arXiv:1406.0307v1 [cond-mat.str-el] 2 Jun 2014 I.Casfiaino PsmercTsi arbitrary in TIs symmetric CP of Classification III. V oooia P hoe n topological and theorem CPT Topological IV. I Dfrinctplgclpae rtce by protected phases topological fermionic 2D II. .Casfiaino DitrcigSTphases: SPT interacting 2D of Classification V. .Introduction I. symmetries dimension topological of T-duality and CPT-equivalence D. .BGsseswt pnU1 n P and U(1) spin with systems BdG D. -arxfruain12 10 formulation K-matrix classes symmetry CPT-equivalent .Otieadmi eut 3 results main and Outline A. .C ymti nuaos4 insulators symmetric CP A. .Casfiainb -hoy7 K-theory by Classification A. .Bl n deKmti hoisincorporated theories K-matrix edge and Bulk A. .Cneto oTsmercisltr 6 insulators symmetric T to Connection C. .Tplgclivrat 8 invariants Topological C. .Casfiaino P hssb K-matrix by phases SPT of Classification C. .BGsseswt pnU1 n P and U(1) spin with systems BdG B. .Drcmdl n P hoe 8 theorem CPT and models Dirac B. .Eg tblt rtrafrSTPae 14 Phases SPT for criteria stability Edge B. .Tplgclcasfiaino other of classification Topological E. P hoe n lsicto ftplgclisltr an insulators topological of classification and theorem CPT 1 ymtis5 symmetries hss neape7 example an phases: ymtis10 8 symmetries symmetries ulSTpae 15 phases SPT and dual phases SPT CPT-equivalent thoeries: ihsmere 12 symmetries with .Eg hoy5 4 theory Edge 2. model tight-binding 1. .Kmti lsicto ffrincSPT fermionic of classification K-matrix 2. bosonic of classification K-matrix 1. eateto hsc,Uiest fIlni tUrbana-Ch at Illinois of University Physics, of Department hss17 16 phases phases SPT non-chiral usdbfr r eae yCTtheorem. CPT ro parti by the related in are clarify show, before l also and cussed is We phases, approach topological interactions. second symmetry-protected the of approa K-mat while effects first the dimensions The incorporates of spatial terms theories. general in bulk in classifi other (2+1)-dimensional K-theory the of and of theory symmetries, terms with in theories one co approaches: and parity, complementary particle-hole, time-reversal, by protected epeetasseai oooia lsicto ffermi of classification topological systematic a present We 2 odne atrTer aoaoy IE,Wk,Saitama, Wako, RIKEN, Laboratory, Theory Matter Condensed CONTENTS hn-s Hsieh, Chang-Tse 1 Ωae:Jn ,2014) 3, June (ΩDated: aaioMorimoto, Takahiro 1 4 7 daaial once o ntepeec faset a of states of trivial not presence are topologically the matter. that conditions, in symmetry matter to, of of connected states adiabatically are (TSCs) conductors inae symmetries, Zirnbauer re sc sTSadPSlse nAltland-Zirnbauer in listed PHS and TRS as (such tries discovered. predicted theoretically been have thereof, combinations and (PHS), symmetry particle-hole I Discussion VI. .Etnlmn pcrmadeffective and spectrum Entanglement A. .Pofo oooia P hoe o interacting for theorem CPT topological of Proof C. .Cluain fsmer rnfrain and transformations symmetry of Calculations B. oooia nuaos(I)adtplgclsuper- topological and (TIs) insulators Topological nmr eetyas nepa ewe nst symme- on-site between interplay years, recent more In bnto fteesmere.W s two use We symmetries. these of mbination mag,11 etGenS,Ubn L61801 IL Urbana St, Green West 1110 ampaign, P hssi -arxtere 21 theories K-matrix in phases SPT o-hrlSTpae ntodmnin 27 References dimensions two in phases SPT non-chiral cnweget 19 symmetries Acknowledgments .Dsuso fSTpae 22 22 phases 21 SPT of Discussion elements 3. identity of operators Equations symmetry 2. of relations Algebraic 1. .Etnlmn pcrm20 with spectrum entanglement of Properties 2. spectrum Entanglement 1. 1,2 ua,tplgclsprodcosdis- superconductors topological cular, eo P hoe ncasfiainof classification in theorem CPT of le ymtis21 symmetries .BsncSTpae ihsmer group symmetry with phases SPT Bosonic b. .FrincSTpae ihsymmetries with phases SPT Fermionic d. .BsncSTpae ihsmer group symmetry with phases SPT Bosonic a. .FrincSTpae ihsymmetry with phases SPT Fermionic c. 2 mtdt w pta iesosbut dimensions spatial two to imited hi pcfi ofe emo theories fermion free to specific is ch ncadbsnctplgclphases topological bosonic and onic aino apdqartcfermion quadratic gapped of cation n hne Ryu Shinsei and Z Z {T 13–20 I n Sscaatrzdb Altland- by characterized TSCs and TIs i hoydsrpino h edge the of description theory rix 2 2 CP T , × CP} .INTRODUCTION I. Z 2 C × 5-18 Japan 351-0198, Z 3 2 P iervra ymty(TRS), symmetry time-reversal superconductors d 1 4–12 n experimentally and CP 23 24 28 19 19 26 25 2 symmetry classes) and non-on-site symmetries (such as T sym. TI CPT-equiv. space group symmetries) has enriched the topological CP sym. TCI phases of matters. A novel class of topological matter (QSHE) characterized (additionally) by non-on-site symmetries, such as topological crystalline insulators (TCIs)21 and topological crystalline superconductors (TSCSs), have T-duality T-duality been discovered.22–29 The topological classification, orig- inally studied in the presence/absence of various on-site symmetries in Altland-Zirnbauer classes,10,11,30 are also Chiral sym. TSC CPT-equiv. P sym.TCSC extended to include the non-on-site symmetries such as with Sz conserv. with Sz conserv. reflection symmetry,31,32 inversion symmetry,33,34 and (crystal) point group symmetries,35–38 recently. (a) Motivated by these recent works, in this paper, we fur- ther study TIs and TSCs protected by a wider set of sym- metries than symmetries in Altland-Zirnbauer classes by T sym. TSC including, in particular, a parity symmetry (PS), which is a symmetry under the reflection of an odd number of spatial coordinates. One of our focuses is, in addition to the cases where parity is conserved, on situations where a CPT-equiv. CPT-equiv. combination of parity with some other symmetries, such as CP (product of PHS and PS) or PT (product of PS CPT-equiv. and TRS), are preserved. For earlier related works, see, P sym. TCSC T and P sym. TCI* for example, Refs. 30, 39–41. Another issue we will discuss in this paper is the effect of interactions on the classification of those topological (b) phases protected by parity and other symmetries (such as combination of parity and other symmetries). It has FIG. 1. Two sets of topological (crystalline) insulators and su- been demonstrated, in various examples, that there are perconductors related by CPT-equivalence and/or T-duality: phases that appear to be topologically distinct from triv- (a) T symmetric TI (QSHE), CP symmetric TCI, chiral (TC) symmetric TSC with S conservation, and P symmetric ial phases at non-interacting level, which, in fact, can z TCSC with S conservation; (b) T symmetric TSC, P sym- adiabatically be deformable into a trivial state of matter z 42–47 metric TCSC, and T and P symmetric TCI that can support in the presence of interactions. For example, Ref. gapless edge states even in the absence of charge U(1) sym- 47 discusses (2+1)-dimensional [or 2 spatial-dimensional metry. (2D)] superconducting systems in the presence of parity and time-reversal symmetries, which are classified, at the quadratic level, by an integer topological invariant, while systems at long wavelength limit, such as the band topol- once inter-particle interactions are included, states with ogy or the electromagnetic response, can be encoded in an integer multiple of eight units of the non-interacting the so-called topological field theory, which respects the topological invariant are shown to be unstable. Focusing Lorentz symmetry. When these topological properties on 2D bulk topological states that support an edge state are protected (or determined) by some symmetry, TRS, described by a K-matrix theory, i.e., Abelian states, we say, they can also be protected solely by CP symme- will study the stability of the edge state (and hence the try, which is a “CPT-equivalent” partner to TRS. For bulk state) in the presence of parity symmetry or parity example, the magnetoelectric effect in 3D time-reversal symmetry combined with other symmetries. symmetric TIs48 is also expected to be observed in a CP We will also show that, once parity symmetry or parity symmetric TI, because they are both described by the ax- symmetry combined with other discrete symmetries is ion term (effective action for electromagnetic response) included into our consideration, CPT theorem plays an with the same nontrivial (quantized) value of θ angle. important role in classifying topological states of matter. In addition, from the prospect of topological classifica- CPT theorem holds in Lorentz invariant quantum field tion, classifying symmetry-protected topological (SPT) theories, which says, C, P, T, when combined into CPT, phases of a free fermion systems [characterized either by is always conserved, i.e., CPT = 1, schematically. For a gapped Hamiltonian of the (d + 1)-dimensional bulk example, a Lorentz invariant CP symmetric field theory or by a gapless Hamiltonian of d-dimensional boundary also possesses TRS, and vice versa. with symmetries] is equivalent to classifying the corre- In condensed matter systems, however, such relations sponding Dirac operators with symmetry restrictions11. between these discrete symmetries (T, C, and P) do not It is thus natural to associate TIs protected by TRS with arise since we are not to be restricted to relativistic sys- TIs protected by CP symmetry, as the Dirac Hamiltonian tems; symmetries can be imposed independently. Never- has a CPT invariant form. theless, some physical properties of these non-relativistic In this paper, by going through classification prob- 3 lems of non-interacting fermions in the presence of vari- As inferred from the CPT theorem and the existence of ous symmetry conditions, and also microscopic stability time-reversal symmetric topological insulators in two di- analysis of interacting edge theories, we will demonstrate mensions, we will show that there are two topologically explicitly such CPT theorem holds at the level of topolog- distinct classes of insulators with CP symmetry in two ical classification for all cases that we studied. Through dimensions, i.e., Z2 classification. We will present in Sec. this analysis, we can see, for example, that 2D TSCs pro- II A a simple example (tight-binding model) of CP sym- tected by spin parity conservation,45,46 and 2D TSCs by metric band insulators, which is constructed from two parity and time-reversal,47 both of which are classified in copies of two-band Chern insulators with opposite chi- terms of Z8, are related by CPT theorem. ralities. While we defer a systematic classification of CP As mentioned above, CPT theorem (i.e., topological symmetric insulators until Sec. III, we discuss the edge classification problems with different set of symmetries theory of the CP symmetric topological insulators, and related by CPT relations) may largely be expected, for perturbations to it such as mass terms. example, once we anticipate description of SPT phases We will also show, in Sec. II B, that systems with CP by an underlying topological field theories. However, symmetry and charge conservation [charge U(1) symme- perhaps more fundamentally, we will also discuss that try] can also be interpreted as BdG systems that preserve while physical Hamiltonians may not obey CPT theorem, parity and one component of SU(2) spin, Sz, say [spin their entanglement Hamiltonians obey a form of CPT U(1) symmetry]. That is, CP symmetric topological in- theorem.33,49,50 sulators can be realized as topological crystalline super- We will also discuss yet another duality relation, “T- conductors with Sz conservation. In terms of edge the- duality”, for a wide range of topological insulators and ories, the relation between CP symmetric insulators and superconductors. T-duality is a duality that exchanges a P symmetric BdG systems is nothing but T-duality or phase field (φ ∼ ϕL + ϕR) and its dual (θ ∼ ϕL − ϕR) in the Kramers-Wannier duality. To the best of our knowl- the (1+1)-dimensional boson theory or in string theory. edge, this topological superconductor protected by spin [Here, φ and θ are the compact boson fields in the (1+1)- U(1) and parity symmetry has not been discussed in the dimensional boson theory and ϕL/R are their left/right- literature. moving parts.] Similarly to CPT theorem, this duality In Sec. II C, following Ref. 33, we make a further relation (and its proper generalization to K-matrix the- connection between CP symmetric insulators and time- ory with multi-component boson fields) relates topologi- reversal (T) symmetric insulators by considering entan- cal classification of (2+1)-dimensional fermionic systems glement Hamiltonians and effective symmetries thereof. with CP symmetry and charge U(1) symmetry to topo- Then we introduce the ideas of CPT-equivalence and T- logical classification with parity symmetry and spin U(1) duality of topological phases in Sec. II D, taking the symmetry. The latter system is a Bogoliubov-de Genne CP symmetric TI and its related systems [shown in FIG. (BdG) system with conserved parity and spin U(1) sym- 1(a)] as an example that shows such equivalence. metry. Therefore, topological classification of (i) time- In Sec. III, we use K-theory to classify noninteracting reversal symmetric insulators conserving charge U(1) CP symmetric TI in arbitrary dimensions. We found the (the quantum spin Hall effect), (ii) CP-symmetric insula- topological classification of this symmetry class is exactly tors with conserving charge U(1), (iii) parity-symmetric the same as that of the symmetry class possessing TRS. BdG systems with conserved spin U(1), and (iv) TC- An explicit construction of the ”effective” TRS opera- symmetric BdG system with conserved spin U(1), are all tor from the Dirac Hamiltonian with CP symmetry is related (equivalent); all these systems are classified by a given. Then the topological invariants of CP symmetric Z 2 topological number. Such relation is shown pictori- TI are also constructed. On the other hand, using the ally in FIG. 1(a). FIG. 1(b) shows another example for extension problem of the Clifford algebra, the BdG sys- CPT-equivalent insulators and BdG systems with related tems with spin U(1) and P symmetries are also shown symmetries. to fall in the classification equivalent to the symmetry While we were preparing the draft, a preprint that is class possessing TRS in any dimensions. We thus extend 51 related to this paper appeared on arXiv. While our CPT-equivalence and T-duality that we observe in Sec. analysis in terms the K-theory largely overlaps with this II in terms of 2D fermionic systems to all dimensions. preprint, our analysis in terms of K-matrix theories and With the same idea, in Sec. III E, we also study topo- our discussion in terms of CPT-theorem and T-duality logical phases protected by PT symmetry, which are were not discussed therein. CPT-equivalent partners of topological phases protected by PHS. While the latter is usually implemented by a BdG Hamiltonian that breaks charge U(1) symmetry (su- A. Outline and main results perconducting system), it is interesting to find a sys- tem of insulator with nontrivial topology protected by The structure of the paper and the main results are PT that manifests the same topological features as TSCs summarized as follows. with PHS. While there is no nontrivial topological phase In Sec. II, we start our discussion by considering 2D protected by PT in 3D and 2D (here we are not inter- fermionic topological phases protected by CP symmetry. ested in the chiral topological phases in 2D), a nontrivial 4
PT symmetric TI, which is characterized by a Z2 class, where the two-component fermion annihilation operator exists in a 1D (and 0D) system. at site r on the two-dimensional square lattice, ψ(r), is In Sec. IV we discuss CPT-equivalence for more gen- given in terms of the electron annihilation operators with T eral symmetry classes. It can be stated as ”topological spin up and down, cr,1/2, as ψ (r) = (cr,1,cr,2), and we CPT theorem” for non-interacting fermionic systems in take t = ∆ = 1. There are four phases separated by three arbitrary dimensions. Furthermore, the complete classi- quantum critical points at µ = 0, ±4, which are labeled fication of TIs and TSCs (and TCIs and TCSCs if spa- by the Chern number as Ch = 0 (|µ| > 4), Ch = −1 tial symmetries are present) for non-interacting fermionic (−4 <µ< 0), and Ch = +1 (0 <µ< +4). In the systems with T, C , P, and/or their combinations is ob- following, we are interested in the phase with Ch = ±1. tained by considering symmetry classes ”AZ+CPT”: (a) † In momentum space, T = k∈BZ ψ (k)[~n(k) · ~σ] ψ(k), CPT-equivalent symmetry classes ”generated” from AZ classes by a trivial CPT symmetry; (b) Other symmetry −P2∆ sin kx ~n(k)= −2∆ sin ky . (2) classes ”generated” from AZ classes by nontrivial CPT symmetries. The result is summarized in TABLE II. 2t(cos kx + cos ky)+ µ In Sec. V we use (Abelian) K-matrix theory to classify We will mostly focus on the case of Ch = ±1. 2D interacting topological phases protected by T, C, P A lattice model of the topological insulator with CP the combined symmetries, and/or U(1) symmetries, for symmetry can be constructed by taking two copies of the either bosonic or fermionic systems. The results are sum- above two-band Chern insulator with opposite chiralities. marized in TABLE III and IV. Comparing with the case Consider the Hamiltonian in momentum space, of non-interacting fermions, we also give an interacting † version of ”topological CPT theorem” for 2D interacting H = ψs(k)[~ns(k) · ~σ] ψs(k) bosonic and fermionic topological phases. The key point k∈BZ s=↑,↓ X X is that any perturbations (not necessary Lorentz scalars) = Ψ†(k)H(k)Ψ(k), (3) that can gap the edges of a 2D bulk must be invariant un- ∈ der a ”trivial” CPT symmetry. We also discuss T-duality kXBZ in the K-matrix formalism. Both CPT-equivalence and where s =↑, ↓ represent “pseudo spin” degrees of free- T-duality for 2D interacting topological phases can be dom, Ψ(k) is a four-component fermion field, and ~ns(k) seen manifestly in Table III and IV. is given, in terms of ~n as ~n↑(k)= ~n(k), ~n↓(k)= ~n↑(k˜)= ~n(k˜), where k˜ = (−k1, k2). I.e., the single particle Hamil- tonian in momentum space is given in terms of the 4 × 4 II. 2D FERMIONIC TOPOLOGICAL PHASES matrix, PROTECTED BY SYMMETRIES H(k)= nx(k)τzσx + ny(k)τ0σy + nz(k)τ0σz, (4)
In this section, we start our discussion by considering where τ0,z is the Pauli matrix acting on the pseudo spin a simple fermionic tight-binding model which is invariant index. under CP symmetry. We will also note that the fermionic The Hamiltonian is invariant under the following CP system can also be interpreted as a topological supercon- transformations: ductor (BdG system) that conserves the z-component of −1 † spin. Later, we will comment on the connection between UΨ(x)U = UCPΨ (˜x), (5) CP symmetric TIs and T symmetric TIs (QSHE), and in- wherex ˜ := (−x1, x2) and UCP is given either by troduce the ideas of CPT-equivalence and T-duality for U = τ σ U T =+U , (η = +1), topological phases. CP x x CP CP T UCP = τyσx,UCP = −UCP, (η = −1). (6) To ditinguish these two cases, we introduced an index A. CP symmetric insulators η; η = ±1 refers to the first/second case. We will also use notation η = e2πiǫ where ǫ = 0, 1/2 for η = 1, −1, 1. tight-binding model respectively. It turns out imposing UCP = τxσx leads to the CP sym- Let us consider the following tight-binding Hamilto- metric topological insulator. This can be seen by looking nian: at the stability of the edge mode that can appear when we terminate the system in y-direction (i.e., the edge is t i∆ T = ψ†(r) ψ(r +ˆx) + h.c. along the x-direction.) One can check, numerically, and i∆ −t r also in terms of the continuum edge theory (see below), X t ∆ UCP = τxσx protects the edge state while UCP = τyσx + ψ†(r) ψ(r +ˆy) + h.c. −∆ −t does not. In the following, these CP transformations will be combined with charge U(1) gauge transformation, and µ 0 + ψ†(r) ψ(r), (1) the corresponding transformation will be denoted by U 0 −µ (See below). 5
2. Edge theory be either column or row vector depending on the con- text.] The matrix elements obey Ξ = Ξ† (hermiticity) T We now develop a continuum theory for the edge state and ∆ = −∆ (Fermi statistics). The presence of SU(2) that exists when the system is terminated in y direction. spin rotation symmetry is represented by I.e., the edge is along x direction. Let us consider the sa 0 free fermion: H4,Ja =0, Ja := T , (13) 0 −sa v H = dx ψ† i∂ ψ − ψ† i∂ ψ . 2π L x L R x R where a = x,y,z. Z v † ψL With conservation of one component of spin, say, z- = dxΨ i∂xσzΨ , Ψ= . (7) 2π ψR component, we have a U(1) symmetry associated with Z rotation around z-axis. With this Sz conservation, one We consider two types of CP symmetry operation can reduce this BdG Hamiltonian into the following form: −1 iα † c UψL(x)U = e ψR(−x), ξ δ H = c† , c H ↑ , H = ↑ , (14) −1 iα † ↑ ↓ 2 c† 2 † T Uψ (x)U = ηe ψ (−x), (8) ↓ δ −ξ↓ R L where CP transformation is combined with the EM U(1) up to a term which is proportional to the identity matrix. charge twist with an arbitray phase factor α. The sign At the quadratic level, this Hamiltonian is a member of η = ± is +/− for topological/non-topological cases: symmetry class A (unitary symmetry class in AZ classes). [with c → c† , one can “convert” spin U(1) to fictitious +1 topological ↓ ↓ η = e2πiǫ = (9) −1 trivial charge U(1)]. This can be seen as follows. Let us consider BdG Hamiltonians which are invariant under rotations This is CP symmetry with twisting by the charge opera- about the z- (or any fixed) axis in spin space, yielding tor, to the condition [H4,Jz] = 0, which implies that the Hamiltonian can be brought into the form U = e−iαFV CP, (10) a 0 0 b where FV is the total charge operator, 0 a′ −bT 0 H = , a† = a, a′† = a′.(15) 4 0 −b∗ −aT 0 † † † ′T FV := FR + FL = dx ψ ψR + ψ ψL . (11) b 0 0 −a R L Z There are two fermion mass bilinears that are consis- Due to the sparse structure of H4, we can rearrange the tent with the charge U(1) symmetry: These masses are elements of this 4N × 4N matrix into the form of a 2N × odd under CP when η = +1. We thus conclude that 2N matrix H2 above. the edge theory is at least at the quadratic level stable Let us now consider parity symmetry. For simplicity, (ingappable). we assume orbitals transform trivially under parity, and hence assume the following form:
−iα B. BdG systems with spin U(1) and P symmetries c↑(r) −1 e c↓(˜r) P P = iα . (16) c↓(r) ηe c↑(˜r) In this section, we show that systems with charge U(1) and CP symmetry can be derived from BdG systems with Within the reduced 2N × 2N basis, parity symmetry conserved one component of spin (Sz, say) and parity looks like CP symmetry. To see this, let us write out symmetry. the Hamiltonian in the following form: The system of our interest preserves spin U(1) but not ξ δ c charge U(1). At the quadratic level, this situation is de- H = c† , c ↑ ↑ (17) scribed by the BdG Hamiltonian. Following Altland and ↑ ↓ δ† −ξT c† ↓ ↓ Zirnbauer,3 we consider the following general form of a = c† (r)ξ (r, r′)c (r′) − c (r)ξT (r, r′)c† (r′) BdG Hamiltonian for the dynamics of quasiparticles ↑a ↑ab ↑b ↓a ↓ab ↓b + c† (r)δ (r, r′)c† (r′)+ c (r)δ† (r, r′)c (r′) 1 c ↑a ab ↓b ↓a ab ↑b H = c†, c H , 2 4 c† (summation over repeated indices are implicit). Then, Ξ ∆ H4 = ∗ T , (12) T −∆ −Ξ −1 † ξ↓ −ηδ c↑ PHP = c , c↓ ∗ T † ↑ ′ −ηδ −ξ br˜ ↑ br′,ar c↓ where H4 is a 4N × 4N matrix for a system with N ar˜ † (18) orbitals (lattice sites), and c = (c↑, c↓). [c and c can 6
(The transpose T here acts both a and r). Thus, the a. T-duality (Kramers-Wannier duality) By taking invariance under P implies T-dual or Kramers-Wannier-dual of the above setting, † T ψ (x) → ψ (x), (27) ξ↓ −ηδ ξ↑ δ L L ∗ T = † T (19) −ηδ −ξ↑ ′ δ −ξ↓ ′ we obtain the P symmetric system. In the bosonized ar,b˜ r˜ ar,br language, this amounts to exchange phase field φ and its With the transformation or relabeling dual θ. Also, if we decompose the complex fermion ψL in 1,2 1 terms of two real (Majorana) fermions χL , ψL = χL + † iχ2 , the above transformation amounts to χ2 → −χ2 c↑ =: Ψ↑, c↓ =: Ψ↓, (20) L L L while keeping the right moving intact. This is nothing we can write the Hamiltonian as but the Kramers-Wannier duality in the Ising model. The P symmetry, dualized from CP symmetry above, † ′ ′ is given by H = Ψ (r) H2(r, r ) Ψ(r ). (21) r,r′ −1 −iα X UψL(x)U = e ψR(−x) −1 iα Provided the system has translational symmetry, UψR(x)U = ηe ψL(−x). (28) H (r, r′) = H (r − r′), with periodic boundary condi- 2 2 This is P symmetry with Sz twisting, tions in each spatial direction (i.e., the system is defined − on a torus T d), we can perform the Fourier transforma- U = e iαFA P, (29) tion and obtain in momentum space where FA is the axial charge,
† H = Ψ (k) H2(k) Ψ(k), (22) † † FA := FR − FL = dx ψRψR − ψLψL . (30) k∈Bz X Z where the crystal momentum k runs over the first Brillouin zone (Bz), and the Fourier component of C. Connection to T symmetric insulators the fermion operator and the Hamiltonian are given −1/2 ik·r by Ψ(r) = V k∈Bz e Ψ(k) and H2(k) = The CP symmetric model introduced above is in fact e−ik·rH (r), respectively. also time-reversal invariant in the absence of perturba- r 2 P Then the P invariance demands tions. If there is Lorentz invariance, because of CPT P theorem, any perturbation to the model that is CP sym- T ξ↓ −ηδ ξ↑ δ metric is also T symmetric. Hence, within Lorentz invari- ∗ T = † T . (23) ant theories, the same set of perturbations is prohibited −ηδ −ξ↑ ˜ δ −ξ↓ −k k by CP and T symmetries. The topological phase pro- Observing that tected by CP symmetry can thus be also viewed as a T symmetric topological phase. T T However, the above argument based on CPT theorem 0 1 −ξ ηδ 0 η ξ↑ δ ↓ = η of course raises a question as we do not want to be con- η 0 ηδ∗ ξT 1 0 δ† −ξT ↑ ↓ fined to relativistic systems, and Lorentz invariance is ab- (24) sent in the lattice model. Note, however, the following: (i) CPT theorem tells us the presence of antiparticles. we then conclude that, when η = 1, This seems a necessary ingredient to have a topological phase (topologically non-trivial “vacua”). (ii) Topologi- ˜ T T τxH2(−k) τx = −H2(k), τx =+τx (25) cal phases that are characterized by a term of topological origin in the response theory, such as the Chern-Simons whereas when η = −1, term or the axion term for the external (background) U(1) gauge field, are Lorentz invariant. This in particu- ˜ T T τyH2(−k) τy = −H2(k), τy = −τy. (26) lar means CP symmetry dictates the theta angle to be 0 or π (mod 2π), just as TRS does. I.e., the single-particle Hamiltonian H2 is CP symmetric. Finally, while Hamiltonians may violate Lorentz in- It should be noted that P symmetry with η = +1 variance, and hence CPT theorem, a version of CPT is somewhat unusual. When acting twice on spinors, like theorem applies to wavefunctions (= projection op- P2 = +1, whereas we usually expect P2 = −1. This is so erators), or the “entanglement Hamiltonian”. In other since parity should reverse the sign of angular momen- words, wavefunctions or the entanglement Hamiltonian tum, either of orbital or spin origin. The P symmetry have more symmetries than the physical Hamiltonian. with η = 1 can be considered as a composition of a P Due to this, for any CP symmetric system, one can de- symmetry with η = −1 and spin parity (−1)N↑ where fine “effective” time-reversal symmetry for the projector N↑ is the number operator associated to up spins. or the entanglement Hamiltonian. See Appendix A. 7
D. CPT-equivalence and T-duality of topological FIG. I shows some examples about CPT-equivalence phases: an example and T-duality among topological (crystalline) insulators and superconductors. Especially, FIG. 1(a) shows the The above discussion reveals a ”CPT-equivalence” be- connection between T symmetric TIs, CP symmetric TIs, tween CP and T symmetric topological phases. Further- and their dual realizations in BdG systems with Sz con- more, from the fact that, both CP symmetric TI with servations introduced in this section. Another example, (CP)2 = 1 [(CP)2 = (−1)Nf ] and T symmetric TI with as shown in FIG. 1(b), is the CPT-equivalence between T 2 Nf 2 T = (−1) (T = 1), where Nf is the total fermion symmetric TSCs, P symmetric TCSCs, and T and P sym- number operator, possess the same nontrivial Z2 (trivial) metric TCIs that can support gapless edge states even in classification in two dimensions, we expect a specific cor- the absence of charge U(1) symmetry. respondence between these two ”CPT-equivalent” topo- In the following section, we make a more precise dis- logical phases. In general, such correspondence can be cussion for the idea of CPT-equivalence and T-duality observed among topological phases protected by discrete introduced here, focusing on non-interacting fermionic symmetries T, C, P, and/or their combinations. We will CP symmetric TIs in arbitrary dimensions discuss it in the following sections. On the other hand, there is a duality – which we call ”T-duality” in this paper – between topological phases III. CLASSIFICATION OF CP SYMMETRIC of insulating and superconducting systems with corre- TIS IN ARBITRARY DIMENSION sponding symmetries. Imposing a symmetry g on a BdG system with Sz conservation (14) will result a constraint In this section we consider systems of non-interacting on the reduced BdG Hamiltonian H2 by the dual sym- fermions with CP symmetry and classify CP symmetric metry g, which is exactly in the same symmetry class as TIs in arbitrary dimensions using K-theory. a tight-binding Hamiltonian H constrained by the sym- Relevant symmetries are written as constraints on the Hamiltonian matrix H as follows. The particle-hole sym- metry ge in a insulating system (with charge conservation implicitly). For example, as we discussed in Sec. II B, metry (PHS) is an anti-unitary operator C that anti- commutes with the Hamiltonian as {C, H} = 0, which is the CPe symmetric topological phase can also be realized in a BdG system with P symmetry and Sz conservation. equivalently written using an unitary operator UC as Another known example is that a chiral symmetric topo- ∗ −1 logical phases (class AIII in AZ class) can also be in- UCH (−k1,..., −kd)UC = −H(k1,...,kd). (34) terpreted as a BdG Hamiltonian possessing TRS and Sz conservation.10,52 Interestingly (and expectedly), a 2D T The parity symmetry P is a symmetry that swaps left- symmetric TI, i.e., the QSHE, also has a dual realization handed and right-handed coordinates, which can be im- in a superconducting system – a BdG system with chi- plemented as a mirror symmetry with respect to a par- ticular direction (here we take k1) as ral symmetry and Sz conservation. This can be seen by a similar discussion from Sec. II B. For a reduced BdG −1 P H(−k1, k2,...,kd)P = H(k1, k2,...,kd), (35) Hamiltonian (by Sz conservation) (14), if we impose a ”chiral” symmetry S (which is defined as a combination with a unitary operator P . Combining these two sym- of T and C symmetries) as metries C and P , we define CP symmetry by an unitary † † c e−iαc c operator UCP satisfying S ↑ S−1 = ↓ = e−iατz U ↑ , c iαc† c† ↓ ηe ↑ ! ↓ ! ∗ −1 UCPH (k1, −k2,..., −kd)UCP = −H(k1, k2,...,kd). τ for η =1 (36) U = x (31) iτy for η = −1, A CP symmetric TI is a topological insulator that does then not possess C nor P symmetry but is characterized with c a combined CP symmetry. −1 c† c ↑ −1 H = SHS = S ↑, ↓ H2 c† S ↓ c c† c † ∗ ↑ A. Classification by K-theory = ↑, ↓ U H2U c† (32) ↓ † T In non-interacting fermion systems, CP symmetric TIs (note that S = TC is antiunitary) implies U H U = H2, 2 are classified using K-theory in a way similar to the i.e., the single-particle Hamiltonian H2 is TRS. In conclu- classification of topological defects discussed by Teo and sion, dual symmetries between the tight-binding Hamil- 53 tonian (with charge conservation) and the BdG reduced Kane. Hamiltonian (with S conservation) have the following A TI with CP symmetry (36) is regardedas a TSC with z ˜ ˜ correspondences: PHS C = CP in the d = d − 1 dimensions with momenta k2,...,kd (that are flipped by an action of C˜), containing T ↔TC, P ↔CP. (33) a defect with a co-dimension D = 1 parameterized with 8 k1 (that is not flipped by C˜). When we have PHS C˜ with with a complex conjugation K. Now we can construct 2 2 C˜ = +1 or C˜ = −1, the symmetry class is class D or an effective TRS from CP symmetry as T = γ1UCPK, class C and the associated classifying spaces Rq are given satisfying 11,32 as ∗ −1 γ1UCPH (−k1,..., −kd)(γ1UCP) = H(k1,...,kd). 2 class D C˜ = 1 : R2 (q = 2), (43) 2 class C C˜ = −1 : R6 (q = 6). (37) The existence of γ1 in the Dirac model enables us to convert the CP symmetry into a TRS, which is not the Then the classification for CP symmetric TI is given by case for a general lattice model where a kinetic term 53 a homotopy group along reflected coordinate is not necessarily written by a gamma matrix γ1. πD(Rq−d˜) ≃ π0(Rq−d˜+D)= π0(Rq+2−d). (38) This can be interpreted that the relevant classifying space changes from Rq to Rq+2, which corresponds to the sym- C. Topological invariants metry class AII (R4) orclass AI(R8 ≃ R0), both possess- ing TRS. Thus CP symmetric TI behaves similar to TR Topological invariants of CP symmetric TIs are con- symmetric TI in terms of topological classification and structed in the same way as those for topological corresponding edge states. This is consistent with the defects.53 For q+2−d =0, 4 in Eq. (38), we have topolog- CPT theorem for Lorentz-invariant systems where CP ical invariants Z. Due to q =2, 6 [Eq. (37)], the topolog- symmetry can be effectively converted into time reversal ical invariants Z are realized in even dimensions d, where symmetry. we can define the Chern number over the Brillouin zone. While we adopted the parity symmetry (35) that flips The Chern number gives the topological invariants, which only one momentum k1, we can generally reverse 2n +1 is written, by putting d =2n, as coordinates for the parity as n 1 2n iF −1 Chn = d k tr , P H(−k1,..., −k2n+1, k2n+2,...,kd)P n! 2π Z = H(k1,...,k2n+1, k2n+2,...,kd). (39) F = dA + A ∧ A, A = huk|d|uki, (44)
The TI with CP symmetry constructed from above P with valence bands |uki and a derivative d with respect can be regarded as a TSC with PHS in the d˜= d − 2n − to momenta k. 1 dimensions with momenta k2n+2,...,kd, containing a Next, first descendant Z2 is given by a Chern-Simons defect with a co-dimension D = 2n + 1 parameterized form, which takes place for q +2 − d = 1 in Eq. (38), with k1,...,k2n+1. Then the classification is given by a so that the dimension d is odd. When we have the first homotopy group descendant Z2, we can choose a continuous gauge A over the entire Brillouin zone, and an integration of the Chern- π (R )= π (R ), (40) D q−d˜ 0 q+4n+2−d Simons form, which is defined for odd dimensions, gives Z where the relevant classifying space looks like Rq+4n+2. topological invariant 2. Z Since we have q = 2 or q = 6, Rq+4n+2 becomes R0 or Second descendant 2 is given by a dimension reduc- Z R4, i.e, the classifying space associated with the symme- tion of the above 2. We consider a one parameter try class with TRS, which is consistent with the CPT family of the Hamiltonian H˜(θ, k) connecting the orig- theorem. inal Hamiltonian H˜(0, k) = H(k) and a reference CP symmetric Hamiltonian H˜(π, k) = H0 with a parameter 0 ≤ θ ≤ π. If we extend a range of θ into −π ≤ θ ≤ π by B. Dirac models and CPT theorem a relation H˜(θ, k , k ,...,k ) While we cannot explicitly construct an anti-unitary 1 2 d ˜∗ −1 operator T for TRS from CP symmetry in general cases, = −UCPH (−θ, k1, −k2,..., −kd)UCP, (45) we can construct T operator from CP symmetry in a ˜ Dirac Hamiltonian we can define a CP symmetric Hamiltonian H(θ, k) over −π ≤ θ ≤ π and k. Then the second descendant Z2 d characterizing the Hamiltonian H(k) is given by an inte- H(k)= mγ + k γ , (41) 0 i i gration of Chern-Simons form for H˜(θ, k). i=1 X where γi’s are anti-commuting gamma matrices, m is a mass, and ki’s are momenta. The CP symmetry (36) D. BdG systems with spin U(1) and P symmetries then leads to relations As we have seen in Sec. IIB, CP symmetric TI can {U K,γ } =0, i =0, 1, CP i be realized by a BdG system with a reflection symme- [U K,γ ]=0, 2 ≤ i ≤ d, (42) CP i try and spin U(1) symmetry (Sz conservation). Here we 9 interpret their equivalence to class AII TIs in terms of unique mass compatible with the CP symmetry. Thus K-theory and Clifford algebras. When we have a unitary Hamiltonians with different signs of the unique mass term operator commuting with the Hamiltonian, we should are topologically distinct. If we try to double the system block-diagonalize the Hamiltonian when we consider a where doubled 2 by 2 degrees of freedom is described by topological classification. When the Sz anti-commutes Pauli matrices ρ, allowed mass terms are not unique since with the PHS C and the reflection symmetry P , the we have σxτxρy and σxτyρy in addition to σz. Then we block-diagonalized Hamiltonian does not possess C nor can adiabatically connect two states in the doubled sys- P any further, while the combined CP still remains as tem by appropriately rotating in the space of mass term, a symmetry of the block Hamiltonian. The situation is which indicates that the classification of CP symmetric summarized as follows, TI in 2D is Z2. Next we show that the above classification for class D {C, H} =0, [Sz, H]=0, {C,Sz} =0, {P,Sz} =0, accompanied with spin U(1) and parity P is equivalent 2 2 to that in class AII, by adopting Clifford algebras classi- Sz =1, P =1, (46) fication for Dirac models. The original classification for along with a parity symmetry (35). [Note: we can choose class D in d-dimensions is given by a Clifford algebra32 2 P = 1 by appropriately multiplying “i”, which may {γ ,C,CJ,Jγ ,...,Jγ } (52) change a commutation/anticommutation relation with 0 1 d C.] and its extension problem with respect to the mass term Now let us look at a topologically non-trivial example γ0 is of this construction. We start with a BdG Hamiltonian Cl → Cl , (53) in class D and two dimensions as d,2 d,3 where the topological index is given by π0(R2−d), espe- H(k)= nx(k)σx + ny(k)σy + nz(k)σz , (47) cially Z for d = 2. The spin U(1) symmetry (Sz) and the parity symmetry (P ), satisfying (46) and [C, P ] = 0, can where ~n(k) is defined in Eq. (2) and we have PHS of C = be included in the Clifford algebra as σxK. A parity symmetry (35) is implemented by taking two copies of the above BdG Hamiltonian (denoted by {γ0,C,CJ,Jγ1,...,Jγd,Jγ1P,γ1PSz}, (54) τ) and a spin U(1) symmetry is implemented by taking for which the extension problem for γ is written as two copies representing spin degrees of freedom (denoted 0 by s), which yields Cld,4 → Cld,5, (55)
and the classification is given by π0(R4−d). In our exam- H(k)= nx(k)σxτzsx + ny(k)σysx + nz(k)σz, (48) ple in d = 2, we have Z2 topological number. Above clas- where we have the spin U(1) symmetry Sz = σzsz. We sification is the same as that for class AII in d-dimensions, have two parity symmetries P (a reflection symmetry which shows that CP symmetric TI and TR symmetric with respect to x-direction) written as TI are equivalent in the level of Dirac models. Indeed, the effective TRS is given from CP symmetry and a kinetic gamma matrix as (43). τxsx, (topological) P = (49) On the other hand, in the case of the parity symmetry τ s . (trivial) ( y x anti-commuting with PHS ({C, P } = 0), the relevant Clifford algebra is A choice of parity P = τxsx, commuting with PHS ([C, P ] = 0), leads to a topologically non-trivial insu- {γ0,C,CJ,Jγ1,...,Jγd,γ1P,Jγ1PSz}, (56) lator, as explained later with Clifford algebras. We can choose either of parity symmetries by adding appropriate and the extension problem for γ0 is terms to the Hamiltonian. Cld+2,2 → Cld+2,3. (57) Block diagonalization with respect to Sz becomes Then the topological invariant is π (R− ), where we have clear, if we change bases as (σxsz, σysz,sx,sy, σzsz) → 0 d a trivial insulator for d = 2 as π (R− ) = 0, that is (σxsx, σysx, σzsx,sy,sz), 0 2 equivalent to class AI in d = 2. This is the reason why
H(k)= −nx(k)σxτzsz − ny(k)σysz + nz(k)σz, we have a trivial insulator if we choose a parity symmetry S = s . (50) P = τysx in Eq. (49). z z While we have so far discussed the system in class D with S and P in two dimensions, we note that a sys- The block Hamiltonian with sz = −1 is given by z tem in three dimensions possesses non-trivial Z2 topo- H(k)= nx(k)σxτz + ny(k)σy + nz(k)σz, (51) logical invariant. This is interesting, since the original class D system in three dimensions is trivial and a block- characterized by a CP symmetry with UCP = τxσx. This diagonalized system with Sz (class A system) is also triv- corresponds to a non-trivial CP symmetric TI given in ial, while the CP symmetry gives rise to a non-trivial Eq. (6) with η = +1. In (51), the mass term σz is the insulator. 10
E. Topological classification of other symmetries CPT operator W = W0 such that the system transforms ”topologically-trivially” under W0. Therefore, we have In a similar manner as CP symmetric TIs, we can de- the following statement: fine PT symmetric TIs. PT symmetry can be defined by Topological CPT theorem for noninteracting fermionic systems: a unitary operator UPT satisfying Let {gi} be a set of symmetries (can be a null set) composed of T , C, P , and/or their ∗ −1 UPTH (k1, −k2,..., −kd)UPT = H(k1, k2,...,kd). (58) combinations. Then for non-interacting fermionic sys- tems there is a ”trivial” CPT operator W = W0, which Then PT symmetric TI is a topological insulator that anticommutes with T and P and commutes with C (from does not possess P nor T symmetry but is characterized which other commutation relations between gi and W0 with a combined PT symmetry. In an analogous way can also be deduced), such that the system with symme- for CP symmetric TI, a classification of PT symmetric tries {gi} and the system of with symmetries {gi, W0} TI in d-dimensions is obtained by considering a system possess the same classifying space or topological classifi- with TRS in d − 1-dimensions containing a topological cation. defect with co-dimension 1. We assume that classifying The proof of the above theorem is straightforward as space for the TR symmetric TI with T˜ = UPTK in 0- we consider the Dirac Hamiltonian (41) (the idea here is 2 dimensions is Rq. We have q = 0 for (UPTK) = +1, similar to the discussion in Sec. III B) : 2 and q = 4 for (UPTK) = −1. Then the classification for PT symmetric TI is given by d H(k)= mγ0 + kiγi, π (R ) ≃ π (R ). (59) i=1 1 q−(d−1) 0 q+2−d X Non-trivial PT symmetric TIs are found in 2-dimensional where γi’s are anti-commuting gamma matrices. The systems in class AI or AII with a reflection symmetry, symmetries T , C, and P (if present) satisfy where TRS and P are broken by a diagonalization with respect to some unitary symmetry but the combined PT [T,γ0]=0, {T,γi=06 } =0, remains, which is characterized by a non-trivial topolog- {C,γ0} =0, [C,γi=06 ]=0, Z ical number . A shift of classifying space by 2 is in- {P,γ1} =0, [P,γi=16 ]=0, (61) terpreted as a change of an effective symmetry class into that with PHS, which is again consistent with the CPT while the CPT symmetry W satisfies theorem. {W, γ0} =0, [W, γi=1]=0, {W, γi=06 ,1} =0. (62)
IV. TOPOLOGICAL CPT THEOREM AND Define M = γ1W , we then have [M,γi]=0 ∀i and thus TOPOLOGICAL CPT-EQUIVALENT M is an unitary symmetry commuting with H: SYMMETRY CLASSES [M, H]=0,M 2 =1. (63) Actually, as we discussed in previous sections, such CPT-equivalence holds ”topologically” for more general For the system with symmetries {gi}, composed of T , symmetry classes (not just for cases discussed in the last C, P , and/or their combinations, if the additional sym- sections). In this section, we discuss the ”topological metry M = M0 commutes with all gi, or equivalently if CPT theorem” and topological CPT-equivalent symme- W = W0 satisfies (if {gi} includes some of the following try classes in noninteracting fermionic systems. symmetries) Combining symmetries T , C, and P , we define the CPT symmetry by an unitary operator W satisfying {W0,T } =0, [W0, C]=0, {W0, P } =0, {W0, CP } =0, [W0, T P ]=0, {W0, T C} =0, (64) W H(k˜)W −1 = −H(k), W 2 =1, (60) we can block diagonalize H with respect to M such that where H(k) is a d-dimensional (d ≥ 1) single particle all symmetries gi are still preserved in each eigenspace of Hamiltonian and k˜ = (−k1, k2,...,kd). Note that here, M. Therefore, the symmetry class and hence the clas- 2 as W is unitary, W can alway be fixed to be 1 by the sification would not change as the symmetry M0 or W0 ′ iα redefinition W = e W with any phase factor (such is added to the original set of symmetries {gi} of the redefinition is also accompanied by changing the com- system. This completes the proof. mutation relations with other existed symmetries at the We would like to point out that, though the topologi- same time). Now, if the system already has some sym- cal CPT theorem is ”proved” (or argued) by considering metries, adding the CPT symmetry constraint [Eq. (60)] the Dirac model (as a representative model of Clifford on H would or would not change the classifying space algebras that capture the topology of classifying spaces), or the topological classification with respect to existing which seems obviously to be invariant under a (trivial) symmetries. That is, in the latter case, there exists a CPT symmetry because of its Lorentz invariance, the 11 same conclusion can be reached by more (mathemati- all have the same classification. In Refs. 31 and 32 the cally) rigorous ways, such as topological K-theory, which first four symmetry classes are denoted respectively as is irrelevant to Lorentz invariance. Actually, this is what classes DIII, D+R+, AII+R−, and DIII+R−+, which we did (in Sec. III) in the discussion for equivalence be- have the same zero-dimensional classifying space R3 and tween T and CP, and C and PT, as part of topological thus the same classification in any dimension (using K- CPT theorem discussed here, using K-theory in a way theory). Moreover, it can be checked that, from the re- similar to topological defects discussed in Ref. 53. sults of Refs. 31 and 32, the topological CPT theorem indeed holds. We note that a natural choice of C, P , and T for spin- Class SM,TC SW,TC Classifying space 1/2 fermions leads to a trivial CPT as expected. This AIII − + C0 can be explicitly seen in the CPT-equivalence of class + ++− Class SM,T or SM,C SW,T or SW,C Classifying space DIII [Γ−+(T, C)] and class DIII+R−+ [Γ−++(T,C,P )]. For spin-1/2 fermions, the symmetry operators are given AI,AII − + C0 as D, C − − C0
Class (SM,T,SM,C) (SW,T,SW,C) Classifying space C = σxK, P = sx,T = isyK, (69)
(−, +) (+, +) Rq → Rq+1 BDI, CII where σi,si are Pauli matrices acting on particle-hole and (+, −) (−, −) Rq → Rq−1 spin degrees of freedom. The parity symmetry P is a (−, −) (+, −) C1 reflection along x-direction and involves a π-rotation of (+, −) (−, −) Rq → Rq+1 spin around x-axis, which is denoted by R−+ in classi- DIII, CI (−, +) (+, +) Rq → Rq−1 fication in Refs. 31 and 32. Then if we consider CPT
(−, −) (+, −) C1 symmetry given as W = −iCPT = σxsz, W satisfies commutation relations in Eq. (64) and is a trivial CPT. TABLE I. Classification of AZ symmetry classes in the Thus an addition of a trivial CPT W changes class DIII presence of a nontrivial CPT symmetry W or an addi- + ++− [Γ−+(T, C)] to class DIII+R−+ [Γ−++(T,C,P )], while it tional unitary symmetry M (commuting with H) in zero does not change the topological classification. dimension. SM,g and SW,g dictate commutation (+) or On the other hand, adding a nontrivial CPT symmetry anticommutation(−) relation of symmetry g, which can be W [which can be represented as a combination of a triv- T , C, or TC. ”Rq” in the last column denotes the origi- nal classifying spaces for the corresponding symmetry classes ial CPT symmetry and some (onsite) order-two unitary (before adding M or W ). symmetry commuting with H such that W changes the commutation relations (64)] to a symmetry class would change the original classifying space. Using the result of Ref. 32, we can directly obtain the change of classifying Based on topological CPT theorem, some symmetry spaces for AZ symmetry classes in the presence of extra classes, defined as topological CPT-equivalent symmetry unitary symmetry M = γ W (commuting with H). The classes here, possess the same classification. For example, 1 result is summarized in TABLE I. symmetry classes The complete classification of TIs and TSCs (and TCIs − Γ−(T ), Γ+(CP ), Γ−+(T,CP ), (65) and TCSCs if spatial symmetries such as P or CP are present) for non-interacting fermionic systems with T, C, all have the same classification. Here we use notations P, and/or their combinations, instead of studying these S g1,g2 symmetries separately, can also be obtained by the one ΓSg (g), ΓSg Sg (g1,g2), 1 2 with symmetry classes ”AZ+CPT” [in Refs. 31 and 32 Sg1,g2 Sg2,g3 Sg3,g1 Γ (g1,g2,g3), (66) Sg1 Sg2 Sg3 classification for symmetry classes ”AZ+P (or reflection R)” have been discussed, but some combined symmetries to denote the symmetry classes composed of {g }, with i like CP are not included there]. Two cases are involved: signs S dictating the commutation (+) or anticom- gi,gj (a) CPT-equivalent symmetry classes ”generated” from mutation (−) relation between g and g . (65) can be i j AZ classes by a trivial CPT symmetry; (b) Other sym- deduced from the following CPT-equivalent symmetry metry classes ”generated” from AZ classes by nontrivial classes : CPT symmetries (based on the result in Table I). The ⊜ − − − Γ−(T ) Γ−+(T, W0)=Γ−+(T, W0T )=Γ++(W0, W0T ) result is summarized in Table II. Generally, we can also reverse an odd number of spatial ⊜ Γ+(W0T )=Γ+(CP ), (67) coordinates as the parity P and the corresponding CPT ⊜ where ” ” represents the CPT-equivalence relations for symmetry W , with k˜ = (−k1,..., −k2n+1, k2n+2,...,kd) symmetry classes. Similarly, as another example, sym- in (60). In this situation, the ”effective” unitary sym- n metry classes metry M can be defined as M = i γ1 ··· γ2n+1W , and + + − ++− the commutation relations between the trivial CPT sym- Γ−+(T, C), Γ++(C, P ), Γ−+(T, P ), Γ−++(T,C,P ), − − − metry W0 and other symmetries (64) will also change if Γ++(CP,PT )=Γ−+(TC,PT )=Γ−+(TC,CP ), (68) n is odd (only commutation relations with antiunitary 12
(a)
CPT-equiv. sym. classes ”generated” from AZ classes by trivial CPT CqorRq π0(CqorRq)
”None” (A), Γ+(CPT ) C0 Z − Γ+(TC) (AIII), Γ+(P ), Γ++(TC,P ) C1 0 + Z Γ+(T ) (AI), Γ−(CP ), Γ+−(T,CP ) R0 + − + +−+ + Z Γ++(T,C) (BDI), Γ++(C, P ), Γ++(T, P ), Γ+++(T,C,P ), Γ−+(CP,PT ) R1 2 + Z Γ+(C) (D), Γ+(PT ), Γ++(C,PT ) R2 2 + + − ++− + Γ−+(T,C) (DIII), Γ++(C, P ), Γ−+(T, P ), Γ−++(T,C,P ), Γ++(CP,PT ) R3 0 + Z Γ−(T ) (AII), Γ+(CP ), Γ−+(T,CP ) R4 + − + +−+ + Γ−−(T,C) (CII), Γ−+(C, P ), Γ−+(T, P ), Γ−−+(T,C,P ), Γ+−(CP,PT ) R5 0 + Γ−(C) (C), Γ−(PT ), Γ−−(C,PT ) R6 0 + + − ++− + Γ+−(T,C) (CI), Γ−+(C, P ), Γ++(T, P ), Γ+−+(T,C,P ), Γ−−(CP,PT ) R7 0 (b)
Other sym. classes ”generated” from AZ classes by nontrivial CPT CqorRq π0(CqorRq ) + + + + + Z Γ++(TC,P ), Γ++(T,CP ), Γ−−(T,CP ), Γ+−(C,PT ), Γ−+(C,PT ) C0 ++− +−+ ++− +−+ Γ+++(T,C,P ), Γ−++(T,C,P ), Γ−−+(T,C,P ), Γ+−+(T,C,P ) C1 0 +++ +−− Z Γ+−+(T,C,P ), Γ+++(T,C,P ) R0 +++ +−− Z Γ+++(T,C,P ), Γ−++(T,C,P ) R2 2 +++ +−− Z Γ−++(T,C,P ), Γ−−+(T,C,P ) R4 +++ +−− Γ−−+(T,C,P ), Γ+−+(T,C,P ) R6 0
TABLE II. Classification of TIs and TSCs for non-interacting fermion systems with symmetry classes composed of T , C, P , and/or their combinations in zero dimension. This can be obtained by adding the CPT symmetry (either trivial or nontrivial ones) to the AZ classes: (a) CPT-equivalent symmetry classes ”generated” from AZ classes by trivial CPT; (b) Other symmetry classes ”generated” from AZ classes by nontrivial CPT (based on the result in Table I). In this table we have fixed [Ai,Aj ] = 0 2 and Ui = 1 (other choices are equivalent), where Ai/j and Ui represent antiunitary and unitary symmetries, respectively. Classification in arbitrary dimensions d is given by π0(Cq−d) or π0(Rq−d), as deduced from zero-dimensional classifying spaces Cq or Rq by K-theory. symmetries such as T and C will change). Nevertheless, interacting systems of either fermions or bosons, as the previous discussions on the case for n = 0 (the same for original CPT theorem applies to Lorentz invariant quan- even n) can be straightforwardly applied to the case for tum field theories with interactions. As a simple but odd n. instructive demonstration, in this section we discuss in- As related to the results in this section, a similar but teracting SPT phases (without topological order) in two more general discussion can also be found in Ref. 51. dimensions by using (Abelian) K-matrix Chern-Simons The CPT symmetry defined in (60) here is one kind of theory. order-two spatial symmetries defined there (on a system without defects). Therefore, classification of AZ classes in the presence of either trivial CPT (related by topo- A. Bulk and edge K-matrix theories incorporated logical CPT theorem) or nontrivial CPT (result shifts of with symmetries the classifying spaces shown in TABLE I) discussed here can also be deduced from the general properties of the K- We begin with the bulk K-matrix action Sbulk = 2 0 ex groups for the additional order-two spatial symmetries, dtd x Lbulk, Lbulk = Lbulk + Lbulk: as derived in Ref. 51. R 1 L0 = ǫµνλK a ∂ a , bulk 4π IJ Iµ ν Jλ ex eQI µνλ sSI µνλ V. CLASSIFICATION OF 2D INTERACTING Lbulk = − ǫ Aµ∂ν aIλ − ǫ Bµ∂ν aIλ, (70) SPT PHASES: K-MATRIX FORMULATION 2π 2π
where aµ represents the N-flavors of dynamical Chern- In the previous sections we have discussed topological Simons (CS) gauge fields, Aµ and Bµ are the external phases protected by T, C, P and/or corresponding com- gauge potentials coupling to the electric charges and spin bined symmetries and classification related by topological degrees of freedom (along some quantization axis), K is CPT theorem in non-interacting fermionic systems. Ac- an integer-valued N × N matrix (symmetric and invert- tually, such CPT-equivalence is expected to hold even for ible), and Q and S are integer-valued N-components vec- 13 tors representing electric charges (in unit of the electric a general symmetry group G that has elements as com- charge e) and spin charges (in unit of the spin charge s), binations of T , C, and P, and/or U(1) symmetries, we respectively. The currents in the bulk are have
e s −1 J µ = ǫµνλQ ∂ a , J µ = ǫµνλS ∂ a , (71) GSedgeG = Sedge, ∀G ∈ G, (77) c 2π I ν Iλ s 2π I ν Iλ with the chiral boson fields transformed as where Jc and Js are the total charge and spin currents, −1 respectively. GφG = αGUGφ + δφG, ∀G ∈ G, (78) In the bulk, we have the transformation laws under symmetries such as TRS (T ), PHS (C), and PS (P) in where αG = 1 (−1) represents an unitary (antiunitary) the x-direction [gµν = diag (+, −, −)]: operator G. Specifically, for TRS, PHS, and PS, (77) gives the constraints for the matrices UT, UC, UP, and µ ν µ ν T : Jc → gµν Jc , Js →−gµν Js , (t, x) → (−t, x), charge and spin vectors Q, S µ µ µ µ C : Jc →−Jc , Js →−Js , T TRS : UT KUT = −K, µ ˜µ µ ˜µ µ µ P : Jc → Jc , Js →−Js , x → x˜ , (72) T T IN + UT Q =0, IN − UT S =0, T where we have defined X˜ µ ≡ (X0, −X1,X2)T for any PHS : UC KUC = K, vector Xµ. We assume that the gauge fields aµ (flavor T T IN + UC Q =0, IN + UC S =0, index is suppressed) obey the following transformation T laws: PS: UP KUP = −K, T T IN + UP Q =0, IN − UP S =0, (79) µ −1 ν T a (t, x)T = gµν UTa (−t, x), where I is the N× N identity matrix. Cases of the Caµ(t, x)C−1 = U aµ(t, x), N C combined symmetries like CP are straightforward. µ −1 µ Pa (x)P = UPa˜ (˜x), (73) For the charge and spin U(1) symmetries of the system,
−1 where UT,UC, and UP are integer-valued N ×N matrices, Ucφ(t, x)Uc = φ(t, x)+ δφc, then we can find these matrices of transformations by the U φ(t, x)U −1 = φ(t, x)+ δφ , (80) symmetries of the theory. However, the above symmetry s s s transformation law does not fully specify the symmetry 0 0 where U ≡ eiθc R dx jc /e and U ≡ eiθs R dx js /s are the properties of charged excitations.54 c s charge and spin U(1) transformations, respectively, and A convenient way to complete the description of the the corresponding phase shifts are given by symmetries is to consider the action at the edge Sedge = dtdx L , L = L0 + Lex : −1 −1 edge edge edge edge δφc = θcK Q, δφs = θsK S. (81)
R 1 On the other hand, the phases δφ in Eq. (78) are L0 = (K ∂ φ ∂ φ − V ∂ φ ∂ φ ) , edge 4π IJ t I x J IJ x I x J determined by how the local quasiparticle excitations, ex e µν s µν which are described by normal-ordered vertex operators Ledge = ǫ QI ∂µφI Aν + ǫ SI ∂µφI Bν , (74) T T 2π 2π ‡eil φ‡ = ‡eiΛ Kφ‡ ≡‡eiΘ(Λ)‡, with l = KΛ and Λ being integer N-components vectors, under the symme- which is derived from the usual bulk-edge correspondence try transformations. That is, the transformation law for of the bulk Chern-Simons theory (70). Now the currents Θ(Λ) is determined by the algebraic relations of the un- in the edge theory are derlying symmetry operators. To classify these discrete e s Z symmetries for interacting systems (beyond the single- jµ = ǫµν Q ∂ φ , jµ = ǫµν S ∂ φ . (75) 2 c 2π I ν I s 2π I ν I particle picture), we constrain the symmetry operators by the following algebraic relations: Under T , C, and P, the edge currents transform similarly 2 Nf as the bulk currents. The transformation law for the Gi = S , ∀Gi ∈ G; µ Gi bosonic fields φI is translated from the gauge fields a G G G−1G−1 = SNf , ∀G , G ∈ G, (82) (73), with additional (constant) phases: i j i j Gi,Gj i j
−1 where S has values ±1. In a bosonic system, the opera- T φ(t, x)T = −UTφ(−t, x)+ δφT, Nf −1 tor S (subscript omitted) is just the identity 1. In a Cφ(t, x)C = UCφ(t, x)+ δφC, fermionic system, SNf can be either the identity 1 or the −1 Nf Pφ(t, x)P = UPφ(t, −x)+ δφP. (76) fermion number parity operator Pf ≡ (−1) (i.e., sym- metries are realized projectively) , where Nf is the total The minus sign in front of UT is just a convention for fermion number operator. Since all T , C, and P (and of a antiuntary operator. For the edge theory (74) with course the combined symmetries) commute with Pf , we 14
have SG1,G2 = SG2,G1 for any two symmetry operators where NIJ is the component of some integer matrix. T G1 and G2. For any local quasiparticle excitation ‡ exp iΛ Kφ‡, T In the presence of U(1) symmetries, the algebraic rela- with Λ Kφ = I ΛI (Kφ)I ≡ I θI , the symmetry tions (82) for fermionic systems might be ”gauge equiva- transformation G acts as P P lent” through the redefinition of the discrete symmetry G T − − G‡ eiΛ Kφ ‡ G 1 = G‡ ePI iθI ‡ G 1 to UαG, where Uα can be charge or spin U(1) with some ′ iθ − 1 [iθ ,iθ ] −1 phase α. Denoting G and G the discrete symmetries with = G‡ e I · e 2 PI