A Fusion rule symmetries of CFTs

# Quantum criticality in many-body parafermion chains

Quantum criticality in many-body parafermion chains

V. Lahtinen1, T. Månsson2, E. Ardonne3,*

1 Dahlem Center for Complex Quantum Systems, Freie Universität Berlin, 14195 Berlin, Germany

2 Department of Theoretical Physics, School of Engineering Sciences, Royal Institute of Technology (KTH), Roslagstullsbacken 21, SE-106 91 Stockholm, Sweden

3 Department of Physics, Stockholm University, AlbaNova University Center, SE-106 91 Stockholm, Sweden

* ardonne@fysik.su.se

March 21, 2018

## Abstract

We construct local generalizations of 3-state Potts models with exotic critical points. We analytically show that these are described by non-diagonal modular invariant partition functions of products of parafermion or conformal field theories (CFTs). These correspond either to non-trivial permutation invariants or block diagonal invariants, that one can understand in terms of anyon condensation. In terms of lattice parafermion operators, the constructed models correspond to parafermion chains with many-body terms. Our construction is based on how the partition function of a CFT depends on symmetry sectors and boundary conditions. This enables to write the partition function corresponding to one modular invariant as a linear combination of another over different sectors and boundary conditions, which translates to a general recipe how to write down a microscopic model, tuned to criticality. We show that the scheme can also be extended to construct critical generalizations of -state Potts models.

## 1 Introduction

At a quantum critical point two distinct phases of matter coexist. A remarkable feature of 1D systems is that such special points in the phase diagram are in general described by a field theory with conformal symmetry – a conformal field theory (CFT) [1, 2]. In other words, the system exhibits a universal behavior regardless of the underlying microscopic model, i.e. what are the local degrees of freedom and how they interact. This universal description at the critical point enables to determine what phases of matter appear in the vicinity of the critical point when the system is perturbed away from it. However, the relation of the universality class to criticality of a microscopic model is a one-way problem. If a given microscopic model exhibits a critical point with diverging correlation length, given a catalog of possible CFTs it is comparatively easy to make ansatzes to verify which CFT describes it. The converse is not true though. While symmetries present in the CFTs constrain the possible microscopic models, there is in general no recipe to write down a model that exhibits a critical point described by a given CFT.

The lack of framework to write down a microscopic model for a given CFT is an outstanding problem from several perspectives. From an academic perspective, it is of interest to understand the minimal microscopic conditions that can give rise to a given universal behavior. This enables to experimentally search for such behavior in existing or synthesized materials. While CFT is a standard theoretical tool, experiments probing CFT predictions beyond measuring critical exponents are still few [3, 4]. Due to the coexistence of phases of matter at the critical points, critical models with universal behavior that are perturbed away from the critical point can also serve as starting points for models of gapped, possibly topological phases of matter. Recently, the interplay of symmetry-protected topological order and quantum criticality has been explored in spin-1/2 chains [5, 6, 7], with a unified picture emerging how the protecting symmetries and the possible fractionalized edge states determine the universality classes of transitions [8, 9].

While the universal description of a critical point is a property of 1D systems (the nature of conformal critical points in 2D is an open question and a subject of cutting-edge numerical studies [10, 11]), it has been proposed that critical 1D systems can serve as building blocks of 2D topologically ordered states of matter [12, 13, 14, 15]. Critical 1D systems can be arranged into 2D array and when coupled together in a designed fashion, the system can become a gapped topologically ordered state whose nature depends only on the couplings and on the CFT describing the critical 1D systems. Thus new microscopic models with exotic critical points enable also the construction of new models for exotic 2D topological states of matter. In particular, since the advent of topological quantum computation, there is much interest to construct states that harbour non-Abelian anyons that could be employed for topologically protected quantum information processing [16].

Motivated by these open questions, in this work we advance the program initiated in our earlier works, where we constructed exactly solvable spin-1/2 chains for all criticalities in the universality class [17, 18, 6]. Instead of spin chains, we focus here on 3-state Potts models, where the local degrees of freedom are not spins, but clock variables, and construct generalizations that exhibit critical points whose critical behavior has not been previously discussed in the literature. 3-state Potts models themselves have recently attracted attention due to their relation to parafermion modes [19], that have been proposed to be realized, following the same principles that lead to the recent experimental discovery of Majorana modes [20, 21], by inducing superconductivity via the proximity effect on fractional quantum Hall edge states, such as the Laughlin state [22, 23]. A uniform array of such parafermion modes is unitarily equivalent to a 3-state Potts model via the parafermion version of the Jordan-Wigner transformation, the Fradkin-Kadanoff transformation [24]. The recent focus on parafermions arises not from themselves though, but from their collective behavior. Were they to hybridize in a 2D array, they could realize a state that hosts the coveted Fibonacci anyons that are universal for quantum computation [14]. This prospect, while speculative, has motivated research into 1D collective states of parafermion modes. Different parafermion phases have been understood from the symmetry protection perspective [25], the microscopics of parafermion CFT describing the criticality of the 3-state Potts model has been analyzed [26] and the phase diagram in the presence of longer range parafermion tunneling has been studied [27].

While our primary motivation is the construction of new Potts-like models with exotic criticalities, our models turn out to be unitarily equivalent to parafermion chains with many-body interactions between the parafermion modes. Thus as a by-product of our construction we address the nature of the critical behavior of parafermion chains in the presence of such many-body terms that may also arise when parafermion modes hybridize. To our understanding, the effect of such many-body terms has not been considered previously in literature. We find that when many-body terms are comparable to local chemical potential-like terms, the chains are critical and described by the non-diagonal modular invariant partition functions of a CFT that is product of some number of parafermion or CFTs.

Usually, when talking about a critical point being described by a given CFT, one refers to the diagonal invariant. Our work explicitly shows that this is too restrictive. We demonstrate that a particular class of non-diagonal modular invariant partition functions – the permutation invariants – that are usually overlooked when talking about physical systems, also have local microscopic models that realize them. Before proceeding to the actual constructions and the physics underlying them, we summarize our main results – the microscopic generalizations of the 3-state Potts models and the exotic critical points they exhibit.

### 1.1 Summary of results: Generalized critical 3-state Potts models

The starting point for our construction is the critical nearest neighbour 3-state Potts model that is described by the Hamiltonian (see for instance Ref. [28])

 H±=±L∑i=1(XiX†i+1+Zi+h.c.) . (1)

The local clock operators and commute on different sites, while on the same site they obey

 Z3=X3=1,Z2=Z†,X2=X†andZX=ωXZ,whereω=e2πi/3. (2)

For future reference, we also define . Like a spin-1/2 chain can be written in terms of fermionic operators via a Jordan-Wigner transformation, 3-state Potts models can be written in terms of parafermion operators via a Fradkin-Kadanoff transformation [24, 14]. Introducing the lattice parafermion operators

 Missing or unrecognized delimiter for \left (3)

that satisfy , and , the Hamiltonian (1) takes the form

 H±=±2L∑i=1(ωα†2i+1α2i+h.c.)+(ωα†2iα2i−1+h.c.). (4)

While this model is quadratic in terms of the parafermion operators, it is strongly interacting and not solvable with a Fourier transformation. Due to the formal similarity to free, quadratic fermions, we refer to it describing ‘quadratic’ parafermions.1 When higher order terms terms appear, such as four or six parafermion terms in the models we construct, we refer to the parafermions exhibiting many-body interactions.

The nature of the criticality of the 3-state Potts model depends on the overall sign. As shown in Table 1, is in the universality class of parafermion CFT with central charge [29, 30], while the criticality of is described by CFT with [31, 32]. The table also summarizes the new critical models we analytically construct and the CFTs that occur in them. Since they all involve terms such as , or , that involve four or more parafermion operators, all the new critical models correspond to parafermions with many-body interactions.

To explain how these models were constructed, the paper is structured as follows. In Section 2.1 we introduce the key concept of modular invariant partition functions of CFTs that characterize the possible distinct critical behavior. To illustrate these concepts and pave the way for constructing the generalized 3-state Potts models, in Section 2.2 we revisit the relation between two critical transverse field Ising chains and the XY chain from the perspective of modular invariant partition functions. This example, as well as its generalizations, has been studied in our previous works [17, 18] from the anyon condensation perspective, whose connection to the current approach of modular invariants is explained in Section 2.3. In Section 3 we review the parafermion and criticalities that appear in the 3-state Potts model and then in Section 4 we provide the detailed derivations of the new models listed in Table 1. Section 5 discusses the possible generalization of our construction to -state Potts models and we conclude with Section 6. A detailed discussion on the fusion rule symmetries that are useful in constructing non-diagonal permutation invariants can be found in Appendix A.

## 2 The Method: Modular invariant partition functions

Our construction is based on the properties of the partition function of the CFT describing a critical point. The partition function of every two-dimensional CFT must be invariant under symmetry operations known as modular transformations [2, 28]. It is possible that a given CFT admits several distinct modular invariant partition functions, in which case each describes critical behavior of distinct type. Often in literature, due to these appearing most often in physically relevant systems, one implicitly refers to the diagonal partition function when referring to a particular critical point being described by a given CFT. However, in general this is too restrictive. There can be also other non-diagonal modular invariant partition functions, to which we refer to permutation or condensation invariants for reasons to be explained below. We show that also these can in fact be realized in microscopic critical models.

To do this, we start from a known critical microscopic model described by the diagonal invariant of a CFT and find all the non-diagonal modular invariant partition functions by employing additional symmetries of the CFT. Then we express these partition functions as linear combinations of the diagonal partition function when it is restricted to different symmetry sectors and have different boundary conditions. This linear combination is subsequently translated into a Hamiltonian term that, when added to the initial critical microscopic system described by the diagonal invariant, changes the nature of the criticality to the non-diagonal invariant. While the resulting system is in general non-local, in all cases we are able to find canonical duality transformations that give a local and translational invariant representative of the new critical model.

To demonstrate this method in detail, in this section we first review the modular invariance of CFT partition functions. Then, by revisiting our earlier work [17, 18], we illustrate the method by showing how the critical XY model can be derived from two decoupled critical transverse field Ising chains. We also review the connection to anyon condensation that provides a simple criterium to predict when a given CFT admits particular type of non-diagonal invariant known as a condensation invariant.

### 2.1 Modular invariant partition functions

The partition function of a system described by the Hamiltonian is given by . When the system is critical, the Hamiltonian can be expanded in the Virasoro algebra generators satisfying the Virasoro algebra with central charge [2]. A special role is played by the operator . Its eigenvalues take the form , where a non-negative integer, that gives the energies of each state in the spectrum. All the energies are shifted away from integer values by , the scaling dimensions of a each chiral primary field . The distinct primary fields , whose number is finite in all the cases we consider, are specific to the CFT describing the critical point. Akin to usual symmetry sectors, every state that descends from belongs to the same ‘conformal tower’ that can be viewed as a sector of the CFT. The partition functions of these sectors are the chiral characters associated with different primary fields . Formally they are defined as the polynomials

 χϕi=q−c/24TrϕiqL0=q−c/24qhϕi∑n=0,1,…anqn , (5)

where the trace is over all the eigenstates of belonging to the conformal tower associated with the primary field . Here is a formal variable of the modular parameter whose powers are the energies of the states and the non-negative integers encode their degeneracy.

The primary field operators are in general chiral, which means that in 1+1D they can either propagate left or right. This means that single chiral CFT can describe chiral states, such as edge states of 2+1D topologically ordered states, but not genuine 1D states that arise from local Hamiltonians (chiral operators do not have local representations). Instead, 1+1D critical systems are described by combining the two chiral halves of the CFT. The full partition function describing such systems can be written as , where the bar denotes Hermitian conjugation. In terms of the left and right chiral characters and , the partition function of a 1+1D critical system takes the general form

 Z=∑ϕ1,ϕ2Mϕ1,ϕ2χϕ1¯χϕ2 , (6)

where and are summed over all primary fields of the CFT and are non-negative integers.

Possible reparametrizations of the torus set stringent constraints on the allowed partition functions of a given CFT described by the matrices with elements . These modular transformations formally transform the modular parameter as

 S:τ→−1/τ,T:τ→τ+1. (7)

Modular invariance of the partition function is then equivalent to demanding that

 [M,S]=[M,T]=0, (8)

where and are modular matrices specific to a given CFT with primary fields. Formally, they are obtained by studying the behavior of the chiral characters (5) under the modular transformations (7), but for most CFTs they can be found in the literature. The matrix is always diagonal with entries , while takes a form that does not have a simple expression in terms of only the scaling dimensions . To find all matrices subject to the constraints (8) amounts to classifying all modular invariant partition functions and hence all distinct critical behaviors that can occur in 1+1D systems. This, however, is a challenging task. A complete classification has been achieved only in a limited set of cases, such as the minimal models [33], but a general classification for arbitrary CFTs is still lacking. For a CFT with a small number of primary fields, one can perform a numerical brute-force search, which is the approach we take in the here.

The different possible modular invariant partition functions fall into different classes. First, clearly the identity matrix commutes with both and . This is known as the diagonal modular invariant, which is the most common partition function associated with a given CFT. In literature, a 1+1D critical systems being described by a CFT usually refers to this invariant. The second case corresponds to the matrix being block-diagonal. Using the original characters , one constructs new ones, , which are combined ‘in a diagonal way’ to form a new partition function . Typically, not all the original characters appear in the block diagonal partition function, and it can happen that certain characters appear several times. For the purpose of this paper, we call these block-diagonal invariants condensation modular invariants due to their connection to anyon condensation explained in Section 2.3. One can view the new chiral characters as corresponding to a primary field of a CFT with the same central charge as the original one, but with a different field content with less primary fields. Thus criticality described by a condensation invariant of some CFT is always equivalent to the diagonal invariant of some other CFT.

Given the diagonal invariant, it is sometimes possible to construct a permutation modular invariant. It is of the form , where denotes a permutation of the primary fields, which leaves the fusion rules of the primary fields invariant (we review fusion rule symmetries in Appendix A). In other words, all chiral characters of both chiral halves appear, but is no longer the identity matrix, but a permutation matrix. Depending on the symmetries of the CFT, it may or may not give rise to the same partition function as the diagonal one when (6) is written out as a polynomial in and . If not, then while the primary field content is each chiral half is the same as for the diagonal invariant, the local physical observables and the energy spectrum are different due to them being constructed from the left and right moving components of different primary fields. While criticality corresponding to permutation invariants has been little studied in the context of physical models, in this work we show that they can indeed also arise in local systems. We note that one does not have to start from the diagonal invariant, one can also obtain different invariants by performing a permutation of the fields on a block-diagonal (or condensation) invariant.

### 2.2 Example: Modular invariants for products of Ising CFTs

To illustrate these concepts, we give an explicit example of a case where distinct modular invariants appear – a product theory of two Ising CFTs. This example also illustrates how to deal with product CFTs and how the partition functions depend on twisted boundary conditions, which are the key methods for constructing new critical Potts-like models later.

The Ising CFT has three primary fields, denoted by , and , with scaling dimensions , and . The and matrices are represented by

 Unknown environment '% (9)

For the Ising CFT, there is only a a single modular invariant partition function given by the diagonal invariant .

The product of two Ising CFTs, that we call Ising CFT, has and nine primary fields that we will denote by pair of labels spanning all the possible combinations of the primary fields. The scaling dimensions for these product primary fields are additive in the constituent fields, i.e. . Similarly, their chiral characters are simply products and the representations of the modular matrices are given as the tensor products and . The matrix is thus now a matrix, but it is still straightforward to find all the modular invariant partition functions satisfying (8) by a brute force calculation. It turns out that there are three solutions given by

 ZIsing×2 =∑ϕ1,ϕ2∈Ising|χ(ϕ1,ϕ2)|2, (10) ZπIsing×2 =∑ϕ∈Ising|χ(ϕ,ϕ)|2+∑ϕ1≠ϕ2∈Isingχ(ϕ1,ϕ2)¯χ(ϕ2,ϕ1), (11) Zu(1)4Ising×2 =|χ(1,1)+χ(ψ,ψ)|2+|χ(1,ψ)+χ(ψ,1)|2+2|χ(σ,σ)|2. (12)

The is the diagonal invariant partition function of the Ising CFT, while is a permutation invariant. In the case of Ising it is not independent, but gives rise to a partition function that is identical to the diagonal invariant. This is due to the ‘layer’ symmetry of Ising that leaves the theory invariant under relabeling of the primary fields , i.e. , so that .

On the other hand, the block diagonality and the absence of some primary fields in means that it is a condensation invariant. If one identifies

 χ~1=χ(1,1)+χ(ψ,ψ),χ~ψ=χ(1,ψ)+χ(ψ,1)andχλ=χ¯λ=χ(σ,σ), (13)

one finds that it should correspond to a diagonal invariant of a theory with three non-trivial primary fields with scaling dimensions and . This CFT, which is a CFT of a compactified boson, is called . In fact, this was to be expected since it is well known that the CFT is obtained from the CFT by means of an orbifold construction [34]. Going in the other direction, one can obtain the CFT form the theory by ’adding the bosonic field to the chiral algebra’. We describe this relation between these CFTs from the perspective of anyon condensation below.

We have thus found that starting from the chiral Ising CFT, one can construct both the diagonal invariant and condensation invariant partition functions. Since it is well known that the critical transverse field Ising (TFI) chain is described by Ising CFT [28], the criticality corresponding to the diagonal invariant of Ising is clearly realized in a system of two decoupled TFI chains. The corresponding microscopic Hamiltonian is given, for instance, as the critical TFI chain with nest-nearest exchange interactions

 HIsing×2=L∑i=1(σxiσxi+2+σzi), (14)

where are the usual spin-1/2 Pauli matrices.

To write down a microscopic model for the condensation invariant, we begin by studying how the diagonal partition function of single Ising CFT depends on the symmetry sectors and boundary conditions. The symmetry sectors are inherited from the TFI chain that has spin flip symmetry and thus two symmetry sectors that we labeled by . A TFI chain can also have either periodic or anti-periodic boundary conditions, which we denote by , respectively. Denoting the partition function in symmetry sector with boundary conditions by , it is well known that [35]

 Z00 =χ1¯χ1+χψ¯χψ Z01 =χσ¯χσ, Z10 =χσ¯χσ Z11 =χ1¯χψ+χψ¯χ1 , (15)

where there holds as required by the duality between the symmetry sectors and boundary conditions when solving the transverse field Ising model with a Jordan-Wigner transformation [36]. Clearly, summing over the symmetry sectors with periodic boundary conditions gives the diagonal invariant partition function of the Ising CFT.

Likewise, the Hamiltonian (14) for two decoupled TFI chains has symmetry and the boundary conditions can be independently chosen for both chains. Employing the property for product CFTs, one can then verify that the diagonal invariant of the Ising CFT is obtained by summing over the four symmetry sectors with periodic boundary conditions

 ZIsing×2 =Z00Z00+Z00Z01+Z01Z00+Z01Z01 . (16)

Instead of decoupled TFI chains, consider a coupled system where the boundary condition of one the Ising CFTs is given by the symmetry sector of the other, and vice versa. Employing the explicit forms of the partition functions (2.2), one can verify that one obtains now precisely the condensation invariant partition function (12) when summing over all symmetry sectors

 Zu(1)4Ising×2=Z00Z00+Z10Z01+Z01Z10+Z11Z11. (17)

To implement this coupling in the decoupled Hamiltonian (14), we introduce the coupling boundary Hamiltonian

 HB=(Pe−1)σxL−1σx1+(Po−1)σxLσx2, (18)

where and are the independent symmetry operators on even and odd sites, respectively. The Hamiltonian correlates then the boundary conditions and symmetry sectors of the two chains precisely as done in the linear combination (17) for the condensation invariant partition function. By construction the resulting system must thus be critical and described by the CFT. While this Hamiltonian is manifestly non-local, we have shown in earlier work [17, 18] that by using non-local duality transformations it can be mapped precisely to the critical XY chain that is known to be described by the diagonal invariant of the CFT.

We have demonstrated how finding out all possible modular invariant partition functions enables to determine whether a given critical system enables to construct new microscopic models for the non-diagonal partition functions. Before applying this strategy to construct microscopic models for criticality in Potts-type models, we review briefly the connection to anyon condensation that provides an easy way to predict when such constructions might be possible without knowing all the modular invariant partition functions (which in general is a hard task for CFTs with increasing number of primary fields).

### 2.3 Anyon condensation perspective

There is an intriguing connection between the modular invariant partition functions and anyon condensation [37, 38]. Chiral CFTs describe also the 1D edge states of 2D topologically ordered states of matter, with the possible anyonic quasi-particle excitations being in one-to-one correspondence with the primary fields of the CFT. Via this bulk-edge correspondence the anyonic statistics of the quasi-particles are given by the scaling dimensions that are interpreted as their topological spins. If for some quasi-particle the spin is an integer, then it may condense, which results in general in a transition to another topologically ordered state of matter. As this state has a different quasi-particle content in the bulk, it must also have a different CFT describing the edge and hence the framework of anyon condensation can in general be expected to also predict relations between CFTs themselves without referring to a particular realization of 2D topological order.

Indeed, in our previous works [17, 18] we have shown that there is a precise counterpart. This insight enabled us to derive microscopic spin chains for all criticalities in the universality class. The example considered in Section 2.2 is the simplest case of this hierarchy ) that also has a realization in generalized cluster models [6]. In particular, we showed that by correlating boundary conditions with symmetry sectors, those primary fields that are predicted to be confined in a condensation transition can be removed from the spectrum and that this changes the critical description precisely as predicted by the framework of anyon condensation [37, 38]. In our example above, we employed this same correlating of symmetry sectors and boundary conditions to construct the modular invariant partition function (12). In the condensation language the block diagonality of resulting in the diagonal partition function of CFT via the identifications (13) can be understood as follows. The boson with condensed and thus became indistinguishable from the vacuum . This results in the confinement of the primary fields , , and that do not appear in the block-diagonal partition function. The consistency of the resulting theory requires that and are treated as the same new fermionic field and that the field must split into two distinct fields and .

The advantage of the condensation picture is that even if not all modular invariant partition functions are known, the existence of a primary field with an integer scaling dimension that is also a simple current implies that a condensation is possible and hence a block-diagonal modular invariant partition function exists2. As it is easy to figure out which fields should be confined [37, 38], this gives a principle how to construct condensation invariant partition functions by correlating symmetry sectors and boundary conditions and subsequently verify that modular invariance is indeed satisfied. We turn now to apply these ideas to construct generalized 3-state Potts models whose criticalities are described by non-diagonal modular invariants of either the permutation or the condensation type.

## 3 Building block: Critical behavior of the 3-state Potts model

In the example of Section 2.2, we started from decoupled TFI chains to construct the XY model corresponding to the condensation invariant. To construct generalized critical 3-state Potts models, we follow the same strategy, but start from some number of decoupled 3-state Potts models. In this section we review the parafermion and criticality that appear in the 3-state Potts model (1) for different overall signs. We also summarize how the partition function depends on symmetry sectors and boundary conditions, which is needed for our construction.

For negative overall sign, the 3-state Potts model is described by the diagonal invariant of the parafermion CFT [30], while for positive sign it is described by the diagonal invariant of the CFT [32]. For both cases these are the only independent modular invariants. To describe the field content of these CFTs on equal footing, it is convenient to view the parafermion CFT as the coset , which holds for any positive integer [29]. This enables to describe the primary field content of theories in terms of the simpler field content of and CFTs. For generality, we provide these properties for generic , although in this work we mainly focus on the case .

The CFT has the central charge and contains primary fields [2]. We denote these by for . The scaling dimension of each primary field is given by and they obey the fusion rules

 φl1×φl2=φ|l1−l2|+φ|l1−l2|+1+⋯+φmax(l1+l2,2k−l1−l2) . (19)

For example, one of the new models we construct has a critical point, as shown in Table 1. This CFT has four primary fields with scaling dimensions .

On the other hand, CFT is the compactified chiral boson theory with primary fields [2]. The primary fields are labeled , where is an integer defined modulo . Here we choose a convention that lies in the range , which means that the scaling dimensions are given by and the fusion rules can be written as

 φm1×φm2=φm1+m2 . (20)

For example, for the CFT describing the 3-state Potts model for a positive overall sign, there are six primary fields with scaling dimensions .

The parafermion CFT has the central charge and it is equivalent to the coset . Its primary fields are then labeled by the labels of the and theories, namely . In addition to the constraints on and we already introduced, due to the coset the labels also have to satisfy the additional constraint and the identification . Using this, one can choose labels such that , in which case there are primary fields whose scaling dimensions are given by . The fusion rules follow directly from the fusion rules of the and theories and are given by

 φl1m1×φl2m2=φ|l1−l2|m1+m2+φ|l1−l2|+1m1+m2+⋯+φmax(l1+l2,2k−l1−l2)m1+m2 . (21)

For the parafermion CFT describing the 3-state Potts model for a negative overall sign, there are six primary fields with scaling dimensions . In the literature these are often also labeled as . For completeness, we note that the modular matrix is determined by the scaling dimensions as discussed in Section 2.1, and the matrix is given by

 S=1√3+6ϕ⎛⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜⎝111ϕϕϕ1ωω2ϕωϕω2ϕ1ω2ωϕω2ϕωϕϕϕϕ−1−1−1ϕωϕω2ϕ−1−ω−ω2ϕω2ϕωϕ−1−ω2−ω⎞⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟⎠ (22)

where and . The ordering of the fields is .

While the fusion rules are not of paramount importance to our construction, we have have included them to highlight a possible ambiguity in using the labels and to label the primary fields . If a permutation of the labels leaves the fusion rules invariant, i.e. there is a fusion rule symmetry, then the partition function corresponding to the permuted labeling may or may not be identical to the diagonal one. This can be checked by writing out the partition functions (6) as polynomials in and and see whether two different labelings give rise to the same partition function or not. Thus the fusion rule symmetries provide easy ansatzes to try and see whether a given CFT admits non-diagonal partition functions. For instance, for the parafermion CFT with primary fields , where and , the only permutation of the fields that leaves the fusion rules invariant is . However, because the characters associated with and are identical, this permutation gives again the diagonal partition function. The fusion rule symmetries for and CFTs, and for their product theories relevant to the present work, are discussed in more detail in Appendix A.

### 3.1 Partition functions for different symmetry sectors and boundary conditions

The dependence of the partition functions on the symmetry sectors and twisted boundary conditions is known for the parafermion and the CFTs [35, 41]. A -state Potts model always has an on-site symmetry described by the operators , where and the phase factor appearing in the commutation relations of the 3-state clock variables is replaced by . This means that there are always symmetry sectors, that we label by , corresponding to the eigenvalues of . Similarly, there are possible twisted boundary conditions that we label by .

Using this notation, the partition function for symmetry sector with twisted boundary conditions is given compactly by

 Z~QQ=12∑0≤l≤k−k

where the factor accounts for the double counting due to the identification . We note that the effect of twisting the boundary conditions only involves changing the labels of the parafermionic fields . In the case of the CFT, the partition functions for fixed symmetry sector and boundary condition take a completely analogous form. One simply drops the label from the field to obtain , where the sum is constraint by .

Analogous to the example presented in Section (2.2), to construct critical generalizations of the Potts models we start from several copies of either the 3-state Potts models at either or critical point, or mixtures of them. The criticality of such systems is then described by products of the respective CFTs. The primary field content of product theories is obtained simply by taking all possible products of the primary fields of each CFT in the product. The scaling dimensions of these product are given as the sum of the scaling dimensions of the constituent fields and the fusion rules are just direct products of their fusion rules. Since the symmetries sectors and boundary conditions of each CFT in the product are independent, the partition function of a product theory of CFTs in a given symmetry sector and for given boundary conditions is given simply as the product . For instance, a product of two parafermion CFTs has primary fields with scaling dimensions . There are nine symmetry sectors labeled by with nine independent boundary conditions . For each choice of them the partition function is given by .

## 4 Generalized 3-Potts models for non-diagonal modular invariants

This section contains the detailed derivation of the critical generalizations of 3-state Potts models listed in Table 1. We consider products of up to three parafermion and CFTs that can be realized by the 3-state Potts model for different overall signs. To obtain the non-diagonal modular invariant partition functions, for products of two CFTs we determine all the distinct invariants by numerically solving the equations (8). For products of three CFTs, brute force solution becomes challenging due to the size of the modular matrices. Instead, guided by the anyon condensation perspective (see Section 2.3) and the fusion rule symmetries (see Appendix A), we consider all possible boundary terms and determine which of these give rise to non-diagonal modular invariants. For each case, we write the invariant as the linear combination over symmetry sectors and boundary conditions (see Section 3.1) and, similar to the example discussed in Section 2.3, construct a Hamiltonian term that implements the change in the nature of criticality in the microscopic setting. Finally, we introduce canonical duality transformations to find unitarily equivalent local models for each non-diagonal modular invariant.

### 4.1 The permutation invariants of Z×23 and u(1)×26 CFTs

We begin by considering the doubled parafermion CFT that is realized by the next-nearest neighbour Hamiltonian

 HZ×23=−L∑i=1(XiX†i+2+Zi+h.c.) , (24)

that describes two critical decoupled 3-state Potts chains. By brute force calculation we find that there are altogether 16 different modular invariants . As expected from CFT containing no primary fields with integer scaling dimensions (so there are no bosons to condense from the anyon condensation perspective), all the invariants are permutation invariants. However, because of the fusion rule symmetries, there are only two distinct modular invariant partition functions (see Appendix A for details). All invariants related by the 8 permutations

 π(φl1m1,φl2m2) =(φl1s1m1,φl2s2m2) π(φl1m1,φl2m2) =(φl2s1m2,φl1s2m1) , (25)

where and , give rise to the same a partition function as the diagonal invariant of the CFT. On the other hand, the 8 permutations

 π(φl1m1,φl2m2) =(φl1s1m2,φl2s2m1) π(φl1m1,φl2m2) =(φl2s1m1,φl1s2m2) , (26)

give rise to a permutation invariant corresponding to the partition function

 ZπZ×23=∑l1,l2=0,…,3−3

where again .

Employing the knowledge how the partition functions depend on the symmetry sectors and boundary conditions (23), it is straightforward to verify that this permutation invariant can be expressed as the linear combination

 ZπZ×23=∑Q1,Q2=0,1,2Z−Q2Q1ZQ1Q2 . (28)

In other words, the boundary condition of the first chain is given by the symmetry sector of the second chain, and vice versa, but in a twisted manner. The Hamiltonian term implementing this coupling between the two 3-state Potts chains is given by

 HB=−(P†2−1)XL−1X†1−(P1−1)XLX†2+h.c., (29)

where are the independent symmetry operators for each chain. While the Hamiltonian is manifestly non-local and breaks translation invariance, it can be brought into a translationally invariant form with the duality transformations

 Z2j−1 =~X†2j−1~X2j Z2j =~Z2j−1~Z2j X2j−1 =(∏kj~X2k−1~X†2k)P1 (30)

where . It is straightforward to verify that the dual operators and satisfy again the clock algebra (2). In terms of them one obtains the Hamiltonian

 HπZ×23=HZ×23+HB=−∑i(~Xi~X†i+1+~Zi~Zi+1+h.c.), (31)

which is critical by construction and described by the non-trivial permutation invariant (27) of the doubled parafermion CFT.

There is a clear difference in the energy spectrum of the criticality described by the diagonal or the permutation invariants . Recalling that the lowest energy states of each conformal tower labelled by the scaling dimensions have energy , Table 2 shows the degeneracies of the lowest lying states. For the diagonal invariant there are four conformal towers with scaling dimensions and four with . These are absent for the permutation invariant and replaced by eight towers with and , that correspond to non-diagonal combination of the chiral halves. This means that if one compares the spectra of these models, the lowest lying excitations that are present in are absent in . We confirmed this behaviour by explicitly diagonalizing the Hamiltonian (31) for system sizes up to sites, showing that the rescaled finite-size spectrum indeed clearly differs from the diagonal invariant and precisely corresponds to the non-trivial permutation invariant (27).

#### The permutation invariant of u(1)×26 Cft

The CFT is realized by two decoupled critical Potts chains described by the Hamiltonian , i.e. just by changing the overall sign of the Hamiltonian (24). Again, there are no fields with integer scaling dimensions, so there are no condensation invariants, but after accounting for the fusion rule symmetries we find again a single independent permutation invariant partition function . While writing it explicitly out is not particularly illustrative, its form can be seen from Table 2, which shows how the energy spectrum of a critical system corresponding to it differs from one described by the diagonal invariant. As above, some of the lowest lying states of the diagonal invariant are missing for the permutation invariant, which exhibits also conformal towers corresponding to non-diagonal combinations of the chiral halves.

To construct a microscopic model that realizes the permutation invariant, we find that the same coupling (28) between the two chains is required. Since changing the overall sign in (24) does not affect the duality transformations, the construction in the previous subsection applies directly also here. Thus the permutation invariant partition function is realized by the Hamiltonian .