Introduction
Although quantum matter is microscopically composed of electrons and ions treated as point particles, its low-energy behavior is governed by emergent quasiparticles. Examples include phonons, magnons, and excitons on the bosonic side, and Landau quasiparticles on the fermionic side. In three-dimensional space such point-like quasiparticles only come in two kinds, bosons or fermions [1 ] , for which the many-body wavefunction is symmetric or antisymmetric under the exchange of identical excitations, respectively.
In lower spatial dimensions, point-like excitations need not obey exclusively bosonic or fermionic statistics, but may instead exhibit fractional statistics [1 , 2 ] . For example, point-like excitations in two-dimensional topologically ordered phases, known as anyons, may obey either Abelian or non-Abelian exchange statistics. In the Abelian case, exchanging two identical anyons multiplies the many-body wavefunction by a phase that need not be restricted to ± 1 \pm1 ± 1 [3 , 4 ] . It has been proposed in Refs. [5 , 6 , 7 , 8 ] that one-dimensional interfaces between two-dimensional topologically ordered states coupled by proximity effect to superconductors can host degrees of freedom that provide one-dimensional avatars of anyonic degrees of freedom. These so-called order-N N N parafermionic degrees of freedom obey exchange relations characterized by an N N N th root of unity. Since then, various platforms have been proposed to host parafermionic degrees of freedom [9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 ] . More recently, the experimental realizations [20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 , 28 ] of fractional Chern insulators (FCIs) [29 , 30 , 31 , 32 , 33 ] , open a new pathway to realize excitations with fractionalized statistics. Unlike the earlier proposals, FCIs do not require an external magnetic field to break time-reversal symmetry. This removes the tension between proximity superconductivity and the magnetic field.
In one-dimensional space, order-N N N parafermions can be thought of as generalizations of Majorana fermions, the latter corresponding to the case of N = 2 N = 2 N = 2 . In one-dimensional space, just as any Hamiltonian built out of Majorana operators is related to a Hamiltonian built out of spin-1/2 via the Jordan-Wigner (JW) transformation [34 ] , any Hamiltonian built out of order-N N N parafermions is related to one built out of Z N \mathbb{Z}_N Z N clock and shift operators via the Fradkin-Kadanoff (FK) transformation [35 , 36 ] .
Motivated by proposals for realizing parafermionic degrees of freedom in one-dimensional systems, substantial attention has been devoted to the phases of parafermionic Hamiltonians [37 , 39 , 40 , 41 , 42 , 43 , 44 , 45 , 46 , 47 , 48 , 49 ] [38]. Of particular interest are invertible topological phases of matter [50 , 51 , 52 , 53 ] , whose local Hamiltonians have a non-degenerate gapped ground state on any closed spatial manifold.
An invertible topological phase is called trivial if it admits a representative whose non-degenerate ground state is a product of localized states. By contrast, under open boundary conditions, a Hamiltonian in a non-trivial invertible topological phase supports protected degrees of freedom localized at its boundaries. These boundary degrees of freedom give rise to a ground-state degeneracy in the thermodynamic limit, with any finite-size splitting vanishing as the boundaries are taken infinitely far apart.
We define a continuous deformation between two local Hamiltonians as a continuous change of their short-range couplings, allowing also for the addition or removal of decoupled local degrees of freedom in product states. Two Hamiltonians whose ground states realize invertible topological phases are then said to be equivalent if they can be connected by such a deformation without closing the spectral gap.
In addition to this notion of equivalence, invertible topological phases are equipped with a composition law called stacking , under which they form a group. Stacking two phases amounts to combining their degrees of freedom without coupling them. A representative Hamiltonian for the resulting phase is given by
H s t a c k = H 1 ⊗ 1 2 + 1 1 ⊗ H 2 , (1) H_{\mathrm{stack}} = H_1 \otimes\mathbb{1}_2 + \mathbb{1}_1 \otimes H_2, \tag*{(1)} H stack = H 1 ⊗ 1 2 + 1 1 ⊗ H 2 , (1)
where H 1 H_1 H 1 and H 2 H_2 H 2 are representative Hamiltonians for the two phases. The trivial phase is the identity element under stacking, and two phases are inverses of one another if their stacking yields the trivial phase.
Additional symmetries can refine the classification of invertible topological phases.1 In the presence of a symmetry, two phases are equivalent only if they can be connected by a continuous deformation that neither closes the spectral gap nor breaks the symmetry, explicitly or spontaneously. An invertible topological phase that becomes trivial once the protecting symmetry is relaxed is called a symmetry-protected topological (SPT) phase [50 , 61 ] .
The classification of invertible topological phases of bosons and fermions is well understood in one-dimensional space [62 , 63 , 64 , 65 , 66 , 67 , 68 ] . In this work, we extend this paradigm to the invertible phases of matter realized by Hamiltonians built out of order-N N N parafermions in one-dimensional space, which we call invertible parafermionic topological (IPT) phases. Following the strategy introduced in Ref. [68 ] , we systematically classify IPT phases in one dimension and derive explicitly their stacking rules given a global symmetry group G \mathcal{G} G . We briefly summarize our strategy and results in § 1 A.
Summary of the results
Throughout this paper, space is one-dimensional. We consider local Hamiltonians built out of order-N N N parafermionic operators that realize the parafermion algebra P N 2 L \mathcal{P}^{2L}_N P N 2 L [see Eq. (2.1)]. We consider parafermions of prime-power order, i.e., N = p k N = p^k N = p k with p p p a prime and k k k a positive integer. We discuss in § 1.2 D, a generalization to the case of arbitrary integer N N N .
We argue that any local Hamiltonian realizing IPT phases of order-N N N parafermions must have a global Z N p f \mathbb{Z}^{\mathrm{pf}}_N Z N pf symmetry (Sec. II A), whose eigenvalues label the parafermion charge of the states in the global Hilbert space. This is a generalization of the Z 2 f \mathbb{Z}_2^f Z 2 f fermion parity symmetry. Any additional symmetry group, which we denote by G b G_b G b , is extended by Z N p f \mathbb{Z}^{\mathrm{pf}}_N Z N pf to give rise to the global parafermionic symmetry group G \mathcal{G} G , such that Z N p f ⊂ G \mathbb{Z}^{\mathrm{pf}}_N \subset\mathcal{G} Z N pf ⊂ G is a normal subgroup.
Following Ref. [68 ] , we posit that, for any Hamiltonian realizing an IPT phase under open boundary conditions, there exists a set of protected boundary degrees of freedom such that (i) they form an algebra of order-N N N parafermions P N η \mathcal{P}^{\eta}_N P N η for some integer η \eta η , (ii) and they render the ground state degenerate in the thermodynamic limit. If so, the distinct IPT phases are characterized by inequivalent boundary projective representations of the global symmetry group G \mathcal{G} G on the algebra P N η \mathcal{P}^{\eta}_N P N η . To classify such boundary projective representations we introduce a triplet of cohomological indices ( § 1.2 )
Table 1. The groups of IPT phases of order-N N N parafermions Inv N p f ( G ) \operatorname{Inv}^{\mathrm{pf}}_{N}(\mathcal{G}) Inv N pf ( G ) and bosonic SPT phases SPT b ( G ) \operatorname{SPT}^{\mathrm{b}}(\mathcal{G}) SPT b ( G ) with symmetry G \mathcal{G} G . We always consider N N N to be a prime power, i.e., N = p k N=p^{k} N = p k , except for the last row, where N N N is a composite with the prime decomposition N = N 1 × N 2 × ⋯ × N n N=N_{1}\times N_{2}\times\cdots\times N_{n} N = N 1 × N 2 × ⋯ × N n . In each row, we compare the groups of IPT phases of order-N N N parafermions and SPT phases of boson, both with G \mathcal{G} G symmetry. We denote these groups by Inv N p f ( G ) \operatorname{Inv}^{\mathrm{pf}}_{N}(\mathcal{G}) Inv N pf ( G ) and SPT b ( G ) \operatorname{SPT}^{\mathrm{b}}(\mathcal{G}) SPT b ( G ) , respectively. When the group G \mathcal{G} G is used for bosonic SPT phases, we treat the subgroup Z N p f \mathbb{Z}^{\mathrm{pf}}_{N} Z N pf as an ordinary Z N \mathbb{Z}_{N} Z N symmetry. When N N N is a prime power, the IPT phases always have the structure Inv N p f ( G ) = SPT N p f ( G ) ⋊ Z 2 \operatorname{Inv}^{\mathrm{pf}}_{N}(\mathcal{G})=\operatorname{SPT}^{\mathrm{pf}}_{N}(\mathcal{G})\rtimes\mathbb{Z}_{2} Inv N pf ( G ) = SPT N pf ( G ) ⋊ Z 2 , where SPT N p f ( G ) \operatorname{SPT}^{\mathrm{pf}}_{N}(\mathcal{G}) SPT N pf ( G ) denotes the SPT phases of order-N N N parafermions. Using the FK duality, the SPT phases of parafermions can be mapped to the corresponding SPT phases of bosons. While the two groups have the same cardinality ∣ SPT N p f ( G ) ∣ = ∣ SPT N b ( G ) ∣ |\operatorname{SPT}^{\mathrm{pf}}_{N}(\mathcal{G})|=|\operatorname{SPT}^{\mathrm{b}}_{N}(\mathcal{G})| ∣ SPT N pf ( G ) ∣ = ∣ SPT N b ( G ) ∣ they are different groups as the stacking rules differ. In the second to last row, we take p p p to be an odd integer p = 2 l − 1 p=2l-1 p = 2 l − 1 with l = 1 , 2 , ⋯ l=1,2,\cdots l = 1 , 2 , ⋯ . Therein, we denote by Heis ( Z N ) \operatorname{Heis}(\mathbb{Z}_{N}) Heis ( Z N ) the non-Abelian Heisenberg group of order N 3 N^{3} N 3 .
( μ , ( κ , ν ) ) ∈ Z 2 × C 1 ( G b , Z N ) × C 2 ( G b , U ( 1 ) ) . (2) (\mu,(\kappa,\nu))\in\mathbb{Z}_{2}\times C^{1}(G_{b},\mathbb{Z}_{N})\times C^{2}(G_{b},U(1)). \tag*{(2)} ( μ , ( κ , ν )) ∈ Z 2 × C 1 ( G b , Z N ) × C 2 ( G b , U ( 1 )) . (2)
The index μ = 0 , 1 \mu=0,1 μ = 0 , 1 denotes whether there is an even or odd number of parafermionic degrees of freedom associated with the boundary (μ = n mod 2 \mu=n\ \operatorname{mod}\ 2 μ = n mod 2 ). The case of μ = 1 \mu=1 μ = 1 describes the IPT phase that is beyond SPTs, i.e., the non-trivial IPT phase whose ground state cannot be deformed into the product state under any local deformation. When the group G \mathcal{G} G is a non-split extension of G b G_{b} G b by Z N p f \mathbb{Z}^{\mathrm{pf}}_{N} Z N pf (see Appendix B), then the only compatible IPT phases are SPTs and the index μ \mu μ always takes the value μ = 0 \mu=0 μ = 0 .
The remaining pair of indices ( κ , ν ) (\kappa,\nu) ( κ , ν ) characterizes the SPT portion of the IPT phases. More specifically, the index κ \kappa κ is related to the parafermion charge of the boundary representations of elements in G b G_{b} G b , while the index ν \nu ν encodes the projective representations of the elements in G b G_{b} G b . The two indices ν \nu ν and κ \kappa κ correspond to the bosonic part and the intrinsically parafermionic part of the SPTs realized by order-N N N parafermions.
We derive the stacking rules (Sec. IV) of IPT phases with G \mathcal{G} G symmetry by considering two boundary projective representations associated with indices ( μ 1 , ( κ 1 , ν 1 ) ) (\mu_{1},(\kappa_{1},\nu_{1})) ( μ 1 , ( κ 1 , ν 1 )) and ( μ 2 , ( κ 2 , ν 2 ) ) (\mu_{2},(\kappa_{2},\nu_{2})) ( μ 2 , ( κ 2 , ν 2 )) and constructing stacked indices
( μ 1 ∧ 2 , ( κ 1 ∧ 2 , ν 1 ∧ 2 ) ) = ( μ 1 , ( κ 1 , ν 1 ) ) ∧ ( μ 2 , ( κ 2 , ν 2 ) ) , (3) (\mu_{1\wedge2},(\kappa_{1\wedge2},\nu_{1\wedge2}))=(\mu_{1},(\kappa_{1},\nu_{1}))\wedge(\mu_{2},(\kappa_{2},\nu_{2})), \tag*{(3)} ( μ 1 ∧ 2 , ( κ 1 ∧ 2 , ν 1 ∧ 2 )) = ( μ 1 , ( κ 1 , ν 1 )) ∧ ( μ 2 , ( κ 2 , ν 2 )) , (3)
by explicitly constructing the stacked representation that acts on the degrees of freedom of the parafermion algebras P N n 1 \mathfrak{P}_{N}^{n_{1}} P N n 1 and P N n 2 \mathfrak{P}_{N}^{n_{2}} P N n 2 in the same way as the projective representations associated with the indices ( μ 1 , ( κ 1 , ν 1 ) ) (\mu_{1},(\kappa_{1},\nu_{1})) ( μ 1 , ( κ 1 , ν 1 )) and ( μ 2 , ( κ 2 , ν 2 ) ) (\mu_{2},(\kappa_{2},\nu_{2})) ( μ 2 , ( κ 2 , ν 2 )) , respectively. We find that the stacked indices are given by
μ 1 ∧ 2 = μ 1 + μ 2 ( m o d 2 ) , \mu_{1\wedge2}=\mu_{1}+\mu_{2}\pmod2, μ 1 ∧ 2 = μ 1 + μ 2 ( mod 2 ) ,
κ 1 ∧ 2 = κ 1 + ( − 1 ) μ 1 κ 2 , \kappa_{1\wedge2}=\kappa_{1}+(-1)^{\mu_{1}}\kappa_{2}, κ 1 ∧ 2 = κ 1 + ( − 1 ) μ 1 κ 2 ,
ν 1 ∧ 2 = ν 1 + ν 2 + ( − 1 ) μ 1 + 1 2 π N κ 1 ∪ κ 2 . (4) \nu_{1\wedge2}=\nu_{1}+\nu_{2}+(-1)^{\mu_{1}+1}\frac{2\pi}{N}\kappa_{1}\cup\kappa_{2}. \tag*{(4)} ν 1 ∧ 2 = ν 1 + ν 2 + ( − 1 ) μ 1 + 1 N 2 π κ 1 ∪ κ 2 . (4)
Here, ∪ \cup ∪ denotes the cup product of the Z N \mathbb{Z}_{N} Z N -valued 1-cochains κ 1 \kappa_{1} κ 1 and κ 2 \kappa_{2} κ 2 , which produces a Z N \mathbb{Z}_{N} Z N -valued 2-cochain, as is explained in Appendix A. Remarkably, these stacking rules are not Abelian in that
( μ 1 ∧ 2 , ( κ 1 ∧ 2 , ν 1 ∧ 2 ) ) ≠ ( μ 2 ∧ 1 , ( κ 2 ∧ 1 , ν 2 ∧ 1 ) ) . (5) (\mu_{1\wedge2},(\kappa_{1\wedge2},\nu_{1\wedge2}))\ne(\mu_{2\wedge1},(\kappa_{2\wedge1},\nu_{2\wedge1})). \tag*{(5)} ( μ 1 ∧ 2 , ( κ 1 ∧ 2 , ν 1 ∧ 2 )) = ( μ 2 ∧ 1 , ( κ 2 ∧ 1 , ν 2 ∧ 1 )) . (5)
Physically, this non-commutativity stems from the fact that parafermionic degrees of freedom require a total ordering and are therefore intrinsically one-dimensional. When stacking two IPT phases, one must consequently prescribe the relative ordering of the parafermionic degrees of freedom belonging to the two phases.
In the stacking rules (1.4a), there are two sources of non-commutativity. The first arises from the factor ( − 1 ) μ 1 (-1)^{\mu_{1}} ( − 1 ) μ 1 . The second arises from the fact that
κ 1 ∪ κ 2 ≠ κ 2 ∪ κ 1 , (6) \kappa_{1}\cup\kappa_{2}\ne\kappa_{2}\cup\kappa_{1}, \tag*{(6)} κ 1 ∪ κ 2 = κ 2 ∪ κ 1 , (6)
in general. We denote the group of IPT phases of order-N N N parafermions with G \mathcal{G} G by Inv N p f ( G ) \operatorname{Inv}^{\mathrm{pf}}_{N}(\mathcal{G}) Inv N pf ( G ) and the group of SPT phases by SPT N p f ( G ) \operatorname{SPT}^{\mathrm{pf}}_{N}(\mathcal{G}) SPT N pf ( G ) . We find the relation
Inv N p f ( G ) = SPT N p f ( G ) ⋊ Z 2 . (7) \operatorname{Inv}^{\mathrm{pf}}_{N}(\mathcal{G})=\operatorname{SPT}^{\mathrm{pf}}_{N}(\mathcal{G})\rtimes\mathbb{Z}_{2}. \tag*{(7)} Inv N pf ( G ) = SPT N pf ( G ) ⋊ Z 2 . (7)
The semidirect product by Z 2 \mathbb{Z}_{2} Z 2 is due to the factor of ( − 1 ) μ 1 (-1)^{\mu_{1}} ( − 1 ) μ 1 . Eq. (1.5) leads to the group SPT N p f ( G ) \operatorname{SPT}^{\mathrm{pf}}_{N}(\mathcal{G}) SPT N pf ( G ) of SPT phases being non-Abelian in general. The fact that a semidirect product by Z 2 \mathbb{Z}_{2} Z 2 arises when stacking IPT phases was first pointed out in Ref. [37 ] . We extend this by showing that even the group of SPT phases of parafermions can be non-Abelian. In Table I, we compute the classification of IPT phases of order-N N N parafermions with a given symmetry group G \mathcal{G} G . We contrast this with the classification of bosonic SPT phases with G \mathcal{G} G symmetry2 . We note that the group of SPT phases of parafermions S P T N p f ( G ) \mathrm{SPT}^{\mathrm{pf}}_N(\mathcal{G}) SPT N pf ( G ) is always of the same order as those of bosons. However, they differ in their group structure which can be non-Abelian.
In the last row of Table I, we consider IPT phases for parafermions of composite order N = N 1 × N 2 × ⋯ × N n N=N_1\times N_2\times\cdots\times N_n N = N 1 × N 2 × ⋯ × N n , with no additional symmetry. Interestingly, in this case there is a distinct IPT phase that is not an SPT for each prime factor (also see § 1.2 D).
Finally, we explicitly construct (Sec. V) lattice Hamiltonians that realize IPT phases of order-N N N parafermions with Z N × Z N × Z N \mathbb{Z}_N\times\mathbb{Z}_N\times\mathbb{Z}_N Z N × Z N × Z N symmetry (with N N N odd prime power). We show that there are 2 N 3 2N^3 2 N 3 IPT phases for each of which the Hamiltonians can be generated by stacking the four Hamiltonians defined in Eq. (5.6). The group of IPT phases is then given by Heis ( Z N ) ⋊ Z 2 \operatorname{Heis}(\mathbb{Z}_N)\rtimes\mathbb{Z}_2 Heis ( Z N ) ⋊ Z 2 where Heis ( Z N ) \operatorname{Heis}(\mathbb{Z}_N) Heis ( Z N ) is the non-Abelian Heisenberg group of order N 3 N^3 N 3 . The group of SPT phases is given by the subgroup Heis ( Z N ) \operatorname{Heis}(\mathbb{Z}_N) Heis ( Z N ) which is non-Abelian as opposed to the case of bosons or fermions.
Outline
The rest of the paper is organized as follows. In Sec. II, we define the order-N N N parafermion algebra, the Hilbert space on which the parafermionic operators act and the protecting symmetries. In § 1.2 , we consider the projective representations realized at the boundaries of IPT phases and introduce cohomological indices that classify them. Section IV presents a derivation of the stacking rules of IPT phases, which, unlike invertible topological phases of bosons and fermions, can form non-Abelian groups under stacking. In Sec. V we provide an explicit example of Hamiltonians that realize IPT phases with a G ≅ Z N 3 \mathcal{G}\cong\mathbb{Z}_N^3 G ≅ Z N 3 symmetry. These phases form a non-Abelian group of order 2 N 3 2N^3 2 N 3 under stacking (the corresponding SPT phases form a non-Abelian group of order N 3 N^3 N 3 under stacking). Sec. VI concludes the paper. Appendices A, B, and C provide summaries of group cohomology, properties of symmetry group G \mathcal{G} G , and proofs of Eq. (52) , respectively.
Definitions and Assumptions
We consider a one-dimensional lattice of L L L sites, each of which supports a pair of order-N N N parafermionic degrees of freedom. For simplicity, we focus on parafermions of prime-power order, i.e., N = p k N=p^k N = p k with p p p a prime and k k k a positive integer. The case of general N N N is discussed in § 1.2 D.
Without loss of generality, we associate an N N N -dimensional local Hilbert space H j \mathcal{H}_j H j to each site j = 1 , … , L j=1,\ldots,L j = 1 , … , L of the lattice. Thus, for a lattice of length L L L , the total Hilbert space H \mathcal{H} H is of dimensionality N L N^L N L . We define 2 L 2L 2 L unitary parafermion operators
{ γ 1 , γ 2 , … , γ 2 L − 1 , γ 2 L } (8) \{\gamma_1,\gamma_2,\ldots,\gamma_{2L-1},\gamma_{2L}\} \tag*{(8)} { γ 1 , γ 2 , … , γ 2 L − 1 , γ 2 L } (8)
acting on the global Hilbert space H \mathcal{H} H such that they satisfy the algebra
γ j N = 1 , γ j γ k = ω N sgn ( k − j ) γ k γ j , ω N : = e i 2 π N , (9) \gamma_j^N=1,\qquad\gamma_j\gamma_k=\omega_N^{\operatorname{sgn}(k-j)}\gamma_k\gamma_j,\qquad\omega_N:=e^{i\frac{2\pi}{N}}, \tag*{(9)} γ j N = 1 , γ j γ k = ω N sgn ( k − j ) γ k γ j , ω N := e i N 2 π , (9)
where sgn ( x ) \operatorname{sgn}(x) sgn ( x ) is the signum function that is 0 0 0 when x = 0 x=0 x = 0 and x / ∣ x ∣ x/|x| x /∣ x ∣ otherwise. We denote by P N 2 L \mathcal{P}^{2L}_N P N 2 L the algebra generated by the 2 L 2L 2 L order-N N N parafermions obeying Eq. (2.1b).
To define a local basis of the N L N^L N L -dimensional global Hilbert space H \mathcal{H} H , it is convenient to pair up the 2 L 2L 2 L parafermion operators using the notation
ξ j ≡ γ 2 j − 1 , η j ≡ γ 2 j , (10) \xi_j\equiv\gamma_{2j-1},\qquad\eta_j\equiv\gamma_{2j}, \tag*{(10)} ξ j ≡ γ 2 j − 1 , η j ≡ γ 2 j , (10)
for j = 1 , 2 , … , L j=1,2,\ldots,L j = 1 , 2 , … , L . We can then define for each site j j j the pairwise commuting local operators
Q j = e i π N + 1 N ξ j η j † , [ Q j , Q k ] = 0 , Q j N = 1. (11) Q_j=e^{i\pi\frac{N+1}{N}}\xi_j\eta_j^\dagger,\qquad[Q_j,Q_k]=0,\qquad Q_j^N=1. \tag*{(11)} Q j = e iπ N N + 1 ξ j η j † , [ Q j , Q k ] = 0 , Q j N = 1. (11)
An orthonormal basis spanning the global Hilbert space H \mathcal{H} H is constructed from the simultaneous eigenstates
∣ α ⟩ ≡ ∣ α 1 , α 2 , … , α L ⟩ (12) |\alpha\rangle\equiv|\alpha_1,\alpha_2,\ldots,\alpha_L\rangle \tag*{(12)} ∣ α ⟩ ≡ ∣ α 1 , α 2 , … , α L ⟩ (12)
of the Q j Q_j Q j operators, such that
Q j ∣ α ⟩ = ω N α j ∣ α ⟩ , (13) Q_j|\alpha\rangle=\omega_N^{\alpha_j}|\alpha\rangle, \tag*{(13)} Q j ∣ α ⟩ = ω N α j ∣ α ⟩ , (13)
where α j = 0 , 1 , … , N − 1 \alpha_j=0,1,\ldots,N-1 α j = 0 , 1 , … , N − 1 and j = 1 , … , L j=1,\ldots,L j = 1 , … , L .
Parafermion charge symmetry
From the local operators (11) we construct the global operator
Q : = ∏ j = 1 L Q j , Q ∣ α ⟩ = ( ω N ) ∑ j = 1 L α j ∣ α ⟩ . (14) Q:=\prod_{j=1}^{L}Q_j,\qquad Q|\alpha\rangle=(\omega_N)^{\sum_{j=1}^{L}\alpha_j}|\alpha\rangle. \tag*{(14)} Q := j = 1 ∏ L Q j , Q ∣ α ⟩ = ( ω N ) ∑ j = 1 L α j ∣ α ⟩ . (14)
The operator Q Q Q is thus diagonal in the basis (12) and its eigenstate ∣ α ⟩ |\alpha\rangle ∣ α ⟩ has the eigenvalue ω N q α \omega_N^{q_\alpha} ω N q α , where
q α = ∑ j = 1 L α j ( m o d N ) . (15) q_\alpha=\sum_{j=1}^{L}\alpha_j\pmod N. \tag*{(15)} q α = j = 1 ∑ L α j ( mod N ) . (15)
Moreover, conjugation of any individual parafermion γ j \gamma_j γ j by Q Q Q gives
Q γ j Q † = ω N γ j , j = 1 , … , 2 L . (16) Q\gamma_j Q^\dagger=\omega_N\gamma_j,\qquad j=1,\ldots,2L. \tag*{(16)} Q γ j Q † = ω N γ j , j = 1 , … , 2 L . (16)
Hence, we call Q Q Q the global parafermion charge operator . A basis state ∣ α ⟩ \lvert\alpha\rangle ∣ α ⟩ carries the parafermion charge q α q_\alpha q α mod N N N , while individual parafermion operators γ j \gamma_j γ j carry unit parafermion charge. The parafermion charge operator (2.3) generalizes global fermion parity, which is recovered when N = 2 N=2 N = 2 . One can show that such an operator can only exist if there is an even total number of parafermion operators.
Any IPT phase built out of order-N N N parafermions must have a Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf parafermion charge symmetry implemented by Q Q Q . On the one hand, just like fermion parity, Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf symmetry is more than an ordinary Z N \mathbb{Z}_N Z N symmetry, as it dictates the mutual statistics obeyed by local operators. On the other hand, unlike fermion parity, Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf symmetry can be broken, either spontaneously or explicitly.
To appreciate why Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf symmetry is more than an ordinary Z N \mathbb{Z}_N Z N symmetry, consider two operators A j A_j A j and B k B_k B k from P N 2 L \mathcal{P}_N^{2L} P N 2 L with definite parafermion charges q A q_A q A and q B q_B q B , respectively. Assume that A j A_j A j and B k B_k B k only act non-trivially on the finite non-overlapping intervals I j I_j I j and I k I_k I k centered around sites j j j and k k k , respectively. It then follows that
A j B k = ω N q A q B sgn ( k − j ) B k A j . (17) A_jB_k=\omega_N^{q_Aq_B\operatorname{sgn}(k-j)}B_kA_j. \tag*{(17)} A j B k = ω N q A q B sgn ( k − j ) B k A j . (17)
Importantly, the algebra (17) holds irrespective of the separation between the sites j j j and k k k . This means that the two well-separated operators A j A_j A j and B k B_k B k can be simultaneously measured only when q A q B = 0 mod N q_Aq_B=0\ \operatorname{mod}N q A q B = 0 mod N .
This observation can be applied to any local Hamiltonian built out of parafermion operators
H = ∑ j h j , (18) H=\sum_j h_j, \tag*{(18)} H = j ∑ h j , (18)
where each summand h j h_j h j acts non-trivially only on an interval centered around site j j j containing no more than r r r sites. The local energy density is then a well-defined local observable if and only if the two Hamiltonian terms h j h_j h j and h k h_k h k can be simultaneously measured for any j j j and k k k that are sufficiently separated, i.e.,
[ h j , h k ] = 0 (19) [h_j,h_k]=0 \tag*{(19)} [ h j , h k ] = 0 (19)
for any j , k j,k j , k such that ∣ j − k ∣ > 2 r \lvert j-k\rvert>2r ∣ j − k ∣ > 2 r . The algebra (17) then implies that if each h j h_j h j has vanishing parafermion charge, i.e., [ Q , H ] = 0 [Q,H]=0 [ Q , H ] = 0 , then the local energy density becomes a well-defined observable. In other words, the condition (19) is satisfied when Hamiltonian (18) is Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf symmetric. The converse is not true, i.e., the condition (19) can be satisfied without the full Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf symmetry.
Unlike fermion parity, the Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf parafermion charge symmetry can be broken spontaneously, e.g., Refs. [44 ] [38]. To see this, consider the parafermion operators γ j p a \gamma_j^{p^a} γ j p a with the integer a ≥ k / 2 a\ge k/2 a ≥ k /2 (recall that N = p k N=p^k N = p k ). These operators obey the mutual bosonic statistics
γ j p a γ k p a = γ k p a γ j p a (20) \gamma_j^{p^a}\gamma_k^{p^a}=\gamma_k^{p^a}\gamma_j^{p^a} \tag*{(20)} γ j p a γ k p a = γ k p a γ j p a (20)
for any j , k = 1 , … , 2 L j,k=1,\ldots,2L j , k = 1 , … , 2 L . In other words, such parafermionic operators can play the role of local order parameters. If such an operator gains a non-vanishing expectation value, it leads to the spontaneous breaking of the parafermion charge symmetry from Z p k p f \mathbb{Z}_{p^k}^{\mathrm{pf}} Z p k pf down to Z p a p f \mathbb{Z}_{p^a}^{\mathrm{pf}} Z p a pf . The ground-state manifold is then p k − a p^{k-a} p k − a -fold degenerate and does not realize an invertible phase of matter.
Unlike fermion parity, the Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf parafermion charge symmetry can also be explicitly broken. To see this, consider the terms γ j p a \gamma_j^{p^a} γ j p a with a ≥ k / 2 a\ge k/2 a ≥ k /2 , which are local operators that commute on distinct sites, i.e., they are local bosons. Adding such a term to a Hamiltonian breaks explicitly the Z p k p f \mathbb{Z}_{p^k}^{\mathrm{pf}} Z p k pf down to Z p a p f \mathbb{Z}_{p^a}^{\mathrm{pf}} Z p a pf symmetry. More generally, one can add any local operator of charge p a p^a p a which results in a explicit breaking of the parafermion charge symmetry. We then expect the problem of classifying IPT phases realized in the presence of such terms to reduce to that with Z p a p f \mathbb{Z}_{p^a}^{\mathrm{pf}} Z p a pf symmetry.
Thus, in our classification of IPT phases of order-N N N parafermions, we will always assume that the Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf parafermion charge symmetry is present and not broken, neither spontaneously nor explicitly.
Parafermionic symmetry groups
An invertible parafermionic phase of matter can have more symmetries than just the Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf parafermion charge symmetry. We denote by G b G_b G b the group of unitary symmetries other than Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf . The elements of the group G b G_b G b are also called bosonic symmetries in the literature [66 ] , hence the subscript b b b .
Demanding that the additional symmetries in group G b G_b G b are locality preserving implies that every symmetry transformation must preserve the Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf grading, i.e., under a symmetry transformation a definite parafermion charge sector must be mapped to another definite parafermion charge sector. This means that the Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf symmetry must be a normal subgroup of the total symmetry group G G G , see Appendix B.§ 2.5.1 . The group G G G is then given as a group extension of G b G_b G b by Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf specified by the short exact sequence (see Appendix B.2)
1 ⟶ Z N p f → ι G → π G b ⟶ 1 , (21) 1\longrightarrow\mathbb{Z}_N^{\mathrm{pf}}\xrightarrow{\iota}G\xrightarrow{\pi}G_b\longrightarrow1, \tag*{(21)} 1 ⟶ Z N pf ι G π G b ⟶ 1 , (21)
where ι \iota ι is an injective inclusion map and π \pi π a surjective quotient map such that im ι = ker π \operatorname{im}\iota=\ker\pi im ι = ker π . The group isomorphism theorems then imply that im ι \operatorname{im}\iota im ι is a normal subgroup and G b ≅ G / im ι ≡ G / Z N p f G_b\cong G/\operatorname{im}\iota\equiv G/\mathbb{Z}_N^{\mathrm{pf}} G b ≅ G / im ι ≡ G / Z N pf . Given a group G b G_b G b , a general group extension is specified by two pieces of data:
a group action ρ ( ⋅ ) : G b → Aut ( Z N p f ) \rho(\cdot):G_b\to\operatorname{Aut}(\mathbb{Z}_N^{\mathrm{pf}}) ρ ( ⋅ ) : G b → Aut ( Z N pf ) of G b G_b G b on the parafermion charge symmetry Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf ,
and an extension class [ c ] [c] [ c ] in the second cohomology group H 2 ( G b , [ Z N p f ] ρ ) H^2\left(G_b,[\mathbb{Z}_N^{\mathrm{pf}}]_\rho\right) H 2 ( G b , [ Z N pf ] ρ ) .
We denote the elements of the total symmetry group ( G , ∘ ) (G,\circ) ( G , ∘ ) by g ~ \tilde{g} g ~ . We can alternatively represent an element g ~ ∈ G \tilde{g}\in G g ~ ∈ G as the pair ( z , g ) (z,g) ( z , g ) , where z ∈ ( Z N p f , + ) z\in(\mathbb{Z}_N^{\mathrm{pf}},+) z ∈ ( Z N pf , + ) and g ∈ ( G b , ⋅ ) g\in(G_b,\cdot) g ∈ ( G b , ⋅ ) with the respective group identity elements 0 ∈ Z N p f 0\in\mathbb{Z}_N^{\mathrm{pf}} 0 ∈ Z N pf and e ∈ G b e\in G_b e ∈ G b . In this representation, the group composition of two elements g ~ ≡ ( z , g ) \tilde{g}\equiv(z,g) g ~ ≡ ( z , g ) and h ~ ≡ ( u , h ) \tilde{h}\equiv(u,h) h ~ ≡ ( u , h ) in G G G becomes
( z , g ) ∘ ( u , h ) = ( z + ρ g u + c ( g , h ) , g h ) , (22) (z,g)\circ(u,h)=(z+\rho_g u+c(g,h),gh), \tag*{(22)} ( z , g ) ∘ ( u , h ) = ( z + ρ g u + c ( g , h ) , g h ) , (22)
where the 2-cochain c c c is a representative of the extension class [ c ] [c] [ c ] . Note that the group action ρ \rho ρ on Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf reduces to the multiplication by an integer modulo N N N , whence we write ρ g ( z ) ≡ ρ g z \rho_g(z)\equiv\rho_g z ρ g ( z ) ≡ ρ g z .
The parafermion charge symmetry Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf can be viewed as a normal subgroup of G G G such that
G b ≅ G / Z N p f . (23) G_b\cong G/\mathbb{Z}_N^{\mathrm{pf}}. \tag*{(23)} G b ≅ G / Z N pf . (23)
When the group action ρ \rho ρ is trivial, i.e., ρ g = 1 \rho_g=1 ρ g = 1 for all g ∈ G b g\in G_b g ∈ G b , Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf lies in the center of G G G .
The possible group actions ρ \rho ρ are severely restricted. As shown in Appendix B § 2.5.3 every automorphism in the image of ρ \rho ρ must be involutions. For N = p k N=p^k N = p k , the group of involutive automorphisms of Z p k p f \mathbb{Z}_{p^k}^{\mathrm{pf}} Z p k pf is
I ( Z p k p f ) = { Z 1 , if p = 2 and k = 1 , Z 2 , if p = 2 and k = 2 , Z 2 × Z 2 , if p = 2 and k > 2 , Z 2 , if p > 2. (24) \mathcal{I}\left(\mathbb{Z}_{p^k}^{\mathrm{pf}}\right)=
\begin{cases}
\mathbb{Z}_1, & \text{if }p=2\text{ and }k=1,\\
\mathbb{Z}_2, & \text{if }p=2\text{ and }k=2,\\
\mathbb{Z}_2\times\mathbb{Z}_2, & \text{if }p=2\text{ and }k>2,\\
\mathbb{Z}_2, & \text{if }p>2.
\end{cases} \tag*{(24)} I ( Z p k pf ) = ⎩ ⎨ ⎧ Z 1 , Z 2 , Z 2 × Z 2 , Z 2 , if p = 2 and k = 1 , if p = 2 and k = 2 , if p = 2 and k > 2 , if p > 2. (24)
The result for generic N N N follows from Eq. (24) by applying the Chinese remainder theorem.
For any prime p p p and positive integer k k k , Eq. (B17) admits the solutions ρ g = ± 1 \rho_g=\pm1 ρ g = ± 1 . They generate a Z 2 \mathbb{Z}_2 Z 2 subgroup, except when p = 2 p=2 p = 2 and k = 1 k=1 k = 1 , where these two solutions coincide modulo 2. When p = 2 p=2 p = 2 and k > 2 k>2 k > 2 , there are two additional solutions, ρ g = 2 k − 1 ± 1 \rho_g=2^{k-1}\pm1 ρ g = 2 k − 1 ± 1 . Together, the four solutions form the group Z 2 × Z 2 \mathbb{Z}_2\times\mathbb{Z}_2 Z 2 × Z 2 .3
Physically, the solution ρ g = − 1 \rho_g=-1 ρ g = − 1 implements charge conjugation , as it maps an operator of parafermion charge q q q to one of charge − q -q − q . Its action on an individual parafermion can be represented as
γ j ↦ γ j † (25) \gamma_j\mapsto\gamma_j^\dagger \tag*{(25)} γ j ↦ γ j † (25)
When p = 2 p=2 p = 2 and k > 2 k>2 k > 2 , the two additional solutions ρ g = 2 k − 1 ± 1 \rho_g=2^{k-1}\pm1 ρ g = 2 k − 1 ± 1 define what we call special charge conjugation symmetries . Consider a single parafermion operator γ j \gamma_j γ j which has unit charge q = 1 q=1 q = 1 . The solution ρ g = 2 k − 1 − 1 \rho_g=2^{k-1}-1 ρ g = 2 k − 1 − 1 combines charge conjugation with the attachment of a parafermion charge 2 k − 1 2^{k-1} 2 k − 1 . By contrast, the solution ρ g = 2 k − 1 + 1 \rho_g=2^{k-1}+1 ρ g = 2 k − 1 + 1 attaches a parafermion charge 2 k − 1 2^{k-1} 2 k − 1 without implementing charge conjugation.
For Majorana fermions, corresponding to N = 2 N=2 N = 2 , one has − 1 = 1 mod 2 -1=1\ \text{mod}\ 2 − 1 = 1 mod 2 . Consequently, charge conjugation acts trivially on the Z 2 p f \mathbb{Z}_2^{\mathrm{pf}} Z 2 pf charge. A non-trivial action of charge conjugation on the parafermion charge is therefore possible only for N ≥ 3 N\ge3 N ≥ 3 .
Boundary projective representations
An invertible parafermionic topological (IPT) phase is realized by a local Hamiltonian
H = ∑ j = 1 L h j , (26) H=\sum_{j=1}^{L}h_j, \tag*{(26)} H = j = 1 ∑ L h j , (26)
such that
each local term h j h_j h j is built out of order-N N N parafermion operators and has finite support,
the ground state is non-degenerate and gapped in the thermodynamic limit L → ∞ L\to\infty L → ∞ ,
and H H H is G G G -invariant, i.e.,
[ U ( z , g ) , h j ] = 0 , j = 1 , … , L , (27) [U(z,g),h_j]=0,\qquad j=1,\ldots,L, \tag*{(27)} [ U ( z , g ) , h j ] = 0 , j = 1 , … , L , (27)
where U ( z , g ) ∈ H N 2 L U(z,g)\in\mathcal{H}_N^{2L} U ( z , g ) ∈ H N 2 L is a unitary operator and the set of such operators labeled by ( z , g ) ∈ G (z,g)\in G ( z , g ) ∈ G realizes a faithful representation of G G G .
Since the ground state ∣ ψ ⟩ |\psi\rangle ∣ ψ ⟩ of the Hamiltonian (26) is non-degenerate and symmetric, the representation U ( z , g ) U(z,g) U ( z , g ) can be chosen to be linear, i.e.,
U ( z , g ) U ( u , h ) = U ( ( z , g ) ∘ ( u , h ) ) , U(z,g)U(u,h)=U((z,g)\circ(u,h)), U ( z , g ) U ( u , h ) = U (( z , g ) ∘ ( u , h )) ,
U ( z , g ) ∣ ψ ⟩ = e i φ z , g ∣ ψ ⟩ , φ z , g ∈ [ 0 , 2 π ) . (28) U(z,g)|\psi\rangle=e^{i\varphi_{z,g}}|\psi\rangle,\qquad\varphi_{z,g}\in[0,2\pi). \tag*{(28)} U ( z , g ) ∣ ψ ⟩ = e i φ z , g ∣ ψ ⟩ , φ z , g ∈ [ 0 , 2 π ) . (28)
When open boundary conditions are imposed, the terms h j h_j h j in Eq. (26) that are simultaneously supported around the sites j = 1 j=1 j = 1 and j = L j=L j = L will be absent. In that case, there will be operators that can be associated with the left and right ends of the open chain that commute with the Hamiltonian. Such operators describe the local degrees of freedom associated with the right and the left boundaries of the open chain. Accordingly, we assume the existence of the sets
O L = { γ L , 1 , γ L , 2 , ⋯ , γ L , n L } , \mathcal{O}_{L}=\left\{\gamma_{L,1},\gamma_{L,2},\cdots,\gamma_{L,n_L}\right\}, O L = { γ L , 1 , γ L , 2 , ⋯ , γ L , n L } ,
O R = { γ R , 1 , γ R , 2 , ⋯ , γ R , n R } , (29) \mathcal{O}_{R}=\left\{\gamma_{R,1},\gamma_{R,2},\cdots,\gamma_{R,n_R}\right\}, \tag*{(29)} O R = { γ R , 1 , γ R , 2 , ⋯ , γ R , n R } , (29)
that are associated with the left and right boundaries, respectively. The operators in the sets O L \mathcal{O}_L O L and O R \mathcal{O}_R O R satisfy the order-N N N parafermion algebra (9) and, thus, generate the parafermion algebras P N n L \mathcal{P}^{n_{\mathrm{L}}}_{N} P N n L and P N n R \mathcal{P}^{n_{\mathrm{R}}}_{N} P N n R , respectively. By assumption, every operator in the set O R \mathcal{O}_{\mathrm{R}} O R lies to the right of those in the set O L \mathcal{O}_{\mathrm{L}} O L , whence the union O L ⊔ O R \mathcal{O}_{\mathrm{L}} \sqcup\mathcal{O}_{\mathrm{R}} O L ⊔ O R generates the parafermion algebra P N n L + n R \mathcal{P}^{n_{\mathrm{L}}+n_{\mathrm{R}}}_{N} P N n L + n R .
We demand that the degrees of freedom on the left and right boundaries together act on a well-defined global Hilbert space. 4 In other words, the total number of boundary parafermion operators must be even, i.e.,
n L + n R = 0 ( m o d 2 ) . (30) n_{\mathrm{L}} + n_{\mathrm{R}} = 0 \pmod{2}. \tag*{(30)} n L + n R = 0 ( mod 2 ) . (30)
We also demand that the boundary algebras generated by the sets of operators (29) are compatible with the bulk symmetries U ( g ~ ) U(\tilde{g}) U ( g ~ ) in the sense that
U ( g ~ ) P N n B U † ( g ~ ) = P N n B (31) U(\tilde{g}) \mathcal{P}^{n_{\mathrm{B}}}_{N} U^{\dagger}(\tilde{g}) = \mathcal{P}^{n_{\mathrm{B}}}_{N} \tag*{(31)} U ( g ~ ) P N n B U † ( g ~ ) = P N n B (31)
for all elements g ~ ∈ G \tilde{g} \in\mathcal{G} g ~ ∈ G and B = L , R \mathrm{B} = \mathrm{L}, \mathrm{R} B = L , R . A boundary that does not satisfy this compatibility condition explicitly breaks the G \mathcal{G} G symmetry which we do not consider.
The bulk symmetry operators (28) induce an action on the boundary degrees of freedom through conjugation. We assume that the boundary degrees of freedom (29) from the set O B \mathcal{O}_{\mathrm{B}} O B realize a faithful representation U B U_{\mathrm{B}} U B of the symmetry group G \mathcal{G} G such that 5
U ( g ~ ) γ B , j U † ( g ~ ) = U B ( g ~ ) γ B , j U B † ( g ~ ) , (32) U(\tilde{g}) \gamma_{\mathrm{B},j} U^{\dagger}(\tilde{g}) = U_{\mathrm{B}}(\tilde{g}) \gamma_{\mathrm{B},j} U_{\mathrm{B}}^{\dagger}(\tilde{g}), \tag*{(32)} U ( g ~ ) γ B , j U † ( g ~ ) = U B ( g ~ ) γ B , j U B † ( g ~ ) , (32)
for all g ~ ∈ G \tilde{g} \in\mathcal{G} g ~ ∈ G and j = 1 , … , n B j = 1,\ldots,n_{\mathrm{B}} j = 1 , … , n B . Physically, the condition (32) means that the boundary representation U B ( g ~ ) U_{\mathrm{B}}(\tilde{g}) U B ( g ~ ) implements the same transformation rules as the bulk representation U ( g ~ ) U(\tilde{g}) U ( g ~ ) . In practice, Eq. (32) is the key relation that we use to explicitly construct the boundary representations of G \mathcal{G} G .
While the bulk representation (28) is linear by definition, the boundary symmetry operators U B ( z , g ) U_{\mathrm{B}}(z,g) U B ( z , g ) may realize a non-trivial projective representation. Distinct IPT phases are characterized by distinct equivalence classes of boundary projective representations. In what follows, we introduce cohomological indices ( μ , [ ( κ , ν ) ] ) (\mu,[(\kappa,\nu)]) ( μ , [( κ , ν )]) that characterize the boundary projective representations that can be realized by order-N N N parafermions given the group G b \mathcal{G}_{b} G b . Without loss of generality, we will consider the degrees of freedom on a single boundary B \mathrm{B} B . We will treat the cases of even and odd n B n_{\mathrm{B}} n B separately. To distinguish these, we introduce a Z 2 \mathbb{Z}_{2} Z 2 -valued index μ \mu μ whose values μ = 0 , 1 \mu= 0,1 μ = 0 , 1 correspond to the case of an even and odd number of boundary parafermions, respectively.
Even number of boundary parafermions
When the number of boundary parafermion operators is even (μ = 0 \mu= 0 μ = 0 ), the parafermionic operators from the set O B \mathcal{O}_{\mathrm{B}} O B act on a local Hilbert space H B \mathcal{H}_{\mathrm{B}} H B associated with the boundary B \mathrm{B} B . We can label the states in H B \mathcal{H}_{\mathrm{B}} H B using the eigenvalues of the boundary parafermion charge operators
Q B , a = e i π N + 1 N γ B , 2 a − 1 γ B , 2 a † , (33) Q_{\mathrm{B},a} = e^{i\pi\frac{N+1}{N}\gamma_{\mathrm{B},2a-1}\gamma_{\mathrm{B},2a}^{\dagger}}, \tag*{(33)} Q B , a = e iπ N N + 1 γ B , 2 a − 1 γ B , 2 a † , (33)
with a = 1 , … , n B / 2 a = 1,\ldots,n_{\mathrm{B}}/2 a = 1 , … , n B /2 . The boundary parafermion charge symmetry is then implemented by the unitary operator
Q B = ∏ a = 1 n B / 2 Q B , a . (34) Q_{\mathrm{B}} = \prod_{a=1}^{n_{\mathrm{B}}/2} Q_{\mathrm{B},a}. \tag*{(34)} Q B = a = 1 ∏ n B /2 Q B , a . (34)
For a general element g ~ = ( z , g ) \tilde{g} = (z,g) g ~ = ( z , g ) in G \mathcal{G} G , we parameterize its boundary representation as
U B ( z , g ) = Q B z C B ( g ) V B ( g ) . (35) U_{\mathrm{B}}(z,g) = Q_{\mathrm{B}}^{z} C_{\mathrm{B}}(g) V_{\mathrm{B}}(g). \tag*{(35)} U B ( z , g ) = Q B z C B ( g ) V B ( g ) . (35)
Here, C B ( g ) C_{\mathrm{B}}(g) C B ( g ) is the unitary operator that implements the (special) charge conjugation symmetry if the group action ρ g ≠ 1 \rho_g \ne1 ρ g = 1 for the element g ∈ G b g \in\mathcal{G}_{b} g ∈ G b . The unitary operator V B ( g ) V_{\mathrm{B}}(g) V B ( g ) carries a definite parafermion charge, i.e.,
V B ( g ) Q B = ω N κ ( g ) Q B V B ( g ) , (36) V_{\mathrm{B}}(g) Q_{\mathrm{B}} = \omega_{N}^{\kappa(g)} Q_{\mathrm{B}} V_{\mathrm{B}}(g), \tag*{(36)} V B ( g ) Q B = ω N κ ( g ) Q B V B ( g ) , (36)
with the 1-cochain κ ∈ C 1 ( G b , Z N ) \kappa\in C^{1}(\mathcal{G}_{b},\mathbb{Z}_{N}) κ ∈ C 1 ( G b , Z N ) encoding the parafermion charge of V B ( g ) V_{\mathrm{B}}(g) V B ( g ) .
The boundary representation U B U_{\mathrm{B}} U B is projective in that it obeys the composition rule
U B ( g ~ ) U B ( h ~ ) = e i ϕ ( g ~ , h ~ ) U B ( g ~ ∘ h ~ ) . (37) U_{\mathrm{B} }(\tilde{g}) U_{\mathrm{B}}(\tilde{h}) = e^{i\phi(\tilde{g},\tilde{h})} U_{\mathrm{B}}(\tilde{g}\mathbin{\circ}\tilde{h}). \tag*{(37)} U B ( g ~ ) U B ( h ~ ) = e i ϕ ( g ~ , h ~ ) U B ( g ~ ∘ h ~ ) . (37)
The projective phase ϕ ∈ Z 2 ( G , U ( 1 ) ) \phi\in Z^{2}(\mathcal{G},U(1)) ϕ ∈ Z 2 ( G , U ( 1 )) is a representative 2-cocycle of the second group cohomology equivalence class [ ϕ ] ∈ H 2 ( G , U ( 1 ) ) [\phi] \in H^{2}(\mathcal{G},U(1)) [ ϕ ] ∈ H 2 ( G , U ( 1 )) , see Appendix A for a review. Thus, the cohomology group H 2 ( G , U ( 1 ) ) H^{2}(\mathcal{G},U(1)) H 2 ( G , U ( 1 )) classifies the distinct projective representations for an even number of boundary parafermions.
One can decompose the 2-cocycle ϕ \phi ϕ in terms of a tuple ( κ , ν ) ∈ C 1 ( G b , Z N ) × C 2 ( G b , U ( 1 ) ) (\kappa,\nu) \in C^{1}(\mathcal{G}_{b},\mathbb{Z}_{N}) \times C^{2}(\mathcal{G}_{b},U(1)) ( κ , ν ) ∈ C 1 ( G b , Z N ) × C 2 ( G b , U ( 1 )) as follows. Recall that κ \kappa κ was already introduced in Eq. (36) . We define ν \nu ν as the restriction of ϕ \phi ϕ on the bosonic symmetries G b ≅ G / Z N p f \mathcal{G}_{b} \cong\mathcal{G}/\mathbb{Z}_{N}^{\mathrm{pf}} G b ≅ G / Z N pf , i.e.,
U B ( 0 , g ) U B ( 0 , h ) = e i ν ( g , h ) U B ( c ( g , h ) , g h ) . (38) U_{\mathrm{B}}(0,g) U_{\mathrm{B}}(0,h) = e^{i\nu(g,h)} U_{\mathrm{B}}(c(g,h),gh). \tag*{(38)} U B ( 0 , g ) U B ( 0 , h ) = e i ν ( g , h ) U B ( c ( g , h ) , g h ) . (38)
Inserting the parameterization (35) into Eq. (37) yields
U B ( g ~ ) U B ( h ~ ) = ω N κ ( g ) u e i ν ( g , h ) U B ( g ~ ∘ h ~ ) . (39) U_{\mathrm{B}}(\tilde{g}) U_{\mathrm{B}}(\tilde{h}) = \omega_{N}^{\kappa(g)u} e^{i\nu(g,h)} U_{\mathrm{B}}(\tilde{g}\mathbin{\circ}\tilde{h}). \tag*{(39)} U B ( g ~ ) U B ( h ~ ) = ω N κ ( g ) u e i ν ( g , h ) U B ( g ~ ∘ h ~ ) . (39)
Comparison with Eq. (37) produces the identity
ϕ ( g ~ , h ~ ) = ν ( g , h ) + 2 π N κ ( g ) u . (40) \phi(\tilde{g},\tilde{h}) = \nu(g,h) + \frac{2\pi}{N}\kappa(g)u. \tag*{(40)} ϕ ( g ~ , h ~ ) = ν ( g , h ) + N 2 π κ ( g ) u . (40)
Thus, the tuple ( κ , ν ) (\kappa,\nu) ( κ , ν ) carries the same information as the projective phase ϕ \phi ϕ . However, while ϕ \phi ϕ is a 2-cocycle, ν \nu ν does not have a vanishing coboundary. Instead, the pair ( κ , ν ) (\kappa,\nu) ( κ , ν ) satisfies a modified 2-coboundary condition
D ( 2 ) ( κ , ν ) = 0 , \mathcal{D}^{(2)}(\kappa,\nu)=0, D ( 2 ) ( κ , ν ) = 0 ,
D ( 2 ) ( κ , ν ) : = ( δ ρ ( 1 ) κ , δ ( 2 ) ν − 2 π N κ ∪ c ) . (41) \mathcal{D}^{(2)}(\kappa,\nu):=\left(\delta_{\rho}^{(1)}\kappa,\delta^{(2)}\nu-\frac{2\pi}{N}\kappa\cup c\right). \tag*{(41)} D ( 2 ) ( κ , ν ) := ( δ ρ ( 1 ) κ , δ ( 2 ) ν − N 2 π κ ∪ c ) . (41)
Similarly, the tuple ( κ , ν ) (\kappa,\nu) ( κ , ν ) is equivalent to another tuple ( κ ′ , ν ′ ) (\kappa',\nu') ( κ ′ , ν ′ ) up to a modified 1-coboundary
( κ ′ , ν ′ ) = ( κ , ν ) + D ( 1 ) ( β , α ) , (\kappa',\nu')=(\kappa,\nu)+\mathcal{D}^{(1)}(\beta,\alpha), ( κ ′ , ν ′ ) = ( κ , ν ) + D ( 1 ) ( β , α ) ,
D ( 1 ) ( β , α ) : = ( 0 , δ ( 1 ) α − 2 π N β ∪ c ) . (42) \mathcal{D}^{(1)}(\beta,\alpha):=\left(0,\delta^{(1)}\alpha-\frac{2\pi}{N}\beta\cup c\right). \tag*{(42)} D ( 1 ) ( β , α ) := ( 0 , δ ( 1 ) α − N 2 π β ∪ c ) . (42)
where α ∈ C 1 ( G b , U ( 1 ) ) \alpha\in C^{1}(G_b,U(1)) α ∈ C 1 ( G b , U ( 1 )) and β ∈ C 0 ( G b , Z N ) \beta\in C^{0}(G_b,\mathbb{Z}_N) β ∈ C 0 ( G b , Z N ) . One verifies that D ( 2 ) ∘ D ( 1 ) = 0 \mathcal{D}^{(2)}\circ\mathcal{D}^{(1)}=0 D ( 2 ) ∘ D ( 1 ) = 0 , and, thus, in terms of the pair ( κ , ν ) (\kappa,\nu) ( κ , ν ) , the second cohomology group H 2 ( G , U ( 1 ) ) H^{2}(G,U(1)) H 2 ( G , U ( 1 )) can be recast as
H 2 ( G , U ( 1 ) ) ≅ ker D ( 2 ) im D ( 1 ) . (43) H^{2}(G,U(1))\cong\frac{\ker\mathcal{D}^{(2)}}{\operatorname{im}\mathcal{D}^{(1)}}. \tag*{(43)} H 2 ( G , U ( 1 )) ≅ im D ( 1 ) ker D ( 2 ) . (43)
The index ν \nu ν in the decomposition (3.12) encodes how the bosonic symmetry G b G_b G b is realized projectively, while the index κ \kappa κ encodes the parafermion charge carried by the representations of G b G_b G b . The triplet of indices
( μ , [ ( κ , ν ) ] ) = ( 0 , [ ( κ , ν ) ] ) , (44) (\mu,[(\kappa,\nu)])=(0,[(\kappa,\nu)]), \tag*{(44)} ( μ , [( κ , ν )]) = ( 0 , [( κ , ν )]) , (44)
thus characterizes the boundary projective representations given an even number of boundary parafermions.
Odd number of boundary parafermions
When the number of boundary parafermion operators is odd (μ = 1 \mu=1 μ = 1 ), one cannot define a local Hilbert space using only the operators from the boundary set O B \mathcal{O}_B O B . For example, trying to a construct parafermion charge operators as in Eq. (3.7a) will always leave one parafermion operator unpaired.
To define a Hilbert space on which the boundary parafermion operators act, we add an auxiliary parafermion operator γ ∞ \gamma_\infty γ ∞ to the set O B \mathcal{O}_B O B of boundary parafermion operators. We assume, without loss of generality, that this auxiliary parafermion operator satisfies the algebra
γ B , j γ ∞ = ω N γ ∞ γ B , j , j = 1 , ⋯ , n B , (45) \gamma_{B,j}\gamma_\infty=\omega_N\gamma_\infty\gamma_{B,j},\qquad j=1,\cdots,n_B, \tag*{(45)} γ B , j γ ∞ = ω N γ ∞ γ B , j , j = 1 , ⋯ , n B , (45)
and define the extended set
O B e x t = { γ B , 1 , γ B , 2 , ⋯ , γ B , n B , γ ∞ } . (46) \mathcal{O}_B^{\mathrm{ext}}=\left\{\gamma_{B,1},\gamma_{B,2},\cdots,\gamma_{B,n_B},\gamma_\infty\right\}. \tag*{(46)} O B ext = { γ B , 1 , γ B , 2 , ⋯ , γ B , n B , γ ∞ } . (46)
Physically, such an auxiliary parafermion is provided by the unpaired boundary parafermion operator on the opposite boundary, since by Eq. (30) both boundaries support an odd number of parafermion degrees of freedom.
The parafermion algebras P N n B \mathcal{P}_N^{n_B} P N n B and P N n B + 1 \mathcal{P}_N^{n_B+1} P N n B + 1 are generated by the set O B \mathcal{O}_B O B and O B e x t \mathcal{O}_B^{\mathrm{ext}} O B ext , respectively.
Using the extended set of parafermion operators (46) , we define ( n B + 1 ) / 2 (n_B+1)/2 ( n B + 1 ) /2 boundary parafermion charge operators
Q B , a = e i π N + 1 N γ B , 2 a − 1 γ B , 2 a † , Q_{B,a}=e^{i\pi\frac{N+1}{N}\gamma_{B,2a-1}\gamma_{B,2a}^{\dagger},} Q B , a = e iπ N N + 1 γ B , 2 a − 1 γ B , 2 a † ,
Q B , ( n B + 1 ) / 2 = e i π N + 1 N γ B , n B γ ∞ † , (47) Q_{B,(n_B+1)/2}=e^{i\pi\frac{N+1}{N}\gamma_{B,n_B}\gamma_{\infty}^{\dagger},} \tag*{(47)} Q B , ( n B + 1 ) /2 = e iπ N N + 1 γ B , n B γ ∞ † , (47)
with a = 1 , 2 , ⋯ , ( n B − 1 ) / 2 a=1,2,\cdots,(n_B-1)/2 a = 1 , 2 , ⋯ , ( n B − 1 ) /2 . We can then label the states in the Hilbert space H B \mathcal{H}_B H B using the eigenvalues of these operators. Unlike the μ = 0 \mu=0 μ = 0 case, H B \mathcal{H}_B H B is non-local as some of the operators acting on it are supported on both boundaries of an open chain. The boundary parafermion charge symmetry is then implemented by the unitary operator
Q B = ∏ a = 1 ( n B + 1 ) / 2 Q B , a . (48) Q_B=\prod_{a=1}^{(n_B+1)/2}Q_{B,a}. \tag*{(48)} Q B = a = 1 ∏ ( n B + 1 ) /2 Q B , a . (48)
When constructing the boundary representations for a general element ( z , g ) ∈ G (z,g)\in\mathcal{G} ( z , g ) ∈ G , we demand that the key relation (32) applies only to the parafermions from the set O B \mathcal{O}_B O B and not to the auxiliary parafermion γ ∞ \gamma_\infty γ ∞ . Since this is a less restrictive requirement than demanding that Eq. (32) applies to the extended set (46) , there is an additional redundancy when constructing the boundary representations in comparison to the μ = 0 \mu=0 μ = 0 case. This redundancy is rooted in the fact that the algebra P N n B \mathcal{P}_N^{n_B} P N n B has a non-trivial center when n B n_B n B is odd. This N N N -dimensional center is generated by the operator
Y = ( ∏ a = 1 ( n B − 1 ) / 2 Q B , a ) γ n B , (49) Y=\left(\prod_{a=1}^{(n_B-1)/2}Q_{B,a}\right)\gamma_{n_B}, \tag*{(49)} Y = a = 1 ∏ ( n B − 1 ) /2 Q B , a γ n B , (49)
which commutes with all the boundary parafermion operators in the set O B \mathcal{O}_B O B . Note that the operator Y Y Y is a product of all boundary parafermion operators and does not contain the auxiliary parafermion γ ∞ \gamma_\infty γ ∞ . It carries unit parafermion charge,
Q B Y Q B † = ω N Y . (50) Q_BYQ_B^{\dagger}=\omega_NY. \tag*{(50)} Q B Y Q B † = ω N Y . (50)
Since Y Y Y commutes with all boundary parafermion operators from the set O B \mathcal{O}_B O B , for any boundary representation U B ( g ~ ) U_B(\widetilde{g}) U B ( g ) satisfying Eq. (32) , the operators
U B ′ ( g ~ ) = Y U B ( g ~ ) (51) U'_B(\widetilde{g})=YU_B(\widetilde{g}) \tag*{(51)} U B ′ ( g ) = Y U B ( g ) (51)
satisfy the same equation. In other words, U B ′ ( g ~ ) U'_B(\widetilde{g}) U B ′ ( g ) is also a valid boundary representation of G \mathcal{G} G . However, U B ′ ( g ~ ) U'_B(\widetilde{g}) U B ′ ( g ) and U B ( g ~ ) U_B(\widetilde{g}) U B ( g ) differ in their algebra with the parafermion charge operator Q B Q_B Q B .
For a general element g ~ = ( z , g ) ∈ G \widetilde{g}=(z,g)\in\mathcal{G} g = ( z , g ) ∈ G , we parameterize its boundary representation as
U B ( z , g ) = Q B z C B ( g ) V B ( g ) γ ∞ − κ ( g ) , (52) U_B(z,g)=Q_B^{z}C_B(g)V_B(g)\gamma_\infty^{-\kappa(g)}, \tag*{(52)} U B ( z , g ) = Q B z C B ( g ) V B ( g ) γ ∞ − κ ( g ) , (52)
see Appendix C for a derivation. Here, C B ( g ) \mathcal{C}_{B}(g) C B ( g ) is the unitary operator that implements the (special) charge conjugation symmetry if the group action ρ g ≠ 1 \rho_{g} \ne1 ρ g = 1 for the element g ∈ G b g \in G_{b} g ∈ G b . The exponent κ ( g ) \kappa(g) κ ( g ) is a 1-cochain κ ∈ C 1 ( G b , Z N ) \kappa\in C^{1}(G_{b},\mathbb{Z}_{N}) κ ∈ C 1 ( G b , Z N ) , the interpretation of which will be given below. The unitary operator V B ( g ) V_{B}(g) V B ( g ) commutes with both Y Y Y and γ ∞ \gamma_{\infty} γ ∞ such that the boundary representation satisfies
U B ( z , g ) Q B = ω κ ( g ) Q B ρ g U B ( z , g ) . (53) U_{B}(z,g)Q_{B}=\omega^{\kappa(g)}Q_{B}^{\rho_{g}}U_{B}(z,g). \tag*{(53)} U B ( z , g ) Q B = ω κ ( g ) Q B ρ g U B ( z , g ) . (53)
In other words, the unitary operator V B ( g ) γ ∞ − κ ( g ) V_{B}(g)\gamma_{\infty}^{-\kappa(g)} V B ( g ) γ ∞ − κ ( g ) carries parafermion charge − κ ( g ) -\kappa(g) − κ ( g ) . Since the unitary V B ( g ) V_{B}(g) V B ( g ) commutes with Y Y Y , it is an element in P N n B \mathcal{P}_{N}^{n_{B}} P N n B while U B ( z , g ) ∈ P N n B + 1 U_{B}(z,g)\in\mathcal{P}_{N}^{n_{B}+1} U B ( z , g ) ∈ P N n B + 1 . The fact that γ ∞ \gamma_{\infty} γ ∞ factorizes as in Eq. (52) stems from the requirement that the algebra P N n B \mathcal{P}_{N}^{n_{B}} P N n B must be closed under the boundary representation U B ( z , g ) U_{B}(z,g) U B ( z , g ) [recall Eq. (31) ].
The presence of the non-trivial center puts restrictions on which parafermionic symmetries G \mathcal{G} G are compatible with an odd number of parafermions. Conjugation of both sides of
U B ( 0 , g ) U B ( 0 , h ) = e i ϕ ( g ~ , h ~ ) Q B c ( g , h ) U B ( 0 , g h ) (54) U_{B}(0,g)U_{B}(0,h)=e^{i\phi(\tilde{g},\tilde{h})}Q_{B}^{c(g,h)}U_{B}(0,gh) \tag*{(54)} U B ( 0 , g ) U B ( 0 , h ) = e i ϕ ( g ~ , h ~ ) Q B c ( g , h ) U B ( 0 , g h ) (54)
by Y Y Y implies that
c = δ ρ ( 1 ) κ = 0. (55) c=\delta_{\rho}^{(1)}\kappa=0. \tag*{(55)} c = δ ρ ( 1 ) κ = 0. (55)
Hence, the only symmetry groups that are compatible with μ = 1 \mu=1 μ = 1 are those that split, i.e., of the form
G ≅ Z N p f ⋊ G b . (56) \mathcal{G}\cong\mathbb{Z}_{N}^{\mathrm{pf}}\rtimes G_{b}. \tag*{(56)} G ≅ Z N pf ⋊ G b . (56)
Conversely, when G \mathcal{G} G is a nonsplit extension, only boundaries with an even number of parafermions (μ = 0 \mu=0 μ = 0 ) are possible.
The composition of two elements g ~ , h ~ ∈ G \tilde{g},\tilde{h}\in\mathcal{G} g ~ , h ~ ∈ G follows Eq. (3.12a) just as in μ = 0 \mu=0 μ = 0 case. Thus, the projective phase in the group composition
U B ( g ~ ) U B ( h ~ ) = e i ϕ ( g ~ , h ~ ) U B ( g ~ ∘ h ~ ) (57) U_{B}(\tilde{g})U_{B}(\tilde{h})=e^{i\phi(\tilde{g},\tilde{h})}U_{B}(\tilde{g}\circ\tilde{h}) \tag*{(57)} U B ( g ~ ) U B ( h ~ ) = e i ϕ ( g ~ , h ~ ) U B ( g ~ ∘ h ~ ) (57)
has the same decomposition
ϕ ( g ~ , h ~ ) = ν ( g , h ) + 2 π N κ ( g ) u . (58) \phi(\tilde{g},\tilde{h})=\nu(g,h)+\frac{2\pi}{N}\kappa(g)u. \tag*{(58)} ϕ ( g ~ , h ~ ) = ν ( g , h ) + N 2 π κ ( g ) u . (58)
However, because the group G \mathcal{G} G splits (c = 0 c=0 c = 0 ), Eq. (3.13a) implies that both ν \nu ν and κ \kappa κ are proper cocycles. The index ν ( g , h ) \nu(g,h) ν ( g , h ) can be computed by restricting ϕ \phi ϕ to the bosonic symmetries G b G_{b} G b ,
U B ( 0 , g ) U B ( 0 , h ) = e i ν ( g , h ) U B ( 0 , g h ) . (59) U_{B}(0,g)U_{B}(0,h)=e^{i\nu(g,h)}U_{B}(0,gh). \tag*{(59)} U B ( 0 , g ) U B ( 0 , h ) = e i ν ( g , h ) U B ( 0 , g h ) . (59)
While the index κ ( g ) \kappa(g) κ ( g ) encodes the parafermion charge of the product V B ( g ) γ ∞ − κ ( g ) V_{B}(g)\gamma_{\infty}^{-\kappa(g)} V B ( g ) γ ∞ − κ ( g ) in the parameterization (52) , this definition is not invariant under the redefinition of U B ( z , g ) U_{B}(z,g) U B ( z , g ) in Eq. (51) . We thus define the index κ ( g ) \kappa(g) κ ( g ) through the relation
U B ( 0 , g ) Y = ω N κ ( g ) Y ρ g U B ( 0 , g ) . (60) U_{B}(0,g)Y=\omega_{N}^{\kappa(g)}Y^{\rho_{g}}U_{B}(0,g). \tag*{(60)} U B ( 0 , g ) Y = ω N κ ( g ) Y ρ g U B ( 0 , g ) . (60)
To summarize, when μ = 1 \mu=1 μ = 1 , the group must split, i.e.,
G ≅ Z N p f ⋊ G b , (61) \mathcal{G}\cong\mathbb{Z}_{N}^{\mathrm{pf}}\rtimes G_{b}, \tag*{(61)} G ≅ Z N pf ⋊ G b , (61)
and the triplet of indices
( μ , [ ( κ , ν ) ] ) = ( 1 , [ κ ] , [ ν ] ) , (62) (\mu,[(\kappa,\nu)])=(1,[\kappa],[\nu]), \tag*{(62)} ( μ , [( κ , ν )]) = ( 1 , [ κ ] , [ ν ]) , (62)
characterizes the boundary projective representations of given an odd number of boundary parafermions.
Ground state degeneracies
We can relate the triplet of indices ( μ , [ ( κ , ν ) ] ) (\mu,[(\kappa,\nu)]) ( μ , [( κ , ν )]) to (protected) ground state degeneracy (GSD) of an IPT phase when open boundary conditions are imposed. When the indices ( μ , [ ( κ , ν ) ] ) (\mu,[(\kappa,\nu)]) ( μ , [( κ , ν )]) are trivial, i.e.,
( μ , [ ( κ , ν ) ] ) = ( 0 , [ ( 0 , 0 ) ] ) , (63) (\mu,[(\kappa,\nu)])=(0,[(0,0)]), \tag*{(63)} ( μ , [( κ , ν )]) = ( 0 , [( 0 , 0 )]) , (63)
there is no constraint on the GSD associated with the boundary modes. In other words, a singlet representation of G \mathcal{G} G is compatible with the trivial IPT phase. When Eq. (63) does not hold, there is a minimal GSD larger than one that cannot be lifted as we shall describe.
When open boundary conditions are imposed, there are two sets of indices ( μ L , [ ( κ L , ν L ) ] ) (\mu_{L},[(\kappa_{L},\nu_{L})]) ( μ L , [( κ L , ν L )]) and ( μ R , [ ( κ R , ν R ) ] ) (\mu_{R},[(\kappa_{R},\nu_{R})]) ( μ R , [( κ R , ν R )]) associated with the left and the right boundaries, respectively. We distinguish two cases, μ L = μ R = 0 , 1 \mu_{L}=\mu_{R}=0,1 μ L = μ R = 0 , 1 that corresponds to having even and odd number of parafermionic degrees of freedom at the two boundaries (recall Eq. (30) ), respectively.
When μ L = μ R = 0 \mu_{L}=\mu_{R}=0 μ L = μ R = 0 , there are well-defined local Hilbert spaces associated with the two ends. In this case, the total Hilbert space associated with the boundaries factorizes
H b d = H L ⊗ H R . (64) \mathcal{H}_{\mathrm{bd}}=\mathcal{H}_{L}\otimes\mathcal{H}_{R}. \tag*{(64)} H bd = H L ⊗ H R . (64)
We can consider the boundary Hamiltonians H L ∈ P N n L H_{L}\in\mathcal{P}_{N}^{n_{L}} H L ∈ P N n L and H R ∈ P N n R H_{R}\in\mathcal{P}_{N}^{n_{R}} H R ∈ P N n R that act on the Hilbert spaces H L \mathcal{H}_{L} H L and H R \mathcal{H}_{R} H R , respectively. By definition, the Hamiltonians commute with the boundary projective representations U L ( z , g ) U_{L}(z,g) U L ( z , g ) and U R ( z , g ) U_{R}(z,g) U R ( z , g ) , respectively. The minimal global GSD is then the minimal GSD of the Hamiltonian
H b d = H L ⊗ 1 H R + 1 H L ⊗ H R . (65) H_{\mathrm{bd}}=H_{L}\otimes\mathbb{1}_{H_{R}}+\mathbb{1}_{H_{L}}\otimes H_{R}. \tag*{(65)} H bd = H L ⊗ 1 H R + 1 H L ⊗ H R . (65)
The minimal degeneracy associated with the left and the right boundaries then can be deduced by computing the minimum dimensional projective irreducible representation (irrep) that is compatible with the indices [ ( κ L , ν L ) ] [(\kappa_{L},\nu_{L})] [( κ L , ν L )] and [ ( κ R , ν R ) ] [(\kappa_{R},\nu_{R})] [( κ R , ν R )] . Let d [ ( κ L , ν L ) ] d_{[(\kappa_{L},\nu_{L})]} d [( κ L , ν L )] and d [ ( κ R , ν R ) ] d_{[(\kappa_{R},\nu_{R})]} d [( κ R , ν R )] be the smallest dimensions of the irreps associated with the left and the right boundary, respectively. We then have global GSD given by
G S D b d μ = 0 = d [ ( κ L , ν L ) ] × d [ ( κ R , ν R ) ] . (66) \mathrm{GSD}_{\mathrm{bd}}^{\mu=0}=d_{[(\kappa_{L},\nu_{L})]}\times d_{[(\kappa_{R},\nu_{R})]}. \tag*{(66)} GSD bd μ = 0 = d [( κ L , ν L )] × d [( κ R , ν R )] . (66)
For a given boundary B = L , R \mathrm{B}=\mathrm{L},\mathrm{R} B = L , R the dimension d [ ( κ B , ν B ) ] d_{[(\kappa_{\mathrm{B}},\nu_{\mathrm{B}})]} d [( κ B , ν B )] depends on the details of the group G \mathcal{G} G and the indices [ ( κ B , ν B ) ] [(\kappa_{\mathrm{B}},\nu_{\mathrm{B}})] [( κ B , ν B )] .
When μ L = μ R = 1 \mu_{\mathrm{L}}=\mu_{\mathrm{R}}=1 μ L = μ R = 1 , the Hilbert space associated with the boundaries are necessarily non-local. The total Hilbert space associated with the boundaries then factorizes as
H b d = H L ⊗ H L R ⊗ H R , (67) \mathcal{H}_{\mathrm{bd}}=\mathcal{H}_{\mathrm{L}}\otimes\mathcal{H}_{\mathrm{LR}}\otimes\mathcal{H}_{\mathrm{R}}, \tag*{(67)} H bd = H L ⊗ H LR ⊗ H R , (67)
where H L \mathcal{H}_{\mathrm{L}} H L and H R \mathcal{H}_{\mathrm{R}} H R are spanned by the eigenstates of the parafermion charge operators localized at the left and right boundaries respectively. The states at the non-local Hilbert space H L R \mathcal{H}_{\mathrm{LR}} H LR are spanned by the eigenstates of the single parafermion charge operator constructed from the unpaired parafermion operators in P N L \mathcal{P}_{N}^{\mathrm{L}} P N L and P N R \mathcal{P}_{N}^{\mathrm{R}} P N R . The boundary Hamiltonians H L ∈ P N L H_{\mathrm{L}}\in\mathcal{P}_{N}^{\mathrm{L}} H L ∈ P N L and H R ∈ P N R H_{\mathrm{R}}\in\mathcal{P}_{N}^{\mathrm{R}} H R ∈ P N R act on the Hilbert spaces
H L ⊗ H L R , (68) \mathcal{H}_{\mathrm{L}}\otimes\mathcal{H}_{\mathrm{LR}}, \tag*{(68)} H L ⊗ H LR , (68)
and
H R ⊗ H L R , (69) \mathcal{H}_{\mathrm{R}}\otimes\mathcal{H}_{\mathrm{LR}}, \tag*{(69)} H R ⊗ H LR , (69)
respectively. Since μ L = μ R = 1 \mu_{\mathrm{L}}=\mu_{\mathrm{R}}=1 μ L = μ R = 1 , there exists operators Y L Y_{\mathrm{L}} Y L and Y R Y_{\mathrm{R}} Y R that commute with the Hamiltonians H L H_{\mathrm{L}} H L and H R H_{\mathrm{R}} H R , respectively. However, the central elements do not commute with each other as well as the global parafermion charge operator. In particular, we have
Y L Y R = ω N Y R Y L Y_{\mathrm{L}}Y_{\mathrm{R}}=\omega_{N}Y_{\mathrm{R}}Y_{\mathrm{L}} Y L Y R = ω N Y R Y L
Q Y B = ω N Y B Q , (70) QY_{\mathrm{B}}=\omega_{N}Y_{\mathrm{B}}Q, \tag*{(70)} Q Y B = ω N Y B Q , (70)
where Q Q Q is the global parafermion charge operator. Because of this global algebra, H b d \mathcal{H}_{\mathrm{bd}} H bd must be at least N N N -fold degenerate, which is associated with the non-trivial indices μ L = μ R = 1 \mu_{\mathrm{L}}=\mu_{\mathrm{R}}=1 μ L = μ R = 1 . Importantly, this N N N -fold degeneracy can be lifted by explicitly breaking the Z N p f \mathbb{Z}_{N}^{\mathrm{pf}} Z N pf symmetry (recall that N = p k N=p^{k} N = p k ). However, a p p p -fold degeneracy cannot be lifted by any local perturbation since Z N p f \mathbb{Z}_{N}^{\mathrm{pf}} Z N pf can be broken down at most down to Z p p f \mathbb{Z}_{p}^{\mathrm{pf}} Z p pf symmetry by a local term.
There can be additional degeneracies protected by the non-trivial values of the indices [ ( κ B , ν B ) ] [(\kappa_{\mathrm{B}},\nu_{\mathrm{B}})] [( κ B , ν B )] . As in the case of μ L = μ R = 0 \mu_{\mathrm{L}}=\mu_{\mathrm{R}}=0 μ L = μ R = 0 , we associate denote by d [ ( κ B , ν B ) ] d_{[(\kappa_{\mathrm{B}},\nu_{\mathrm{B}})]} d [( κ B , ν B )] the smallest dimension of an irrep that is compatible with the indices [ ( κ B , ν B ) ] [(\kappa_{\mathrm{B}},\nu_{\mathrm{B}})] [( κ B , ν B )] at boundary B \mathrm{B} B can have. The global GSD is then given by
G S D b d μ = 1 = 1 N × d [ ( κ L , ν L ) ] × d [ ( κ R , ν R ) ] , (71) \mathrm{GSD}_{\mathrm{bd}}^{\mu=1}=\frac{1}{N}\times d_{[(\kappa_{\mathrm{L}},\nu_{\mathrm{L}})]}\times d_{[(\kappa_{\mathrm{R}},\nu_{\mathrm{R}})]}, \tag*{(71)} GSD bd μ = 1 = N 1 × d [( κ L , ν L )] × d [( κ R , ν R )] , (71)
where we divide by N N N to account for double counting the global degeneracy associated with the non-local Hilbert space H L R \mathcal{H}_{\mathrm{LR}} H LR .
Classification for general N N N
We have so far considered parafermions of prime-power order, i.e., N = p k N=p^{k} N = p k , where p p p is a prime and k k k is a positive integer. We seek to generalize the triplet ( μ , ( κ , ν ) ) (\mu,(\kappa,\nu)) ( μ , ( κ , ν )) to any finite N N N . To this end, consider the prime-power decomposition
N = p 1 k 1 p 2 k 2 ⋯ p n k n . (72) N=p_{1}^{k_{1}}p_{2}^{k_{2}}\cdots p_{n}^{k_{n}}. \tag*{(72)} N = p 1 k 1 p 2 k 2 ⋯ p n k n . (72)
The global parafermion charge symmetry factorizes accordingly,
Z N p f = Z N 1 p f × Z N 2 p f × ⋯ × Z N n p f , (73) \mathbb{Z}_{N}^{\mathrm{pf}}=\mathbb{Z}_{N_{1}}^{\mathrm{pf}}\times\mathbb{Z}_{N_{2}}^{\mathrm{pf}}\times\cdots\times\mathbb{Z}_{N_{n}}^{\mathrm{pf}}, \tag*{(73)} Z N pf = Z N 1 pf × Z N 2 pf × ⋯ × Z N n pf , (73)
where N a = p a k a N_{a}=p_{a}^{k_{a}} N a = p a k a for a = 1 , … , n a=1,\ldots,n a = 1 , … , n . Let us consider the operators
γ j ( a ) : = γ j N / N a (74) \gamma_{j}^{(a)}:=\gamma_{j}^{N/N_{a}} \tag*{(74)} γ j ( a ) := γ j N / N a (74)
for all j = 1 , … , 2 L j=1,\ldots,2L j = 1 , … , 2 L . Note that each operator (74) is of order N a N_{a} N a and has the parafermion charge
Q γ j ( a ) Q † = ω N N / N a γ j ( a ) = ω N a γ j ( a ) . (75) \begin{aligned}
Q\gamma_{j}^{(a)}Q^{\dagger}=\omega_{N}^{N/N_{a}}\gamma_{j}^{(a)} \\
&=\omega_{N_{a}}\gamma_{j}^{(a)}. \tag*{(75)}
\end{aligned} Q γ j ( a ) Q † = ω N N / N a γ j ( a ) = ω N a γ j ( a ) . (75)
These operators satisfy the algebra
γ j ( a ) γ k ( b ) = e i 2 π N N 2 N a N b sgn ( k − j ) γ k ( b ) γ j ( a ) = e i 2 π N a s n sgn ( k − j ) γ k ( b ) γ j ( a ) . (76) \begin{aligned}
\gamma_{j}^{(a)}\gamma_{k}^{(b)}=e^{\mathrm{i}\frac{2\pi}{N}\frac{N^{2}}{N_{a}N_{b}}\operatorname{sgn}(k-j)}\gamma_{k}^{(b)}\gamma_{j}^{(a)} \\
&=e^{\mathrm{i}\frac{2\pi}{N_{a}}s_{n}\operatorname{sgn}(k-j)}\gamma_{k}^{(b)}\gamma_{j}^{(a)}. \tag*{(76)}
\end{aligned} γ j ( a ) γ k ( b ) = e i N 2 π N a N b N 2 sgn ( k − j ) γ k ( b ) γ j ( a ) = e i N a 2 π s n sgn ( k − j ) γ k ( b ) γ j ( a ) . (76)
Due to the prime decomposition (72) , we have
N N b = s a δ a , b ( m o d N a ) , (77) \frac{N}{N_{b}}=s_{a}\delta_{a,b}\pmod{N_{a}}, \tag*{(77)} N b N = s a δ a , b ( mod N a ) , (77)
where s a = N / N a ( m o d N a ) s_{a}=N/N_{a}\pmod{N_{a}} s a = N / N a ( mod N a ) is an integer coprime to N a N_{a} N a . We note that when a ≠ b a\ne b a = b , the set of operators γ j ( a ) \gamma_{j}^{(a)} γ j ( a ) and γ k ( b ) \gamma_{k}^{(b)} γ k ( b ) commute for any j j j and k k k . When a = b a=b a = b , the algebra (3.38) reduces to
γ j ( a ) γ k ( a ) = ω N a s a sgn ( k − j ) γ k ( a ) γ j ( a ) . (78) \gamma_{j}^{(a)}\gamma_{k}^{(a)}=\omega_{N_{a}}^{s_{a}\operatorname{sgn}(k-j)}\gamma_{k}^{(a)}\gamma_{j}^{(a)}. \tag*{(78)} γ j ( a ) γ k ( a ) = ω N a s a sgn ( k − j ) γ k ( a ) γ j ( a ) . (78)
While this is reminiscent of the parafermion algebra (9) , the phase factor on the right-hand side is not the principal N a N_{a} N a th root of unity, i.e., ω N a \omega_{N_{a}} ω N a (it is, however, still a primitive root). Interestingly, the algebra (3.40) is in general distinct from the parafermion algebra (2.1b) with the principal N a N_{a} N a th root of unity in the sense that they are not related by a local redefinition of the parafermion operators.6
We interpret for each a = 1 , … , n a=1,\ldots,n a = 1 , … , n the set of operators { γ j ( a ) } j = 1 2 L \{\gamma_j^{(a)}\}_{j=1}^{2L} { γ j ( a ) } j = 1 2 L from Eq. (74) as a decoupled flavor of order-N a N_a N a parafermions. In particular, the total Hilbert space factorizes into subspaces,
H = ⨂ a = 1 n H a , (79) \mathcal{H}=\bigotimes_{a=1}^{n}\mathcal{H}_a, \tag*{(79)} H = a = 1 ⨂ n H a , (79)
where each factor H a \mathcal{H}_a H a is spanned by the eigenstates of the order-N a N_a N a parafermion charge operator7
Q ( a ) = ∏ j = 1 L e i π s a ( N a + 1 ) / N a ξ j ( a ) ( η j ( a ) ) † . (80) Q^{(a)}=\prod_{j=1}^{L}e^{i\pi s_a(N_a+1)/N_a}\xi_j^{(a)}\left(\eta_j^{(a)}\right)^\dagger. \tag*{(80)} Q ( a ) = j = 1 ∏ L e iπ s a ( N a + 1 ) / N a ξ j ( a ) ( η j ( a ) ) † . (80)
We note that the action of the operator Q ( a ) Q^{(a)} Q ( a ) on the parafermion operators γ j ( a ) \gamma_j^{(a)} γ j ( a ) do not match the action (75) of the operator Q Q Q since
Q ( a ) γ j ( a ) ( Q ( a ) ) † = ω N a s a γ j ( a ) . (81) Q^{(a)}\gamma_j^{(a)}\left(Q^{(a)}\right)^\dagger=\omega_{N_a}^{s_a}\gamma_j^{(a)}. \tag*{(81)} Q ( a ) γ j ( a ) ( Q ( a ) ) † = ω N a s a γ j ( a ) . (81)
Nevertheless, there exists a factorization
Q = e i φ ∏ a = 1 n ( Q ( a ) ) b a , (82) Q=e^{i\varphi}\prod_{a=1}^{n}\left(Q^{(a)}\right)^{b_a}, \tag*{(82)} Q = e i φ a = 1 ∏ n ( Q ( a ) ) b a , (82)
with integers b a = 1 , … , N a − 1 b_a=1,\ldots,N_a-1 b a = 1 , … , N a − 1 . This is guaranteed by the fact that s a s_a s a and N a N_a N a are coprime, i.e., there exist integers b a b_a b a such that b a s a = 1 m o d N a b_as_a=1\ \mathrm{mod}\ N_a b a s a = 1 mod N a for all a = 1 , … , n a=1,\ldots,n a = 1 , … , n . Since the left-hand and right-hand sides of Eq. (82) have identical action for all flavors a a a , they must be equal up to a U ( 1 ) U(1) U ( 1 ) phase which we denote by e i φ e^{i\varphi} e i φ .
For symmetry operators other than Q Q Q , a factorization similar to the one in Eq. (82) is not guaranteed. The factorization (82) means that the parafermion charge of each flavor is separately conserved. However, a symmetry operator can still mix two flavors a a a and b b b provided that the parafermion operator γ j ( a ) \gamma_j^{(a)} γ j ( a ) is mixed with operators from flavor b b b with zero Z N b p f \mathbb{Z}_{N_b}^{\mathrm{pf}} Z N b pf parafermion charge.
Nevertheless, the classification of IPT phases for general N N N simplifies when a factorization similar to Eq. (82) holds for any symmetry operators8 , i.e.,
U ( g ~ ) = e i φ ∏ a = 1 n U ( a ) ( g ~ ) . (83) U(\tilde{g})=e^{i\varphi}\prod_{a=1}^{n}U^{(a)}(\tilde{g}). \tag*{(83)} U ( g ~ ) = e i φ a = 1 ∏ n U ( a ) ( g ~ ) . (83)
In this case, the unitary representations U ( a ) ( g ~ ) U^{(a)}(\tilde{g}) U ( a ) ( g ~ ) in general form a subgroup of the full symmetry group G \mathcal{G} G , i.e., G \mathcal{G} G in general acts as a subgroup G ( a ) \mathcal{G}^{(a)} G ( a ) on the flavor a a a . For example, this is true for Z N p f \mathbb{Z}_{N}^{\mathrm{pf}} Z N pf parafermion charge symmetry which acts as the Z N a p f \mathbb{Z}_{N_a}^{\mathrm{pf}} Z N a pf subgroup.
Parafermionic stacking rules
When the factorization (83) holds, the classification of IPT phases for order-N N N parafermions thus reduces to the classification of IPT phases for each order-N a N_a N a parafermion flavor with G ( a ) ⊂ G \mathcal{G}^{(a)}\subset\mathcal{G} G ( a ) ⊂ G symmetry, which has been achieved in § 1.2 . We introduce a triplet of indices ( μ ( a ) , [ κ ( a ) , ν ( a ) ] ) (\mu^{(a)},[\kappa^{(a)},\nu^{(a)}]) ( μ ( a ) , [ κ ( a ) , ν ( a ) ]) such that
μ ( a ) = 0 , 1 , κ ( a ) ∈ C 1 ( G ( a ) / Z N a p f , Z N a ) , ν ( a ) ∈ C 2 ( G ( a ) / Z N a p f , U ( 1 ) ) , (84) \begin{aligned}
\mu^{(a)}&=0,1,\\
\kappa^{(a)}&\in C^1(\mathcal{G}^{(a)}/\mathbb{Z}_{N_a}^{\mathrm{pf}},\mathbb{Z}_{N_a}),\\
\nu^{(a)}&\in C^2(\mathcal{G}^{(a)}/\mathbb{Z}_{N_a}^{\mathrm{pf}},U(1)), \tag*{(84)}
\end{aligned} μ ( a ) κ ( a ) ν ( a ) = 0 , 1 , ∈ C 1 ( G ( a ) / Z N a pf , Z N a ) , ∈ C 2 ( G ( a ) / Z N a pf , U ( 1 )) , (84)
that can take values independently for each flavor. The complete classification for order-N N N parafermions is then given by the collection of triplets
{ ( μ ( a ) , [ κ ( a ) , ν ( a ) ] ) } . (85) \left\{(\mu^{(a)},[\kappa^{(a)},\nu^{(a)}])\right\}. \tag*{(85)} { ( μ ( a ) , [ κ ( a ) , ν ( a ) ]) } . (85)
We note that when only the Z N p f \mathbb{Z}_{N}^{\mathrm{pf}} Z N pf parafermion charge symmetry is present, only the indices μ ( a ) = 0 , 1 \mu^{(a)}=0,1 μ ( a ) = 0 , 1 in the decomposition of indices (3.44) for each flavor can be non-trivial. We find that when no additional bosonic symmetries are present there is a non-trivial IPT phase for each prime factor in N N N , with the classification given by Z 2 n \mathbb{Z}_2^n Z 2 n .
Parafermionic stacking rules
We now describe the so-called parafermionic stacking operation, that is, a composition rule between two IPT phases that produces a third IPT phase. We begin by motivating and defining the stacking operation, before computing the corresponding stacking rule for the classifying indices. We do so by distinguishing four cases according to the parity of the number of boundary parafermions, or in other words, the value of the classifying index μ \mu μ for each phase.
We assume N N N to be a prime power. Consider the pair of IPT phases realized by the local Hamiltonians9
H 1 : = ∑ j = 1 L h 1 , j , H 2 : = ∑ j = 1 L h 2 , j , (86) H_1:=\sum_{j=1}^{L}h_{1,j},\qquad H_2:=\sum_{j=1}^{L}h_{2,j}, \tag*{(86)} H 1 := j = 1 ∑ L h 1 , j , H 2 := j = 1 ∑ L h 2 , j , (86)
acting on the global Hilbert spaces H 1 \mathcal{H}_1 H 1 and H 2 \mathcal{H}_2 H 2 , respectively. Here, for any j = 1 , … , L j=1,\ldots,L j = 1 , … , L and a = 1 , 2 a=1,2 a = 1 , 2 , the energy density h a , j h_{a,j} h a , j is built out of most r a r_a r a order-N N N parafermion operators from the subset10
{ ξ a , k ≡ γ a , 2 k − 1 , η a , k ≡ γ a , 2 k ∣ k ∈ I a , j } . (87) \{\xi_{a,k}\equiv\gamma_{a,2k-1},\eta_{a,k}\equiv\gamma_{a,2k}\mid k\in I_{a,j}\}. \tag*{(87)} { ξ a , k ≡ γ a , 2 k − 1 , η a , k ≡ γ a , 2 k ∣ k ∈ I a , j } . (87)
of the associated parafermion algebra P N , a 2 L \mathcal{P}^{2L}_{N,a} P N , a 2 L . We associate the triplets of indices
Φ a L : = ( μ a , [ ( κ a , ν a ) ] ) (88) \Phi^{\mathrm{L}}_a := \left(\mu_a, [(\kappa_a,\nu_a)]\right) \tag*{(88)} Φ a L := ( μ a , [( κ a , ν a )] ) (88)
to the left boundaries of the Hamiltonians H a H_a H a in Eq. (4.1) when open boundary conditions are imposed, where we used Φ a L \Phi^{\mathrm{L}}_a Φ a L to denote the triplet collectively. Recall that consistency demands that the triplets of indices associated with the right edge must be the inverse of those associated with the left edge, i.e.,
Φ a R = Φ a L ‾ , (89) \Phi^{\mathrm{R}}_a = \overline{\Phi^{\mathrm{L}}_a}, \tag*{(89)} Φ a R = Φ a L , (89)
where the inverse is defined with respect to the stacking operation we are going to describe. The IPT phases realized by the Hamiltonians H a H_a H a are then fully characterized by the indices Φ a L \Phi^{\mathrm{L}}_a Φ a L , a = 1 , 2 a = 1,2 a = 1 , 2 .
To define the stacking operation, we must prescribe how the combined parafermionic degrees of freedom
{ ξ a , k ≡ γ a , 2 k − 1 , η a , k ≡ γ a , 2 k ∣ a = 1 , 2 , k = 1 , … , L } (90) \{\xi_{a,k} \equiv\gamma_{a,2k-1},\eta_{a,k} \equiv\gamma_{a,2k}\mid a=1,2,\ k=1,\ldots,L\} \tag*{(90)} { ξ a , k ≡ γ a , 2 k − 1 , η a , k ≡ γ a , 2 k ∣ a = 1 , 2 , k = 1 , … , L } (90)
act on the stacked Hilbert space
H 1 ∧ 2 : = H 1 ⊗ H 2 . (91) \mathcal{H}_{1\wedge2}:=\mathcal{H}_1\otimes\mathcal{H}_2. \tag*{(91)} H 1 ∧ 2 := H 1 ⊗ H 2 . (91)
This amounts to defining the stacked parafermion algebra P N , 1 ∧ 2 4 L \mathcal{P}^{4L}_{N,1\wedge2} P N , 1 ∧ 2 4 L and requires us to choose an ordering of the parafermion operators Eq. (4.4a). Once this done, we define the stacked Hamiltonian
H 1 ∧ 2 : = H 1 ⊗ 1 H 2 + 1 H 1 ⊗ H 2 (92) H_{1\wedge2}:=H_1\otimes1_{\mathcal{H}_2}+1_{\mathcal{H}_1}\otimes H_2 \tag*{(92)} H 1 ∧ 2 := H 1 ⊗ 1 H 2 + 1 H 1 ⊗ H 2 (92)
acting on the Hilbert space H 1 ∧ 2 \mathcal{H}_{1\wedge2} H 1 ∧ 2 . The stacking operation is relevant to two scenarios, shown in Figure 1 , each of which corresponds to a different choice of ordering of the parafermion operators.
Figure 1. Two scenarios for stacking parafermionic chain 1 (orange) with parafermionic chain 2 (green). Prior to stacking, all parafermionic boundary states are assumed to be localized near the first and last sites of the respective chains marked by the red circles. (a) Stacking rules at the left boundary under the convention that all boundary states of chain 1 are ordered to the left of those of chain 2. The labels Φ 1 L \Phi^{\mathrm{L}}_1 Φ 1 L and Φ 2 L \Phi^{\mathrm{L}}_2 Φ 2 L denote the left-boundary projective representations of chains 1 and 2, respectively, while Φ 1 ∧ 2 L \Phi^{\mathrm{L}}_{1\wedge2} Φ 1 ∧ 2 L denotes their corresponding projective stacked representation. (b) Chain 1 is placed to the left of chain 2. The labels Φ 1 R \Phi^{\mathrm{R}}_1 Φ 1 R and Φ 2 L \Phi^{\mathrm{L}}_2 Φ 2 L denote the right-boundary projective representation of chain 1 and the left-boundary projective representation of chain 2, respectively. The label Φ 1 ∧ 2 I \Phi^{\mathrm{I}}_{1\wedge2} Φ 1 ∧ 2 I denotes the projective stacked representation at the interface between the two chains.
In scenario 1, shown in Figure 1 (a), we consider physically putting the two one-dimensional chains that support the γ 1 , j \gamma_{1,j} γ 1 , j and γ 2 , j \gamma_{2,j} γ 2 , j degrees of freedom, shown in orange and green respectively, on top of each other. The boundary degrees of freedom realizing the indices Eq. (88) are schematically marked with red circles. Crucially, we choose the boundary degrees of freedom associated with the Hamiltonian H 1 H_1 H 1 to be to the left of the boundary degrees of freedom associated with Hamiltonian H 2 H_2 H 2 . This choice of ordering will be important. An analogous stacking operation was first defined in [37 ] . The stacking results in a new IPT phase, whose degrees of freedom are shown in purple, is characterized by the left index
Φ 1 ∧ 2 L = Φ 1 L ∧ Φ 2 L . (93) \Phi^{\mathrm{L}}_{1\wedge2}=\Phi^{\mathrm{L}}_1\wedge\Phi^{\mathrm{L}}_2. \tag*{(93)} Φ 1 ∧ 2 L = Φ 1 L ∧ Φ 2 L . (93)
In scenario 2, shown in Figure 1 (b), we consider physically putting the two one-dimensional chains that support the γ 1 , j \gamma_{1,j} γ 1 , j and γ 2 , j \gamma_{2,j} γ 2 , j degrees of freedom, shown in orange and green respectively, next to each other. Crucially, we choose all γ 1 , j \gamma_{1,j} γ 1 , j operators to be to the left of all γ 2 , j \gamma_{2,j} γ 2 , j operators. In this case, the stacking does not result in a uniform third IPT phase. Instead, it results in an interface between the right edge of the IPT phase described by Hamiltonian H 1 H_1 H 1 and the left edge of the IPT phase described by Hamiltonian H 2 H_2 H 2 . The interface is associated with the indices
Φ 1 ∧ 2 I = Φ 1 R ∧ Φ 2 L = Φ 1 L ‾ ∧ Φ 2 L , (94) \Phi^{\mathrm{I}}_{1\wedge2}=\Phi^{\mathrm{R}}_1\wedge\Phi^{\mathrm{L}}_2=\overline{\Phi^{\mathrm{L}}_1}\wedge\Phi^{\mathrm{L}}_2, \tag*{(94)} Φ 1 ∧ 2 I = Φ 1 R ∧ Φ 2 L = Φ 1 L ∧ Φ 2 L , (94)
where we used Eq. (89) in the last equality. We note that the interface describes a relative index between the two IPT phases. Importantly, when Φ 1 L = Φ 2 L \Phi^{\mathrm{L}}_1=\Phi^{\mathrm{L}}_2 Φ 1 L = Φ 2 L , the index Φ 1 ∧ 2 I \Phi^{\mathrm{I}}_{1\wedge2} Φ 1 ∧ 2 I must be trivial
Φ 1 ∧ 2 I = ( 0 , [ ( 0 , 0 ) ] ) , (95) \Phi^{\mathrm{I}}_{1\wedge2}=(0,[(0,0)]), \tag*{(95)} Φ 1 ∧ 2 I = ( 0 , [( 0 , 0 )]) , (95)
meaning that the interface can be fully gapped. This is expected, since in this case we are just spatially expanding an IPT phase to the right which should be gapped in the bulk. Conversely, when Φ 1 L ≠ Φ 2 L \Phi^{\mathrm{L}}_1 \ne\Phi^{\mathrm{L}}_2 Φ 1 L = Φ 2 L , the interface corresponds to a non-trivial index Φ 1 ∧ 2 I \Phi^{\mathrm{I}}_{1\wedge2} Φ 1 ∧ 2 I . There must be gapless degrees of freedom localized at the interface whose degeneracy is described by Φ 1 ∧ 2 I \Phi^{\mathrm{I}}_{1\wedge2} Φ 1 ∧ 2 I . If we associate the trivial indices ( 0 , [ ( 0 , 0 ) ] ) (0, [(0,0)]) ( 0 , [( 0 , 0 )]) with the vacuum (denoted as vac), then the open boundary conditions can be thought of as an interface between an IPT phase and the vacuum. In this case, one finds
Φ 1 ∧ v a c I = Φ 1 R , Φ v a c ∧ 2 I = Φ 2 L , (96) \Phi^{\mathrm{I}}_{1\wedge\mathrm{vac}}=\Phi^{\mathrm{R}}_1,\qquad\Phi^{\mathrm{I}}_{\mathrm{vac}\wedge2}=\Phi^{\mathrm{L}}_2, \tag*{(96)} Φ 1 ∧ vac I = Φ 1 R , Φ vac ∧ 2 I = Φ 2 L , (96)
which delivers the left and right indices associated with an IPT phase, as expected.
In both scenarios described above, we can deduce the stacked indices ( μ 1 ∧ 2 , [ ( κ 1 ∧ 2 , ν 1 ∧ 2 ) ] ) (\mu_{1\wedge2},[(\kappa_{1\wedge2},\nu_{1\wedge2})]) ( μ 1 ∧ 2 , [( κ 1 ∧ 2 , ν 1 ∧ 2 )]) by the indices associated with zero-dimensional parafermionic representations of the symmetry group G \mathcal{G} G . In particular, the gapped bulk degrees of freedom are immaterial to the result of the stacking operation. Without loss of generality, we assume that the (single-side) boundary degrees of freedom of one stacking component lie either completely to the left or completely to the right of the (single-side) boundary degrees of freedom of the other stacking component. More precisely, let us denote by the sets for a = 1 , 2 a=1,2 a = 1 , 2
O a = { γ a , 1 , γ a , 2 , ⋯ γ a , n a } , (97) \mathcal{O}_a=\{\gamma_{a,1},\gamma_{a,2},\cdots\gamma_{a,n_a}\}, \tag*{(97)} O a = { γ a , 1 , γ a , 2 , ⋯ γ a , n a } , (97)
The boundary degrees of freedom that realize the projective representations U a ( g ~ ) U_a(\tilde{g}) U a ( g ~ ) of G \mathcal{G} G . We denote by 1 ∧ 2 1\wedge2 1 ∧ 2 the stacking operation where the entirety of the boundary degrees of freedom O 1 \mathcal{O}_1 O 1 are chosen to be to the left of the entirety of the boundary degrees of freedom O 2 \mathcal{O}_2 O 2 (the opposite case being 2 ∧ 1 2\wedge1 2 ∧ 1 ).
We then define the stacked representation U 1 ∧ 2 U_{1\wedge2} U 1 ∧ 2 of the symmetry group G \mathcal{G} G by demanding that it acts on each stacking-component O a \mathcal{O}_a O a in accordance with its respective representation U a U_a U a , as expressed by the following consistency equations
U 1 ∧ 2 ( g ~ ) γ 1 , j U 1 ∧ 2 † ( g ~ ) = U 1 ( g ~ ) γ 1 , j U 1 † ( g ~ ) , (98) U_{1\wedge2}(\tilde{g})\gamma_{1,j}U^\dagger_{1\wedge2}(\tilde{g})=U_1(\tilde{g})\gamma_{1,j}U^\dagger_1(\tilde{g}), \tag*{(98)} U 1 ∧ 2 ( g ~ ) γ 1 , j U 1 ∧ 2 † ( g ~ ) = U 1 ( g ~ ) γ 1 , j U 1 † ( g ~ ) , (98)
U 1 ∧ 2 ( g ~ ) γ 2 , k U 1 ∧ 2 † ( g ~ ) = U 2 ( g ~ ) γ 2 , k U 2 † ( g ~ ) , (99) U_{1\wedge2}(\tilde{g})\gamma_{2,k}U^\dagger_{1\wedge2}(\tilde{g})=U_2(\tilde{g})\gamma_{2,k}U^\dagger_2(\tilde{g}), \tag*{(99)} U 1 ∧ 2 ( g ~ ) γ 2 , k U 1 ∧ 2 † ( g ~ ) = U 2 ( g ~ ) γ 2 , k U 2 † ( g ~ ) , (99)
where j = 1 , ⋯ , n 1 j=1,\cdots,n_1 j = 1 , ⋯ , n 1 and k = 1 , ⋯ , n 2 k=1,\cdots,n_2 k = 1 , ⋯ , n 2 .
In what follows, we consider the stacking for four cases where μ 1 = 0 , 1 \mu_1=0,1 μ 1 = 0 , 1 and μ 2 = 0 , 1 \mu_2=0,1 μ 2 = 0 , 1 . In each case, we denote by Q 1 ∧ 2 Q_{1\wedge2} Q 1 ∧ 2 , C 1 ∧ 2 ( g ) C_{1\wedge2}(g) C 1 ∧ 2 ( g ) , Y 1 ∧ 2 Y_{1\wedge2} Y 1 ∧ 2 the stacked parafermion charge operator, the charge conjugation operator associated with element g ∈ G b g\in G_b g ∈ G b , and the generator of the center when the stacked index μ 1 ∧ 2 = 1 \mu_{1\wedge2}=1 μ 1 ∧ 2 = 1 , according to the definitions introduced in § 1.2 .
Even-even stacking
For even-even stacking, we have μ 1 = μ 2 = 0 \mu_1=\mu_2=0 μ 1 = μ 2 = 0 . We thus compute the stacked index
μ 1 ∧ 2 = μ 1 + μ 2 = 0 ( m o d 2 ) . (100) \mu_{1\wedge2}=\mu_1+\mu_2=0\pmod{2}. \tag*{(100)} μ 1 ∧ 2 = μ 1 + μ 2 = 0 ( mod 2 ) . (100)
The boundary projective representations U 1 ( z , g ) U_1(z,g) U 1 ( z , g ) and U 2 ( z , g ) U_2(z,g) U 2 ( z , g ) are of the form (35) . We define for any ( z , g ) ∈ G (z,g)\in\mathcal{G} ( z , g ) ∈ G the stacked representation ansatz
U 1 ∧ 2 ( z , g ) : = Q 1 ∧ 2 z C 1 ∧ 2 ( g ) V 1 ∧ 2 ( g ) , (101) U_{1\wedge2}(z,g):=Q^z_{1\wedge2}C_{1\wedge2}(g)V_{1\wedge2}(g), \tag*{(101)} U 1 ∧ 2 ( z , g ) := Q 1 ∧ 2 z C 1 ∧ 2 ( g ) V 1 ∧ 2 ( g ) , (101)
where
V 1 ∧ 2 ( g ) : = V 1 ( g ) V 2 ( g ) Q 1 − κ 2 ( g ) Q 2 + κ 1 ( g ) . (102) V_{1\wedge2}(g):=V_1(g)V_2(g)Q_1^{-\kappa_2(g)}Q_2^{+\kappa_1(g)}. \tag*{(102)} V 1 ∧ 2 ( g ) := V 1 ( g ) V 2 ( g ) Q 1 − κ 2 ( g ) Q 2 + κ 1 ( g ) . (102)
One verifies that the stacked representation (4.12) satisfies the consistency relations (4.10). In particular, the parafermion charge operators Q 1 Q_1 Q 1 and Q 2 Q_2 Q 2 are needed in Eq. (4.12b) to compensate for the multiplicative phase ω N − κ 1 ( g ) \omega_N^{-\kappa_1(g)} ω N − κ 1 ( g ) that appears when conjugating γ 2 , k \gamma_{2,k} γ 2 , k by V 1 ( g ) V_1(g) V 1 ( g ) and the multiplicative phase ω N + κ 2 ( g ) \omega_N^{+\kappa_2(g)} ω N + κ 2 ( g ) that appears when conjugating γ 1 , j \gamma_{1,j} γ 1 , j by V 2 ( g ) V_2(g) V 2 ( g ) .
To compute the indices κ 1 ∧ 2 \kappa_{1\wedge2} κ 1 ∧ 2 and ν 1 ∧ 2 \nu_{1\wedge2} ν 1 ∧ 2 associated with the stacked representation (4.12), it is sufficient to consider the composition of two elements g ~ ≡ ( z , g ) \tilde{g}\equiv(z,g) g ~ ≡ ( z , g ) and h ~ ≡ ( u , h ) \tilde{h}\equiv(u,h) h ~ ≡ ( u , h ) in G \mathcal{G} G , i.e.,
U 1 ∧ 2 ( g ~ ) U 1 ∧ 2 ( h ~ ) = Q 1 ∧ 2 z + u C 1 ∧ 2 ( g ) V 1 ∧ 2 ( g ) × Q 1 ∧ 2 u C 1 ∧ 2 ( h ) V 1 ∧ 2 ( h ) . (103) U_{1\wedge2}(\tilde{g})U_{1\wedge2}(\tilde{h})=Q^{z+u}_{1\wedge2}C_{1\wedge2}(g)V_{1\wedge2}(g)
\times Q^u_{1\wedge2}C_{1\wedge2}(h)V_{1\wedge2}(h). \tag*{(103)} U 1 ∧ 2 ( g ~ ) U 1 ∧ 2 ( h ~ ) = Q 1 ∧ 2 z + u C 1 ∧ 2 ( g ) V 1 ∧ 2 ( g ) × Q 1 ∧ 2 u C 1 ∧ 2 ( h ) V 1 ∧ 2 ( h ) . (103)
Rearranging the terms produces phase factors dependent on the parafermion charges of each operator according to Eq. (17) . The operators Q 1 Q_1 Q 1 and Q 2 Q_2 Q 2 carry no parafermion charge, while the operators V 1 ( g ) V_1(g) V 1 ( g ) and V 2 ( g ) V_2(g) V 2 ( g ) carry parafermion charges − κ 1 ( g ) -\kappa_1(g) − κ 1 ( g ) and − κ 2 ( g ) -\kappa_2(g) − κ 2 ( g ) , respectively. The charge conjugation operator C 1 ∧ 2 ( g ) C_{1\wedge2}(g) C 1 ∧ 2 ( g ) does not have a definite parafermion charge. Finally, the parafermion charge of the operator
[ V a ( g ) ] ρ h = C 1 ∧ 2 ( h ) V a ( g ) C 1 ∧ 2 ( h ) , (104) [V_a(g)]^{\rho_h} = C_{1\wedge2}(h)V_a(g)C_{1\wedge2}(h), \tag*{(104)} [ V a ( g ) ] ρ h = C 1 ∧ 2 ( h ) V a ( g ) C 1 ∧ 2 ( h ) , (104)
is simply − ρ h κ a ( g ) -\rho_h\kappa_a(g) − ρ h κ a ( g ) , with a = 1 , 2 a=1,2 a = 1 , 2 . After rearranging the terms, we obtain
U 1 ∧ 2 ( g ~ ) U 1 ∧ 2 ( h ~ ) = ω N [ κ 1 ( g ) + κ 2 ( g ) ] u − ρ h κ 1 ( g ) κ 2 ( h ) × Q 1 ∧ 2 z + ρ g u C 1 ∧ 2 ( g h ) × [ V 1 ( g ) ] ρ h V 1 ( h ) [ V 2 ( g ) ] ρ h V 2 ( h ) × Q 1 − ρ h κ 2 ( g ) − κ 2 ( h ) Q 2 + ρ h κ 1 ( g ) + κ 1 ( h ) . (105) \begin{aligned}
U_{1\wedge2}(\tilde{g})U_{1\wedge2}(\tilde{h})={}&\omega_N^{[\kappa_1(g)+\kappa_2(g)]u-\rho_h\kappa_1(g)\kappa_2(h)}\\
&\times Q_{1\wedge2}^{z+\rho_g u}C_{1\wedge2}(gh)\\
&\times[V_1(g)]^{\rho_h}V_1(h)[V_2(g)]^{\rho_h}V_2(h)\\
&\times Q_1^{-\rho_h\kappa_2(g)-\kappa_2(h)}Q_2^{+\rho_h\kappa_1(g)+\kappa_1(h)}. \tag*{(105)}
\end{aligned} U 1 ∧ 2 ( g ~ ) U 1 ∧ 2 ( h ~ ) = ω N [ κ 1 ( g ) + κ 2 ( g )] u − ρ h κ 1 ( g ) κ 2 ( h ) × Q 1 ∧ 2 z + ρ g u C 1 ∧ 2 ( g h ) × [ V 1 ( g ) ] ρ h V 1 ( h ) [ V 2 ( g ) ] ρ h V 2 ( h ) × Q 1 − ρ h κ 2 ( g ) − κ 2 ( h ) Q 2 + ρ h κ 1 ( g ) + κ 1 ( h ) . (105)
Now, we use the following two identities for a = 1 , 2 a=1,2 a = 1 , 2 . First,
[ V a ( g ) ] ρ h V a ( h ) = e i ν a ( g , h ) Q a ρ g κ a ( g , h ) V a ( g h ) , (106) [V_a(g)]^{\rho_h}V_a(h)=e^{i\nu_a(g,h)}Q_a^{\rho_g\kappa_a(g,h)}V_a(gh), \tag*{(106)} [ V a ( g ) ] ρ h V a ( h ) = e i ν a ( g , h ) Q a ρ g κ a ( g , h ) V a ( g h ) , (106)
which follows immediately from Eq. (38) . Second,
ρ h κ a ( g ) + κ a ( h ) = κ a ( g h ) , (107) \rho_h\kappa_a(g)+\kappa_a(h)=\kappa_a(gh), \tag*{(107)} ρ h κ a ( g ) + κ a ( h ) = κ a ( g h ) , (107)
which follows from the cocycle condition δ ρ ( 1 ) κ = 0 \delta^{(1)}_\rho\kappa=0 δ ρ ( 1 ) κ = 0 . These two identities deliver
U 1 ∧ 2 ( g ~ ) U 1 ∧ 2 ( h ~ ) = ω N [ κ 1 ( g ) + κ 2 ( g ) ] u e i ν 1 ( g , h ) + i ν 2 ( g , h ) − i 2 π N ρ h κ 1 ( g ) κ 2 ( h ) × Q 1 ∧ 2 z + ρ g u + c ( g , h ) C 1 ∧ 2 ( g h ) × V 1 ( g h ) V 2 ( g h ) Q 2 − κ 2 ( g h ) Q 1 + κ 1 ( g h ) = ω N [ κ 1 ( g ) + κ 2 ( g ) ] u e i ν 1 ( g , h ) + i ν 2 ( g , h ) − i 2 π N ρ h κ 1 ( g ) κ 2 ( h ) × U 1 ∧ 2 ( g ~ ∘ h ~ ) . (108) \begin{aligned}
U_{1\wedge2}(\tilde{g})U_{1\wedge2}(\tilde{h})={}&\omega_N^{[\kappa_1(g)+\kappa_2(g)]u}e^{i\nu_1(g,h)+i\nu_2(g,h)-i\frac{2\pi}{N}\rho_h\kappa_1(g)\kappa_2(h)}\\
&\times Q_{1\wedge2}^{z+\rho_g u+c(g,h)}C_{1\wedge2}(gh)\\
&\times V_1(gh)V_2(gh)Q_2^{-\kappa_2(gh)}Q_1^{+\kappa_1(gh)}\\
={}&\omega_N^{[\kappa_1(g)+\kappa_2(g)]u}e^{i\nu_1(g,h)+i\nu_2(g,h)-i\frac{2\pi}{N}\rho_h\kappa_1(g)\kappa_2(h)}\\
&\times U_{1\wedge2}(\tilde{g}\circ\tilde{h}). \tag*{(108)}
\end{aligned} U 1 ∧ 2 ( g ~ ) U 1 ∧ 2 ( h ~ ) = = ω N [ κ 1 ( g ) + κ 2 ( g )] u e i ν 1 ( g , h ) + i ν 2 ( g , h ) − i N 2 π ρ h κ 1 ( g ) κ 2 ( h ) × Q 1 ∧ 2 z + ρ g u + c ( g , h ) C 1 ∧ 2 ( g h ) × V 1 ( g h ) V 2 ( g h ) Q 2 − κ 2 ( g h ) Q 1 + κ 1 ( g h ) ω N [ κ 1 ( g ) + κ 2 ( g )] u e i ν 1 ( g , h ) + i ν 2 ( g , h ) − i N 2 π ρ h κ 1 ( g ) κ 2 ( h ) × U 1 ∧ 2 ( g ~ ∘ h ~ ) . (108)
Comparison with Eq. (3.12) yields the stacked indices
κ 1 ∧ 2 ( g ) = κ 1 ( g ) + κ 2 ( g ) , (109) \kappa_{1\wedge2}(g)=\kappa_1(g)+\kappa_2(g), \tag*{(109)} κ 1 ∧ 2 ( g ) = κ 1 ( g ) + κ 2 ( g ) , (109)
ν 1 ∧ 2 ( g , h ) = ν 1 ( g , h ) + ν 2 ( g , h ) − 2 π N ρ h κ 1 ( g ) κ 2 ( h ) , (110) \nu_{1\wedge2}(g,h)=\nu_1(g,h)+\nu_2(g,h)-\frac{2\pi}{N}\rho_h\kappa_1(g)\kappa_2(h), \tag*{(110)} ν 1 ∧ 2 ( g , h ) = ν 1 ( g , h ) + ν 2 ( g , h ) − N 2 π ρ h κ 1 ( g ) κ 2 ( h ) , (110)
which we write compactly as
κ 1 ∧ 2 = κ 1 + κ 2 , (111) \kappa_{1\wedge2}=\kappa_1+\kappa_2, \tag*{(111)} κ 1 ∧ 2 = κ 1 + κ 2 , (111)
ν 1 ∧ 2 = ν 1 + ν 2 − 2 π N κ 1 ∪ κ 2 . (112) \nu_{1\wedge2}=\nu_1+\nu_2-\frac{2\pi}{N}\kappa_1\cup\kappa_2. \tag*{(112)} ν 1 ∧ 2 = ν 1 + ν 2 − N 2 π κ 1 ∪ κ 2 . (112)
The index κ 1 ∧ 2 \kappa_{1\wedge2} κ 1 ∧ 2 quantifies the combined parafermion charge of the unitary operators V 1 V_1 V 1 and V 2 V_2 V 2 . The stacked index ν 1 ∧ 2 \nu_{1\wedge2} ν 1 ∧ 2 acquires an additional contribution 2 π N κ 1 ∪ κ 2 \frac{2\pi}{N}\kappa_1\cup\kappa_2 N 2 π κ 1 ∪ κ 2 due to the parafermionic nature of the underlying degrees of freedom. When N = 2 N=2 N = 2 , this additional contribution reduces to the case of Majorana degrees of freedom [68 ] . In contrast to the Majorana case, the stacking rule (4.18) can lead to a non-Abelian stacking group of IPT phases, see Sec. V for such an example. This is because for any two 1-cocycles κ , κ ′ ∈ Z 1 ( G b , [ Z N ] ρ ) \kappa,\kappa'\in Z^1(G_b,[\mathbb{Z}_N]_\rho) κ , κ ′ ∈ Z 1 ( G b , [ Z N ] ρ ) , one has the identity
ρ h κ ( g ) κ ′ ( h ) ∼ − ρ h κ ′ ( g ) κ ( h ) . (113) \rho_h\kappa(g)\kappa'(h)\sim-\rho_h\kappa'(g)\kappa(h). \tag*{(113)} ρ h κ ( g ) κ ′ ( h ) ∼ − ρ h κ ′ ( g ) κ ( h ) . (113)
In other words, the cup product of the two 1-cocycles is antisymmetric up to a 2-coboundary, i.e.,
κ 1 ∪ κ 2 ∼ − κ 2 ∪ κ 1 ⟹ ν 1 ∧ 2 ≠ ν 2 ∧ 1 . (114) \kappa_1\cup\kappa_2\sim-\kappa_2\cup\kappa_1\Longrightarrow\nu_{1\wedge2}\ne\nu_{2\wedge1}. \tag*{(114)} κ 1 ∪ κ 2 ∼ − κ 2 ∪ κ 1 ⟹ ν 1 ∧ 2 = ν 2 ∧ 1 . (114)
Odd-odd stacking
For odd-odd stacking, we have both μ 1 = μ 2 = 1 \mu_1=\mu_2=1 μ 1 = μ 2 = 1 . We thus compute the stacked index
μ 1 ∧ 2 = μ 1 + μ 2 = 0 m o d 2. (115) \mu_{1\wedge2}=\mu_1+\mu_2=0\ \mathop{\rm mod}\ 2. \tag*{(115)} μ 1 ∧ 2 = μ 1 + μ 2 = 0 mod 2. (115)
The boundary projective representations U 1 ( z , g ) U_1(z,g) U 1 ( z , g ) and U 2 ( z , g ) U_2(z,g) U 2 ( z , g ) are of the form (52) . Since μ 1 ∧ 2 = 0 \mu_{1\wedge2}=0 μ 1 ∧ 2 = 0 , the stacked representation must act on a local Hilbert space and must be built out of n 1 + n 2 n_1+n_2 n 1 + n 2 local parafermion operators. In other words, the stacked representation cannot contain the auxiliary parafermions γ 1 , ∞ \gamma_{1,\infty} γ 1 , ∞ and γ 2 , ∞ \gamma_{2,\infty} γ 2 , ∞ . As we shall see, the key idea is that the central elements Y 2 Y_2 Y 2 and Y 1 Y_1 Y 1 act just like γ 1 , ∞ \gamma_{1,\infty} γ 1 , ∞ and γ 2 , ∞ † \gamma^\dagger_{2,\infty} γ 2 , ∞ † on the parafermion operators γ 1 , j \gamma_{1,j} γ 1 , j and γ 2 , k \gamma_{2,k} γ 2 , k , respectively.
We define for any ( z , g ) ∈ G (z,g)\in\mathcal{G} ( z , g ) ∈ G the stacked representation ansatz
U 1 ∧ 2 ( z , g ) : = Q 1 ∧ 2 z C 1 ∧ 2 ( g ) V 1 ∧ 2 ( g ) , (116) U_{1\wedge2}(z,g):=Q_{1\wedge2}^{z}C_{1\wedge2}(g)V_{1\wedge2}(g), \tag*{(116)} U 1 ∧ 2 ( z , g ) := Q 1 ∧ 2 z C 1 ∧ 2 ( g ) V 1 ∧ 2 ( g ) , (116)
where
V 1 ∧ 2 ( g ) : = V 1 ( g ) V 2 ( g ) Y 1 + κ 2 ( g ) Y 2 − κ 1 ( g ) . (117) V_{1\wedge2}(g):=V_1(g)V_2(g)Y_1^{+\kappa_2(g)}Y_2^{-\kappa_1(g)}. \tag*{(117)} V 1 ∧ 2 ( g ) := V 1 ( g ) V 2 ( g ) Y 1 + κ 2 ( g ) Y 2 − κ 1 ( g ) . (117)
One verifies that the stacked representation (4.22) satisfies the consistency relations (4.10). Note that, by definition, the operators V 1 ( g ) V_1(g) V 1 ( g ) and V 2 ( g ) V_2(g) V 2 ( g ) do not carry parafermion charges. Hence, for any g g g , the product V 1 ( g ) Y 2 + κ 1 ( g ) V_1(g)Y_2^{+\kappa_1(g)} V 1 ( g ) Y 2 + κ 1 ( g ) commutes with all γ 2 , k \gamma_{2,k} γ 2 , k parafermions. Similarly, the product V 2 ( g ) Y 1 − κ 2 ( g ) V_2(g)Y_1^{-\kappa_2(g)} V 2 ( g ) Y 1 − κ 2 ( g ) commutes with all γ 1 , j \gamma_{1,j} γ 1 , j parafermions.
To compute the indices κ 1 ∧ 2 \kappa_{1\wedge2} κ 1 ∧ 2 and ν 1 ∧ 2 \nu_{1\wedge2} ν 1 ∧ 2 associated with the representation (4.22), it is sufficient to consider the composition of two elements g ~ ≡ ( z , g ) \tilde{g}\equiv(z,g) g ~ ≡ ( z , g ) and h ~ ≡ ( u , h ) \tilde{h}\equiv(u,h) h ~ ≡ ( u , h ) in G \mathcal{G} G , i.e.,
U 1 ∧ 2 ( g ~ ) U 1 ∧ 2 ( h ~ ) = Q 1 ∧ 2 z C 1 ∧ 2 ( g ) V 1 ∧ 2 ( g ) Q 1 ∧ 2 u C 1 ∧ 2 ( h ) V 1 ∧ 2 ( h ) . (118) U_{1\wedge2}(\tilde{g})U_{1\wedge2}(\tilde{h})=Q_{1\wedge2}^{z}C_{1\wedge2}(g)V_{1\wedge2}(g)Q_{1\wedge2}^{u}C_{1\wedge2}(h)V_{1\wedge2}(h). \tag*{(118)} U 1 ∧ 2 ( g ~ ) U 1 ∧ 2 ( h ~ ) = Q 1 ∧ 2 z C 1 ∧ 2 ( g ) V 1 ∧ 2 ( g ) Q 1 ∧ 2 u C 1 ∧ 2 ( h ) V 1 ∧ 2 ( h ) . (118)
Upon reordering of the terms, we find
U 1 ∧ 2 ( g ~ ) U 1 ∧ 2 ( h ~ ) = ω N [ κ 1 ( g ) − κ 2 ( g ) ] u + ρ h κ 1 ( g ) κ 2 ( h ) × Q 1 ∧ 2 z + ρ g u C 1 ∧ 2 ( g h ) × [ V 1 ( g ) ] ρ h V 1 ( h ) [ V 2 ( g ) ] ρ h V 2 ( h ) × Y 1 + ρ h κ 2 ( g ) + κ 2 ( h ) Y 2 − ρ h κ 1 ( g ) − κ 1 ( h ) . (119) \begin{aligned}
U_{1\wedge2}(\tilde{g})U_{1\wedge2}(\tilde{h})={}&\omega_N^{[\kappa_1(g)-\kappa_2(g)]u+\rho_h\kappa_1(g)\kappa_2(h)}\\
&\times Q_{1\wedge2}^{z+\rho_g u}C_{1\wedge2}(gh)\\
&\times[V_1(g)]^{\rho_h}V_1(h)[V_2(g)]^{\rho_h}V_2(h)\\
&\times Y_1^{+\rho_h\kappa_2(g)+\kappa_2(h)}Y_2^{-\rho_h\kappa_1(g)-\kappa_1(h)}. \tag*{(119)}
\end{aligned} U 1 ∧ 2 ( g ~ ) U 1 ∧ 2 ( h ~ ) = ω N [ κ 1 ( g ) − κ 2 ( g )] u + ρ h κ 1 ( g ) κ 2 ( h ) × Q 1 ∧ 2 z + ρ g u C 1 ∧ 2 ( g h ) × [ V 1 ( g ) ] ρ h V 1 ( h ) [ V 2 ( g ) ] ρ h V 2 ( h ) × Y 1 + ρ h κ 2 ( g ) + κ 2 ( h ) Y 2 − ρ h κ 1 ( g ) − κ 1 ( h ) . (119)
where the multiplicative phase factors arise when we pass Q 1 ∧ 2 u Q_{1\wedge2}^{u} Q 1 ∧ 2 u through Y 1 Y_1 Y 1 and Y 2 Y_2 Y 2 , and when Y 2 − ρ h κ 1 ( g ) Y_2^{-\rho_h\kappa_1(g)} Y 2 − ρ h κ 1 ( g ) is passed through Y 1 + κ 2 ( h ) Y_1^{+\kappa_2(h)} Y 1 + κ 2 ( h ) .
Using the identity following from Eq. (59)
[ V a ( g ) ] ρ h V a ( h ) = e i ν a ( g , h ) V a ( g h ) , (120) [V_a(g)]^{\rho_h}V_a(h)=e^{i\nu_a(g,h)}V_a(gh), \tag*{(120)} [ V a ( g ) ] ρ h V a ( h ) = e i ν a ( g , h ) V a ( g h ) , (120)
in combination with the cocycle condition (107) delivers the final result,
U 1 ∧ 2 ( g ~ ) U 1 ∧ 2 ( h ~ ) = ω N [ κ 1 ( g ) − κ 2 ( g ) ] u e i ν 1 ( g , h ) + i ν 2 ( g , h ) + i 2 π N ρ h κ 1 ( g ) κ 2 ( h ) × Q 1 ∧ 2 z + ρ g u C 1 ∧ 2 ( g h ) × V 1 ( g h ) V 2 ( g h ) Y 1 + κ 2 ( g h ) Y 2 − κ 1 ( g h ) = ω N [ κ 1 ( g ) − κ 2 ( g ) ] u e i ν 1 ( g , h ) + i ν 2 ( g , h ) + i 2 π N ρ h κ 1 ( g ) κ 2 ( h ) × U 1 ∧ 2 ( g ~ ∘ h ~ ) . (121) \begin{aligned}
U_{1\wedge2}(\tilde{g})U_{1\wedge2}(\tilde{h})={}&\omega_N^{[\kappa_1(g)-\kappa_2(g)]u}e^{i\nu_1(g,h)+i\nu_2(g,h)+i\frac{2\pi}{N}\rho_h\kappa_1(g)\kappa_2(h)}\\
&\times Q_{1\wedge2}^{z+\rho_g u}C_{1\wedge2}(gh)\\
&\times V_1(gh)V_2(gh)Y_1^{+\kappa_2(gh)}Y_2^{-\kappa_1(gh)}\\
={}&\omega_N^{[\kappa_1(g)-\kappa_2(g)]u}e^{i\nu_1(g,h)+i\nu_2(g,h)+i\frac{2\pi}{N}\rho_h\kappa_1(g)\kappa_2(h)}\\
&\times U_{1\wedge2}(\tilde{g}\circ\tilde{h}). \tag*{(121)}
\end{aligned} U 1 ∧ 2 ( g ~ ) U 1 ∧ 2 ( h ~ ) = = ω N [ κ 1 ( g ) − κ 2 ( g )] u e i ν 1 ( g , h ) + i ν 2 ( g , h ) + i N 2 π ρ h κ 1 ( g ) κ 2 ( h ) × Q 1 ∧ 2 z + ρ g u C 1 ∧ 2 ( g h ) × V 1 ( g h ) V 2 ( g h ) Y 1 + κ 2 ( g h ) Y 2 − κ 1 ( g h ) ω N [ κ 1 ( g ) − κ 2 ( g )] u e i ν 1 ( g , h ) + i ν 2 ( g , h ) + i N 2 π ρ h κ 1 ( g ) κ 2 ( h ) × U 1 ∧ 2 ( g ~ ∘ h ~ ) . (121)
Comparison with Eq. (3.12) yields the stacked indices
κ 1 ∧ 2 ( g ) = κ 1 ( g ) − κ 2 ( g ) , (122) \kappa_{1\wedge2}(g)=\kappa_1(g)-\kappa_2(g), \tag*{(122)} κ 1 ∧ 2 ( g ) = κ 1 ( g ) − κ 2 ( g ) , (122)
ν 1 ∧ 2 ( g , h ) = ν 1 ( g , h ) + ν 2 ( g , h ) + 2 π N ρ h κ 1 ( g ) κ 2 ( h ) , (123) \nu_{1\wedge2}(g,h)=\nu_1(g,h)+\nu_2(g,h)+\frac{2\pi}{N}\rho_h\kappa_1(g)\kappa_2(h), \tag*{(123)} ν 1 ∧ 2 ( g , h ) = ν 1 ( g , h ) + ν 2 ( g , h ) + N 2 π ρ h κ 1 ( g ) κ 2 ( h ) , (123)
which we write compactly as
κ 1 ∧ 2 = κ 1 − κ 2 , (124) \kappa_{1\wedge2}=\kappa_1-\kappa_2, \tag*{(124)} κ 1 ∧ 2 = κ 1 − κ 2 , (124)
ν 1 ∧ 2 = ν 1 + ν 2 + 2 π N κ 1 ∪ κ 2 . (125) \nu_{1\wedge2}=\nu_1+\nu_2+\frac{2\pi}{N}\kappa_1\cup\kappa_2. \tag*{(125)} ν 1 ∧ 2 = ν 1 + ν 2 + N 2 π κ 1 ∪ κ 2 . (125)
The index κ 1 ∧ 2 \kappa_{1\wedge2} κ 1 ∧ 2 quantifies the total parafermion charge of the unitary operator V 1 ∧ 2 V_{1\wedge2} V 1 ∧ 2 . The stacked index ν 1 ∧ 2 \nu_{1\wedge2} ν 1 ∧ 2 acquires an additional contribution 2 π N κ 1 ∪ κ 2 \frac{2\pi}{N}\kappa_1\cup\kappa_2 N 2 π κ 1 ∪ κ 2 due to the parafermionic nature of the underlying degrees of freedom. Note the sign differences compared to the even-even stacking case (4.18). Here too, the non-commutativity (114) holds.
Even-odd stacking
For even-odd stacking, we have μ 1 = 0 \mu_1=0 μ 1 = 0 and μ 2 = 1 \mu_2=1 μ 2 = 1 . We thus compute the stacked index
μ 1 ∧ 2 = μ 1 + μ 2 = 1 m o d 2. (126) \mu_{1\wedge2}=\mu_1+\mu_2=1\ \mathrm{mod}\ 2. \tag*{(126)} μ 1 ∧ 2 = μ 1 + μ 2 = 1 mod 2. (126)
The boundary projective representations U 1 ( z , g ) U_1(z,g) U 1 ( z , g ) and U 2 ( z , g ) U_2(z,g) U 2 ( z , g ) are of the forms (35) and (52) , respectively. The stacked representation must be of the form (52) , where the auxiliary parafermion is supplied by the operator γ 2 , ∞ \gamma_{2,\infty} γ 2 , ∞ . Hence, we identify γ 1 ∧ 2 , ∞ ≡ γ 2 , ∞ \gamma_{1\wedge2,\infty}\equiv\gamma_{2,\infty} γ 1 ∧ 2 , ∞ ≡ γ 2 , ∞ .
We define for any ( z , g ) ∈ G (z,g)\in\mathcal{G} ( z , g ) ∈ G the stacked representation ansatz
U 1 ∧ 2 ( z , g ) : = Q 1 ∧ 2 z C 1 ∧ 2 ( g ) V 1 ∧ 2 ( g ) γ 1 ∧ 2 , ∞ − ( κ 1 ( g ) + κ 2 ( g ) ) , (127) U_{1\wedge2}(z,g):=Q_{1\wedge2}^{z}C_{1\wedge2}(g)V_{1\wedge2}(g)\gamma_{1\wedge2,\infty}^{-(\kappa_1(g)+\kappa_2(g))}, \tag*{(127)} U 1 ∧ 2 ( z , g ) := Q 1 ∧ 2 z C 1 ∧ 2 ( g ) V 1 ∧ 2 ( g ) γ 1 ∧ 2 , ∞ − ( κ 1 ( g ) + κ 2 ( g )) , (127)
where
V 1 ∧ 2 ( g ) : = V 1 ( g ) V 2 ( g ) Q 1 − κ 2 ( g ) Y 2 + κ 1 ( g ) . (128) V_{1\wedge2}(g):=V_1(g)V_2(g)Q_1^{-\kappa_2(g)}Y_2^{+\kappa_1(g)}. \tag*{(128)} V 1 ∧ 2 ( g ) := V 1 ( g ) V 2 ( g ) Q 1 − κ 2 ( g ) Y 2 + κ 1 ( g ) . (128)
One verifies that the stacked representation (4.22) satisfies the consistency relations (4.10). Note that the operator V 1 ∧ 2 ( g ) V_{1\wedge2}(g) V 1 ∧ 2 ( g ) carries no parafermion charge. This is because both V 2 ( g ) V_2(g) V 2 ( g ) and Q 1 Q_1 Q 1 carry no parafermion charges, while the parafermion charge of Y 2 + κ 1 ( g ) Y_2^{+\kappa_1(g)} Y 2 + κ 1 ( g ) compensates that of the operator V 1 ( g ) V_1(g) V 1 ( g ) . Similarly, one observes that the operator V 1 ∧ 2 ( g ) V_{1\wedge2}(g) V 1 ∧ 2 ( g ) commutes with both the central element Y 1 ∧ 2 Y_{1\wedge2} Y 1 ∧ 2 and the auxiliary parafermion γ 1 ∧ 2 , ∞ \gamma_{1\wedge2,\infty} γ 1 ∧ 2 , ∞ .
Comparing with Eq. (52) , one finds that the stacked index κ 1 ∧ 2 \kappa_{1\wedge2} κ 1 ∧ 2 is given by
κ 1 ∧ 2 ( g ) = κ 1 ( g ) + κ 2 ( g ) . (129) \kappa_{1\wedge2}(g)=\kappa_1(g)+\kappa_2(g). \tag*{(129)} κ 1 ∧ 2 ( g ) = κ 1 ( g ) + κ 2 ( g ) . (129)
The stacked index ν 1 ∧ 2 \nu_{1\wedge2} ν 1 ∧ 2 can then be obtained by computing
U 1 ∧ 2 ( 0 , g ) U 1 ∧ 2 ( 0 , h ) = C 1 ∧ 2 ( g ) V 1 ∧ 2 ( g ) γ 1 ∧ 2 , ∞ − κ 1 ∧ 2 ( g ) × C 1 ∧ 2 ( h ) V 1 ∧ 2 ( h ) γ 1 ∧ 2 , ∞ − κ 1 ∧ 2 ( h ) = C 1 ∧ 2 ( g h ) [ V 1 ∧ 2 ( g ) ] ρ h V 1 ∧ 2 ( h ) γ 1 ∧ 2 , ∞ − ρ h κ 1 ∧ 2 ( g ) − κ 1 ∧ 2 ( h ) . (130) \begin{aligned}
U_{1\wedge2}(0,g)U_{1\wedge2}(0,h)
&=C_{1\wedge2}(g)V_{1\wedge2}(g)\gamma_{1\wedge2,\infty}^{-\kappa_{1\wedge2}(g)}\\
&\quad\times C_{1\wedge2}(h)V_{1\wedge2}(h)\gamma_{1\wedge2,\infty}^{-\kappa_{1\wedge2}(h)}\\
&=C_{1\wedge2}(gh)[V_{1\wedge2}(g)]^{\rho_h}V_{1\wedge2}(h)\gamma_{1\wedge2,\infty}^{-\rho_h\kappa_{1\wedge2}(g)-\kappa_{1\wedge2}(h)}. \tag*{(130)}
\end{aligned} U 1 ∧ 2 ( 0 , g ) U 1 ∧ 2 ( 0 , h ) = C 1 ∧ 2 ( g ) V 1 ∧ 2 ( g ) γ 1 ∧ 2 , ∞ − κ 1 ∧ 2 ( g ) × C 1 ∧ 2 ( h ) V 1 ∧ 2 ( h ) γ 1 ∧ 2 , ∞ − κ 1 ∧ 2 ( h ) = C 1 ∧ 2 ( g h ) [ V 1 ∧ 2 ( g ) ] ρ h V 1 ∧ 2 ( h ) γ 1 ∧ 2 , ∞ − ρ h κ 1 ∧ 2 ( g ) − κ 1 ∧ 2 ( h ) . (130)
where in reaching the last line we used the fact that V 1 ∧ 2 ( g ) V_{1\wedge2}(g) V 1 ∧ 2 ( g ) commutes with γ 1 ∧ 2 , ∞ \gamma_{1\wedge2,\infty} γ 1 ∧ 2 , ∞ . We now rearrange the terms in the product [ V 1 ∧ 2 ( g ) ] ρ h [V_{1\wedge2}(g)]^{\rho_h} [ V 1 ∧ 2 ( g ) ] ρ h as
[ V 1 ∧ 2 ( g ) ] ρ h V 1 ∧ 2 ( h ) = [ V 1 ( g ) ] ρ h [ V 2 ( g ) ] ρ h Q 1 − ρ h κ 2 ( g ) Y 2 + ρ h κ 1 ( g ) × V 1 ( h ) V 2 ( h ) Q 1 − κ 2 ( h ) Y 2 + κ 1 ( h ) = ω N ρ h κ 1 ( g ) κ 1 ( h ) + ρ h κ 2 ( g ) κ 1 ( h ) e i ν 1 ( g , h ) + i ν 2 ( g , h ) × V 1 ( g h ) V 2 ( g h ) Q 1 − κ 2 ( g h ) Y 2 + κ 1 ( g h ) , (131) \begin{aligned}
[V_{1\wedge2}(g)]^{\rho_h}V_{1\wedge2}(h)
={}&[V_1(g)]^{\rho_h}[V_2(g)]^{\rho_h}Q_1^{-\rho_h\kappa_2(g)}Y_2^{+\rho_h\kappa_1(g)}\\
&\times V_1(h)V_2(h)Q_1^{-\kappa_2(h)}Y_2^{+\kappa_1(h)}\\
={}&\omega_N^{\rho_h\kappa_1(g)\kappa_1(h)+\rho_h\kappa_2(g)\kappa_1(h)}
e^{i\nu_1(g,h)+i\nu_2(g,h)}\\
&\times V_1(gh)V_2(gh)Q_1^{-\kappa_2(gh)}Y_2^{+\kappa_1(gh)}, \tag*{(131)}
\end{aligned} [ V 1 ∧ 2 ( g ) ] ρ h V 1 ∧ 2 ( h ) = = [ V 1 ( g ) ] ρ h [ V 2 ( g ) ] ρ h Q 1 − ρ h κ 2 ( g ) Y 2 + ρ h κ 1 ( g ) × V 1 ( h ) V 2 ( h ) Q 1 − κ 2 ( h ) Y 2 + κ 1 ( h ) ω N ρ h κ 1 ( g ) κ 1 ( h ) + ρ h κ 2 ( g ) κ 1 ( h ) e i ν 1 ( g , h ) + i ν 2 ( g , h ) × V 1 ( g h ) V 2 ( g h ) Q 1 − κ 2 ( g h ) Y 2 + κ 1 ( g h ) , (131)
where we used Eq. (106) with c = 0 c=0 c = 0 for the unitary V 1 V_1 V 1 and Eq. (120) for the unitary V 2 V_2 V 2 .11 The last line is precisely the unitary V 1 ∧ 2 ( g h ) V_{1\wedge2}(gh) V 1 ∧ 2 ( g h ) . We thus find
U 1 ∧ 2 ( 0 , g ) U 1 ∧ 2 ( 0 , h ) = ω N ρ h κ 1 ( g ) κ 1 ( h ) + ρ h κ 2 ( g ) κ 1 ( h ) e i ν 1 ( g , h ) + i ν 2 ( g , h ) × U 1 ∧ 2 ( 0 , g h ) , (132) \begin{aligned}
U_{1\wedge2}(0,g)U_{1\wedge2}(0,h)
={}&\omega_N^{\rho_h\kappa_1(g)\kappa_1(h)+\rho_h\kappa_2(g)\kappa_1(h)}
e^{i\nu_1(g,h)+i\nu_2(g,h)}\\
&\times U_{1\wedge2}(0,gh), \tag*{(132)}
\end{aligned} U 1 ∧ 2 ( 0 , g ) U 1 ∧ 2 ( 0 , h ) = ω N ρ h κ 1 ( g ) κ 1 ( h ) + ρ h κ 2 ( g ) κ 1 ( h ) e i ν 1 ( g , h ) + i ν 2 ( g , h ) × U 1 ∧ 2 ( 0 , g h ) , (132)
and the stacked index ν 1 ∧ 2 ( g , h ) \nu_{1\wedge2}(g,h) ν 1 ∧ 2 ( g , h ) is given by
ν 1 ∧ 2 ( g , h ) = ν 1 ( g , h ) + ν 2 ( g , h ) + 2 π N ρ h ( κ 1 ( g ) κ 1 ( h ) + κ 2 ( g ) κ 1 ( h ) ) . (133) \nu_{1\wedge2}(g,h)=\nu_1(g,h)+\nu_2(g,h)+\frac{2\pi}{N}\rho_h\left(\kappa_1(g)\kappa_1(h)+\kappa_2(g)\kappa_1(h)\right). \tag*{(133)} ν 1 ∧ 2 ( g , h ) = ν 1 ( g , h ) + ν 2 ( g , h ) + N 2 π ρ h ( κ 1 ( g ) κ 1 ( h ) + κ 2 ( g ) κ 1 ( h ) ) . (133)
Here, we can simplify the contributions from the 1-cocycles κ 1 \kappa_1 κ 1 and κ 2 \kappa_2 κ 2 by using the identity (113) to finally obtain
ν 1 ∧ 2 ( g , h ) = ν 1 ( g , h ) + ν 2 ( g , h ) − 2 π N ρ h κ 1 ( g ) κ 2 ( h ) . (134) \nu_{1\wedge2}(g,h)=\nu_1(g,h)+\nu_2(g,h)-\frac{2\pi}{N}\rho_h\kappa_1(g)\kappa_2(h). \tag*{(134)} ν 1 ∧ 2 ( g , h ) = ν 1 ( g , h ) + ν 2 ( g , h ) − N 2 π ρ h κ 1 ( g ) κ 2 ( h ) . (134)
We write the stacked indices compactly as
κ 1 ∧ 2 = κ 1 + κ 2 , (135) \kappa_{1\wedge2}=\kappa_1+\kappa_2, \tag*{(135)} κ 1 ∧ 2 = κ 1 + κ 2 , (135)
ν 1 ∧ 2 = ν 1 + ν 2 − 2 π N κ 1 ∪ κ 2 . (136) \nu_{1\wedge2}=\nu_1+\nu_2-\frac{2\pi}{N}\kappa_1\cup\kappa_2. \tag*{(136)} ν 1 ∧ 2 = ν 1 + ν 2 − N 2 π κ 1 ∪ κ 2 . (136)
Here too, the non-commutativity (114) holds.
Odd-even stacking
For odd-even stacking, we have μ 1 = 1 \mu_1=1 μ 1 = 1 and μ 2 = 0 \mu_2=0 μ 2 = 0 . We thus compute the stacked index
μ 1 ∧ 2 = μ 1 + μ 2 = 1 m o d 2. (137) \mu_{1\wedge2}=\mu_1+\mu_2=1\ \mathrm{mod}\ 2. \tag*{(137)} μ 1 ∧ 2 = μ 1 + μ 2 = 1 mod 2. (137)
The boundary projective representations U 1 ( z , g ) U_1(z,g) U 1 ( z , g ) and U 2 ( z , g ) U_2(z,g) U 2 ( z , g ) are of the forms (52) and (35) , respectively. The stacked representation must be of the form (52) , where the auxiliary parafermion is supplied by the operator γ 1 , ∞ \gamma_{1,\infty} γ 1 , ∞ . Hence, we identify γ 1 ∧ 2 , ∞ ≡ γ 1 , ∞ \gamma_{1\wedge2,\infty}\equiv\gamma_{1,\infty} γ 1 ∧ 2 , ∞ ≡ γ 1 , ∞ . Recall that in the even-odd stacking case of Sec. IV C, the operators γ 2 , k \gamma_{2,k} γ 2 , k realizing an odd parafermion algebra lay to the “right” of those realizing an even parafermion algebra. In contrast, in odd-even stacking the operators γ 1 , j \gamma_{1,j} γ 1 , j realizing an odd parafermion algebra lie to the “left” of those realizing an even parafermion algebra. This will lead to different stacking rules.
We define for any ( z , g ) ∈ G (z,g)\in G ( z , g ) ∈ G the stacked representation ansatz
U 1 ∧ 2 ( z , g ) : = Q 1 ∧ 2 z C 1 ∧ 2 ( g ) V 1 ∧ 2 ( g ) γ 1 ∧ 2 , ∞ − ( κ 1 ( g ) − κ 2 ( g ) ) , (138) U_{1\wedge2}(z,g):=Q_{1\wedge2}^{z}C_{1\wedge2}(g)V_{1\wedge2}(g)\gamma_{1\wedge2,\infty}^{-(\kappa_1(g)-\kappa_2(g))}, \tag*{(138)} U 1 ∧ 2 ( z , g ) := Q 1 ∧ 2 z C 1 ∧ 2 ( g ) V 1 ∧ 2 ( g ) γ 1 ∧ 2 , ∞ − ( κ 1 ( g ) − κ 2 ( g )) , (138)
where
V 1 ∧ 2 ( g ) : = V 1 ( g ) V 2 ( g ) Q 2 − κ 1 ( g ) Y 1 + κ 2 ( g ) . (139) V_{1\wedge2}(g):=V_1(g)V_2(g)Q_2^{-\kappa_1(g)}Y_1^{+\kappa_2(g)}. \tag*{(139)} V 1 ∧ 2 ( g ) := V 1 ( g ) V 2 ( g ) Q 2 − κ 1 ( g ) Y 1 + κ 2 ( g ) . (139)
One verifies that the stacked representation (4.22) satisfies the consistency relations (4.10). Note that the operator V 1 ∧ 2 ( g ) V_{1\wedge2}(g) V 1 ∧ 2 ( g ) carries no parafermion charge. This is because both V 1 ( g ) V_1(g) V 1 ( g ) and Q 1 Q_1 Q 1 carry no parafermion charges, while the parafermion charge of Y 1 + κ 2 ( g ) Y_1^{+\kappa_2(g)} Y 1 + κ 2 ( g ) compensates that of the operator V 2 ( g ) V_2(g) V 2 ( g ) . Similarly, one observes that the operator V 1 ∧ 2 ( g ) V_{1\wedge2}(g) V 1 ∧ 2 ( g ) commutes with both the central element Y 1 ∧ 2 Y_{1\wedge2} Y 1 ∧ 2 and the auxiliary parafermion γ 1 ∧ 2 , ∞ \gamma_{1\wedge2,\infty} γ 1 ∧ 2 , ∞ .
Comparing with Eq. (52) , one finds that the stacked index κ 1 ∧ 2 \kappa_{1\wedge2} κ 1 ∧ 2 is given by
κ 1 ∧ 2 ( g ) = κ 1 ( g ) − κ 2 ( g ) . (140) \kappa_{1\wedge2}(g)=\kappa_1(g)-\kappa_2(g). \tag*{(140)} κ 1 ∧ 2 ( g ) = κ 1 ( g ) − κ 2 ( g ) . (140)
The stacked index ν 1 ∧ 2 \nu_{1\wedge2} ν 1 ∧ 2 can then be obtained by computing
U 1 ∧ 2 ( 0 , g ) U 1 ∧ 2 ( 0 , h ) = C 1 ∧ 2 ( g ) V 1 ∧ 2 ( g ) γ 1 ∧ 2 , ∞ − κ 1 ∧ 2 ( g ) × C 1 ∧ 2 ( h ) V 1 ∧ 2 ( h ) γ 1 ∧ 2 , ∞ − κ 1 ∧ 2 ( h ) = C 1 ∧ 2 ( g h ) [ V 1 ∧ 2 ( g ) ] ρ h V 1 ∧ 2 ( h ) γ 1 ∧ 2 , ∞ − ρ h κ 1 ∧ 2 ( g ) − κ 1 ∧ 2 ( h ) . (141) \begin{aligned}
U_{1\wedge2}(0,g)U_{1\wedge2}(0,h)
&=C_{1\wedge2}(g)V_{1\wedge2}(g)\gamma_{1\wedge2,\infty}^{-\kappa_{1\wedge2}(g)}\\
&\quad\times C_{1\wedge2}(h)V_{1\wedge2}(h)\gamma_{1\wedge2,\infty}^{-\kappa_{1\wedge2}(h)}\\
&=C_{1\wedge2}(gh)[V_{1\wedge2}(g)]^{\rho_h}V_{1\wedge2}(h)
\gamma_{1\wedge2,\infty}^{-\rho_h\kappa_{1\wedge2}(g)-\kappa_{1\wedge2}(h)}. \tag*{(141)}
\end{aligned} U 1 ∧ 2 ( 0 , g ) U 1 ∧ 2 ( 0 , h ) = C 1 ∧ 2 ( g ) V 1 ∧ 2 ( g ) γ 1 ∧ 2 , ∞ − κ 1 ∧ 2 ( g ) × C 1 ∧ 2 ( h ) V 1 ∧ 2 ( h ) γ 1 ∧ 2 , ∞ − κ 1 ∧ 2 ( h ) = C 1 ∧ 2 ( g h ) [ V 1 ∧ 2 ( g ) ] ρ h V 1 ∧ 2 ( h ) γ 1 ∧ 2 , ∞ − ρ h κ 1 ∧ 2 ( g ) − κ 1 ∧ 2 ( h ) . (141)
where in reaching the last line we used the fact that V 1 ∧ 2 ( g ) V_{1\wedge2}(g) V 1 ∧ 2 ( g ) commutes with γ 1 ∧ 2 , ∞ \gamma_{1\wedge2,\infty} γ 1 ∧ 2 , ∞ . We now rearrange the terms in the product [ V 1 ∧ 2 ( g ) ] ρ h [V_{1\wedge2}(g)]^{\rho_h} [ V 1 ∧ 2 ( g ) ] ρ h as
[ V 1 ∧ 2 ( g ) ] ρ h V 1 ∧ 2 ( h ) = [ V 1 ( g ) ] ρ h [ V 2 ( g ) ] ρ h Q 2 − ρ h κ 1 ( g ) + ρ h κ 2 ( g ) × V 1 ( h ) V 2 ( h ) Q 2 − κ 1 ( h ) + κ 2 ( h ) = ω N ρ h κ 2 ( g ) κ 2 ( h ) + ρ h κ 1 ( g ) κ 2 ( h ) e i ν 1 ( g , h ) + i ν 2 ( g , h ) × V 1 ( g h ) V 2 ( g h ) Q 2 − κ 1 ( g h ) Y 1 + κ 2 ( g h ) . (142) \begin{aligned}
[V_{1\wedge2}(g)]^{\rho_h}V_{1\wedge2}(h)
&=[V_1(g)]^{\rho_h}[V_2(g)]^{\rho_h}Q_2^{-\rho_h\kappa_1(g)+\rho_h\kappa_2(g)}\\
&\quad\times V_1(h)V_2(h)Q_2^{-\kappa_1(h)+\kappa_2(h)}\\
&=\omega_N^{\rho_h\kappa_2(g)\kappa_2(h)+\rho_h\kappa_1(g)\kappa_2(h)}
e^{i\nu_1(g,h)+i\nu_2(g,h)}\\
&\quad\times V_1(gh)V_2(gh)Q_2^{-\kappa_1(gh)}Y_1^{+\kappa_2(gh)}. \tag*{(142)}
\end{aligned} [ V 1 ∧ 2 ( g ) ] ρ h V 1 ∧ 2 ( h ) = [ V 1 ( g ) ] ρ h [ V 2 ( g ) ] ρ h Q 2 − ρ h κ 1 ( g ) + ρ h κ 2 ( g ) × V 1 ( h ) V 2 ( h ) Q 2 − κ 1 ( h ) + κ 2 ( h ) = ω N ρ h κ 2 ( g ) κ 2 ( h ) + ρ h κ 1 ( g ) κ 2 ( h ) e i ν 1 ( g , h ) + i ν 2 ( g , h ) × V 1 ( g h ) V 2 ( g h ) Q 2 − κ 1 ( g h ) Y 1 + κ 2 ( g h ) . (142)
where we used Eq. (4.16a) with c = 0 c=0 c = 0 for the unitary V 1 V_1 V 1 and Eq. (120) for the unitary V 2 V_2 V 2 .12 The last line is precisely the unitary V 1 ∧ 2 ( g h ) V_{1\wedge2}(gh) V 1 ∧ 2 ( g h ) . We thus find
U 1 ∧ 2 ( 0 , g ) U 1 ∧ 2 ( 0 , h ) = ω N ρ h κ 2 ( g ) κ 2 ( h ) − ρ h κ 1 ( g ) κ 2 ( h ) e i ν 1 ( g , h ) + i ν 2 ( g , h ) × U 1 ∧ 2 ( 0 , g h ) , (143) U_{1\wedge2}(0,g)U_{1\wedge2}(0,h)
=\omega_N^{\rho_h\kappa_2(g)\kappa_2(h)-\rho_h\kappa_1(g)\kappa_2(h)}
e^{i\nu_1(g,h)+i\nu_2(g,h)}
\times U_{1\wedge2}(0,gh), \tag*{(143)} U 1 ∧ 2 ( 0 , g ) U 1 ∧ 2 ( 0 , h ) = ω N ρ h κ 2 ( g ) κ 2 ( h ) − ρ h κ 1 ( g ) κ 2 ( h ) e i ν 1 ( g , h ) + i ν 2 ( g , h ) × U 1 ∧ 2 ( 0 , g h ) , (143)
and the stacked index ν 1 ∧ 2 ( g , h ) \nu_{1\wedge2}(g,h) ν 1 ∧ 2 ( g , h ) is given by
ν 1 ∧ 2 ( g , h ) = ν 1 ( g , h ) + ν 2 ( g , h ) + 2 π N ρ h ( κ 2 ( g ) κ 2 ( h ) + κ 1 ( g ) κ 2 ( h ) ) . (144) \nu_{1\wedge2}(g,h)=\nu_1(g,h)+\nu_2(g,h)
+\frac{2\pi}{N}\rho_h\left(\kappa_2(g)\kappa_2(h)+\kappa_1(g)\kappa_2(h)\right). \tag*{(144)} ν 1 ∧ 2 ( g , h ) = ν 1 ( g , h ) + ν 2 ( g , h ) + N 2 π ρ h ( κ 2 ( g ) κ 2 ( h ) + κ 1 ( g ) κ 2 ( h ) ) . (144)
Using the identity (113) delivers
ν 1 ∧ 2 ( g , h ) = ν 1 ( g , h ) + ν 2 ( g , h ) + 2 π N ρ h κ 1 ( g ) κ 2 ( h ) . (145) \nu_{1\wedge2}(g,h)=\nu_1(g,h)+\nu_2(g,h)+\frac{2\pi}{N}\rho_h\kappa_1(g)\kappa_2(h). \tag*{(145)} ν 1 ∧ 2 ( g , h ) = ν 1 ( g , h ) + ν 2 ( g , h ) + N 2 π ρ h κ 1 ( g ) κ 2 ( h ) . (145)
We write the stacked indices compactly as
κ 1 ∧ 2 = κ 1 − κ 2 , (146) \kappa_{1\wedge2}=\kappa_1-\kappa_2, \tag*{(146)} κ 1 ∧ 2 = κ 1 − κ 2 , (146)
ν 1 ∧ 2 = ν 1 + ν 2 + 2 π N κ 1 ∪ κ 2 . (147) \nu_{1\wedge2}=\nu_1+\nu_2+\frac{2\pi}{N}\kappa_1\cup\kappa_2. \tag*{(147)} ν 1 ∧ 2 = ν 1 + ν 2 + N 2 π κ 1 ∪ κ 2 . (147)
Here too, the non-commutativity (114) holds.
Summary of the stacking rules
We considered the stacking of two IPT phases respectively characterized by the classifying indices ( μ 1 , [ κ 1 , ν 1 ] ) (\mu_1,[\kappa_1,\nu_1]) ( μ 1 , [ κ 1 , ν 1 ]) and ( μ 2 , [ κ 2 , ν 2 ] ) (\mu_2,[\kappa_2,\nu_2]) ( μ 2 , [ κ 2 , ν 2 ]) by distinguishing four cases according to the μ 1 , μ 2 ∈ { 0 , 1 } 2 \mu_1,\mu_2\in\{0,1\}^2 μ 1 , μ 2 ∈ { 0 , 1 } 2 . We obtained the stacking rules (4.18), (4.27), (4.36), and (4.44). These can be combined into a single stacking rule:
μ 1 ∧ 2 = μ 1 + μ 2 ( m o d 2 ) , \mu_{1\wedge2}=\mu_1+\mu_2\pmod2, μ 1 ∧ 2 = μ 1 + μ 2 ( mod 2 ) ,
κ 1 ∧ 2 = κ 1 + ( − 1 ) μ 1 κ 2 , (148) \kappa_{1\wedge2}=\kappa_1+(-1)^{\mu_1}\kappa_2, \tag*{(148)} κ 1 ∧ 2 = κ 1 + ( − 1 ) μ 1 κ 2 , (148)
ν 1 ∧ 2 = ν 1 + ν 2 + ( − 1 ) μ 1 + 1 2 π N κ 1 ∪ κ 2 . \nu_{1\wedge2}=\nu_1+\nu_2+(-1)^{\mu_1+1}\frac{2\pi}{N}\kappa_1\cup\kappa_2. ν 1 ∧ 2 = ν 1 + ν 2 + ( − 1 ) μ 1 + 1 N 2 π κ 1 ∪ κ 2 .
One can identify from these formulae two sources of non-Abelian stacking: non-trivial μ 1 \mu_1 μ 1 and the (graded) anti-symmetry of the cup product.
Since the index μ = 0 \mu=0 μ = 0 identifies symmetry protected topological (SPT) phases, we can write the general stacking group structure of invertible parafermionic topological (IPT) phases as
Inv N p f ( G ) ≅ SPT N p f ( G ) ⋊ Z 2 (149) \operatorname{Inv}^{\mathrm{pf}}_N(G)\cong\operatorname{SPT}^{\mathrm{pf}}_N(G)\rtimes\mathbb{Z}_2 \tag*{(149)} Inv N pf ( G ) ≅ SPT N pf ( G ) ⋊ Z 2 (149)
The semi-direct product captures the non-Abelian stacking due to non-trivial μ 1 \mu_1 μ 1 . In stark contrast to bosons and fermions, parafermionic SPTs can themselves realize non-Abelian stacking. This is a consequence of the (graded) anti-symmetry of the cup-product, which we demonstrate explicitly in the example of Sec. V.
IPT phases with Z N × Z N × Z N p f \mathbb{Z}_N \times\mathbb{Z}_N \times\mathbb{Z}_N^{\mathrm{pf}} Z N × Z N × Z N pf symmetry
Let N N N be an odd prime-power and assume that the protecting symmetry group is given by the direct product
G ≅ Z N p f × G b , G b : = ( Z N × Z N ) . (150) \mathcal{G} \cong\mathbb{Z}_N^{\mathrm{pf}} \times G_b,\qquad G_b := (\mathbb{Z}_N \times\mathbb{Z}_N). \tag*{(150)} G ≅ Z N pf × G b , G b := ( Z N × Z N ) . (150)
We are going to introduce a family of local parafermionic Hamiltonians with the symmetry group (150) such that their IPT phases are characterized by the triplets 13
[ μ , ( κ 1 , κ 2 ) , ν ] ∈ { 0 , 1 } × { 0 , 1 , … , N − 1 } 2 × { 0 , 1 , … , N − 1 } (151) [\mu,(\kappa_1,\kappa_2),\nu]\in\{0,1\}\times\{0,1,\ldots,N-1\}^{2}\times\{0,1,\ldots,N-1\} \tag*{(151)} [ μ , ( κ 1 , κ 2 ) , ν ] ∈ { 0 , 1 } × { 0 , 1 , … , N − 1 } 2 × { 0 , 1 , … , N − 1 } (151)
that encode their boundary projective representation and obey the stacking rules
[ μ , ( κ 1 , κ 2 ) , ν ] ∧ [ μ ~ , ( κ ~ 1 , κ ~ 2 ) , ν ~ ] = [ μ + μ ~ , κ + ( − 1 ) μ κ ~ , ν + ν ~ − ( − 1 ) μ ( κ 1 κ ~ 2 − κ 2 κ ~ 1 ) ] . (152) [\mu,(\kappa_1,\kappa_2),\nu]\wedge[\widetilde{\mu},(\widetilde{\kappa}_1,\widetilde{\kappa}_2),\widetilde{\nu}]
=
[\mu+\widetilde{\mu},\kappa+(-1)^\mu\widetilde{\kappa},\nu+\widetilde{\nu}-(-1)^\mu(\kappa_1\widetilde{\kappa}_2-\kappa_2\widetilde{\kappa}_1)] . \tag*{(152)} [ μ , ( κ 1 , κ 2 ) , ν ] ∧ [ μ , ( κ 1 , κ 2 ) , ν ] = [ μ + μ , κ + ( − 1 ) μ κ , ν + ν − ( − 1 ) μ ( κ 1 κ 2 − κ 2 κ 1 )] . (152)
Remarkably, these stacking rules are the composition rules of the semi-direct product
Heis ( Z N ) ⋊ Z 2 , (153) \operatorname{Heis}(\mathbb{Z}_N)\rtimes\mathbb{Z}_2, \tag*{(153)} Heis ( Z N ) ⋊ Z 2 , (153)
where Heis ( Z N ) \operatorname{Heis}(\mathbb{Z}_N) Heis ( Z N ) is the Heisenberg group (i.e., the group of 3-by-3 unitriangular matrices) over the integer modulo ring Z N \mathbb{Z}_N Z N ,
Heis ( Z N ) = { ( 1 a z 0 1 b 0 0 1 ) ∣ a , b , z ∈ Z N } . (154) \operatorname{Heis}(\mathbb{Z}_N)=\left\{\left.
\begin{pmatrix}
1&a&z\\
0&1&b\\
0&0&1
\end{pmatrix}
\right|a,b,z\in\mathbb{Z}_N\right\}. \tag*{(154)} Heis ( Z N ) = ⎩ ⎨ ⎧ 1 0 0 a 1 0 z b 1 a , b , z ∈ Z N ⎭ ⎬ ⎫ . (154)
The group (153) is non-Abelian both because of the semi-direct product and because Heis ( Z N ) \operatorname{Heis}(\mathbb{Z}_N) Heis ( Z N ) has nontrivial commutators.
We first construct the parafermionic Hilbert space as is depicted in Figure 2 . We consider a chain made of 3 L 3L 3 L sites.
Figure 2. The 3 L 3L 3 L sites of a chain are depicted as ellipses colored in orange, green, and purple. Each ellipse contains two dots. The left (right) dot of an orange ellipse represents the parafermion ξ 3 j + 1 \xi_{3j+1} ξ 3 j + 1 (η 3 j + 1 \eta_{3j+1} η 3 j + 1 ). The left (right) dot of a green ellipse represents the parafermion ξ 3 j + 2 \xi_{3j+2} ξ 3 j + 2 (η 3 j + 2 \eta_{3j+2} η 3 j + 2 ). The left (right) dot of a purple ellipse represents the parafermion ξ 3 j + 3 \xi_{3j+3} ξ 3 j + 3 (η 3 j + 3 \eta_{3j+3} η 3 j + 3 ). This rule fixes the ordering of the parafermions and the Hilbert space.
These are represented by ellipses alternatingly colored in orange, green, and purple. Each site carries a pair of order-N N N parafermions ( ξ , η ) (\xi,\eta) ( ξ , η ) , which are denoted by a left (ξ \xi ξ ) and right (η \eta η ) dot within each ellipse. Orange, green, and purple ellipses are labeled 3 j + 1 3j+1 3 j + 1 , 3 j + 2 3j+2 3 j + 2 , and 3 j + 3 3j+3 3 j + 3 , respectively. The ordered set of parafermions for the chain is then
{ ξ 3 j + 1 , η 3 j + 1 , ξ 3 j + 2 , η 3 j + 2 , ξ 3 j + 3 , η 3 j + 3 } j = 0 L − 1 . (155) \{\xi_{3j+1},\eta_{3j+1},\xi_{3j+2},\eta_{3j+2},\xi_{3j+3},\eta_{3j+3}\}_{j=0}^{L-1}. \tag*{(155)} { ξ 3 j + 1 , η 3 j + 1 , ξ 3 j + 2 , η 3 j + 2 , ξ 3 j + 3 , η 3 j + 3 } j = 0 L − 1 . (155)
Define the four Hamiltonians
H [ 0 , ( 0 , 0 ) , 0 ] ( b ) : = − λ ∑ j = 0 L − 1 ( ξ 3 j + 1 η 3 j + 1 † + ξ 3 j + 2 η 3 j + 2 † + ξ 3 j + 3 η 3 j + 3 † + H . c . ) , (156) H^{(\mathrm b)}_{[0,(0,0),0]}:=
-\lambda\sum_{j=0}^{L-1}\left(\xi_{3j+1}\eta^\dagger_{3j+1}+\xi_{3j+2}\eta^\dagger_{3j+2}+\xi_{3j+3}\eta^\dagger_{3j+3}+\mathrm{H.c.}\right), \tag*{(156)} H [ 0 , ( 0 , 0 ) , 0 ] ( b ) := − λ j = 0 ∑ L − 1 ( ξ 3 j + 1 η 3 j + 1 † + ξ 3 j + 2 η 3 j + 2 † + ξ 3 j + 3 η 3 j + 3 † + H.c. ) , (156)
H [ 1 , ( 0 , 0 ) , 0 ] ( b ) : = − λ ∑ j = 0 L − 1 ( ξ 3 j + 1 η 3 j + 1 † + ξ 3 j + 2 η 3 j + 2 † + η 3 j + 3 ξ 3 ( j + 1 ) + 3 † + H . c . ) , (157) H^{(\mathrm b)}_{[1,(0,0),0]}:=
-\lambda\sum_{j=0}^{L-1}\left(\xi_{3j+1}\eta^\dagger_{3j+1}+\xi_{3j+2}\eta^\dagger_{3j+2}+\eta_{3j+3}\xi^\dagger_{3(j+1)+3}+\mathrm{H.c.}\right), \tag*{(157)} H [ 1 , ( 0 , 0 ) , 0 ] ( b ) := − λ j = 0 ∑ L − 1 ( ξ 3 j + 1 η 3 j + 1 † + ξ 3 j + 2 η 3 j + 2 † + η 3 j + 3 ξ 3 ( j + 1 ) + 3 † + H.c. ) , (157)
H [ 1 , ( 0 , 1 ) , 0 ] ( b ) : = − λ ∑ j = 0 L − 1 ( ξ 3 j + 1 η 3 j + 1 † + η 3 j + 2 ξ 3 ( j + 1 ) + 2 † + ξ 3 j + 3 η 3 j + 3 † + H . c . ) , (158) H^{(\mathrm b)}_{[1,(0,1),0]}:=
-\lambda\sum_{j=0}^{L-1}\left(\xi_{3j+1}\eta^\dagger_{3j+1}+\eta_{3j+2}\xi^\dagger_{3(j+1)+2}+\xi_{3j+3}\eta^\dagger_{3j+3}+\mathrm{H.c.}\right), \tag*{(158)} H [ 1 , ( 0 , 1 ) , 0 ] ( b ) := − λ j = 0 ∑ L − 1 ( ξ 3 j + 1 η 3 j + 1 † + η 3 j + 2 ξ 3 ( j + 1 ) + 2 † + ξ 3 j + 3 η 3 j + 3 † + H.c. ) , (158)
H [ 1 , ( 1 , 0 ) , 0 ] ( b ) : = − λ ∑ j = 0 L − 1 ( η 3 j + 1 ξ 3 ( j + 1 ) + 1 † + ξ 3 j + 2 η 3 j + 2 † + ξ 3 j + 3 η 3 j + 3 † + H . c . ) . (159) H^{(\mathrm b)}_{[1,(1,0),0]}:=
-\lambda\sum_{j=0}^{L-1}\left(\eta_{3j+1}\xi^\dagger_{3(j+1)+1}+\xi_{3j+2}\eta^\dagger_{3j+2}+\xi_{3j+3}\eta^\dagger_{3j+3}+\mathrm{H.c.}\right). \tag*{(159)} H [ 1 , ( 1 , 0 ) , 0 ] ( b ) := − λ j = 0 ∑ L − 1 ( η 3 j + 1 ξ 3 ( j + 1 ) + 1 † + ξ 3 j + 2 η 3 j + 2 † + ξ 3 j + 3 η 3 j + 3 † + H.c. ) . (159)
The superscript b = 0 , 1 \mathrm b=0,1 b = 0 , 1 implements periodic and open boundary conditions, respectively. The subscript denotes the classifying indices associated to the corresponding IPT phases, as we will be shown explicitly. The coupling λ > 0 \lambda>0 λ > 0 has units of energy. The summands are grouped according to flavor. These four Hamiltonians are depicted schematically in Figure 3 .
Figure 3. The family of Hamiltonians (5.6) are quadratic forms in the parafermionic operators. Sites of the chain (ellipses) are assigned pairs of parafermionic operators (dots) as explained in Figure 2 . Two dots connected by a bond represent the product of the parafermion operators at its endpoints, with the operator at the left endpoint multiplying the operator at the right endpoint from the left.
Each Hamiltonian is a sum of pairwise commuting terms and has a non-degenerate ground state under periodic boundary conditions (b = 0 b = 0 b = 0 ). Changing the boundary condition leaves H [ 0 , ( 0 , 0 ) , 0 ] ( b ) H^{(b)}_{[0,(0,0),0]} H [ 0 , ( 0 , 0 ) , 0 ] ( b ) unchanged. For H [ 1 , ( 0 , 0 ) , 0 ] ( b ) H^{(b)}_{[1,(0,0),0]} H [ 1 , ( 0 , 0 ) , 0 ] ( b ) , H [ 1 , ( 0 , 1 ) , 0 ] ( b ) H^{(b)}_{[1,(0,1),0]} H [ 1 , ( 0 , 1 ) , 0 ] ( b ) , and H [ 1 , ( 1 , 0 ) , 0 ] ( b ) H^{(b)}_{[1,(1,0),0]} H [ 1 , ( 1 , 0 ) , 0 ] ( b ) , choosing open boundary conditions b = 1 b = 1 b = 1 leads to the respective terms η 3 L ξ 1 † \eta_{3L}\xi_{1}^{\dagger} η 3 L ξ 1 † , η 3 L − 1 ξ 2 † \eta_{3L-1}\xi_{2}^{\dagger} η 3 L − 1 ξ 2 † , and η 3 L − 2 ξ 1 † \eta_{3L-2}\xi_{1}^{\dagger} η 3 L − 2 ξ 1 † dropping out of the sums. These imply an N N N -fold ground-state degeneracy under open boundary conditions. We construct from the local charges Q j Q_j Q j the operators
U k : = ∏ j = 0 3 L − 1 Q 3 j + k , k = 1 , 2 , 3. (160) U_k := \prod_{j=0}^{3L-1} Q_{3j+k}, \qquad k = 1,2,3. \tag*{(160)} U k := j = 0 ∏ 3 L − 1 Q 3 j + k , k = 1 , 2 , 3. (160)
These are of order N N N and commute pairwise. One may think of these as the global charge restricted to each parafermion flavor. They commute with the above Hamiltonians and we fix the protecting symmetry group to be
G = ⟨ Q , U 1 , U 2 ⟩ ≅ Z N p f × ( Z N × Z N ) , (161) \mathcal{G} = \langle Q,U_1,U_2\rangle\cong\mathbb{Z}_N^{\mathrm{pf}} \times(\mathbb{Z}_N \times\mathbb{Z}_N), \tag*{(161)} G = ⟨ Q , U 1 , U 2 ⟩ ≅ Z N pf × ( Z N × Z N ) , (161)
as announced at the beginning of this section. Thus, the Hamiltonians in Eq. (5.6) are indeed representatives of G \mathcal{G} G -symmetry protected topological phases.
We now compute the classifying indices of the Hamiltonians and show that they are correctly labeled. Since the ground state of H [ 0 , ( 0 , 0 ) , 0 ] ( b ) H^{(b)}_{[0,(0,0),0]} H [ 0 , ( 0 , 0 ) , 0 ] ( b ) remains non-degenerate under open boundary conditions b = 1 b = 1 b = 1 , it is clear that it corresponds to the trivial phase classified by the indices [ 0 , ( 0 , 0 ) , 0 ] [0,(0,0),0] [ 0 , ( 0 , 0 ) , 0 ] . For the other Hamiltonians, the single summand dropping out when choosing open boundary conditions causes the boundary algebra to be generated by an odd number of parafermions. This implies that the first classifying index is μ = 1 \mu= 1 μ = 1 for the three Hamiltonians (5.6b)–(5.6d). Using an auxiliary right-infinite parafermion γ ∞ \gamma_{\infty} γ ∞ , the boundary parafermion algebras for the respective Hamiltonians are
P [ 1 , ( 0 , 0 ) , 0 ] = ⟨ ξ 1 , η 1 , η 2 , ξ 2 , ξ 3 , γ ∞ ⟩ , (162) \mathcal{P}_{[1,(0,0),0]} = \langle\xi_1,\eta_1,\eta_2,\xi_2,\xi_3,\gamma_{\infty}\rangle, \tag*{(162)} P [ 1 , ( 0 , 0 ) , 0 ] = ⟨ ξ 1 , η 1 , η 2 , ξ 2 , ξ 3 , γ ∞ ⟩ , (162)
P [ 1 , ( 0 , 1 ) , 0 ] = ⟨ ξ 1 , η 1 , ξ 2 , γ ∞ ⟩ , (163) \mathcal{P}_{[1,(0,1),0]} = \langle\xi_1,\eta_1,\xi_2,\gamma_{\infty}\rangle, \tag*{(163)} P [ 1 , ( 0 , 1 ) , 0 ] = ⟨ ξ 1 , η 1 , ξ 2 , γ ∞ ⟩ , (163)
P [ 1 , ( 1 , 0 ) , 0 ] = ⟨ ξ 1 , ξ 2 , η 2 , γ ∞ ⟩ , (164) \mathcal{P}_{[1,(1,0),0]} = \langle\xi_1,\xi_2,\eta_2,\gamma_{\infty}\rangle, \tag*{(164)} P [ 1 , ( 1 , 0 ) , 0 ] = ⟨ ξ 1 , ξ 2 , η 2 , γ ∞ ⟩ , (164)
where we underlined the parafermion stemming from the dropped summation terms. The boundary algebra P [ 1 , ( 1 , 0 ) , 0 ] \mathcal{P}_{[1,(1,0),0]} P [ 1 , ( 1 , 0 ) , 0 ] was expanded with the neighboring parafermion pair ξ 2 , η 2 \xi_2,\eta_2 ξ 2 , η 2 to avoid the degenerate case of vanishing boundary parafermion charge. We now consider the action of the global symmetries Q Q Q , U 1 U_1 U 1 , and U 2 U_2 U 2 on these boundary degrees of freedom. For the Hamiltonian H [ 1 , ( 0 , 0 ) , 0 ] ( b ) H^{(b)}_{[1,(0,0),0]} H [ 1 , ( 0 , 0 ) , 0 ] ( b ) , they produce for each boundary parafermion (excluding the auxiliary γ ∞ \gamma_{\infty} γ ∞ ) the following phases,
Q : ( ω N , ω N , ω N , ω N , ω N ) , (165) Q : (\omega_N,\omega_N,\omega_N,\omega_N,\omega_N), \tag*{(165)} Q : ( ω N , ω N , ω N , ω N , ω N ) , (165)
U 1 : ( ω N , ω N , 1 , 1 , 1 ) , (166) U_1 : (\omega_N,\omega_N,1,1,1), \tag*{(166)} U 1 : ( ω N , ω N , 1 , 1 , 1 ) , (166)
U 2 : ( 1 , 1 , ω N , ω N , 1 ) . (167) U_2 : (1,1,\omega_N,\omega_N,1). \tag*{(167)} U 2 : ( 1 , 1 , ω N , ω N , 1 ) . (167)
Analogously, for the Hamiltonian H [ 1 , ( 0 , 1 ) , 0 ] ( b ) H^{(b)}_{[1,(0,1),0]} H [ 1 , ( 0 , 1 ) , 0 ] ( b )
Q : ( ω N , ω N , ω N ) , (168) Q : (\omega_N,\omega_N,\omega_N), \tag*{(168)} Q : ( ω N , ω N , ω N ) , (168)
U 1 : ( ω N , ω N , 1 ) , (169) U_1 : (\omega_N,\omega_N,1), \tag*{(169)} U 1 : ( ω N , ω N , 1 ) , (169)
U 2 : ( 1 , 1 , ω N ) . (170) U_2 : (1,1,\omega_N). \tag*{(170)} U 2 : ( 1 , 1 , ω N ) . (170)
and the Hamiltonian H [ 1 , ( 1 , 0 ) , 0 ] ( b ) H^{(b)}_{[1,(1,0),0]} H [ 1 , ( 1 , 0 ) , 0 ] ( b )
Q : ( ω N , ω N , ω N ) , (171) Q : (\omega_N,\omega_N,\omega_N), \tag*{(171)} Q : ( ω N , ω N , ω N ) , (171)
U 1 : ( ω N , 1 , 1 ) , (172) U_1 : (\omega_N,1,1), \tag*{(172)} U 1 : ( ω N , 1 , 1 ) , (172)
U 2 : ( 1 , ω N , ω N ) . (173) U_2 : (1,\omega_N,\omega_N). \tag*{(173)} U 2 : ( 1 , ω N , ω N ) . (173)
We must now find boundary projections Q ( b ) Q^{(b)} Q ( b ) , U 1 ( b ) U_1^{(b)} U 1 ( b ) , and U 2 ( b ) U_2^{(b)} U 2 ( b ) in terms of the boundary parafermions whose action on the degrees of freedom match the global ones recorded above. One may verify that such representations are given: for H [ 1 , ( 0 , 0 ) , 0 ] ( b ) H^{(b)}_{[1,(0,0),0]} H [ 1 , ( 0 , 0 ) , 0 ] ( b ) by
Q ( b ) ≡ Q 1 Q 2 ( ω N N + 1 2 ξ 3 γ ∞ † ) , (174) Q^{(b)} \equiv Q_1Q_2\left(\omega_N^{\frac{N+1}{2}}\xi_3\gamma_{\infty}^{\dagger}\right), \tag*{(174)} Q ( b ) ≡ Q 1 Q 2 ( ω N 2 N + 1 ξ 3 γ ∞ † ) , (174)
U 1 ( b ) ≡ Q 1 , (175) U_1^{(b)} \equiv Q_1, \tag*{(175)} U 1 ( b ) ≡ Q 1 , (175)
U 2 ( b ) ≡ Q 2 . (176) U_2^{(b)} \equiv Q_2. \tag*{(176)} U 2 ( b ) ≡ Q 2 . (176)
for H [ 1 , ( 0 , 1 ) , 0 ] ( b ) H^{(b)}_{[1,(0,1),0]} H [ 1 , ( 0 , 1 ) , 0 ] ( b ) by
Q ( b ) ≡ Q 1 ( ω N N + 1 2 ξ 2 γ ∞ † ) , (177) Q^{(b)} \equiv Q_1\left(\omega_N^{\frac{N+1}{2}}\xi_2\gamma_{\infty}^{\dagger}\right), \tag*{(177)} Q ( b ) ≡ Q 1 ( ω N 2 N + 1 ξ 2 γ ∞ † ) , (177)
U 1 ( b ) ≡ Q 1 , (178) U_1^{(b)} \equiv Q_1, \tag*{(178)} U 1 ( b ) ≡ Q 1 , (178)
U 2 ( b ) ≡ ( ω N N + 1 2 ξ 2 γ ∞ † ) , (179) U_2^{(b)} \equiv\left(\omega_N^{\frac{N+1}{2}}\xi_2\gamma_{\infty}^{\dagger}\right), \tag*{(179)} U 2 ( b ) ≡ ( ω N 2 N + 1 ξ 2 γ ∞ † ) , (179)
and for H [ 1 , ( 1 , 0 ) , 0 ] ( b ) H^{(b)}_{[1,(1,0),0]} H [ 1 , ( 1 , 0 ) , 0 ] ( b ) by
Q ( b ) ≡ ( ω N N + 1 2 ξ 1 ξ 2 † ) ( ω N N + 1 2 η 2 γ ∞ † ) , (180) Q^{(b)} \equiv\left(\omega_N^{\frac{N+1}{2}}\xi_1\xi_2^{\dagger}\right)\left(\omega_N^{\frac{N+1}{2}}\eta_2\gamma_{\infty}^{\dagger}\right), \tag*{(180)} Q ( b ) ≡ ( ω N 2 N + 1 ξ 1 ξ 2 † ) ( ω N 2 N + 1 η 2 γ ∞ † ) , (180)
U 1 ( b ) ≡ ω N N + 1 2 ξ 1 † γ ∞ , (181) U_1^{(b)} \equiv\omega_N^{\frac{N+1}{2}}\xi_1^{\dagger}\gamma_{\infty}, \tag*{(181)} U 1 ( b ) ≡ ω N 2 N + 1 ξ 1 † γ ∞ , (181)
U 2 ( b ) ≡ Q 2 . (182) U_2^{(b)} \equiv Q_2. \tag*{(182)} U 2 ( b ) ≡ Q 2 . (182)
Moreover, all these operators commute with their respective Hamiltonians.
We now use these boundary representations to compute the classifying indices. As mentioned, the first index is always μ = 1 \mu= 1 μ = 1 as there is an odd number of boundary parafermions. The second index κ = ( κ 1 , κ 2 ) \kappa= (\kappa_1,\kappa_2) κ = ( κ 1 , κ 2 ) is an element of
H 1 [ Z N × Z N , Z N ] ≅ Z N × Z N . (183) H^1[\mathbb{Z}_N \times\mathbb{Z}_N,\mathbb{Z}_N] \cong\mathbb{Z}_N \times\mathbb{Z}_N. \tag*{(183)} H 1 [ Z N × Z N , Z N ] ≅ Z N × Z N . (183)
It corresponds to the order of the Hermitian conjugated auxiliary parafermion γ ∞ † \gamma_{\infty}^{\dagger} γ ∞ † in the boundary representations U 1 ( b ) ( κ 1 ) U_1^{(b)}(\kappa_1) U 1 ( b ) ( κ 1 ) and U 2 ( b ) ( κ 2 ) U_2^{(b)}(\kappa_2) U 2 ( b ) ( κ 2 ) . From definitions (5.13), (5.14), and (5.15) one deduces
( κ 1 , κ 2 ) = ( 0 , 0 ) , ( 0 , 1 ) , ( 1 , 0 ) , (184) (\kappa_1,\kappa_2) = (0,0),\ (0,1),\ (1,0), \tag*{(184)} ( κ 1 , κ 2 ) = ( 0 , 0 ) , ( 0 , 1 ) , ( 1 , 0 ) , (184)
respectively. The third index ν \nu ν is an element of
H 2 [ Z N × Z N , Z N ] ≅ Z N . (185) H^2[\mathbb{Z}_N \times\mathbb{Z}_N, \mathbb{Z}_N] \cong\mathbb{Z}_N. \tag*{(185)} H 2 [ Z N × Z N , Z N ] ≅ Z N . (185)
It measures the projective equivalency class of the boundary representation of the internal symmetry Z N × Z N \mathbb{Z}_N \times\mathbb{Z}_N Z N × Z N , which may be computed by considering the commutation relation between U 1 ( b ) U_1^{(b)} U 1 ( b ) and U 2 ( b ) U_2^{(b)} U 2 ( b ) . In fact, in the three cases above they always commute, whence ν = 0 \nu=0 ν = 0 . We have thus verified that the classifying indices match the subscripts of Hamiltonians (5.6).
Recall that N N N is odd. To find the explicit stacking rule (152) from Eq. (148) , one must compute the group cohomologies and write out the cup product. The latter follows from the fact that the non-degenerate alternating bilinear form over two-dimensional spaces is unique up to scalar factors, which may be viewed as the determinant. Since the cup product of two 1-cocycles is anti-symmetric, we can conclude
κ ∪ κ ~ = κ 1 κ ~ 2 − κ 2 κ ~ 1 . (186) \kappa\cup\widetilde{\kappa}=\kappa_1\widetilde{\kappa}_2-\kappa_2\widetilde{\kappa}_1. \tag*{(186)} κ ∪ κ = κ 1 κ 2 − κ 2 κ 1 . (186)
We label the three non-trivial IPT phases by
S : = [ 1 , ( 0 , 0 ) , 0 ] , (187) S := [1,(0,0),0], \tag*{(187)} S := [ 1 , ( 0 , 0 ) , 0 ] , (187)
X ~ : = [ 1 , ( 1 , 0 ) , 0 ] , (188) \widetilde{X} := [1,(1,0),0], \tag*{(188)} X := [ 1 , ( 1 , 0 ) , 0 ] , (188)
Y ~ : = [ 1 , ( 0 , 1 ) , 0 ] . (189) \widetilde{Y} := [1,(0,1),0]. \tag*{(189)} Y := [ 1 , ( 0 , 1 ) , 0 ] . (189)
Using our stacking rules, we see that the phase S S S may be used to flip the first classifying index, giving us the SPT phases, and we define
X : = [ 0 , ( 1 , 0 ) , 0 ] = X ~ ∧ S , (190) X := [0,(1,0),0]=\widetilde{X}\wedge S, \tag*{(190)} X := [ 0 , ( 1 , 0 ) , 0 ] = X ∧ S , (190)
Y : = [ 0 , ( 0 , 1 ) , 0 ] = Y ~ ∧ S . (191) Y := [0,(0,1),0]=\widetilde{Y}\wedge S. \tag*{(191)} Y := [ 0 , ( 0 , 1 ) , 0 ] = Y ∧ S . (191)
In particular, applying the stacking-group commutator yields the phase
Z : = [ X , Y ] φ ( N ) = [ 0 , ( 0 , 0 ) , − 2 ] φ ( N ) = [ 0 , ( 0 , 0 ) , 1 ] , (192) Z := [X,Y]^{\varphi(N)}=[0,(0,0),-2]^{\varphi(N)}=[0,(0,0),1], \tag*{(192)} Z := [ X , Y ] φ ( N ) = [ 0 , ( 0 , 0 ) , − 2 ] φ ( N ) = [ 0 , ( 0 , 0 ) , 1 ] , (192)
where φ \varphi φ is Euler’s totient function and we used the fact that N N N is odd. It is thus clear that the phases X ~ , Y ~ \widetilde{X},\widetilde{Y} X , Y in combinations with the phase S S S generate all other G ≅ Z N p f × ( Z N × Z N ) \mathcal{G}\cong\mathbb{Z}_N^{\mathrm{pf}}\times(\mathbb{Z}_N\times\mathbb{Z}_N) G ≅ Z N pf × ( Z N × Z N ) -protected topological phases. Moreover, the isomorphism of the SPT phases (the μ = 0 \mu=0 μ = 0 stacking subgroup) to the Heisenberg group Heis ( Z N ) \operatorname{Heis}(\mathbb{Z}_N) Heis ( Z N ) is given explicitly by the mapping
[ 0 , ( κ 1 , κ 2 ) , ν ] ⟼ ( 1 κ 1 κ 1 κ 2 − ν 2 0 1 κ 2 0 0 1 ) , (193) [0,(\kappa_1,\kappa_2),\nu]\longmapsto
\begin{pmatrix}
1 & \kappa_1 & \frac{\kappa_1\kappa_2-\nu}{2}\\
0 & 1 & \kappa_2\\
0 & 0 & 1
\end{pmatrix}, \tag*{(193)} [ 0 , ( κ 1 , κ 2 ) , ν ] ⟼ 1 0 0 κ 1 1 0 2 κ 1 κ 2 − ν κ 2 1 , (193)
which is well-defined due to N N N being odd. Finally, stacking the SPT phases from the right with S S S leads to the semi-direct product in the total stacking group,
Heis ( Z N ) ⋊ Z 2 . (194) \operatorname{Heis}(\mathbb{Z}_N)\rtimes\mathbb{Z}_2. \tag*{(194)} Heis ( Z N ) ⋊ Z 2 . (194)
Conclusion
In this work, we have studied the invertible parafermionic topological (IPT) phases in one-dimensional space. Building on Ref. [68 ] , we introduced a triplet of indices ( μ , [ ( κ , ν ) ] ) (\mu,[(\kappa,\nu)]) ( μ , [( κ , ν )]) that characterizes projective representations of global symmetry G \mathcal{G} G acting on order-N N N parafermions. These indices in turn classify distinct one-dimensional IPT phases of matter with symmetry G \mathcal{G} G . While we have only considered unitary symmetries in this work, our results can be extended to anti-unitary symmetries such as time-reversal, see for example Refs. [66 , 68 ] .
We have then derived the so-called stacking rules that prescribes how two IPT phases can be merged to produce a third one. Stacking rules can be thought of as a composition rule that gives the IPT phases a group structure. We have shown that as opposed the invertible topological phases of bosons and fermions, IPT phases can form a non-Abelian group under stacking.
In Sec. Sec. V we explicitly constructed Hamiltonians that realize IPT phases of order odd prime power N N N with Z N × Z N × Z N p f \mathbb{Z}_N\times\mathbb{Z}_N\times\mathbb{Z}_N^{\mathrm{pf}} Z N × Z N × Z N pf symmetry. Therein, we showed that under stacking the IPT phases form the non-Abelian group Heis ( Z N ) ⋊ Z 2 \operatorname{Heis}(\mathbb{Z}_N)\rtimes\mathbb{Z}_2 Heis ( Z N ) ⋊ Z 2 of order 2 N 3 2N^3 2 N 3 , where Heis ( Z N ) \operatorname{Heis}(\mathbb{Z}_N) Heis ( Z N ) is the Heisenberg group of order N 3 N^3 N 3 .
In Sec. § 1.2 D, we discussed the classification of IPT phases for general integer N N N . Given the prime decomposition of N N N , we have shown that the degrees of freedom split into flavors, one for each prime factor. When the global symmetries factorize on each flavor as in Eq. (83) , the classification simplifies and is given in Eq. (3.44). A natural question to ask would be if the classification in terms of indices (3.44) also holds in the more general case when Eq. (83) does not hold, which we leave for future work.
One can relate the IPT phases of order-N N N parafermions to bosonic gapped phases realized by Z N \mathbb{Z}_N Z N clock and shift variables [35 , 36 , 43 ] . While, parafermionic SPT phases are generally mapped to bosonic SPT phases, the IPT phases that are not SPTs are mapped to spontaneous symmetry breaking phases. It would be interesting to match the stacking rules obeyed by IPT phases with the maps between bosonic gapped phases.
Acknowledgments
We thank Christopher Mudry and Claudio Chamon for discussions and collaboration in the early stages of this work. We thank Christopher Mudry for comments on the manuscript and and for identifying an issue in Sec. § 1.2 D. A.L.G.M. acknowledges support from the Swiss Study Foundation and the Swiss State Secretariat for Education, Research and Innovation. Ö.M.A. was supported in parts by SNSF Grant No. 214429, by NSF Grant No. DMR2022428, and by the Simons Collaboration on Ultra-Quantum Matter, which is a grant from the Simons Foundation (651444, SHS).
Group cohomology
Given two groups G G G and M M M , an n n n -cochain is the map
ϕ : G n → M , ( g 1 , g 2 , ⋯ , g n ) ↦ ϕ ( g 1 , g 2 , ⋯ , g n ) , (195) \phi: G^n \to M,
\qquad
(g_1,g_2,\cdots,g_n) \mapsto\phi(g_1,g_2,\cdots,g_n), \tag*{(195)} ϕ : G n → M , ( g 1 , g 2 , ⋯ , g n ) ↦ ϕ ( g 1 , g 2 , ⋯ , g n ) , (195)
that maps an n n n -tuple ( g 1 , g 2 , ⋯ , g n ) (g_1,g_2,\cdots,g_n) ( g 1 , g 2 , ⋯ , g n ) to an element ϕ ( g 1 , g 2 , ⋯ , g n ) ∈ M \phi(g_1,g_2,\cdots,g_n) \in M ϕ ( g 1 , g 2 , ⋯ , g n ) ∈ M . The set of all n n n -cochains from G n G^n G n to M M M is denoted by C n ( G , M ) C^n(G,M) C n ( G , M ) . We define an M M M -valued 0-cochain to be an element of the group M M M itself, i.e., C 0 ( G , M ) = M C^0(G,M)=M C 0 ( G , M ) = M . Henceforth, we will denote the group composition rule in G G G by ⋅ \cdot ⋅ and the group composition rule in M M M additively by + + + (− - − denoting the inverse element). For our purposes, the Abelian group M M M will be either U ( 1 ) U(1) U ( 1 ) or the cyclic group Z N \mathbb{Z}_N Z N .
In the latter case, we allow a non-trivial group action of G G G on M = Z N M=\mathbb{Z}_N M = Z N . Given a group homomorphism ρ g : G → Z N \rho_g:G\to\mathbb{Z}_N ρ g : G → Z N , we define the group action
ρ g : Z N → Z N , m ↦ ρ g m , (196) \rho_g:\mathbb{Z}_N\to\mathbb{Z}_N,
\qquad
m\mapsto\rho_g m, \tag*{(196)} ρ g : Z N → Z N , m ↦ ρ g m , (196)
which is just multiplication by integers modulo N N N . In our setting, this group action will always be involutive, i.e., ρ g 2 = 1 m o d N \rho_g^2=1\ \mathrm{mod}\ N ρ g 2 = 1 mod N . The map ρ \rho ρ indicates whether an element g ∈ G g\in G g ∈ G implements a (special) charge conjugation or not. Note that since ρ g \rho_g ρ g is a group homomorphism one has ρ g ⋅ h = ρ g ρ h = ρ h ρ g \rho_{g\cdot h}=\rho_g\rho_h=\rho_h\rho_g ρ g ⋅ h = ρ g ρ h = ρ h ρ g . When M = U ( 1 ) M=U(1) M = U ( 1 ) , we view G G G as acting with the trivial group action, which corresponds to our restriction to unitary representations of symmetries.
We define the map δ ρ n \delta_\rho^n δ ρ n
δ ρ n : C n ( G , M ) → C n + 1 ( G , M ) , ϕ ↦ ( δ ρ n ϕ ) , (197) \delta_\rho^n:C^n(G,M)\to C^{n+1}(G,M),
\qquad
\phi\mapsto(\delta_\rho^n\phi), \tag*{(197)} δ ρ n : C n ( G , M ) → C n + 1 ( G , M ) , ϕ ↦ ( δ ρ n ϕ ) , (197)
from n n n -cochains to ( n + 1 ) (n+1) ( n + 1 ) -cochains such that
( δ ρ n ϕ ) ( g 1 , ⋯ , g n + 1 ) : = ϕ ( g 2 , ⋯ , g n , g n + 1 ) + ∑ i = 1 n ( − 1 ) i ϕ ( g 1 , ⋯ , g i ⋅ g i + 1 , ⋯ , g n + 1 ) − ( − 1 ) n ρ g n + 1 ϕ ( g 1 , ⋯ , g n ) . (198) (\delta_\rho^n\phi)(g_1,\cdots,g_{n+1}) := \phi(g_2,\cdots,g_n,g_{n+1})+\sum_{i=1}^{n}(-1)^i\phi(g_1,\cdots,g_i\cdot g_{i+1},\cdots,g_{n+1})-(-1)^n\rho_{g_{n+1}}\phi(g_1,\cdots,g_n). \tag*{(198)} ( δ ρ n ϕ ) ( g 1 , ⋯ , g n + 1 ) := ϕ ( g 2 , ⋯ , g n , g n + 1 ) + i = 1 ∑ n ( − 1 ) i ϕ ( g 1 , ⋯ , g i ⋅ g i + 1 , ⋯ , g n + 1 ) − ( − 1 ) n ρ g n + 1 ϕ ( g 1 , ⋯ , g n ) . (198)
The map δ ρ n \delta_\rho^n δ ρ n is called a coboundary operator.
Example n = 2 n=2 n = 2 : The coboundary operator δ ρ 2 \delta_\rho^2 δ ρ 2 is defined by
( δ ρ 2 ϕ ) ( g 1 , g 2 , g 3 ) = ϕ ( g 2 , g 3 ) + ( − 1 ) 1 ϕ ( g 1 ⋅ g 2 , g 3 ) + ( − 1 ) 2 ϕ ( g 1 , g 2 ⋅ g 3 ) − ( − 1 ) 2 ρ g 3 ϕ ( g 1 , g 2 ) = ϕ ( g 2 , g 3 ) − ϕ ( g 1 ⋅ g 2 , g 3 ) + ϕ ( g 1 , g 2 ⋅ g 3 ) − ρ g 3 ϕ ( g 1 , g 2 ) . (199) \begin{aligned}
(\delta_\rho^2\phi)(g_1,g_2,g_3)=\phi(g_2,g_3)+(-1)^1\phi(g_1\cdot g_2,g_3)+(-1)^2\phi(g_1,g_2\cdot g_3)-(-1)^2\rho_{g_3}\phi(g_1,g_2) \\
&=\phi(g_2,g_3)-\phi(g_1\cdot g_2,g_3)+\phi(g_1,g_2\cdot g_3)-\rho_{g_3}\phi(g_1,g_2). \tag*{(199)}
\end{aligned} ( δ ρ 2 ϕ ) ( g 1 , g 2 , g 3 ) = ϕ ( g 2 , g 3 ) + ( − 1 ) 1 ϕ ( g 1 ⋅ g 2 , g 3 ) + ( − 1 ) 2 ϕ ( g 1 , g 2 ⋅ g 3 ) − ( − 1 ) 2 ρ g 3 ϕ ( g 1 , g 2 ) = ϕ ( g 2 , g 3 ) − ϕ ( g 1 ⋅ g 2 , g 3 ) + ϕ ( g 1 , g 2 ⋅ g 3 ) − ρ g 3 ϕ ( g 1 , g 2 ) . (199)
Example n = 1 n=1 n = 1 : The coboundary operator δ ρ 1 \delta_\rho^1 δ ρ 1 is defined by
( δ ρ 1 ϕ ) ( g 1 , g 2 ) = ϕ ( g 2 ) − ϕ ( g 1 ⋅ g 2 ) + ρ g 2 ϕ ( g 1 ) . (200) (\delta_\rho^1\phi)(g_1,g_2)=\phi(g_2)-\phi(g_1\cdot g_2)+\rho_{g_2}\phi(g_1). \tag*{(200)} ( δ ρ 1 ϕ ) ( g 1 , g 2 ) = ϕ ( g 2 ) − ϕ ( g 1 ⋅ g 2 ) + ρ g 2 ϕ ( g 1 ) . (200)
One verifies the important identity
Φ ( g 1 , g 2 ) : = ( δ ρ 1 ϕ ) ( g 1 , g 2 ) ⟹ ( δ ρ 2 Φ ) ( g 1 , g 2 , g 3 ) = 0. (201) \Phi(g_1,g_2):=(\delta_\rho^1\phi)(g_1,g_2)\implies(\delta_\rho^2\Phi)(g_1,g_2,g_3)=0. \tag*{(201)} Φ ( g 1 , g 2 ) := ( δ ρ 1 ϕ ) ( g 1 , g 2 ) ⟹ ( δ ρ 2 Φ ) ( g 1 , g 2 , g 3 ) = 0. (201)
Using the coboundary operator, we define two the sets
Z n ( G , M ρ ) : = ker ( δ ρ n ) = { ϕ ∈ C n ( G , M ) ∣ δ ρ n ϕ = 0 } , (202) Z^n(G,M_\rho):=\ker(\delta_\rho^n)=\{\phi\in C^n(G,M)\mid\delta_\rho^n\phi=0\}, \tag*{(202)} Z n ( G , M ρ ) := ker ( δ ρ n ) = { ϕ ∈ C n ( G , M ) ∣ δ ρ n ϕ = 0 } , (202)
and
B n ( G , M ρ ) : = im ( δ ρ n − 1 ) = { ϕ ∈ C n ( G , M ) ∣ ϕ = δ ρ n − 1 ϕ ′ , ϕ ′ ∈ C n − 1 ( G , M ) } . (203) B^n(G,M_\rho):=\operatorname{im}(\delta_\rho^{n-1})=\{\phi\in C^n(G,M)\mid\phi=\delta_\rho^{n-1}\phi',\ \phi'\in C^{n-1}(G,M)\}. \tag*{(203)} B n ( G , M ρ ) := im ( δ ρ n − 1 ) = { ϕ ∈ C n ( G , M ) ∣ ϕ = δ ρ n − 1 ϕ ′ , ϕ ′ ∈ C n − 1 ( G , M )} . (203)
The cochains in Z n ( G , M ρ ) Z^{n}(G,M_{\rho}) Z n ( G , M ρ ) are called n n n -cocycles. The cochains in B n ( G , M ρ ) B^{n}(G,M_{\rho}) B n ( G , M ρ ) are called n n n -coboundaries. The action of the boundary operator on the elements of the group M M M is sensitive to the homomorphism ρ \rho ρ . For this reason, we label M M M by ρ \rho ρ in Z n ( G , M ρ ) Z^{n}(G,M_{\rho}) Z n ( G , M ρ ) and B n ( G , M ρ ) B^{n}(G,M_{\rho}) B n ( G , M ρ ) . The importance of the coboundaries is that the identity (A6) generalizes to
ϕ = δ ρ n − 1 ϕ ′ ⟹ δ ρ n ϕ = 0. (204) \phi= \delta_{\rho}^{n-1}\phi' \quad\Longrightarrow\quad\delta_{\rho}^{n}\phi= 0. \tag*{(204)} ϕ = δ ρ n − 1 ϕ ′ ⟹ δ ρ n ϕ = 0. (204)
The n n n th cohomology group is defined as the quotient of the n n n -cocycles by the n n n -coboundaries, i.e.,
H n ( G , M ρ ) : = Z n ( G , M ρ ) / B n ( G , M ρ ) . (205) H^{n}(G,M_{\rho}) := Z^{n}(G,M_{\rho})/B^{n}(G,M_{\rho}). \tag*{(205)} H n ( G , M ρ ) := Z n ( G , M ρ ) / B n ( G , M ρ ) . (205)
The n n n th cohomology group H n ( G , M ρ ) H^{n}(G,M_{\rho}) H n ( G , M ρ ) is an additive Abelian group. We denote its elements by [ ϕ ] ∈ H n ( G , M ρ ) [\phi] \in H^{n}(G,M_{\rho}) [ ϕ ] ∈ H n ( G , M ρ ) , i.e., the equivalence class of the n n n -cocycle ϕ \phi ϕ .
Finally, we define the cup product of a pair of Z N \mathbb{Z}_{N} Z N -valued cochains as follows. Given two cochains ϕ ∈ C n ( G , Z N ) \phi\in C^{n}(G,\mathbb{Z}_{N}) ϕ ∈ C n ( G , Z N ) and θ ∈ C m ( G , Z N ) \theta\in C^{m}(G,\mathbb{Z}_{N}) θ ∈ C m ( G , Z N ) , we produce the ( n + m ) (n+m) ( n + m ) -cochain ( ϕ ∪ θ ) ∈ C n + m ( G , Z N ) (\phi\cup\theta) \in C^{n+m}(G,\mathbb{Z}_{N}) ( ϕ ∪ θ ) ∈ C n + m ( G , Z N ) through
( ϕ ∪ θ ) ( g 1 , ⋯ , g n , g n + 1 , ⋯ , g m ) : = ρ g n + 1 ⋅ g n + 2 ⋯ g n + m ϕ ( g 1 , ⋯ , g n ) θ ( g n + 1 , ⋯ , g n + m ) m o d N , (206) (\phi\cup\theta)(g_{1},\cdots,g_{n},g_{n+1},\cdots,g_{m}) := \rho_{g_{n+1}\cdot g_{n+2}\cdots g_{n+m}}\phi(g_{1},\cdots,g_{n})\theta(g_{n+1},\cdots,g_{n+m}) \mod N, \tag*{(206)} ( ϕ ∪ θ ) ( g 1 , ⋯ , g n , g n + 1 , ⋯ , g m ) := ρ g n + 1 ⋅ g n + 2 ⋯ g n + m ϕ ( g 1 , ⋯ , g n ) θ ( g n + 1 , ⋯ , g n + m ) mod N , (206)
where the multiplication of the cochains ϕ \phi ϕ and θ \theta θ is treated as the multiplication of integers modulo N N N . For instance, the cup product of a 1-cochain α ∈ C 1 ( G , Z N ) \alpha\in C^{1}(G,\mathbb{Z}_{N}) α ∈ C 1 ( G , Z N ) and a 2-cochain β ∈ C 2 ( G , Z N ) \beta\in C^{2}(G,\mathbb{Z}_{N}) β ∈ C 2 ( G , Z N ) is
( α ∪ β ) ( g 1 , g 2 , g 3 ) = ρ g 2 g 3 α ( g 1 ) β ( g 2 , g 3 ) . (207) (\alpha\cup\beta)(g_{1},g_{2},g_{3}) = \rho_{g_{2}g_{3}}\alpha(g_{1})\beta(g_{2},g_{3}). \tag*{(207)} ( α ∪ β ) ( g 1 , g 2 , g 3 ) = ρ g 2 g 3 α ( g 1 ) β ( g 2 , g 3 ) . (207)
The cup product defined in Eq. (206) satisfies
δ ρ n + m ( ϕ ∪ θ ) = ( δ ρ n ϕ ∪ θ ) + ( − 1 ) n ( ϕ ∪ δ ρ m θ ) (208) \delta_{\rho}^{n+m}(\phi\cup\theta) = (\delta_{\rho}^{n}\phi\cup\theta) + (-1)^{n}(\phi\cup\delta_{\rho}^{m}\theta) \tag*{(208)} δ ρ n + m ( ϕ ∪ θ ) = ( δ ρ n ϕ ∪ θ ) + ( − 1 ) n ( ϕ ∪ δ ρ m θ ) (208)
for two cochains ϕ ∈ C n ( G , M ) \phi\in C^{n}(G,M) ϕ ∈ C n ( G , M ) and θ ∈ C m ( G , M ) \theta\in C^{m}(G,M) θ ∈ C m ( G , M ) . Hence, the cup product of two cocycles is again a cocycle, as the right-hand side of Eq. (208) vanishes in that case.
Construction of the parafermionic symmetry group G G G
As argued in Sec. II A, the IPT phases of order-N N N parafermions have a global Z N p f \mathbb{Z}_{N}^{\mathrm{pf}} Z N pf parafermion charge symmetry. In the presence of additional symmetries described by a group G b G_b G b , the global symmetry G G G is a group extension of G b G_b G b by Z N p f \mathbb{Z}_{N}^{\mathrm{pf}} Z N pf , as we shall explain.
Parafermion charge symmetry is a normal subgroup
We demand that any element g ∈ G b g \in G_b g ∈ G b realizes a locality-preserving transformation, i.e., it maps every operator to nearby operators. Let us consider two operators A j A_j A j and B k B_k B k with definite parafermion charges q A q_A q A and q B q_B q B , respectively, whose supports are spatially separated. They satisfy the algebra (17) . Under the action of g ∈ G b g \in G_b g ∈ G b , these operators are mapped to the operators A j ′ A'_j A j ′ and B k ′ B'_k B k ′ which, by g g g being locality-preserving, must also have supports that are spatially separated. In particular, they must satisfy the same algebra (17) , i.e.,
A j ′ B k ′ = ω N q A q B sgn ( k − j ) B k ′ A j ′ . (209) A'_j B'_k = \omega_N^{q_A q_B\operatorname{sgn}(k-j)} B'_k A'_j. \tag*{(209)} A j ′ B k ′ = ω N q A q B sgn ( k − j ) B k ′ A j ′ . (209)
We can write the operators A j ′ A'_j A j ′ and B k ′ B'_k B k ′ in terms of sums of operators with well-defined parafermion charges,
A j ′ = ∑ α = 0 N − 1 A j ′ ( α ) , B k ′ = ∑ β = 0 N − 1 B k ′ ( β ) . (210) A'_j = \sum_{\alpha=0}^{N-1} A_j'^{(\alpha)}, \qquad B'_k = \sum_{\beta=0}^{N-1} B_k'^{(\beta)}. \tag*{(210)} A j ′ = α = 0 ∑ N − 1 A j ′ ( α ) , B k ′ = β = 0 ∑ N − 1 B k ′ ( β ) . (210)
where A j ′ ( α ) A_j'^{(\alpha)} A j ′ ( α ) (B k ′ ( β ) B_k'^{(\beta)} B k ′ ( β ) ) is an operator with definite parafermion charge α \alpha α (β \beta β ) ∈ Z N \in\mathbb{Z}_N ∈ Z N . On the one hand, the left-hand side of Eq. (209) becomes
A j ′ B k ′ = ∑ α , β = 0 N − 1 A j ′ ( α ) B k ′ ( β ) = ∑ α , β = 0 N − 1 ω N α β sgn ( k − j ) B k ′ ( β ) A j ′ ( α ) , (211) \begin{aligned}
A'_j B'_k &= \sum_{\alpha,\beta=0}^{N-1} A_j'^{(\alpha)} B_k'^{(\beta)} \\
&= \sum_{\alpha,\beta=0}^{N-1} \omega_N^{\alpha\beta\operatorname{sgn}(k-j)} B_k'^{(\beta)} A_j'^{(\alpha)}, \tag*{(211)}
\end{aligned} A j ′ B k ′ = α , β = 0 ∑ N − 1 A j ′ ( α ) B k ′ ( β ) = α , β = 0 ∑ N − 1 ω N α β sgn ( k − j ) B k ′ ( β ) A j ′ ( α ) , (211)
where we used the algebra (17) for each definite parafermion-charge summand. On the other hand, the right-hand side of Eq. (209) becomes
ω N q A q B sgn ( k − j ) B k ′ q A A j ′ q B = ω N q A q B sgn ( k − j ) ∑ α , β = 0 N − 1 B k ′ ( β ) A j ′ ( α ) . (212) \omega_N^{q_Aq_B\operatorname{sgn}(k-j)} B_k'{}^{q_A} A_j'{}^{q_B} = \omega_N^{q_Aq_B\operatorname{sgn}(k-j)} \sum_{\alpha,\beta=0}^{N-1} B_k'{}^{(\beta)} A_j'{}^{(\alpha)}. \tag*{(212)} ω N q A q B sgn ( k − j ) B k ′ q A A j ′ q B = ω N q A q B sgn ( k − j ) α , β = 0 ∑ N − 1 B k ′ ( β ) A j ′ ( α ) . (212)
Matching the two sides of the equation and comparing coefficients implies that only the operators with charges satisfying α β = q A q B ( m o d N ) \alpha\beta=q_Aq_B \pmod N α β = q A q B ( mod N ) can appear in the decompositions (B2). By choosing q A = q B = 1 q_A=q_B=1 q A = q B = 1 we find
α β = 1 ( m o d N ) . (213) \alpha\beta=1 \pmod N. \tag*{(213)} α β = 1 ( mod N ) . (213)
For a fixed α \alpha α this equation is satisfied by at most a single β \beta β . Hence, the sums in (B2) reduce to a single term associated to a parafermion charge α \alpha α and β \beta β . In other words, locality dictates that any global internal symmetry operator must map an operator with a definite parafermion charge to another operator with a definite parafermion charge, i.e., the Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf grading must be preserved.
A consequence of the fact that every symmetry preserves the Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf grading is that Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf must be a normal subgroup of G \mathcal{G} G , since its block-diagonal form must be preserved under conjugation by the symmetries up to permutations of blocks. To be precise, a Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf -eigensector with definite parafermion charge q q q can only be mapped to the eigensector
g : q ↦ ρ g q + κ , (214) g:q\mapsto\rho_g q+\kappa, \tag*{(214)} g : q ↦ ρ g q + κ , (214)
for any element g ∈ G b g\in G_b g ∈ G b . Here, ρ g \rho_g ρ g corresponds to the possible action of g g g on Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf while the κ = 0 , 1 , ⋯ , N − 1 \kappa=0,1,\cdots,N-1 κ = 0 , 1 , ⋯ , N − 1 denotes the parafermion charge carried by the representation of g g g .
Parafermionic symmetry group G \mathcal{G} G
We are now equipped to describe the structure of the total symmetry group G \mathcal{G} G . Since Z N p f ⊂ G \mathbb{Z}_N^{\mathrm{pf}}\subset\mathcal{G} Z N pf ⊂ G is a normal subgroup, G \mathcal{G} G is given by a group extension of G b ≅ G / Z N p f G_b\cong\mathcal{G}/\mathbb{Z}_N^{\mathrm{pf}} G b ≅ G / Z N pf . We denote by g ~ \tilde{g} g ~ the elements of group G \mathcal{G} G . Each element g ~ \tilde{g} g ~ can also be alternatively written as the pair ( z , g ) (z,g) ( z , g ) where z ∈ Z N p f z\in\mathbb{Z}_N^{\mathrm{pf}} z ∈ Z N pf and g ∈ G b g\in G_b g ∈ G b . We denote by ∘ \circ ∘ the group composition rule in G \mathcal{G} G while we use the addition + + + and multiplication ⋅ \cdot ⋅ for group composition rules in Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf and G \mathcal{G} G .
The group extension is specified by the short-exact sequence
1 ⟶ Z N p f → ι G → π G b ⟶ 1 , (215) 1\longrightarrow\mathbb{Z}_N^{\mathrm{pf}}\xrightarrow{\iota}\mathcal{G}\xrightarrow{\pi}G_b\longrightarrow1, \tag*{(215)} 1 ⟶ Z N pf ι G π G b ⟶ 1 , (215)
where ι \iota ι is an injective inclusion map and π \pi π a surjective quotient map such that im ι = ker π \operatorname{im}\iota=\operatorname{ker}\pi im ι = ker π . The map ι \iota ι describes how elements of Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf are embedded in the group G \mathcal{G} G , respectively. The map π \pi π specifies the quotient
G b ≅ G / ker π = G / im ι ≅ G / Z N p f . (216) G_b\cong\mathcal{G}/\operatorname{ker}\pi
=\mathcal{G}/\operatorname{im}\iota
\cong\mathcal{G}/\mathbb{Z}_N^{\mathrm{pf}}. \tag*{(216)} G b ≅ G / ker π = G / im ι ≅ G / Z N pf . (216)
Since Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf is a normal subgroup the quotient is a group. We define two more maps
s : G b ⟶ G , τ : G ⟶ Z N p f , (217) s:G_b\longrightarrow\mathcal{G},\qquad\tau:\mathcal{G}\longrightarrow\mathbb{Z}_N^{\mathrm{pf}}, \tag*{(217)} s : G b ⟶ G , τ : G ⟶ Z N pf , (217)
such that
π ( s ) = id G b , τ ( ι ) = id Z N p f , τ ( ι ( z ) ∘ g ~ ) = z τ ( g ~ ) , τ ( s ( g ) ) = 0 , (218) \pi(s)=\operatorname{id}_{G_b},\qquad\tau(\iota)=\operatorname{id}_{\mathbb{Z}_N^{\mathrm{pf}}},\qquad\tau(\iota(z)\circ\tilde{g})=z\tau(\tilde{g}),\qquad\tau(s(g))=0, \tag*{(218)} π ( s ) = id G b , τ ( ι ) = id Z N pf , τ ( ι ( z ) ∘ g ~ ) = z τ ( g ~ ) , τ ( s ( g )) = 0 , (218)
where id G b \operatorname{id}_{G_b} id G b and id Z N p f \operatorname{id}_{\mathbb{Z}_N^{\mathrm{pf}}} id Z N pf are identity isomorphisms on the groups G b G_b G b and Z N p f \mathbb{Z}_N^{\mathrm{pf}} Z N pf . The map s s s is injective but not necessarily a homomorphism and called a section of the projection π \pi π . There is always a choice of a section such that τ ( s ( g ) ) = 0 \tau(s(g))=0 τ ( s ( g )) = 0 , which we make. The map τ \tau τ is surjective and called the trivialization.
There exist a bijection between the groups G \mathcal{G} G and Z N p f × G b \mathbb{Z}_N^{\mathrm{pf}}\times G_b Z N pf × G b as sets but not necessarily as groups. To see this, we note that any element g ~ ∈ G \tilde{g}\in\mathcal{G} g ~ ∈ G is mapped to a pair ( z , g ) ∈ Z N p f × G b (z,g)\in\mathbb{Z}_N^{\mathrm{pf}}\times G_b ( z , g ) ∈ Z N pf × G b by
( τ , π ) : g ~ ↦ ( τ ( g ~ ) , π ( g ~ ) ) = ( z , g ) . (219) (\tau,\pi):\tilde{g}\mapsto\bigl(\tau(\tilde{g}),\pi(\tilde{g})\bigr)=(z,g). \tag*{(219)} ( τ , π ) : g ~ ↦ ( τ ( g ~ ) , π ( g ~ ) ) = ( z , g ) . (219)
Conversely, an element ( z , g ) ∈ Z N p f × G b (z,g) \in\mathbb{Z}^{\mathrm{pf}}_N \times G_b ( z , g ) ∈ Z N pf × G b is mapped to an element g ~ ∈ G \tilde{g} \in\mathcal{G} g ~ ∈ G through the map
ι ∘ s : ( z , g ) ↦ ι ( z ) ∘ s ( g ) ≡ g ~ . (220) \iota\circ s : (z,g) \mapsto\iota(z) \circ s(g) \equiv\tilde{g}. \tag*{(220)} ι ∘ s : ( z , g ) ↦ ι ( z ) ∘ s ( g ) ≡ g ~ . (220)
One verifies that both maps are bijections. In particular, ( τ , π ) (\tau,\pi) ( τ , π ) is the left inverse of ι ∘ s \iota\circ s ι ∘ s .
We would like to write the group composition rule ∘ \circ ∘ in G \mathcal{G} G in terms of the group composition rules ( + , ⋅ ) (+, \cdot) ( + , ⋅ ) in Z N p f × G b \mathbb{Z}^{\mathrm{pf}}_N \times G_b Z N pf × G b . To this end, we observe for two elements g , h ∈ G b g,h \in G_b g , h ∈ G b
π ( s ( g ) ∘ s ( h ) ) = π ( s ( g ) ) π ( s ( h ) ) = g h = π ( s ( g h ) ) , (221) \pi(s(g) \circ s(h)) = \pi(s(g))\pi(s(h)) = gh = \pi(s(gh)), \tag*{(221)} π ( s ( g ) ∘ s ( h )) = π ( s ( g )) π ( s ( h )) = g h = π ( s ( g h )) , (221)
where in the last line we have used the fact that the product g h gh g h can be written as π ( s ( g h ) ) \pi(s(gh)) π ( s ( g h )) . In other words, the composition rule
π ( s ( g ) ∘ s ( h ) ) = π ( s ( g h ) ) , (222) \pi(s(g) \circ s(h)) = \pi(s(gh)), \tag*{(222)} π ( s ( g ) ∘ s ( h )) = π ( s ( g h )) , (222)
holds under the projection π \pi π . However, since ker π = im ι \ker\pi= \operatorname{im}\iota ker π = im ι , this equation implies
s ( g ) ∘ s ( h ) = c ( g , h ) s ( g h ) (223) s(g) \circ s(h) = c(g,h)s(gh) \tag*{(223)} s ( g ) ∘ s ( h ) = c ( g , h ) s ( g h ) (223)
where c ( g , h ) ∈ im ι ≅ Z N p f c(g,h) \in\operatorname{im}\iota\cong\mathbb{Z}^{\mathrm{pf}}_N c ( g , h ) ∈ im ι ≅ Z N pf . In particular, we have the composition rule between the two pairs ( z , g ) (z,g) ( z , g ) and ( u , h ) (u,h) ( u , h ) in Z N p f × G b \mathbb{Z}^{\mathrm{pf}}_N \times G_b Z N pf × G b
ι ( z ) s ( g ) ι ( u ) s ( h ) = ι ( z ) ι ( ρ g u ) s ( g ) s ( h ) = ι ( z + ρ g u + c ( g , h ) ) s ( g h ) , (224) \iota(z)s(g)\iota(u)s(h) = \iota(z)\iota(\rho_g u)s(g)s(h) = \iota(z+\rho_g u+c(g,h))s(gh), \tag*{(224)} ι ( z ) s ( g ) ι ( u ) s ( h ) = ι ( z ) ι ( ρ g u ) s ( g ) s ( h ) = ι ( z + ρ g u + c ( g , h )) s ( g h ) , (224)
where we denote by ρ g \rho_g ρ g the action of g g g on the element u u u which is due to the possible non-commutativity between the elements s ( g ) s(g) s ( g ) and ι ( u ) \iota(u) ι ( u ) in G \mathcal{G} G .
We make the identification ( z , g ) = ι ( z ) s ( g ) ≡ g ~ (z,g)=\iota(z)s(g)\equiv\tilde{g} ( z , g ) = ι ( z ) s ( g ) ≡ g ~ and ( u , h ) = ι ( u ) s ( h ) ≡ h ~ (u,h)=\iota(u)s(h)\equiv\tilde{h} ( u , h ) = ι ( u ) s ( h ) ≡ h ~ and rewrite Eq. (224) as the composition rule
( z , g ) ∘ ( u , h ) = ( z + ρ g u + c ( g , h ) , g h ) . (225) (z,g)\circ(u,h)=(z+\rho_g u+c(g,h),gh). \tag*{(225)} ( z , g ) ∘ ( u , h ) = ( z + ρ g u + c ( g , h ) , g h ) . (225)
Here, c ( g , h ) c(g,h) c ( g , h ) is the representative of an extension class [ c ] [c] [ c ] . More precisely, c ∈ C 2 ( G b , Z N p f ) c \in C^2(G_b,\mathbb{Z}^{\mathrm{pf}}_N) c ∈ C 2 ( G b , Z N pf ) is a 2-cochain such that on the one hand the associativity in gorup G \mathcal{G} G implies that
δ ρ ( 2 ) c = 0 , (226) \delta^{(2)}_\rho c=0, \tag*{(226)} δ ρ ( 2 ) c = 0 , (226)
and on the other hand, a group isomorphism α : G → G \alpha:\mathcal{G}\to\mathcal{G} α : G → G shifts c c c to c ′ c' c ′ given by
c ′ = c + δ ρ ( 1 ) α . (227) c'=c+\delta^{(1)}_\rho\alpha. \tag*{(227)} c ′ = c + δ ρ ( 1 ) α . (227)
Hence, the inequivalent groups G \mathcal{G} G that can be obtained by extending a group G b G_b G b by Z N p f \mathbb{Z}^{\mathrm{pf}}_N Z N pf are specified by the extension classes [ c ] [c] [ c ] in the second cohomology group H 2 ( G b , [ Z N p f ] ρ ) H^2\left(G_b,[\mathbb{Z}^{\mathrm{pf}}_N]_\rho\right) H 2 ( G b , [ Z N pf ] ρ ) .
Group actions on Z N p f \mathbb{Z}^{\mathrm{pf}}_N Z N pf are involutive
The possible group actions ρ \rho ρ are severely restricted. To derive this restriction, consider an element g ∈ G b g \in G_b g ∈ G b and a corresponding element g ~ = ( z , g ) ∈ G \tilde{g}=(z,g)\in\mathcal{G} g ~ = ( z , g ) ∈ G . Consider again the two local operators A j A_j A j and B k B_k B k with definite parafermion charges q A q_A q A and q B q_B q B , respectively, whose supports are spatially separated. They obey the exchange algebra (17) . Under the action of the symmetry g ~ \tilde{g} g ~ , these operators are mapped to A j ′ A'_j A j ′ and B k ′ B'_k B k ′ , respectively. The group action ρ g \rho_g ρ g then maps their parafermion charges to ρ g q A \rho_g q_A ρ g q A and ρ g q B \rho_g q_B ρ g q B , respectively. Consequently, Eq. (17) implies that the transformed operators obey
A j ′ B k ′ = ω N ρ g 2 q A q B sgn ( k − j ) B k ′ A j ′ . (228) A'_j B'_k=\omega_N^{\rho_g^2 q_A q_B\operatorname{sgn}(k-j)}B'_k A'_j. \tag*{(228)} A j ′ B k ′ = ω N ρ g 2 q A q B sgn ( k − j ) B k ′ A j ′ . (228)
At the time, a unitary transformation must preserve the exchange algebra of A j A_j A j and B k B_k B k , i.e.,
A j ′ B k ′ = ω N q A q B sgn ( k − j ) B k ′ A j ′ . (229) A'_j B'_k=\omega_N^{q_A q_B\operatorname{sgn}(k-j)}B'_k A'_j. \tag*{(229)} A j ′ B k ′ = ω N q A q B sgn ( k − j ) B k ′ A j ′ . (229)
Comparing these two exchange relations for operators with q A = q B = 1 q_A=q_B=1 q A = q B = 1 yields
ρ g 2 = 1 m o d N . (230) \rho_g^2=1\ \mathrm{mod}\ N. \tag*{(230)} ρ g 2 = 1 mod N . (230)
Thus, every automorphism in the image of ρ \rho ρ must be involutive. For N = p k N=p^k N = p k , the group of involutive automorphisms of Z p k p f \mathbb{Z}_{p^k}^{\mathrm{pf}} Z p k pf is
I ( Z p k p f ) = { Z 1 , if p = 2 and k = 1 , Z 2 , if p = 2 and k = 2 , Z 2 × Z 2 , if p = 2 and k > 2 , Z 2 , if p > 2. (231) \mathcal{I}\left(\mathbb{Z}_{p^k}^{\mathrm{pf}}\right)=
\begin{cases}
\mathbb{Z}_1, & \text{if }p=2\text{ and }k=1,\\
\mathbb{Z}_2, & \text{if }p=2\text{ and }k=2,\\
\mathbb{Z}_2\times\mathbb{Z}_2, & \text{if }p=2\text{ and }k>2,\\
\mathbb{Z}_2, & \text{if }p>2.
\end{cases} \tag*{(231)} I ( Z p k pf ) = ⎩ ⎨ ⎧ Z 1 , Z 2 , Z 2 × Z 2 , Z 2 , if p = 2 and k = 1 , if p = 2 and k = 2 , if p = 2 and k > 2 , if p > 2. (231)
The result for generic N N N follows from Eq. (231) by applying the Chinese remainder theorem.
Proof of Eq. (3.20)
We derive for phases with odd number of parafermions on the boundaries (μ = 1 \mu=1 μ = 1 ) the parametrization (52) of the boundary representation. Consider without loss of generality an element g ~ = ( 0 , g ) ∈ G \tilde{g}=(0,g)\in\mathcal{G} g ~ = ( 0 , g ) ∈ G such that there is no charge conjugation, i.e., ρ g = 1 \rho_g=1 ρ g = 1 . Equation (52) then reduces to
U B ( 0 , g ) = V B ( g ) γ ∞ − κ ( g ) , (232) U_{\mathrm{B}}(0,g)=V_{\mathrm{B}}(g)\gamma_{\infty}^{-\kappa(g)}, \tag*{(232)} U B ( 0 , g ) = V B ( g ) γ ∞ − κ ( g ) , (232)
where V B ( g ) ∈ P N n B V_{\mathrm{B}}(g)\in\mathcal{P}_{N}^{n_{\mathrm{B}}} V B ( g ) ∈ P N n B . In other words, the auxiliary parafermion may be completely factorized out of U B ( 0 , g ) U_{\mathrm{B}}(0,g) U B ( 0 , g ) . The full parametrization (52) follows from this particular case analogously to the even boundary (μ = 0 \mu=0 μ = 0 ) parametrization (35) .
To prove Eq. (232) , we make the following physical assumption. Let Y Y Y be the generator of the center of the boundary algebra P N n B \mathcal{P}_{N}^{n_{\mathrm{B}}} P N n B as defined in Eq. (49) . We ask that the image of Y Y Y under the action by conjugation of U B ( 0 , g ) U_{\mathrm{B}}(0,g) U B ( 0 , g ) must have well-defined (boundary) parafermion charge.
Since U B ( 0 , g ) U_{\mathrm{B}}(0,g) U B ( 0 , g ) is unitary, and thus preserves the algebra’s commutation relation, it must map Y Y Y to an element of the center. As Y Y Y itself generates the center, we have
U B ( 0 , g ) Y U B ( 0 , g ) † = ∑ a = 0 N − 1 c a Y a , c a ∈ C . (233) U_{\mathrm{B}}(0,g)Y U_{\mathrm{B}}(0,g)^\dagger=\sum_{a=0}^{N-1}c_aY^a,\qquad c_a\in\mathbb{C}. \tag*{(233)} U B ( 0 , g ) Y U B ( 0 , g ) † = a = 0 ∑ N − 1 c a Y a , c a ∈ C . (233)
Because Y Y Y has unit parafermion charge, our assumption implies
U B ( 0 , g ) Y U B ( 0 , g ) † = c a Y a (234) U_{\mathrm{B}}(0,g)Y U_{\mathrm{B}}(0,g)^\dagger=c_aY^a \tag*{(234)} U B ( 0 , g ) Y U B ( 0 , g ) † = c a Y a (234)
for some a = 0 , … , N − 1 a=0,\ldots,N-1 a = 0 , … , N − 1 . Moreover, since g ~ = ( 0 , g ) \tilde{g}=(0,g) g ~ = ( 0 , g ) was chosen to realize no charge conjugation, and Y N = 1 Y^N=1 Y N = 1 , we deduce
U B ( 0 , g ) Y U B ( 0 , g ) † = ω N κ Y , (235) U_{\mathrm{B}}(0,g)Y U_{\mathrm{B}}(0,g)^\dagger=\omega_N^\kappa Y, \tag*{(235)} U B ( 0 , g ) Y U B ( 0 , g ) † = ω N κ Y , (235)
where κ = 0 , … , N − 1 \kappa=0,\ldots,N-1 κ = 0 , … , N − 1 . Inverting this equation yields
Y U B ( 0 , g ) Y † = ω N − κ U B ( 0 , g ) . (236) Y U_{\mathrm{B}}(0,g)Y^\dagger=\omega_N^{-\kappa}U_{\mathrm{B}}(0,g). \tag*{(236)} Y U B ( 0 , g ) Y † = ω N − κ U B ( 0 , g ) . (236)
Since Y Y Y commutes with all parafermion operators on the boundary with the exception of the auxiliary γ ∞ \gamma_\infty γ ∞ , this identity can only hold if equation (C1) is true, where we identified κ ≡ κ ( g ) \kappa\equiv\kappa(g) κ ≡ κ ( g ) .
References