ 𝑙𝑙 𝟐𝟐 A’ 2𝛽𝛽 𝝈𝝈 other non-spatial symmetry, represented by g ∈ G, on 2𝛽𝛽 the cross-cap states. We act with g on the cross-cap

B’ D’ B’ D’ B boundary conditions, 𝛽𝛽 C B C 𝛽𝛽 [Φ(0, σ2 + β) − UP Φ(0, σ2)] |CP i = 0 −1 D D ⇒ [Φ(0, σ + β) − U Φ(0, σ )] |C i = 0 A x1 A G 2 P 2 G G P /2 /2 ⇒ [UgΦ(0, σ2 + β) − UP UgΦ(0, σ2)] G |CP i = 0 𝝈𝝈𝟏𝟏 𝑙𝑙 𝑙𝑙 𝑙𝑙 −1 FIG. 1. Rearrangement of spacetime. (a) The original loop ⇒ Φ(0, σ2 + β) − Ug UP Ug Φ(0, σ2) G |CP i = 0. channel (Euclidean) spacetime. Here the points (A, B) and (18) (A0,B0) in the spacetime are identified due to the boundary conditions Eq. (??). The line segments with the same symbols Thus we deduce the following relation: are also identified by the boundary conditions Eq. (??). (b) We shift the box of the spacetime formed by A0B0D0C0 along G |CP i = |Cg·P ·g−1 i. (19) x2 by β. (c) After shifting the box, we flip the orientation of If the cross-cap condition is invariant under g ∈ G, the box with respect to x1 = 3`/4. (d) We slide the box along −1 UP = Ug UP Ug , then we may expect that so is the x1 by `/2. Then we obtain the crosscap geometry. Notice that cross-cap state, |CP i = G |CP i, classically. However this any x2 ∈ [0, β] is identified with its antipodal point at the slice expected invariance may be broken down quantum me- of “time” at x1 = 0 and x1 = `/2. By renaming the variables x1 → σ1 and x2 → σ2, we obtain the spacetime of the tree chanically. This then signals that the theory is anoma- channel. Here the σ1 direction is taken to be a direction lous and should describe the edge of a SPT phase defined of (fictitious) time-evolution. At the boundary of spacetime in one higher dimension. located at σ1 = 0 and σ1 = `/2, there are cross-caps.

III. BOSONIC SPT PHASES The fields now obey the cross-cap boundary conditions In this section, we apply our strategy above to edge Φ(0, σ2 + β) = UP 0 Φ(0, σ2), theories of bosonic SPT phases consisting of a single component non-chiral boson (i.e., a two-component chiral bo- Φ(`/2, σ2 + β) = UP Φ(`/2, σ2), son with 2 × 2 K-matrix). We consider situations where 0 0−1 Φ(σ1, σ2 + 2β) = G Φ(σ1, σ2)G , (14) parity (P ) or a combination of parity and charge conjugation (CP ) is a part of the full symmetry group, which can where potentially protect the edge theory from gap-opening. P2 = P02 = G 0. (15) In Ref. 26, SPT phases protected by P or CP symmetry together with a continuous U(1) symmetry were ◦ By further rotating (σ1, σ2) coordinates by 90 , we can studied by using a generalization of Laughlin’s argument. take σ1 as a time-coordinate. In this coordinate system, It was found that non-trivial Z2 bosonic SPT phases can the time-evolution is generated by a (fictitious) Hamil- exist in the presence of the following combinations of tonian (denoted by Htree in the following). The fact symmetries: that the system is defined on an unoriented surface is • P U(1)A (“spin U(1)”), now encoded in boundary conditions at σ1 = 0 and o σ = `/2. We then introduce cross-cap states which obey 1 • CP o U(1)V (“charge U(1)”), these boundary conditions as: where the subscript A/V represents the “axial/vectorial” [Φ(0, σ2 + β) − UP 0 Φ(0, σ2)] |CP i = 0, nature of the U(1) symmetry.

[Φ(`/2, σ2 + β) − UP Φ(`/2, σ2)] |CP 0 i = 0. (16) In Ref. 39, following the spirit of Refs. 17 and 45, a microscopic stability analysis for the bosonic edge theory The partition function can be written in terms of the is carried out by enumerating possible gapping potentials cross-cap states as in the presence of various discrete symmetries. It was

K − ` H (2β) found that the bosonic edge theory can be ingappable Z = hCP 0 |e 2 tree |CP i, (17) (protected) in the presence of the following symmetries: 6

• P × C,P × T given by 1 k 1 R • P × TC, CP × T , T × C. p = (p + p ) = , p˜ = (p − p ) = w, 2 L R R 2 L R α0 SPTs protected by these symmetries are all classified by k R k R p = + w, p = − w, (25) Z2. In addition, it was also found that there are SPT L R α0 R R α0 phases protected by where k and w are an integer. In terms of these momen- • T × C × P tum eigenvalues, the compactification conditions on the boson fields are 4 The classification of these SPT phases is Z2. 0 In the following, we will study these SPT phases by us- ϕL(x + `) − ϕL(x) = +πα pL, ing cross-cap states and by identifying quantum anoma- 0 ϕR(x + `) − ϕR(x) = −πα pR, lies. As we will show, this analysis reproduces exactly the φ(x + `) − φ(x) = πα0(p − p ) = 2πRw, same classification as in Ref. 39, and hence it gives us a L R α0 perspective complimentary to the generalized Laughlin’s θ(x + `) − θ(x) = πα0(p + p ) = 2π k. (26) argument on the Klein bottle in the “loop channel” pic- L R R ture, and to microscopic stability analysis. The Hilbert space is constructed as a tensor product of the bosonic oscillator Fock spaces, each of which generated by pairs of creation and annihilation operators A. Free compactified boson {αm, α−m}m>0 and {α˜m, α˜−m}m>0, and the zero mode sector associated to xL,R and pL,R. We will denote states We start from the free boson theory on a spatial ring in the zero mode sector by specifying their momentum of circumference ` defined by the partition function Z = eigenvalues as R D[φ] exp(iS) with the action 0 |p, p˜i = |k/R, Rw/α i, k, w ∈ Z, (27) 1 Z Z ` 1 2 2 or more simply as |k, wi. Alternatively, the Fourier trans- S = 0 dt dx (∂tφ) − v(∂xφ) , (20) 4πα 0 v formation of the momentum eigenkets defines the “position” eigenkets, which we denote by where the spacetime coordinate of the edge theory is de- 0 0 noted by (t, x), v is the velocity, α is the coupling con- |φ0, θ0i 0 < φ0 ≤ 2πR, 0 < θ0 ≤ 2πα /R. (28) stant, and the φ-field is compactified as The two basis are related by φ ∼ φ + 2πR, (21) Z 2πR Z 2πα0/R −ipφ0−ipθ˜ 0 |p, p˜i = dφ0 dθ0e |φ0, θ0i. (29) with the compactification radius R. The canonical com- 0 0 mutation relation is

0 0 X 0 [φ(x, t), ∂tφ(x , t)] = i2πα v δ(x − x − n`). (22) B. Symmetries n∈Z Various symmetries of the single-component compact- We use the chiral decomposition of the boson field, and ified boson theory are listed below. introduce the dual field θ as d. U(1) × U(1) symmetry In the free boson theory, φ = ϕ + ϕ , θ = ϕ − ϕ . (23) when there is no perturbation, there are two conserved L R L R U(1) charges, one for each left- and right-moving sector, The mode expansion of the chiral boson fields is given by defined by ± (x = vt ± x) Z ` 0 NL,R = dx ∂xϕL,R = α πpL,R, (30) n6=0 0 + r 0 + + 0 x α X αn − 2πinx ϕ (x ) = x + πα p + i e ` , L L L ` 2 n They satisfy n∈Z 0 n6=0 [ϕ ,N ] = [ϕ ,N ] = α πi. (31) − r 0 − L L R R − 0 x α X αñ − 2πinx ϕ (x ) = x + πα p + i e ` , R R R ` 2 n Correspondingly to these conserved quantities, the free n∈Z (24) boson theory is invariant under the following U(1)×U(1) symmetry where [αm, α−n] = [˜αm, α˜−n] = mδmn and [xL, pL] = Uδφ,δθ : φ → φ + δφ, θ → θ + δθ, [xR, pR] = i. The compactification condition on the boson fields implies the allowed momentum eiganvalues are : ϕL → ϕL + δϕL, ϕR → ϕR + δϕR. (32) 7

In terms of the conserved charges, the generators of the C. Crosscap States U(1) × U(1) transformations are given by 0 We now move on to the tree channel picture and con- U L = eiδϕLNL/(α π) = eiδϕLpL , δϕL struct cross-cap states by twisting P - and CP - symme- 0 R = eiδϕRNR/(α π) = eiδϕRpR , UδϕR tries. After rearranging spacetime and exchanging the L R i(δφp+δθp˜) role of space and time, the spacetime is, space × time = Uδφ,δθ = U U = e . (33) δϕL δϕR 2β × `/2. We parameterize this Euclidean spacetime by Note also that Uδφ,δθ acts on the momentum eigenkets (σ1, σ2). The original spacetime of [0, `]×[0, β] is param- as eterized by (x1, x2) (for the mapping between the two i(pδφ+˜pδθ) spacetime see Fig. 1). Uδφ,δθ|p, p˜i = e |p, p˜i. (34) e. C-symmetry Particle-hole symmetry or charge conjugation (C-symmetry) is unitary and acts on the 1. P -symmetry bosonic fields as 0 We first consider the cross-cap state obtained by twist- mcπα C : φ → −φ + ncπR, θ → −θ + R ing parity symmetry, defined in Eq. (??). The corresponding cross-cap states are defined by :(x1, x2) → (x1, x2), (35) where (nc, mc) ∈ {0, 1}. From these transformation laws φ(σ2) − φ(σ2 + β) − npπR |Cp(np, mp)i = 0, of the boson fields, we read off the action of C-symmetry 0 on the position basis as mpπα θ(σ2) + θ(σ2 + β) − |Cp(np, mp)i = 0. (41) iδ 0 R C |φ0, θ0i = e |−φ0 + ncπR, −θ0 + mcπα /Ri , (36) By mode-expansion, the cross-cap condition (??) trans- iδ where e is an unknown phase factor. In order to have lates into the corresponding condition for each mode. To the relation C |p, p˜i ∝ | − p, −p˜i, expected from the com- find an explicit form of the cross-cap states, we first focus mutation relation between C and p, p˜, the phase δ has to on the zero mode sector of the boson fields: be a constant (independent of φ0 and θ0). The action of 0 πα pσ˜ 2 C-symmetry on the momentum eigenstates is given by φ(σ2) = xL + xR + + ··· , 0 β mcπα iδ −ipncπR−ip˜ 0 C |p, p˜i = e e R | − p, −p˜i πα pσ2 θ(σ2) = xL − xR + + ··· . (42) = eiδe−iπknc−iπwmc | − p, −p˜i, (37) β where p = k/R andp ˜ = wR/α0. Since δ is constant, Within the zero mode sector, we solve the cross-cap con- the phase ambiguity is fixed once we specify the action ditions (??). The first condition (??) can be reduced to 0 of C on a reference state, e.g., |p, p˜i = |0, 0i. In our anal- πα p˜ − npπR = 0 mod 2πR ysis presented below, the reference state and its charge R conjugation parity eiδ plays an important role. ⇒ p˜ = (2N + n ),N ∈ . (43) α0 p Z f. T -symmetry Antiunitary time-reversal operator T acts on the boson fields as Similarly, the second condition can be solved as: 0 0 0 0 mT πα 2πα pσ2 0 mpπα 2πα T : φ → φ + nT πR, θ → −θ + 2(xL − xR) + + πα p = mod R β R R :(x , x ) → (x , −x ), (38) πα0 1 2 1 2 ⇒ p = 0, (x − x ) = (2N˜ + m ) , N˜ ∈ . (44) L R p 2R Z where nT , mT ∈ {0, 1}. g. P -symmetry Parity P is defined by Solving the conditions (??) and (??), we find the cross- 0 cap state |Cp(np, mp)i in terms of the momentum eigen- mpπα P : φ → φ + npπR, θ → −θ + ket {|p, p˜i} as R X mpN :(x1, x2) → (` − x1, x2), (39) |Cp(np, mp)i = (−1) |0, 2N + npi. (45) N∈Z where np, mp ∈ {0, 1}. h. CP -symmetry The above symmetries can be The full cross-cap state is obtained by including the parts combined. For example, CP -symmetry is a non-local related to the oscillatory modes: unitary symmetry, and defined by " ∞ n # q √ X (−1) 0 R 2 exp − α−nα˜−n |Cp(np, mp)i. (46) mcpπα n CP : φ → −φ + ncpπR, θ → θ + n=1 R For our purpose of diagnosing SPT phases, however, it :(x , x ) → (` − x , x ), (40) 1 2 1 2 turns out that it is enough to focus on the zero-mode where ncp, mcp ∈ {0, 1}. sector. 8

2. CP -symmetry which is invariant when 2πα0 Next we consider CP -symmetry. The corresponding 2δθ = 0 mod . (52) R cross-cap conditions are given by Thus, at least classically, we expect the theory to be invariant when δθ = πα0/R. On the other hand, this invari- φ(σ ) + φ(σ + β) − n πR |C (n , m )i = 0, 2 2 cp cp cp cp ance may not be maintained at the quantum level. The A m πα0 cross-cap state may not be invariant under Uδθ=πα0/R θ(σ ) − θ(σ + β) − cp |C (n , m )i = 0. 2 2 R cp cp cp and can pick up an anomalous phase; one can easily check

A np (47) Uπα0/R|Cp(np, 0)i = (−1) |Cp(np, 0)i. (53) By solving these conditions within the zero mode sector, Thus the edge theory, when enforced parity symme- we obtain, as the cross-cap state, try with np = 1, is anomalous for U(1)A symmetry. This invariance/non-invariance is equivalent to the X ncpN |Ccp(ncp, mcp)i = (−1) |2N + mcp, 0i. (48) invariance/non-invariance of the Klein bottle partition N∈Z function under large gauge transformations discussed in Ref. 26. Observe also that the anomalous phase here is Z2-valued (i.e., a sign), and this implies that the classifi- D. Diagnosis of SPT phases cation is Z2 since the two copies of the theory is trivial. This agrees with the loop channel calculation.26,39 1. P o U(1)A and CP o U(1)V symmetries SPT phases protected by CP o U(1)V can be discussed in the same manner as P o U(1)A. The cross-cap We start with the case of the symmetry group P o state obtained by twisting CP is given by |Ccp(0, mcp)i U(1)A, which has been studied in Ref. 26 by using a in Eq. (??). The U(1)V symmetry is generated by V generalization of Laughln’s gauge argument. In the gen- Uδφ = exp(ipδφ) as eralized Laughlin’s argument, the edge theory is put on V an unoriented spacetime, such as the Klein bottle, and Uδφ : θ → θ, φ → φ + δφ. (54) then the invariance under flux threading (a large gauge Under this symmetry action, the cross-cap condition (??) transformation of U(1)A symmetry) of the partition function of the edge theory is investigated. More specifically, with ncp = 0 is invariant when the analysis in Ref. 26 is performed in the loop-channel 2δφ = 0 mod 2πR. (55) channel picture. In the following, we will reproduce this result in terms of the tree-channel calculations, i.e., by On the other hand, we find using the cross-cap state. V mcp The relevant cross-cap state is |Cp(np, 0)i presented in Uδφ=πR|Ccp(0, mcp)i = (−1) |Ccp(0, mcp)i. (56) Eq. (??). We set mp = 0 since np ∈ {0, 1}, mp = 0 Thus the edge theory with parity symmetry with mcp = 1 is enough to classify the P o U(1)A-symmetric SPT 26,39 phases according to Ref. 39. However, it is straightfor- and U(1)V is anomalous. Here the anomalous phase ward to generalize the analysis below to more general is the Z2 sign, and thus classification is Z2 as in the previous case. sets of {np, mp}. The U(1)A symmetry is generated by A The above analysis in terms of the cross-cap states is a U = exp(ipδθ˜ ): δθ reformulation of Ref. 26; while in Ref. 26 the Klein bot- U A : θ → θ + δθ, φ → φ. (49) tle partition functions are computed in the loop channel δθ picture, here we have considered and studied the cross- A cap states in the tree channel. A few technical remarks Let us act with Uδθ on the cross-cap condition (??). The cross-cap condition for φ is trivially invariant under A. are in order. (i) We have considered adiabatic trans- Uδθ A For the condition written in terms of the dual field θ formations of the cross-cap states by acting with Uδθ or V and by continuously changing δθ or δφ, respec- (with mp = 0), Uδφ tively. In terms of the original, loop channel picture, A A −1 A these twisting parameters should appear as a twisting Uδθ {θ(σ2) + θ(σ2 + β)} (Uδθ ) Uδθ |Cp(np, mp)i = 0 A angle in twisting boundary conditions in the spatial di- ⇒ {θ(σ2) + θ(σ2 + β) + 2δθ} Uδθ |Cp(np, mp)i = 0. rection. This can be seen explicitly as follows. By using (50) the formula G |CP i = |CgP g−1 i, the partition function in the tree channel can be written as, Thus, the cross-cap condition is transformed into ` K − 2 Htree A Z = hCP 0 |e Uδθ |CP i θ(σ2) + θ(σ2 + β) = 0 ` − Htree = hCP 0 |e 2 |C A A i. (57) ⇒ θ(σ2) + θ(σ2 + β) + 2δθ = 0, (51) Uδθ ·P ·U−δθ 9

This can be written in the loop channel as to form the cross-cap state |Ccp, (np, mp)i and then study the action of C on the cross-cap: K −βHloop Z = Tr 0 A −1 A e (58) P Uδθ P U−δθ 0 iδ npmp C |Ccp(np, mp)i = e (−1) |Ccp(np, mp)i, (61) By noting U A P −1U A = P −1U A , and taking P 0 = P , δθ −δθ −2δθ 0 where eiδ is an overall phase factor, which, as in Eq. K −βHloop Z = Tr A e , (59) U−2δθ (??), cannot be determined. This again generates the same classification as Eq. (??). i.e., 2δθ (not δθ) appears, in the loop channel, as a twist- This situation should be contrasted with the cases with ing angle for a spatial boundary condition. When 2δθ a continuous U(1) symmetry; in the latter case, one can is an integer multiple of 2π, the twisted system is large compare the phase acquired by the cross-cap state when gauge equivalent to the system without twisting. V V acting with Uδθ and Uδθ+πR; One can follow the evolu- (ii) The current analysis in the tree-channel picture V tion of U |Ccp(ncp, mcp)i adiabatically. In the present has a few advantages over the loop-channel calculations. δθ case, because of the discrete nature of C-transformation, First of all, in Ref. 26, we rely heavily on the exis- such comparison is not possible. tence of a continuous U(1) symmetry (either U(1)A or U(1)V ) as in Laughlin’s thought experiment in the quantum Hall effect. Therefore, it is not entirely obvious how 3. P × T symmetry the methodology in Ref. 26 can be generalized to SPT phases which lack continuous (U(1)) symmetries. The We now consider the cases where we have both P/CP reformulation in the tree channel and in term of cross- and T/CT symmetries. As in the case of P × C, we can cap states, however, indicates a natural way to deal with first twist P/CP to obtain the corresponding cross-cap SPTs without U(1) symmetry. In fact, the procedure de- state. The action of the remaining symmetry, T/CT , on scribed above can be generalized to cases without U(1) the cross-cap state can be studied. However, unlike C- symmetry. (See the following sections.) Second, in the symmetry, which is a non-spatial unitary symmetry, how current reformulation in the tree channel, there is no need T/CT symmetry acts in the tree-channel is non-trivial. to compute the full partition function, although it is pos- We illustrate this point for the case of P × T symmetry. sible to compute the partition function by using cross-cap Let us consider the bosonic edge theory in the pres- states. ence of P -symmetry with mp = 0 and T -symmetry with nT = 0. We twist P -symmetry and consider the cor- 2. P × C symmetry responding cross-cap state |Cp(np, 0)i in Eq. (??). One can check the cross-cap condition is left invariant under T -symmetry: the fields obey, in the loop channel, the To demonstrate that our methodology works in the ab- following boundary conditions: sence of a continuous U(1) symmetry, we now consider SPT phases protected by P ×C. Here the relevant cross- φ(x1, x2 + β) ≡ φ(` − x1, x2) + npπR, cap state is |Cp(np, mp)i in Eq. (??). We consider C- φ(x1 + `, x2) ≡ φ(x1, x2), symmetry (??) with nc = mc = 0. One can check easily that under the action of C-symmetry, the cross-cap con- θ(x1, x2 + β) ≡ −θ(` − x1, x2), dition (??) is invariant. On the other hand, the cross-cap θ(x1 + `, x2) ≡ θ(x1, x2). (62) state may not, as one can check explicitly as Under time-reversal, these conditions are transformed iδ npmp C |Cp(np, mp)i = e (−1) |Cp(np, mp)i, (60) into: by using Eq. (??). To judge if the cross-cap state is φ(x1, −x2) ≡ φ(` − x1, β − x2) + npπR, anomalous or not, we need to know the overall phase φ(x1 + `, β − x2) ≡ φ(x1, β − x2), eiδ, which originates from our ignorance on the charge πα0 πα0 conjugation parity of the zero mode sector. Depending − θ(x , −x ) + m ≡ θ(` − x , β − x ) − m , 1 2 T R 1 2 T R on our choice of δ, either one of phases with (np, mp) = 0 0 (0, 0), (0, 1), (1, 0) or (n , m ) = (1, 1) is anomalous. A πα πα p p − θ(x1 + `, β − x2) + mT ≡ −θ(x1, β − x2) + mT . natural choice would be eiδ = 1, as this means we as- R R sign the charge conjugation parity +1 to the reference (63) state |p, p˜i = |0, 0i. With this choice, the edge the- For example, we can derive the first line of Eq.(??) from ory with the symmetry group P × C is anomalous when the first line of Eq.(??) by following steps. (np, mp) = (1, 1): this result is consistent with the microscopic analysis given in Ref. 39. The phase acquired by φ(x1, x2 + β) ≡ φ(` − x1, x2) + npπR, the cross-cap state under the action of C-symmetry is Z2 ↔ φ(x1, y2) ≡ φ(` − x1, y2 − β) + npπR, (i.e., sign), and hence the classification is . Further- Z2 → φ(x , −y ) ≡ φ(` − x , −y + β) + n πR, more, one may twist CP symmetry, which can be gener- 1 2 1 2 p ated by the multiplication of C and P , i.e., CP = C ×P , ↔ φ(x1, −x2) ≡ φ(` − x1, β − x2) + npπR, (64) 10 where we renamed the variable y2 = x2 + β inbetween Thus CPg or the original time-reversal symmetry is the first and the second lines, and we acted T -symmetry anomalous. This result is consistent with the gapping inbetween the second and the third lines. With the com- potential analysis in Ref. 39. pactification conditions, these are equivalent to In Appendix A, we present identification of quantum anomalies using cross-cap states for other bosonic SPT φ(x1, −x2) ≡ φ(` − x1, β − x2) + npπR, phases studied in Ref. 39. φ(x1 + `, β − x2) ≡ φ(x1, β − x2), θ(x , −x ) ≡ −θ(` − x , β − x ), 1 2 1 2 IV. REAL FERMIONIC SPTS θ(x1 + `, β − x2) ≡ θ(x1, β − x2), (65) In this section, we discuss edge theories of (2+1)d topo- which, with mere relabeling β − x2 → x2, are equivalent to the original boundary conditions. logical superconductors in the presence of discrete sym- Since T -transformation leaves the P -twisted boundary metries. In particular, we consider topological supercon- condition invariant, we expect the cross-cap state is also ductors with reflection symmetry in symmetry classes invariant under T -transformation, at least up to a phase D+R+ and BDI + R++, and their CPT partners dis- factor. (Recall that we are after this possible anomalous cussed in Refs. 46–48, At quadratic level, i.e., in the ab- phase of the cross-cap state.) To compute this phase, sence of interactions, topological classification for these we need to know how T -transformation looks like in symmetry classes are found to be Z2 and Z, respectively. the tree channel. In the tree channel, T -transformation For the latter, it was found in Ref. 11, the Z classifi- should look like a parity transformation (in fact, CP - cation collapses into Z8 in the presence of interactions. transformation, since T flips the sign of φ): Our discussion in this section can trivially be extended to other fermionic SPT phases supporting complex (Dirac) fermion edge modes. φ(σ1, σ2) → −φ(σ1, 2β − σ2), πα0 θ(σ1, σ2) → θ(σ1, 2β − σ2) + mT . (66) R A. Majorana fermion edge states This CP -transformation in the tree-channel picture, obtained from T -transformation in the loop-channel pic- Consider an edge theory of a (topological) supercon- ductor consisting of N flavors of non-chiral real (Majo- ture, is denoted by CPg in the following. If the phase f field φ and its dual θ obey the standard commutation rana) fermions described by the Hamiltonian relation, this CPg-transformation must be unitary, i.e., Nf Z ` X a a a a it must preserve, in particular, the Heisenberg algebra H = dx [ψL(−vi∂x)ψL + ψR(+vi∂x)ψR] . (70) obeyed by the zero modes, [xL, pL] = [xR, pR] = i. If a=1 0 it were defined in the original coordinates (x , x ), this 1 2 The fermi velocity v is set to be identical for all fermion unitary CPg-transformation is CPT-dual of T -symmetry. flavors for simplicity. The fermion fields obey the canon- Since T -transformation in the loop channel preserves ical anticommutation relations the boundary condition, this should be so in tree channel a b 0 ab X 0 as well. The cross-cap conditions {ψL(x), ψL(x )} = 2πδ δ(x − x + `m), m∈Z [φ(σ2) − φ(σ2 + β) − npπR] |Cp(np, 0)i = 0, a b 0 ab X 0 {ψR(x), ψR(x )} = 2πδ δ(x − x + `m). (71) [θ(σ ) + θ(σ + β)] |C (n , 0)i = 0, (67) 2 2 p p m∈Z The (1+1)d non-chiral fermionic edge theory (??) can be are transformed into, by the CP transformation, g realized at the edge of (2+1)d topological superconductors in symmetry classes DIII, D+R and BDI + R . [−φ(2β − σ ) + φ(β − σ ) − n πR] |C (n , 0)i = 0, + ++ 2 2 p CPg p p The fermionic edge theory (??) is invariant under the h πα0 θ(2β − σ ) + m following three symmetries: 2 T R (i) The fermion number parity conservation, where the πα0 i fermion parity operator is given by + θ(β − σ2) + mT CPg |Cp(np, 0)i = 0. (68) R Nf F X Due to the compactification condition, these cross-cap Gf = (−1) ,F = Fa, (72) conditions are equivalent to the original conditions. a=1

While the crosscap conditions are preserved by CPg, the where Fa is the total fermion number operator for the cross-cap state |Cp(np, 0)i (??) is not invariant when a-th ﬂavor, n = m = 1, p T Z ` 1 a a Fa = dx iψLψR. (73) mT np CPg |Cp(np, 0)i = (−1) |Cp(np, 0)i. (69) 2π 0 11

The fermion number parity conservation is the most fun- Class T P Classification 2 damental symmetry of any fermionic system. In the fol- DIII T = Gf None Z2 lowing, the fermion number parity conservation is as- BDI T 2 = 1 None 0 2 sumed to be always unbroken (at least classically – it D+R+ None P = 1 Z2 may however be anomalous). 2 D+R− None P = −1 0 (ii) Time-reversal symmetry. We consider two-kinds of 2 2 DIII+R−− T = Gf P = −1 Z → Z8 time-reversal: one which squares to +1: 2 2 BDI+R++ T = 1 P = 1 Z → Z8 −1 T ψL(t, x)T = ψR(−t, x), TABLE I. Edge theories of (2+1)d topological superconduc- −1 T ψR(t, x)T = ψL(−t, x), tors with various time-reversal and parity symmetries studied in Sec. IV. T 2 = 1, T iT −1 = −i, (74) and the other which squares to −1: B. Cross-cap states −1 T ψL(t, x)T = ψR(−t, x), −1 a. Cross-cap condition In both symmetry classes T ψR(t, x)T = −ψL(−t, x), D+R+ and DIII+R−−, parity P and fermion number 2 −1 T = Gf , T iT = −i. (75) parity Gf are conserved. Hence, theories in these symmetry classes can be twisted by parity P, and the combi- (The flavor index is suppressed.) nation of fermion number parity and spatial parity Gf P. (iii) Parity symmetries. Similarly to time-reversal, we In the fermionic edge theory (??), twisting boundary con- consider two-kinds of parities: one which squares to +1: ditions in time direction by P or Gf P leads to the fol- −1 lowing conditions on the fermion fields: PψL(t, x)P = ψR(t, ` − x), −1 PψR(t, x)P = ψL(t, ` − x), ψL(x1, x2 + β) = η1ψR(` − x1, x2), 2 UP = σx, P = 1, (76) ψR(x1, x2 + β) = η2ψL(` − x1, x2). (78) and the other which squares to −1: Here, η1,2 are given by

ψ (t, x) −1 = ψ (t, ` − x),  2 P L P R  (+, +) twisting by P with P = 1 −1  2 PψR(t, x)P = −ψL(t, ` − x).  (−, −) twisting by Gf P with P = 1 (η1, η2) = 2 2 UP = iσy, P = −1. (77)  (+, −) twisting by P with P = −1  2  (−, +) twisting by Gf P with P = −1 Note that the phase of parity (and hence parity squared) (79) is arbitrary for systems of complex fermions whereas there is no such arbitrariness for real fermions. In the rearranged geometry (see Fig. 1), these twisted Observe that T 2 = −1 and P2 = 1 are CPT conju- boundary conditions are given by gate to each other. Similarly, T 2 = 1 and P2 = −1 are CPT conjugate to each other. (These can easily be in- ψL(σ1, σ2) = η1ψR(σ1, σ2 + β), ferred from the fact that they both prohibit the same kind ψR(σ1, σ2) = η2ψL(σ1, σ2 + β). (80) of masses.) In the absence of time-reversal and parity symmetries, the non-chiral fermion edge state can easily By making a 90 degree rotation in the (σ1, σ2) plane, be gapped by adding a mass term. To realize a non- 0 0 (σ1, σ2) → (σ1, σ2) = (σ2, −σ1), we exchange the role trivial SPT phase, we need to impose T , P, or both T of time and space coordinates and regard σ1 direction and (see Table I). Imposing 2 = −1 alone leads to 0 P T as a fictitious time direction. The fermion fields ψL,R Z2 topological classification (class DIII) while imposing with respect to the rotated coordinate system (σ0 , σ0 ) 2 1 2 T = 1 alone does not give rise to a SPT. Similarly, im- are related to the original fermion fields as 2 posing P = 1 alone leads to Z2 topological classification (class D + R ) while imposing 2 = −1 alone does not 0 iπ/4 0 −iπ/4 + P ψL = e ψL, ψR = e ψR. (81) 2 give rise to a SPT (class D+R−). Imposing T = −1 2 together with P = −1 (class DIII + R−−) gives rise to The cross-cap states are defined by the cross-cap condi- Z8 topological distinction. Similarly, its CPT conjugate, tions 2 2 P = 1 together with T = +1 (class BDI + R++) gives rise to Z8 topological distinction. In the following, [ψL(σ2) − iη1ψR(σ2 + β)] |C(η1, η2)i = 0, our task is to understand these Z2 (class D+R+) and [ψR(σ2) + iη2ψL(σ2 + β)] |C(η1, η2)i = 0, (82) Z8 (class BDI+R++) classifications of SPTs in terms of quantum anomalies. We will apply our formalism using where for notational simplicity we have removed 0 and 0 0 cross-cap states. simply write ψL,R → ψL,R and σ1,2 → σ1,2. 12

A crucial observation is that for the cross-cap condi- and the Fock vacuum |0f i of the f-fermions. The cross- 2 tions generated by P with P = +1, there are Majo- cap state |C, η = +i is then nothing but |0f i itself, rana fermion zero modes in the mode expansion of the iφ+ fermion fields, while there are no Majorana zero modes |C, η = +i = e |0f i. (86) when P2 = −1. This follows from the general formula 2 On the other hand, the cross-cap state |C, η = −i can be (??), which suggests that when P = 1(−1) the fermion constructed as fields obey periodic (anti-periodic) boundary condition. Nf (This can also be understood by first assuming the exis- Y |C, η = −i = eiφ− f †|0 i. (87) tence of zero modes, ψL0 and ψR0 for the left- and right- a f moving fermion fields, respectively. Then, the cross-cap a=1 condition tells us ψL0 −iη1ψR0 = 0 and ψR0 +iη2ψL0 = 0. [N.B. In the above representation of |C, ηi, the ambiguous With η = −η , these two conditions are not compatible 1 2 phases φ± are not fixed by the cross-cap condition. These with each other.) phases will not affect our later analysis, and hence will be This distinction between P2 = 1 and P2 = −1 are set to zero henceforth. One common convention for the −i π i π † directly related to the Z2 classification in class D+R+ phase is |C, η = +i = e 8 |0f i, |C, η = −i = e 8 f |0f i, and to the absence of SPT phases in class D+R−. For where we focus on the case of single flavor Nf = 1 2 class D+R− (P = −1), the fermion zero modes are for simplicity. Then we notice the following identities −iη π −iη π not allowed, and hence the corresponding cross-cap state ψ0L|C, ηi = e 4 |C, −ηi, ψ0R|C, ηi = e 4 |C, −ηi. is constructed entirely in terms of fermionic oscillator One motivation for this phase convention is that it well modes (non-zero modes). It can be checked easily that compares with the operator product expansions of the 42 i π −i π −i π this cross-cap state is anomaly free. On the other hand, Ising CFT, ψLσ ∼ e 4 µ, ψLµ ∼ e 4 σ, ψRσ ∼ e 4 µ, 2 i π for class D+R+ (P = +1), the construction of the cor- ψRµ ∼ e 4 σ, where σ and µ are the Ising spin operator responding cross-cap state is more non-trivial because of and the disorder operator, respectively. ] the presence of fermion zero modes. We will see the ac- Alternatively, when Nf = even, one can introduce the tion of the fermion number parity operator on the cross- following fermion creation operators (see, for example, cap state, |C(η1, η2)i with η1 = η2, give rise to an anoma- Ref. 49): lous phase (sign), indicative of the expected Z2 classification. † 1 2j−1 2j † 1 2j−1 2j dLj = (ψ0L + iψ0L), dRj = (ψ0R + iψ0R), b. Construction of cross-cap states Let us now con- 2 2 (88) struct the cross-cap state focusing on η1 = η2 = η: and the Fock vacuum |0di annihilated by dLj and dRj. a a † [ψ (σ2) − iηψ (σ2 + β)] |C, ηi = 0, L R We observe that dLj − idRj = f2j−1 − if2j,(djL − a a † [ψR(σ2) + iηψL(σ2 + β)] |C, ηi = 0, (83) idjR)|0f i = 0, and

Y † † where a = 1,...,Nf . When η1 = η2, the cross-cap condi- |C, +i = |0f i = (djL − idjR)|0di. (89) tion is compatible with the periodic boundary condition j for the fermion ﬁeld in σ1 direction. Within the zero mode sector, the cross-cap condition reads Similarly,

Nf [ψa − iηψa ] |C, ηi = 0, (84) Y † Y † † 0L 0R |C, −i = fa|0f i = (djR − idjL)|0di. (90) a=1 j where ψa is the zero mode for the a-th flavor, satis- 0L/R (Similar construction is possible even for N = odd by fying (ψa )2 = 1. f 0L/R adding an extra Majorana fermion as in Ref. 8. We will In the following, we construct the cross-cap states not dwell on this point as the identification of a quantum within the zero-mode sector explicitly, together with a anomaly when Nf = odd is rather straight forward and reference state. [Recall the important role played by the does not depend on the choice of the reference state.) reference state for the case of the bosonic SPT phases One important feature of the second construction is the discussed below Eqs. (??) and (??)]. The choice of the clear factorization of the vacuum |0di into the left- and reference state, however, is far from obvious, due to the right-moving sectors, |0di = |0diL ⊗ |0diR, as it is an- degeneracy in the zero mode sector. We will illustrate nihilated by dLj and dRj separately. The entire Hilbert this point by contrasting two constructions. space built out of |0di also factorizes into the left- and In the first construction, we construct the Hilbert space right-moving sectors. This factorization allows us to in- of the zero modes by considering the following fermion troduce the action of reflection (time-reversal) in a trans- creation/annihilation operators: parent way. Such factorization is also expected for general construction of cross-cap states and boundary states 1 1 f † = (ψa + iψa ), f = (ψa − iψa ), (85) in boundary conformal field theories. (See, for example, a 2 0L 0R a 2 0L 0R Ref. 50.) 13

C. Fermion number parity - Z2 classiﬁcation in and class D+R+ −1 −1 Pfψ0LPf = ψ0R, Pfψ0RPf = −ψ0L. (94) We now ask the properties of the cross-cap states under the action of symmetry generators. As a start, let us Explicitly, they can be written as consider the fermion number parity. In the tree channel, iδ Y 1 a its explicit form within the zero mode sector is given by Pf= e √ (ψ0L + ψ0R) (95) 2 1 1 2 2 Nf Nf a Gf = (iψ0Lψ0R)(iψ0Lψ0R) ··· (iψL ψ0R ). (91) The fermion number parity acting on the cross-cap states and gives ˜ iδ Y 1 a = e √ (1 − ψ0Lψ0R) , (96) Nf P Gf |C, ±i = (±) |C, ±i. (92) a 2

Therefore, when Nf = even, there is no anomaly. On the respectively, where eiδ is an unknown phase factor and other hand, when Nf = odd, one would conclude, |C, −i is anomalous while |C, +i is not. This is consistent with will be discussed in more detail shortly. The first parity operator does not preserve the cross-cap condition with the Z2 classification of (2+1)d topological superconduc- η1 = η2, so we will focus on the second one. One also tors. in symmetry class D+R+. Upon closer inspection, however, Eq. (??) would look verifies strange since the two cross-cap states |C, ±i should be 2 2iδ Nf treated on the equal footing: If the cross-cap |C, ±i is Pf = e (i) Gf , (97) obtained by twisting P, |C, ∓i is obtained from simply from Gf P. Hence, both cross-cap states |C, ±i be- and Gf Pf= PGf f . long/correspond to the same phase. In fact, it should Let us now calculate the action of Pf on the cross- be noted that there is a phase ambiguity in defining the cap states. By using the representation in terms of the cross-cap states and the fermion number parity opera- † f-fermions, ψ0Lψ0R = i(2f f − 1), Pf can be written as tor. In the above analysis, we implicitly made a particular choice where the fermion number parity of the iδ Y 1 Pf= e √ [1 − i(2na − 1)] , (98) ground state |0f i is +1. In principle, one could assign a 2 different fermion number parity eigenvalue, e.g., by mod- a ifying the definition of the fermion number parity oper- † where n = f †f . Then, the action of on the cross-cap ator, Gf → −Gf . Alternatively, instead of using f , f, a a a Pf one could define c := f † and c† := f, which leads to states is given by Q † |C, −i = |0ci and |C, +i = a c |0ci. In this conven- iδ Y 1 tion, it is tempting to claim Gf |C, −i = +|C, −i while Pf|C, +i = e √ [1 − i(2na − 1)] |0f i N Gf |C, +i = (−1) f |C, +i. Thus, there is some ambi- a 2 guity when deducing the fermion number parity eigen- 1 π iδ Y +i Nf iδ = e √ [1 + i] |0f i = e 4 e |C, +i, value. We have encountered similar ambiguity when deal- 2 ing with CP symmetric bosonic SPT phases. Such am- a 1 π biguity of the fermion number parity eigenvalue of the iδ Y −i Nf iδ Pf|C, −i = e √ [1 − i] |C, −i = e 4 e |C, −i. ground state, however, does not affect our conclusion, a 2 since, independent of the phase choice, when Nf = odd, (99) we cannot make both |C, ±i anomaly-free. Hence we conclude the classification of symmetry class D+R . Z2 + The relative phase between Pf|C, +i and Pf|C, −i is e+iπNf /2, which is independent of the choice of eiδ (the choice of the action of on the reference state), and D. Time-reversal - Z8 classification in class BDI + Pf R++ vanishes when Nf = 4 × integer. In other words, one cannot make both cross-cap states anomaly-free unless We now include time-reversal symmetry. After π/2 Nf = 4 × integer. One then immediately concludes the rotation of spacetime, a time-reversal operator is trans- classification is at least Z4. On the other hand, as we have seen for the case of the formed into a parity operator (unitary) (denoted by Pf iδ in the following). Correspondingly to two time-reversal bosonic SPT phases, a proper choice of the phase e , if exists, leads to a refined classification. If we choose operators T 2 = ±1 in the loop channel picture, there are two kinds of parity operators in the tree channel picture. |0f i as the reference state and demand |0f i transform They act on the fermion zero modes as: trivially under Pf, Pf|0f i = |0f i, we obtain Z4 classification. Alternatively, if we choose |0di as the reference −1 −1 Pψ0LP = ψ0R, Pψ0RP = ψ0L, (93) f f f f state, and demand Pf|0di = |0di, which may be obtained 14 from Pf|0diL,R = |0diR,L, and other symmetries. In Laughlin’s thought experiment, quantized charge pumping caused by flux thread- Pf|C, +i = Pf|0f i ing (large gauge transformation) characterizes the quan-

Nf /2 tum Hall eﬀect even in the presence of interactions and Y † † disorder. Within edge theories of the quantum Hall ef- = Pf (djL − idjR)|0di j=1 fect, the charge pumping by a large gauge transformation appears as a quantum anomaly, i.e., non-conservation of Nf /2 Y the electric U(1) charge. = (d† + id† )|0 i jR jL d The edge theories of SPT phases (if exist) diﬀer from j=1 the edge theories of the quantum Hall systems. First, Nf /2 = (i) |0f i. (100) quite generically, edge theories supported by SPT phases are non-chiral (having vanishing thermal Hall conduc- I.e., eiδ = 1. It can also be checked, straightforwardly, tance). Second, edge theories of SPT phases may be

Nf /2 protected by discrete symmetries, and may not have a Pf|C, −i = (i) |C, −i, (101) continuous U(1) symmetry. For these reasons, diagnosing following Eq. (??). Thus, with this choice, the two cross- edge theories of SPT phases requires to identify quantum cap states |C, ηi can be both made anomaly free only anomalies (breaking down of symmetries of SPT phases when Nf = 8 × integer, i.e., Z8 classification. by quantum effects) of a subtler kind than in the quan- As a final comment, we provide a yet another point tum Hall effect. By twisting parity in edge theories of of view by using Eq. (??). Equation (??) suggests, (2+1)d topological phases, we obtain a cross-cap state within the zero mode sector, the symmetry is realized and investigate possible quantum anomalies in terms of projectively. The “unwanted” phase e2iδ(i)Nf can be re- the cross-cap state. This method is applied explicitly moved by choosing eiδ = e−iπNf /4. However, with this to the several examples, including bosonic and fermionic choice, the reference state now acquires an anomalous SPT phases, and reproduces the known classifications in −iπN /4 phase Pf|0di = e f |0di. This conflict between the Refs. 11 and 39. In conclusion, we have provided a way two demands, one to represent the symmetry group non- to gauge the spatial symmetries to efficiently diagnose projectively and the other to make the reference state SPT phases, which is beyond the cohomology classifica- transform trivially under Pf, can be considered as a form tion tables. We close with a few comments. of quantum anomaly. (i) In our approach, it is crucial to fix the action of symmetry operators on a reference state in the tree-channel picture. In particular, possible phase ambiguity should V. CONCLUSION be fixed, although in many cases a partial result is obtained even without fixing this ambiguity. While we have There are three known ways of symmetry breaking in provided a proposal to fix this phase ambiguity by speci- nature. First, a symmetry can be broken explicitly by fying reference states, it is desirable to have a more con- adding a symmetry-breaking perturbation to the Hamil- vincing and convenient method. One possible approach tonian. Second, a symmetry can be broken spontaneously to this problem is to use the state-operator correspon- in many-body systems and in field theories because of dence and assign the phase which is consistent with the interactions. These two forms of symmetry breaking oc- operator algebra of the edge theories. For example, such cur both in classical and quantum systems. The third argument is used in Ref. 51 in somewhat similar context. way of symmetry breaking is more subtle and happens Following this type of approach is left for the future re- only in quantum many-body systems; A symmetry can search. be anomalous, meaning it can be broken because of quan- (ii) We treated SPT phases where parity and time- tum effects. In discussing SPT phases, we are to distin- reversal are in conflict quantum mechanically, in that guish different quantum phases of matter, all respecting enforcing parity by orientifolding leads to breakdown of the same set of symmetries. Therefore, by definition, the time-reversal, e.g., class BDI+R++ topological super- Landau paradigm, while powerful in discussing phases conductors. We choose to twist parity since twisting with spontaneous symmetry breaking, cannot be applied unitary symmetry appears to be more straightforward, to SPT phases. On the other hand, as we demonstrated, while, in principle, we could equally consider twisting anomalous symmetry breaking (quantum anomalies) can by time-reversal. In fact, after exchanging the role of be useful in diagnosing and perhaps even classifying SPT space and time coordinates and when considering cross- phases. cap states, what appears to be time-reversal symme- The idea of using quantum anomalies to character- try in the original, loop-channel picture is represented ize topological phases of matter goes back to Laughlin’s as a parity (or CP ) symmetry in the tree-channel pic- thought experiment (Laughlin’s gauge argument) in the ture. This indicates that by exchanging the role of space quantum Hall states. Our methodology can be consid- and time coordinates, twisting time-reversal is neither ered as a proper generalization of Laughlin’s gauge ar- more nor less than twisting parity, and should lead to an gument to SPT phases protected by parity (reflection) orientifold theory. Our analysis in this paper thus pro- 15 vides an insight into SPT phases protected, not by par- ACKNOWLEDGMENTS ity, but by time-reversal symmetry and other symmetry, e.g., (2+1)d time-reversal symmetric topological insula- We thank Xie Chen and Andreas Ludwig for helpful tors (the quantum spin Hall effect), which are protected discussion. This work is supported by the NSF under by U(1) and time-reversal symmetries. Orientifolds of V Grant No. DMR-1064319 (GYC), No. DMR 1408713 gapless edge states discussed in this work may provide a (GYC), and DMR-1455296 (SR), the Brain Korea 21 complementary view to, e.g., Ref. 28, which discusses a PLUS Project of Korea Government (GYC), and Alfred way to gauge time-reversal in gapped bulk SPT phases P. Sloan foundation (SR). by using the tensor network representation of quantum states.

(iii) In our approach, the use of cross-cap states, upon going from the loop to tree channel picture, most clearly Appendix A: Bosonic SPT phases demonstrates how quantum anomalies may appear. In doing so, we exchange the role of space and time coor- In this appendix, we will complete the classifications of dinates. The fictitious time-evolution and the fictitious (2+1)d bosonic SPT phases protected by P × TC,CP × Hilbert space in the tree channel picture is in fact akin T,T × C and T × C × P . The methodology for the to the row-to-row transfer matrix and to the auxiliary classification is already explained in the main text. In space of matrix product states (MPSs), respectively. The this Appendix, to simplify our analysis, we will make projective representation of the symmetry group of SPT one more assumption: the CPT-theorem. If we limit phases on the auxiliary space of MPSs has been success- our focus to topological classification of relativistic sys- fully used to diagnose and classify (1+1)d gapped SPT tems, the CPT-theorem allows us to relate various SPT phases. Our use of cross-cap states and the associated phases protected by different sets of symmetries. For anomalous phases have some similarities with the classi- example, with the CPT-theorem, one can convert, e.g., fication of (1+1)d gapped SPT phases by MPSs. the classification problem of bosonic SPT phases with T - symmetry into the classification problem of bosonic SPT (iv) In our previous calculations26, the full partition phases with CP -symmetry. While the restriction to rel- functions on the Klein bottle are computed explicitly and ativistic systems may sound too stringent, in Ref. 39, checked for the presence/absence of quantum anomalies. it was shown that classifications of free fermionic SPT In the current reformulation in the tree channel, there is phases by K-theory or the Clifford algebra do satisfy no need to compute the partition function, although it is the CPT-equivalence in general. I.e., different fermionic possible to compute the partition function by using cross- SPT phases whose symmetries are CPT-conjugate to cap states. In addition, we do not rely on the continuous each other are classified by the same topological charge. U(1) symmetry, and hence can treat a wider class of SPT While we do not have such systematic understanding for phases that are not discussed in Ref. 26. the case of bosonic SPT phases, which are necessarily interacting, within microscopic analysis of gapping po- The tree-channel formulation also somewhat liberates tentials performed in Ref. 39, such CPT relations among us from the unwanted reliance on relativistic and confor- different SPT phases still exist for bosonic SPT phases. mal symmetries when discussing topological classification Our analysis presented below, which makes use of cross- of phases of matter. As an MPS can be constructed for cap states and assumes the CPT-theorem, also confirm an arbitrary (1+1)d (gapped as well as gapless) quantum these results, although it should be emphasized that the states, cross-cap states can in principle be constructed for assumption of the CPT-theorem can be avoided if one general edge theories which lack relativistic invariance. If wishes. it comes to compute the full partition function, the lack c. P × TC symmetry We illustrate the use of CPT- of the relativistic invariance may be inconvenient, as the theorem by considering the edge theory of bosonic SPT Hamiltonian in the loop and tree channels do not agree. phases protected by P × TC. As P -symmetry, we con- The partition function in such situations would not be sider the case with mp = 0 (and the corresponding cross- expressed in terms of well-known functions such as the cap state |Cp(np, 0)i in Eq. (??)). On the other hand, Jacobi theta functions. TC-symmetry, which is a ”spatial“ symmetry in time and anti-unitary, is given by The importance, necessity and limitations of relativis- 0 tic and conformal invariance, when following our ap- mTC πα TC : φ → −φ + nTC πR, θ → θ + proach to SPT phases, are however not entirely under- R stood currently, and are left for the future studies. It :(x , x ) → (x , −x ), (A1) would be interesting to note that, in constructing cross- 1 2 1 2 cap states and, similarly, in constructing conformal invariant boundary states in boundary conformal field the- where nTC , mTC ∈ {0, 1}. ories, the Virasoro algebra plays a major role. (i) As the first step, we convert TC-symmetry into its 16

CPT partner, i.e., a parity symmetry P 0: CP and twisting CP 0 (??) invariant. Using the CP 0-

0 0 symmetry, we form another cross-cap state |Ccp (0, mT )i. mTC πα 0 If |C 0 (0, m )i or |C (n , m )i is not invariant under P : φ → φ + nTC πR, θ → −θ + cp T cp cp cp R Z, the corresponding theory with CP × T is anomalous. :(x1, x2) → (−x1, x2), (A2) ncpmT Z |Ccp0 (0, mT )i = (−1) |Ccp0 (0, mT )i, which is unitary. Thus we have the bosonic theory with ncpmcp Z |Ccp(ncp, mcp)i = (−1) |Ccp(ncp, mcp)i. (A8) the two parity symmetries P × P 0 after this mapping. (ii) Next, we form a non-spatial unitary symmetry by Thus the edge theory is anomalous if ncpmT = 1 combining P and P 0, X = P · P 0: or ncpmcp = 1. We conclude the following triplet (mT , ncp, mcp) for the anomalous states. 0 mTC πα X : φ → φ + (np + nTC )πR, θ → θ + (m , n , m ) = (1, 1, 0), (0, 1, 1), (1, 1, 1). (A9) R T cp cp :(x1, x2) → (x1, x2). (A3) This result agrees with Ref. 39. e. P × T This case is analyzed in Sec. III without Because X is non-spatial in (x1, x2)-space, it should be using CPT theorem. The CPT partner of T -symmetry, non-spatial in (σ1, σ2)-space as well (see Fig. 1). Fur- CP 0, is given by thermore, it is straightforward to check that X leaves 0 0 the cross-cap conditions by P and P (??) invariant . 0 mT πα CP : φ → −φ θ → θ + (iii) As the third step, we twist P and P 0 to construct R the cross-cap states and study if they are invariant un- :(x1, x2) → (−x1, x2). (A10) der X or not. By twisting P 0-symmetry, we form the By combining P and CP 0, we form a non-spatial unitary cross-cap state |C 0 (n , m )i. If either |C (n , 0)i or p TC TC p p symmetry, Y ≡ P · CP 0, which acts on the bosonic fields |Cp0 (nTC , mTC )i is not invariant under X, then the corresponding theory with P × TC is anomalous: as 0 mT πα npmTC Y : φ → −φ + npπR, θ → −θ + X |Cp(np, 0)i = (−1) |Cp(np, 0)i, R nTC mTC X |Cp0 (nTC , mTC )i = (−1) |Cp0 (nTC , mTC )i. :(x1, x2) → (x1, x2). (A11) (A4) Y is non-spatial and unitary in (σ1, σ2)-space (Fig. 1). It Thus the edge theory is anomalous if npmTC = 1 or can also be checked that Y leaves the cross-cap conditions 0 0 nTC mTC = 1. To conclude, we have the following triplet by P (??) and CP (??) invariant. Twisting the CP - (np, nTC , mTC ) for the anomalous states. symmetry, we form another cross-cap state |Ccp0 (0, mT )i. The action of Y on the cross-cap states is given by (np, nTC , mTC ) = (1, 0, 1), (0, 1, 1), (1, 1, 1). (A5) npmT Y |Cp(np, 0)i = (−1) |Cp(np, 0)i, With these procedure, we conclude that the theory with npmT Y |Ccp0 (0, mT )i = (−1) |Ccp0 (0, mT )i. (A12) the symmetry P × TC is anomalous if its CPT partner, 0 the state with P ×P symmetry, is anomalous. The result Thus the edge theory is anomalous if npmT = 1, i.e., agrees with Ref. 39, where all the possible perturbations (np, mT ) = (1, 1). This agrees with Ref. 39. which can potentially gap out the edge theory are enu- f. T ×C symmetry Here in this case, there is no ap- merated. parent P -symmetry. We can however map T -symmetry d. CP × T symmetry The cross-cap state for the into its CPT partner, i.e., CP -symmetry. Then we can SPT phase with CP × T is |Ccp(ncp, mcp)i (Eq. (??)), twist CP -symmetry to obtain a cross-cap state and act 0 where (ncp, mcp) ∈ {0, 1}. We find the CPT partner CP C on the cross-cap state to diagnose the stability of the of T -symmetry theory. The CPT partner of T -symmetry, CP -symmetry, is defined as 0 0 mT πα 0 CP : φ → −φ, θ → θ + mT πα R CP : φ → −φ + nT πR, θ → θ + R :(x1, x2) → (−x1, x2), (A6) :(x1, x2) → (−x1, x2) (A13) which is unitary. We again form a non-spatial unitary By twisting this CP -symmetry, we obtain the cross- symmetry by combining CP and CP 0. We call the sym- cap state |C (n , m )i (??). On the other hand, C- metry as Z = CP · CP 0: cp T T symmetry (??) is identified with nc = mc = 0. Under 0 (mT + mcp)πα the action of C-symmetry, the cross-cap condition (??) Z : φ → φ + ncpπR, θ → θ + R is invariant, while we find, by using (??), nT mT :(x1, x2) → (x1, x2) (A7) C |Ccp(nT , mT )i = (−1) |Ccp(nT , mT )i. (A14)

Z is non-spatial in (σ1, σ2)-space (see Fig. 1). We can also Thus the theory with the parity symmetry with nT = 39 show that Y leaves the cross-cap conditions by twisting 1, mT = 1 and C is anomalous. 17

g. T × C × P symmetry The symmetry group T × (??), this occurs when C × P is CPT-equivalent to CP 0 × C × P or P 0 × C × P , where [nT, mT, nP, mP] =[0, 0, 1, 1], [0, 1, 1, 0], [0, 1, 1, 1], [1, 0, 0, 1], [1, 1, 0, 0], [1, 1, 0, 1], 0 0 mT πα [1, 0, 1, 1], [1, 1, 1, 0], [1, 1, 1, 1]. CP : φ → −φ + nT πR, θ → θ + R (A17) :(x , x ) → (−x , x ), 1 2 1 2 Putting two identical copies of each phase above together 0 2 0 mT πα results a trivial phase, i.e., [nT, mT, nP, mP] = [trivial]. P : φ → φ + nT πR, θ → −θ + R Here, when we put two phases together, the result- 1 1 1 1 2 2 2 2 :(x1, x2) → (−x1, x2). (A15) ing phase, say, [nT, mT, nP, mP] ⊕ [nT, mT, nP, mP], has

the corresponding cross-cap states |CΩi (nΩi , mΩi )i = |C1 (n1 , m1 )i ⊗ |C2 (n2 , m2 )i, which are the direct Ωi Ωi Ωi Ωi Ωi Ωi We can choose a set {P,P 0,CP 0} (other choices can (tensor) product of the cross-cap states of the original be {P,P 0,CP }, {P,CP,CP 0}, etc.) that generates the two phases. On the other hand, all phases above are in- symmetry group P 0 × C × P , and consider their cor- equivalent [we can not obtain a trivial phase by putting responding cross-cap states |Cp(np, mp)i, |Cp0 (nT , mT )i any two different elements in the set (??) together]. How- and |Ccp0 (nT , mT )i. To study the SPT phases (and their ever, phases shown in (??) are not all possible nontrivial group structure) with the group P 0 × C × P , we can SPT phases with symmetries T ×C ×P ; putting different check how these cross-cap states transform under the phases above together might result other nontrivial SPT non-spatial symmetries gi which are generated by com- phases that are not listed above. To determine the group bining symmetries from {P,P 0,CP 0}, such as X ≡ P ·P 0, structure of these phases, we observe that putting three Y ≡ P · CP 0, and C. We can choose {X,Y }, {X,C}, or phases in any vertical or horizontal line of the array (??) {Y,C} as a ”complete” set of non-spatial symmetries. together will result a trivial phase, or equivalently, we Here we consider {Y,C}, and their action on the cross- have cap states are given by: [0, 0, 1, 1] ⊕ [0, 1, 1, 0] = [0, 1, 1, 1] ⊕ ⊕ ⊕ npmT Y |Cp(np, mp)i = (−1) |Cp(np, mp)i, npmp [1, 0, 0, 1] ⊕ [1, 1, 0, 0] = [1, 1, 0, 1] C |Cp(np, mp)i = (−1) |Cp(np, mp)i, = = = nT mp Y |Cp0 (nT , mT )i = (−1) |Cp0 (nT , mT )i, nT mT [1, 0, 1, 1] ⊕ [1, 1, 1, 0] = [1, 1, 1, 1]. (A18) C |Cp0 (nT , mT )i = (−1) |Cp0 (nT , mT )i, npmT Y |Ccp0 (nT , mT )i = (−1) |Ccp0 (nT , mT )i, From this, we see that all nontrivial phases can nT mT C |Ccp0 (nT , mT )i = (−1) |Ccp0 (nT , mT )i. (A16) be generated by a specific set of four phases, say, {[0, 0, 1, 1], [0, 1, 1, 0], [1, 0, 0, 1], [1, 1, 0, 0]} (there are 3 × 3 = 9 equivalent choices for these four group gen- The system is in a nontrivial SPT phase if there exists at erators); there are in total fifteen nontrivial SPT phases, 4 39 least one symmetry-noninvariant cross-cap state. From which form a Z2 group.

1 M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 10 X.-L. Qi, arXiv:1202.3983v2 [cond-mat.str-el] (2012). (2010). 11 H. Yao and S. Ryu, Physical Review B 88, 064507 (2013). 2 X.-L. Qi and S.-C. Zhang, Rev. Mod. Phys. 83, 1057 12 C. Wang and T. Senthil, Phys. Rev. B 89, 195124 (2014), (2011). arXiv:1401.1142 [cond-mat.str-el]. 3 A. P. Schnyder, S. Ryu, A. Furusaki, and A. W. W. Lud- 13 L. Fidkowski, X. Chen, and A. Vishwanath, Physical Re- wig, Physical Review B 78, 195125 (2008). view X 3, 041016 (2013), arXiv:1305.5851 [cond-mat.str- 4 A. Kitaev, in AIP Conf. Proc, 1134 (2009) p. 22. el]. 5 A. M. Turner and A. Vishwanath, ArXiv e-prints (2013), 14 M. A. Metlitski, L. Fidkowski, X. Chen, and A. Vish- arXiv:1301.0330 [cond-mat.str-el]. wanath, ArXiv e-prints (2014), arXiv:1406.3032 [cond- 6 T. Senthil, ArXiv e-prints (2014), arXiv:1405.4015 [cond- mat.str-el]. mat.str-el]. 15 A. Vishwanath and T. Senthil, Physical Review X 3, 7 L. Fidkowski and A. Kitaev, Phys. Rev. B 81, 134509 011016 (2013). (2010). 16 S. Geraedts and O. Motrunich, ArXiv e-prints (2014), 8 L. Fidkowski and A. Kitaev, Phys. Rev. B 83, 075103 arXiv:1408.1096 [cond-mat.stat-mech]. (2011). 17 Y.-M. Lu and A. Vishwanath, Phys. Rev. B 86, 125119 9 S. Ryu and S.-C. Zhang, Physical Review B 85, 245132 (2012). (2012). 18

18 G. Y. Cho, J. C. Teo, and S. Ryu, Physical Review B 89, 34 P. Horava, Nuclear physics. B 327, 461 (1989). 235103 (2014). 35 C. Angelantonj and A. Sagnotti, Phys. Rept. 371, 1 (2002). 19 A. Cappelli and E. Randellini, Journal of High Energy 36 I. Brunner and K. Hori, Journal of High Energy Physics Physics 2013, 1 (2013). 2004, 023 (2004). 20 X. Chen, Z.-C. Gu, Z.-X. Liu, and X.-G. Wen, Phys. Rev. 37 I. Brunner and K. Hori, Journal of High Energy Physics B 87, 155114 (2013), arXiv:1106.4772 [cond-mat.str-el]. 2004, 005 (2004). 21 A. Kapustin, ArXiv e-prints (2014), arXiv:1403.1467 38 S. Yamaguchi, Progress of Theoretical Physics 97, 703 [cond-mat.str-el]. (1997), cond-mat/9704039. 22 A. Kapustin, ArXiv e-prints (2014), arXiv:1404.6659 39 C.-T. Hsieh, T. Morimoto, and S. Ryu, ArXiv e-prints [cond-mat.str-el]. (2014), arXiv:1406.0307 [cond-mat.str-el]. 23 A. Kapustin, R. Thorngren, A. Turzillo, and Z. Wang, 40 C. Xu and A. W. Ludwig, arXiv preprint arXiv:1112.5303 ArXiv e-prints (2014), arXiv:1406.7329 [cond-mat.str-el]. (2011). 24 M. Levin and Z.-C. Gu, Phys. Rev. B 86, 115109 (2012). 41 C. Xu, Physical Review B 87, 144421 (2013). 25 O. M. Sule, X. Chen, and S. Ryu, Physical Review B 88, 42 P. Mathieu and D. Senechal, Conformal ﬁeld theory (New 075125 (2013). York: Springer, 1997). 26 C.-T. Hsieh, O. M. Sule, G. Y. Cho, S. Ryu, and R. Leigh, 43 J. G. Polchinski, String theory (Cambridge university arXiv preprint arXiv:1403.6902 (2014). press, 2003). 27 Y.-M. Lu and D.-H. Lee, ArXiv e-prints 1210.0909 (2012), 44 P. H. Ginsparg, arXiv preprint hep-th/9108028 63 (1988). arXiv:1210.0909 [cond-mat.str-el]. 45 Y.-M. Lu and A. Vishwanath, arXiv preprint 28 X. Chen and A. Vishwanath, ArXiv e-prints (2014), arXiv:1302.2634 (2013). arXiv:1401.3736 [cond-mat.str-el]. 46 C.-K. Chiu, H. Yao, and S. Ryu, Phys. Rev. B 88, 075142 29 K. Hori, S. Katz, A. Klemm, R. Pandharipande, (2013), arXiv:1303.1843 [cond-mat.mes-hall]. R. Thomas, et al., Mirror symmetry (2003). 47 T. Morimoto and A. Furusaki, Phys. Rev. B 88, 125129 30 R. Blumenhagen and E. Plauschinn, Lect.Notes Phys. 779, (2013), arXiv:1306.2505 [cond-mat.mes-hall]. 1 (2009). 48 K. Shiozaki and M. Sato, Phys. Rev. B 90, 165114 (2014), 31 C. G. Callan, C. Lovelace, C. R. Nappi, and S. A. Yost, arXiv:1403.3331 [cond-mat.mes-hall]. Nucl. Phys. B 293, 83 (1987). 49 O. Bergman and M. R. Gaberdiel, Nuclear Physics B 499, 32 A. Sagnotti, Cargese ’87, “Non-perturbative Quantum 183 (1997), hep-th/9701137. Field Theory,” ed. G. Mack et al. (Pergamon Press) , 521 50 B. Bates, C. Doran, and K. Schalm, ArXiv High Energy (1988). Physics - Theory e-prints (2006), hep-th/0612228. 33 J. Polchinski and Y. Cai, Nucl. Phys. B 296, 91 (1988). 51 E. G. Gimon and J. Polchinski, Phys. Rev. D 54, 1667 (1996), hep-th/9601038.