Exploring Duality Symmetries, Multicriticality and RG Flows Atc 2

Download as pdf or txt
Download as pdf or txt
You are on page 1of 61

Exploring duality symmetries, multicriticality and RG flows

at c = 2

Jeremias Aguilera Damiaa , Giovanni Galatia ,


arXiv:2401.04166v1 [hep-th] 8 Jan 2024

Ondrej Hulikb , and Salvo Mancania

a Physique Théorique et Mathématique and International Solvay Institutes


Université Libre de Bruxelles, C.P. 231, 1050 Brussels, Belgium
b Theoretische Natuurkunde, Vrije Universiteit Brussel

Pleinlaan 2, B-1050 Brussels, Belgium

Abstract
In this work, we study the realization of non-invertible duality symmetries along the
toroidal branch of the c = 2 conformal manifold. A systematic procedure to construct
symmetry defects is implemented to show that all Rational Conformal Field Theories
along this branch enjoy duality symmetries. Furthermore, we delve into an in-depth
analysis of two representative cases of multicritical theories, were the toroidal branch
meets various orbifold branches. For these particular examples, the categorical data
and the defect Hilbert spaces associated to the duality symmetries are obtained by
resorting to modular covariance. Finally, we study the interplay between these novel
symmetries and the various exactly marginal and relevant deformations, including
some representative examples of Renormalization Group flows where the infrared is
constrained by the non-invertible symmetries and their anomalies.
Contents

1 Introduction 3

2 Review of duality symmetries for the c = 1 bosonic CFT 6

3 c = 2 Toroidal CFT and Duality symmetries 10


3.1 Discrete gauging and duality symmetries . . . . . . . . . . . . . . . . . . . . 16

4 Duality symmetries at multicriticality 20


4.1 The quadri-critical point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.1.1 Duality symmetry and fusion category . . . . . . . . . . . . . . . . . 22
4.1.2 Modular bootstrap at the quadri-critical point . . . . . . . . . . . . 24
4.1.3 Marginal and relevant deformations at the quadri-critical point . . . 26
4.2 The bi-critical point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.2.1 Duality symmetry and fusion category . . . . . . . . . . . . . . . . . 32
4.2.2 Modular bootstrap at the bi-critical point . . . . . . . . . . . . . . . 36
4.2.3 Marginal and relevant deformations at the bi-critical point . . . . . 40

5 Duality symmetries of SU (3)1 WZW 43

A Details on compact bosons 46


A.1 The self-dual point at c = 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
A.2 The SU (3) point at c = 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

B Partition function for RCFTs at c = 2 50

C Tambara-Yamagami categories 52

D Modular functions 56

2
1 Introduction

Quantum Field Theories in two spacetime dimensions often provide a suitable arena to
explore interesting dynamical aspects while still retaining some analytical control. This
becomes more explicit when the field theory enjoys conformal invariance, i.e. it is a Con-
formal Field Theory (CFT). In such a case, the theory dynamics get highly constrained by
the infinite dimensional Virasoro algebra [1, 2]. Moreover, when the spacetime is taken to
be the Eucliedean torus1 , covariance under large diffeomorphisms, encoded in the modular
group, places additional restrictions on the space of states of the theory. For some pur-
poses, this proves enough to determine the structure of the Hilbert space or the behavior
of some observables analytically over certain parametric regimes [4], remarkably the high
temperature density of states as pioneered by Cardy [5] (see also [6] for a novel general-
ization to higher dimensions). More generally, non-trivial solutions of these constraints
can be found either analytically or numerically via more refined conformal bootstrap tech-
niques [7–11]. A special set of conformal field theories in two dimensions corresponds to the
Rational Conformal Field Theories (RCFT) [9]. A key property of these theories is that
the Hilbert space decomposes into a finite set of representations of some chiral algebra.
For c < 1, all unitary conformal field theories are of this kind and the partition function is
accounted by a finite set of representations of the Virasoro algebra. These are the minimal
models. For higher values of the central charge, i.e. c ≥ 1, certain points on the conformal
manifold meet some extra structure due to the occurrence of additional, generically higher
spin, conserved currents assembling an enhanced chiral algebra. Notably, the dynamics of
these remarkable theories is described in terms of a finite set of conformal primaries with
respect to the enhanced chiral algebra, hence corresponding to RCFTs. We will focus on
examples of this kind throughout this work.
Concomitantly, rational CFTs played a prominent role in the discovery and character-
ization of topological defects, leading to novel forms of global symmetry, and the develop-
ment of alternative descriptions in terms of higher dimensional topological field theories
(TQFT) (see e.g. [12–22]). In this work, we will focus on the former aspect, namely the
description of interesting global symmetry structures and the exploration of their conse-
quences. A comprehensive treatment of the general formalism and some applications for
c ≤ 1 CFTs can be found in e.g. [18–20, 23], whereas we will be mainly concerned about
examples at higher values of the central charge, specifically bosonic RCFTs 2 at c = 2.
A prominent role is played by duality symmetries. This type of global symmetry arise in
1
Of course, this notion extends to any Riemann surface of arbitrary genus, with the large diffeomor-
phisms comprised in the Mapping Class Group (see for instance [3]). However, considering the two-
dimensional torus will be enough for the purposes of this work.
2
Recently, non-invertible duality symmetries were constructed for a particular subset of irrational field
theories at c = 2 [24]. We will not consider this type of theories.

3
theories featuring self-duality upon gauging a particular subset of its global symmetry. In
particular, the conformal manifold of c ≥ 1 CFTs accommodates the action of non-trivial
duality groups (see e.g. [25] for a review), making the occurrence of duality symmetries
quite natural3 .
Due to the fact that the action associated to duality symmetries usually involves an
orbifold by some (invertible)4 global symmetry, the fusion realized by its associated topo-
logical operators does not follow a group law. The underlying structure is generically
accounted for by Category Theory. A distinctive feature of conformal field theories in
two dimensions is that the construction of topological defects is highly conditioned by the
modular bootstrap5 , as a natural consequence of modular covariance [12, 13, 18, 19, 33].
This program not only enables the determination of consistent topological defects, but
also allows to obtain the additional topological data involved in the underlying categori-
cal structure [20]. We will apply these tools extensively for the examples studied in this
paper. Remarkably, solutions to the various constraints imposed by associativity of the
OPE and modular covariance belong to a finite discrete set, a fact referred to as Ocneanu
rigidity [34]. This naturally endows categorical symmetries with a notion of robustness,
rendering well defined Renormalization Group (RG) invariants, such as ordinary symme-
tries. Concomitantly, duality symmetries may participate in anomalies [20, 35–39], hence
leading to interesting constraints on RG flows triggered by duality symmetric operators.
In this work we explicitly show how anomalous duality symmetries constraint symmetry
preserving RG flows.
A special loci pertaining to the conformal manifold of CFTs with central charge
c ≥ 1 corresponds to the set of multicritical points. These (generically rational) theo-
ries may connect various branches of the conformal manifold. A prototypical example is
the Kosterlitz-Thouless (KT) theory sitting at the merging point between the circle and
the orbifold branches of the c = 1 bosonic conformal manifold. In particular, it has been
found that the KT point preserves a series of non-invertible duality symmetries some of
which are further preserved by the exactly marginal deformation spanning the orbifold
branch [20]. Inspired by this kind of analysis, we present a characterization of the duality
symmetry structure featured at certain multicritical points along the toroidal branch of
the conformal manifold of c = 2 bosonic CFTs. Contrary to what happens at c = 1, the
richer duality group at c = 2 allows for the occurrence of categorical symmetries graded
3
Let us mention that duality symmetries are not exclusive of lower dimensional theories, see for instance
[26–30] for some constructions and applications in four dimensions
4
See the recent work [31, 32] for some alternative constructions involving gauging non-invertible sym-
metries.
5
A priori, these conditions find a natural generalization in higher dimensions, achieved by putting the
theory on appropriate spacetime manifolds. However, we are not aware of a precise treatment along these
lines for higher dimensional theories.

4
by a non-commutative structure, as we will illustrate in the examples below (see [29, 40]
for recent examples of non-commutative duality symmetries constructed in higher dimen-
sions).
Topological defects in CFT are a subset of the conformal defects. On general grounds,
the characterization of conformal defects plays a significant role in many fields, ranging
from the description of finite size effects in critical systems [41–43] to the classification
of branes in String Theory [44, 45]. Furthermore, recent studies established a connection
between self-duality symmetries in bosonic CFT’s and interesting topological transitions
between fermionic theories by means of a generalized Jordan-Wigner map [46, 47]. By
similar considerations, the structures uncovered in this work may shed light onto more
intricate transitions taking place along the fermionic conformal manifold at higher values
of the central charge. This type of systems have engendered some interest due to their
potential application in designing quantum computing devices [48]. We hope to explore
these aspects in more detail in the near future.
This paper is organized as follows. In section 2 we briefly review some properties of
the c = 1 theory, emphasizing its symmetry structure and the modular properties of its
(twisted) torus partition function that will be important for the discussion of the c = 2
theory. In section 3 we review properties of the c = 2 toroidal branch of the conformal
manifold and we explicitly show that any RCFT point within this branch enjoys non-
invertible duality symmetries arising form the self-duality of these theories under gaugings
of subgroups of their symmetries. To achieve that, we introduce the generalized metric E
as a useful parametrization of the conformal manifold. We rephrase the generic condition
for the presence of a duality symmetry defect D as the existence of an associated matrix
D ∈ O(2, 2, Q), parametrizing a given topological manipulation, such that

(D−1 )⊺ ED−1 = E . (1.1)

In section 4 we delve into the detailed analysis of some multicritical points of the
toroidal branch which we dub quadri-critical and bi-critical point, where the theory pos-
sess extra exactly marginal parameters generating orbifold branches. Being RCFTs, these
points have non-invertible duality symmetries which we analyze in detail, describing the
underlying categorical structure and their consistency conditions coming from the modular
bootstrap. We then describe duality preserving marginal and relevant deformations. In
the quadri-critical point the theory factorize in the product of two c = 1 RCFT. Therefore
we can explicitly check some of the constraints coming from the above-mentioned duality
symmetry, for instance verifying the ones coming from the non-invertible ’t Hooft anoma-
lies. On the other hand, in the bi-critical point the theory is genuinely c = 2, in the sense
that it is not a product of c = 1 theories, and some considerations coming from the duality
symmetry are actually non-trivial predictions.

5
Rorb.

R=1 R=2 R
SU (2)1 pt. KT pt.

Figure 1: Pictorial representation of the c = 1 conformal manifold.

In section 5 we briefly comment on the enhanced symmetry point where the theory is
equivalent to SU (3)1 WZW model. As opposed to the c = 1 case, this point still enjoys
some non-invertible duality symmetries which are not of the Verlinde type and that we
briefly analyze. We end with some appendices containing some technical details.

2 Review of duality symmetries for the c = 1 bosonic CFT

We begin by summarizing the main aspects concerning the dynamics of non-invertible


duality symmetries in bosonic c = 1 CFT’s, for a comprehensive analysis see [18, 20].
These theories are realized in terms of a compact scalar φ ∼ φ + 2π with action
Z
R2
S= d2 x dφ ∧ ⋆dφ (2.1)

where R is the radius of the target space circle. We chose the conventions in which the self-
dual radius is R = 1 and the circle branch in the conformal manifold is parametrized by
R ≥ 1. The connected piece6 of the conformal manifold comprises an additional branch,
namely the orbifold branch, parametrized by an exactly marginal parameter Rorb ≥ 1.
These two branches merge at R = 2 (Rorb = 1), the Kosterlitz-Thouless (KT) point [49],
which is then a multicritical point, see Figure 1. Theories in this conformal manifold
admit a T -dual description in terms of a dual variable φe with radius Re = R−1 . In terms of
the chiral left (right) moving fields X (X), the scalar field and its dual have the following
decomposition
1 R
φ = √ (X + X) , φ̃ = √ (X − X) (2.2)
2R 2
6
In addition, there are three disconnected exceptional orbifold points pertaining to the c = 1 conformal
manifold. See [20] (section 5) for a description of the categorical structures featured by these theories.

6
hence T -duality acts on the right movers as7 X → −X.
A generic point along the circle branch features a Un (1) × Uw (1) symmetry , associated
to shifts of φ and φe respectively8 . The spectrum of (genuine) conformal primaries along
the circle branch contains vertex operators of the form9

e 1 n  1 n 
Vn,w = einφ eiwφ = eipX eipX , p= √ + wR , p= √ − wR (2.3)
2 R 2 R
with n, w ∈ Z. The conformal dimension and spin reads
 
1 1 1 n2
h = p2 , h = p2 , ∆ = h + h = + w 2 2
R , s = h − h = nw . (2.4)
2 2 2 R2

In this presentation, n and w span the charges of Vn,w under the U (1)n and U (1)w respec-
tively and by Dirac quatization condition we have nw ∈ Z. Upon gauging a ZN ⊂ Un (1),
R
the theory gets mapped to one at R′ = N . This is so due to the fact that the ef-
fect of the gauging is twofold. On the one hand, it projects the spectrum to invari-
ant states, namely the ones with n = N n′ (n′ ∈ Z). In addition, it also incorporates
twisted sectors, namely Virasoro multiplets corresponding to vertex operators with frac-

tional winding number w = wN (w′ ∈ Z mod N ). Remarkably, the incorporation of the
twisted sectors is required by modular invariance in two dimensions, see for instance [1].
One can retrieve the charge lattice to its original form by rescaling n → n′ = Nn and
w → w′ = N w, hence leading to the aforementioned rescaling of the radius. Analogously,
gauging ZN × ZM ⊂ Un (1) × Uw (1) leads to R′ → M N R. Notice that the latter combined
gauging is only consistent for gcd(N, M ) = 1 due to the mixed anomaly involving the two
U (1) symmetries10 . In the following, we will use the symbol σN,M to denote this kind of
operations.
p
At the special values R = N/M with N, M ∈ Z, the chiral algebra is enhanced √
to U (1)2K (K = N M ) by additional holomorphic conserved currents J± = e±i 2KX of
spin K. In particular, for N = M = 1, this corresponds to the self-dual point, where
the additional currents have spin one and combine into an SU (2)1 chiral algebra (see
Appendix A). The theory at any of these special points becomes rational and the Hilbert
space decompose into a finite number of representations of the chiral algebra U (1)2K . At

R = N , the partition function is given by a diagonal modular invariant in terms of the
7
There are actually multiple consistent definitions of T -duality. In this article we chose the one that
acts on the right movers and moreover, defines a (non-anomalous) Z2 action at the self-dual point. For a
detailed discussion about alternative choices see [50].
8
Equivalently, there are two chiral symmetries U (1) × U (1) acting respectively on X and X.
9
Throughout this paper, normal ordering is implicit in the definition of vertex operators.
10
By the same token, operators of the form Vn,0 and V0,w are not mutually local except for nw ∈ 2Z.
See Appendix A.

7
characters of such representations. More generally, the partition function on the Euclidean
torus with complex structure τ takes the form
2M
X −1 2N
X −1
Z= χN a+M b χN a−M b , (2.5)
a=0 b=0

with the chiral characters defined as


1 X 1 (k+2Kr)2
χk (τ ) ∼ χ−k (τ ) ∼ χk+2K (τ ) = q 4K , q = e2πiτ . (2.6)
η(τ )
r∈Z

Each block in the sum (2.5) accounts for the contribution of primary states with
U (1)n × U (1)w charges satisfying M n = N a + M N (r + r ′ ) and N w = M b + M N (r − r ′ ),
for r, r ′ ∈ Z 11 .
p
Interestingly, the circle branch theories at R = N/M feature non-trivial duality
symmetries [16,18,20]. These are found by combining T -duality with gauging ZN × ZM ⊂
Um (1) × Uw (1), i.e. T ◦ σN,M . In fact, it is straightforward to check that these two
combined actions leave the radius invariant. Away from the self-dual point, the defects
implementing these global symmetries are non-invertible and, together with the generator
η(η̃) of the invertible ZN (ZM ), describe a Tambara-Yamagami (TY) category with fusion
N
X −1 M
X −1
D2 = η n η̃ m , η N = η̃ M = 1 , Dη = ηD = Dη̃ = η̃D = D (2.7)
n=0 m=0

where D denotes the defect implementing the T ◦σN,M duality symmetry. The TY category
described by the above objects is also characterized by additional topological data, namely
a bicharacter χab (a, b ∈ ZN × ZM ) and Frobenius-Schur (FS) indicator ǫ ∈ {−1, 1}. We
refer the reader to appendix C for a more detailed description of these categories. When
acting on vertex operators, the action reads12

D : Vn,w → K(−1)nw V N w M n , n ∈ N Z , w ∈ M Z (2.8)
M , N

and mapping it to a non-genuine operator otherwise. Note the occurrence of the K
prefactor associated to the quantum dimension of D. The additional phase (−1)nm arises
11
More generally, by equating the conformal weights of the operators involved in the chiral characters
of U (1)2M N with their expressions in terms of the global charges one finds
k + k′ k − k′
χk χk ′ ⇒ M n = + M N (r + r ′ ) , Nw = + M N (r − r ′ )
2 2
where the integers r, r ′ correspond to each term in the sums defining the characters.
12
More generally, there is a continuous set of consistent duality defects constructed by stacking D with
invertible operators implementing chiral rotations and/or charge conjugation. See [20] for a detailed
discussion on this point.

8
due to the fact that momentum and winding modes, which are exchanged by T -duality,
are not mutually local (see appendix A).
A topological defect implementing a symmetry in quantum field theory must obey cer-
tain consistency conditions. In two dimensions, these are imposed by modular invariance
or, more generally, modular covariance, finding their natural implementation in terms of
the modular bootstrap. More precisely, torus partition functions twisted by topological
defects along non-trivial cycles are mapped to each other by the modular group SL(2, Z)

iφ(γ) aτ + b
Z(D,D)
e (γ · τ ) = e Z(De −b Dd ,De a D−c ) (τ ) , γ ·τ = (2.9)
cτ + d
where we allowed for the occurrence of an anomalous phase φ(γ). This phase is not going
to play any role in the following, hence being omitted from now on. Whenever negative
exponents occur in the right hand side of (2.9), those should be interpreted as the insertion
of the orientation reversed defect, as the inverse might not exist. An important implication
of the above is depicted in the following diagram:

τ →− 1
τ
Z(1,D) = D −−−−→ ≡ Z(D,1) (2.10)

where, in the last equality, we emphasized the fact that a consistent topological defect
should lead to a well defined twisted Hilbert space. More precisely, being topological,
it commutes with the stress tensor, hence the defect Hilbert space must have a natural
decomposition in terms of Virasoro characters
X
Z(D,1) = nh,hVh V h (2.11)
h,h

where, by the above considerations, the coefficients nh,h must be positive integers. This
becomes a key test under which any proposed symmetry defect should be contrasted, as
we will show in many examples in this paper. Moreover, similar applications of (2.9)
enable extracting more refined topological data associated to a given symmetry structure
[18–20, 51–54].
As an instructive example, we apply the above analysis to the defect implementing the
symmetry action (2.8). Similar computations can be found in [20], though we are not aware
of an explicit application to the non-diagonal case. By the action of T -duality, namely
X → −X, the insertion of the duality defect D along the spatial cycle projects into the

9
p = 0 subspace. This condition sets N a−M b = 0, hence the twisted partition function gets

contributions only from the identity characters, spanned by the states |p = 2M N r, p =
0, N, Ni, r ∈ Z. Here N (N) denote, as usual, the number of (anti-)holomorphic oscillator
modes. In addition, T -duality induces a (−1) phase for states with an odd number of
right-moving oscillator modes [50]. Finally, there is the phase factor in (2.8) due to the
non-locality between winding and momentum modes. Putting all together one obtains
√ !
NM X M N r M N r2
Z(1,D) (τ ) = (−1) q ϑ4 (2τ ) , (2.12)
|η(τ )|2
r∈Z

where we already performed the sum over the oscillator modes. See appendix D for a list
of the relevant modular functions and their modular properties.
For the case of even M N we get

NM τ →− τ1 1 τ
 τ

Z(1,D) = ϑ 3 (2N M τ )ϑ 4 (2τ ) −−−−→ Z (D,1) (τ ) = ϑ 3 2N M ϑ 2 2 , (2.13)
|η|2 2|η|2

whereas for odd M N one gets instead



NM τ →− τ1 1 τ
 τ

Z(1,D) = 2
ϑ 4 (2N M τ )ϑ 4 (2τ ) −−−−→ Z(D,1) (τ ) = ϑ2 2N M ϑ2 2 . (2.14)
|η| 2|η|2

It can be easily checked that both expressions comply with the condition shown in (2.10),
hence leading to a well defined trace over a defect Hilbert space.

3 c = 2 Toroidal CFT and Duality symmetries

The conformal manifold of c = 2 bosonic CFT’s is significantly richer than the one for
c = 1 described in the previous section. In particular, the connected non-exceptional
component comprises 28 non-equivalent branches, meeting at various multicritical regions.
We refer the reader to [55] for a comprehensive overview. A distinguished branch is the
toroidal branch, where the target space corresponds to a torus with complex structure
τ and Kahler modulus ρ, hence spanning a four-dimensional variety. This branch also
accommodates a collection of special points with enhanced global symmetry, as we will
comment momentarily.
Before delving into particular examples, let us review some aspects pertaining to
generic theories along the c = 2 toroidal branch (for more details we refer to [25, 55]).
The field content of these theories consists on two compact scalar fields φi ∼ φi + 2π
(i = 1, 2) with action
Z Z
1 i
S= δµν Gij ∂µ φi ∂ν φj + ǫµν Bij ∂µ φi ∂ν φj , (3.1)
4π 4π

10
where G, B are respectively the metric and Kalb-Ramond fields parametrizing the target
space two-torus ! !
1 τ 1 0 b
G = R2 B= . (3.2)
τ1 τ12 + τ22 −b 0
It is customary to repackage the geometric data in terms of the complex structure and
Kahler modulus
√ √ 
τ = τ1 + iτ2 , ρ = ρ1 + iρ2 = b + i G , G = R 2 τ2 . (3.3)

These two complex variables parametrize the 4-dimensional toroidal branch of the c = 2
conformal manifold. Motion along this branch is generated by the four exactly marginal
operators of the form ∂φi ∂φj . Note that when τ1 = ρ1 = 0 the theory factorizes as a
tensor product of two c = 1 compact bosons at radii R1 = R and R2 = Rτ2 .
These theories admit equivalent descriptions in terms of dual variables φei ∼ φei + 2π.
In addition, the real fields φi , φei admit the following decomposition into left and right
movers
1  1 2 
φ1 = √ τ2 (X 1 + X ) − τ1 (X 2 + X )
2τ2 ρ2
1 2
φ2 = √ (X 2 + X )
2τ2 ρ2
1  1 2 
φe1 = √ ρ2 (X 1 − X ) − ρ1 (X 2 + X )
2τ2 ρ2
1  1 2
φe2 = √ (ρ1 τ2 + ρ2 τ1 )X 1 + (ρ1 τ2 − ρ2 τ1 )X + (ρ2 τ2 − ρ1 τ1 )X 2 − (ρ1 τ1 + ρ2 τ2 )X .
2τ2 ρ2
(3.4)

At a generic point along this branch, the theory possess a continuous global symmetry of
the form
U (1)n × U (1)w ≡ U (1)n1 × U (1)n2 × U (1)w1 × U (1)w2 (3.5)

with Uni (1) : φi → φi + αi and Uwi (1) : φei → φei + α


ei , (αi , α
ei ∈ [0, 2π)). Note that the
exactly marginal operators spanning the toroidal branch are naturally neutral under these
symmetries. The spectrum of charged conformal primaries is therefore determined by four
integer charges n = (n1 , n2 )T and w = (w1 , w2 )T
Tφ e

Vn,w = ein eiw , (3.6)

where φ = φ1 , φ2 , φe = (φe1 , φe2 ). Equivalently, these primaries are in one to one cor-
respondence with sites in the even, self-dual, integer charge lattice with signature (2, 2)

11
spanned by the following left- and right-moving momenta
!
1 n1 τ2 + ρ1 τ2 w2 + ρ2 (w1 + w2 τ1 )
p= √ (3.7)
2τ2 ρ2 n2 − τ1 n1 − ρ1 (w1 + w2 τ1 ) + ρ2 τ2 w2
!
1 n1 τ2 + ρ1 τ2 w2 − ρ2 (w1 + w2 τ1 )
p= √ .
2τ2 ρ2 n2 − τ1 n1 − ρ1 (w1 + w2 τ1 ) − ρ2 τ2 w2

In terms of these momenta, the conformal dimensions read


1 2 1
h= (p + p22 ) = [n + (G + B)w]⊺ G−1 [n + (G + B)w]
2 1 4 (3.8)
1 1
h = (p21 + p22 ) = [n − (G − B)w]⊺ G−1 [n − (G − B)w]
2 4
and the spins s = h − h = ni wi ∈ Z are integrally quantized since the theory is bosonic.
There are several criteria for defining rationality of a generic CFT. A natural one
states that a theory is rational if its Hilbert space decompose into a finite number of
non-equivalent irreducible representations of a certain (generally enhanced) chiral algebra
[56, 57]13 .
Analogously to the c = 1 theories at R2 ∈ Q, there are distinguished points in the
c > 1 conformal manifold where the theory becomes rational. At c = 2, the criterion for
rationality of a toroidal CFT described by (3.1) can be rephrased in terms of geometric
properties of its charge lattice. Alternatively, it can be proven that a c = 2 toroidal CFT
is rational if G ∈ GL(2, Q) and B ∈ Skew(2, Q). The several criteria just mentioned can
be proven to be equivalent. Geometrically, it has been shown in [58] that rationality is also
achieved if the target space torus admits Complex Multiplication14 . The latter condition
implies that the moduli τ, ρ satisfy
√ 
τ,ρ ∈Q D , (3.9)
√ 
where Q D denotes the imaginary quadratic number field, for some negative integer

D. The √latter
 is obtained from√the field of the rational numbers by introducing
√  D, i.e.
x∈Q D then x = x1 + x2 D with x1 , x2 ∈ Q. The field structure of Q D then
√ 
naturally descends from the one of Q. In particular, if τ belongs to Q D , then there
exist some integers a, b and c such that

aτ 2 + bτ + c = 0 , a, b, c ∈ Z , gcd(a, b, c) = 1 (3.10)
13
More precisely, in such a case we say that the theory is rational with respect to such enhanced chiral
algebra, even if it might not be rational with respect to Virasoro. These are the prototypical examples
that occur at c ≥ 1.
14
Given a torus T 2 = C/Λ determined by some lattice Λ, T 2 admits Complex Multiplication if there
exists a complex number z ∈ C − R such that zΛ ⊂ Λ.

12
hence
b i √
τ =−
+ −D , D = b2 − 4ac (3.11)
2a 2a
and similarly for ρ. Even if the equations for both complex variables may have different
integer coefficients, they must have the same discriminant D for the theory to be an
RCFT. The condition D < 0 is implied in order for the solutions to correspond to a
physical theory, i.e. τ2 , ρ2 > 0. 15 It can be proven that the set of RCFTs is dense within
the toroidal branch, similarly to the rational theories along the c = 1 circle branch.
An RCFT in the toroidal branch features an extended chiral algebra of the form

u(1)2KL × u(1)2KR (3.12)

where KL,R are some matrices constructed out of G and B at each particular rational
point. We refer to appendix B for some details on the determination of these matrices.
The partition function of a rational theory is accounted for by a finite modular invariant
combination of characters of (3.12). Furthermore, the (anti)holomorphic representations
of the chiral algebra are in one to one correspondence with the sites in the following lattices

LL,R ≡ λL,R ∈ Z2 , λL,R ∼ λL,R + KL,R v , v ∈ Z2 . (3.13)

The partition function then reads


X
Z= χλL χωb λL (3.14)
~
λL ∈DL

where ω̂ : LL → LR is a particular group isomorphism. We refer to appendix B for


a detailed account on the construction of the above partition function. The characters
corresponding to the representations labelled λL,R read16

1 X 2 |λL +KL l|2K −1 1 X 2 |λR +KR r|2K −1


1 1

χλL (τ ) = q L , χ̄λR (τ̄ ) = q̄ R . (3.15)


η(τ )2 2
η(τ̄ )2 2
l∈Z r∈Z

Furthermore, the partition function of the theory is given by a diagonal modular invariant
b = 1) whenever there exists a duality frame such that that τ = f aρ with a the same
(i.e. ω
as in (3.10) and f some positive integer [58].
15
p
As an illustrative example, consider the case of a product of two c = 1 RCFTs at R1 = N1 /M1 and
p
R2 = N2 /M2 , for which τ = iR1 /R2 , ρ = iR1 R2 . This situation trivially fits into the class described
above since

N2 M1 τ 2 + N1 M2 = 0 , M1 M2 ρ2 + N1 N2 = 0 , D = −N1 N2 M1 M2 .

16
We denote |v|2M = v i Mij v j .

13
Within the toroidal branch, there are special (generically rational) points where the
global symmetry gets enhanced to higher rank chiral algebra, hence corresponding to
enhanced symmetry points. More precisely, the algebras realized at these special loci at
c = 2 are of the A-type of maximal rank 2, i.e. su(2), su(2)2 or su(3). Theories featuring
these algebras have been classified (see for instance [55])
2
(τ, ρ) = (i, i) : SU (2)2 × SU (2) ,
(τ, ρ) = (ei2π/3 , ei2π/3 ) : SU (3) × SU (3) ,
1
(τ, ρ) = (τ, τ ) 6= {i, ei2π/3 } , τ1 = {0, } : SU (2) × U (1) × SU (2) × U (1) ,
2
1
(τ, ρ) = (τ, τ ) , τ1 6= {0, } : U (1) × U (1) × SU (2) × U (1) ,
2
1
(τ, ρ) = (τ, −τ ) , τ1 6= {0, } : SU (2) × U (1) × U (1) × U (1) . (3.16)
2
Note that the last two lines above correspond to continuous sets of theories, hence com-
prising both rational and irrational theories.
In spite of being interesting by themselves, the enhanced symmetry points play a crucial
role in the characterization of the conformal manifold. In particular, it has been shown that
all multicritical theories, i.e. where different branches of the conformal manifold intersect,
can be obtained from non-trivial quotients of theories with enhanced symmetry [59, 60].
Along these lines, the additional exactly marginal deformations featured by multicritical
theories are easily constructed in terms of conserved currents at enhanced symmetry points
(see appendix A for an example of this construction at the c = 1 KT point). In this work,
we will mainly focus on two particular examples of multicritical theories, namely the
quadri-critical and the bi-critical points, which are obtained by taking certain Z2 and Z3
quotients of the SU (2)2 and the SU (3) theories respectively (see figure 2 for a pictorial
representation of the points we are interested in).
The duality groups along the c > 1 toroidal branch are significantly larger than the
T -duality at c = 1, hence making room for richer structures of duality symmetries featured
at special points. In particular, for c = 2 the duality group is of the form

O(2, 2, Z) ≃ P (SL(2, Z)τ × SL(2, Z)ρ ) ⋊ ZM I
2 × Z2 , (3.17)

with the first two factors SL(2, Z)τ,ρ acting separately on τ and ρ via
!
aτ + b a b
τ→ , ∈ SL(2, Z)τ
cτ + d c d
!
a ′ ρ + b′ a′ b′
ρ→ ′ , ∈ SL(2, Z)ρ . (3.18)
c ρ + d′ c′ d′

14
Zs3 SU (2)21 (i, 2i)
SU (3)1 Zs2
(i, i)
(w, w) (w, α)

(τ, ρ)

Figure 2: Pictorial representation of a slice of the √toroidal branch containing two multi-
2πi
critical points. Here w = e 3 while α = − 21 + 3 23 as in the main text. The red and
blue lines represent respectively the gauging of the Zs3 and Zs2 shift symmetries defined in
(4.42) and (4.3). At the 4-critical point (i, 2i) three orbifold branches (one 4d and two 2d)
join while at the bi-critical point (w, α) we get just one 2d orbifold branch.

These two actions are quotiented by the central element (τ, ρ) ∼ (−τ, −ρ). The remaining
two Z2 actions correspond to mirror symmetry and target space inversion respectively

ZM
2 : (τ, ρ) → (ρ, τ ) , ZI2 : (τ, ρ) → (−τ , −ρ) . (3.19)

The various orbifold branches in the c = 2 conformal manifold generically preserve a


subgroup of the duality group (3.17) of the toroidal branch (see for instance [55]).
The four dimensional representation of the duality group (3.17) has a natural action
by conjugation over the generalized metric E (see e.g. [61, 62]). The latter corresponds to
an alternative description of these theories in terms of a 4 × 4 matrix acting on the vector
e T . More precisely, in terms of the metric G and antisymmetric tensor B, the
Φ ≡ (φ, φ)
generalized metric reads
!
G − B · G−1 · B −B · G−1
E= . (3.20)
G−1 · B G−1

For a given element O ∈ O(2, 2, Z) acting on Φ as Φ′ = OΦ, the parameters of the theory
get transformed according to

O : E → E ′ = (O−1 )⊺ EO−1 . (3.21)

15
In this formulation, the generators of the duality group (3.17) admit the following four
dimensional representation
 
a −b 0 0 !
−c d 0 0 a b
 
  for ∈ SL(2, Z)τ
0 0 d c c d
0 0 b a
 ′ 
d 0 0 c′ !
 0 d′ −c′ 0  a ′ b′
 
  for ∈ SL(2, Z)ρ (3.22)
 0 −b′ a′ 0  c′ d′
b′ 0 0 a′
   
0 0 1 0 −1 0 0 0
0 1 0 0 0 1 0 0
   
M =  ∈ ZM 2 , I =  ∈ ZI2
1 0 0 0 0 0 −1 0
0 0 0 1 0 0 0 1

As we will discuss further below in this section, this description of the theory enables
to construct a matrix representation of topological manipulations, and more generally
duality symmetries. Precisely, to a given defect D implementing such a manipulation,
one associates a 4 × 4 matrix D ∈ O(2, 2, Q) acting on the fields as described around
(3.21). Notice that the coefficients are in Q instead of Z. This is a consequence of the
fact that, as we will verify in several examples below, duality symmetries usually involve
topological manipulations, such as discrete gauging, which cannot be represented by a
matrix in O(2, 2, Z).

3.1 Discrete gauging and duality symmetries

Similarly to the c = 1 case, gauging a non-anomalous subgroup of U (1)2n × U (1)2w connects


two points in the toroidal branch and therefore can be represented as a 4 × 4 matrix σ
acting on the 4-component vector (φ⊺ , φe ⊺ ), or equivalently on the charge (n⊺ , w⊺ ), as
! ! ! !
φ′ φ n′ n
=σ e ⇐⇒ = (σ −1 )⊺ . (3.23)
e′
φ φ w′ w

Dirac quantization condition after the gauging, implies that

n′⊺ w′ ∈ Z =⇒ σ ∈ O(2, 2, Q) . (3.24)

Therefore any gauging of subgroups of U (1)2n × U (1)2w can be represented as a matrix


σ ∈ O(2, 2, Q). This is a particular example of the matrix representation for topological

16
manipulations introduced before. We emphasize that the map between gaugings and
matrices in O(2, 2, Q) is not bijective: there can be different matrices representing the
same gauging operation17 . A duality symmetry defect D then implies the existence of an
O(2, 2, Q) matrix D which leaves the theory parameters invariant. This condition can be
written, using the generalized metric E, as18

(D−1 )⊺ ED−1 − E = 0 . (3.25)

At this point it is important to emphasize the distinction between the actual symmetry
operation over the Hilbert space (or equivalently on the conformal primaries) implemented
by D, which does not need to be invertible, and its matrix representation D acting on the
global charges, for which an inverse matrix can always be defined. The invariance of the
theory parameters in (3.25) is the manifestation of the self-duality of the theory under the
gauging of a global symmetry subgroup G parameterized by D ∈ O(2, 2, Q). Thoughout
this paper, will show in several examples how the actual subgroup G can be extracted from
the matrix D satisfying (3.25). On the contrary, when D ∈ O(2, 2, Z), this corresponds
to an invertible symmetry which permutes the primary operators. The action of the
corresponding symmetry defect on the set of conformal primary operators is
 p
 V |G|(−1)α Vn′ ,w′ if n′i , wi′ ∈ Z
n,w →
D : (3.26)
 Vn,w → non genuine op. if n′ , w′ 6∈ Zi i

where n′ , w′ are the charges obtained upon acting with the associated matrix as
! !
n′ −1 ⊺ n
= (D ) · , (3.27)
w′ w

and the phase α comes by imposing the consistency of this action with locality (see ap-
pendix A for more details).
As mentioned before, matrices D naturally decompose in terms of duality operations in
O(2, 2, Z) and topological manipulations associated to gauging of discrete subgroups of the
global symmetry, represented by matrices σ ∈ O(2, 2, Q). Let us momentarily concentrate
on the latter. Given a matrix σ ∈ O(2, 2, Q) implementing a gauging of a subgroup
G ⊂ U (1)2n × U (1)2w , it becomes instrumental to extract the form of such a subgroup of
the global symmetry. In the following, we present an algorithmic way of achieving that.
! !
n p/q 0 0 p/q o
17
Equivalently in the c=1, the group O(1, 1, Q) = , represents all the
0 q/p q/p 0
possible gauging of subgroups Zp × Zq ⊂ U (1)n × U (1)w but the diagonal and off-diagonal matrices span
the same set of operations.
18
This condition can be easily generalizable to CFTs with integer central charge c > 2.

17
Consider the set of primary operators
 
σ (σ−1 )⊺
S ≡ Vn,w | {(n, w)} −−−−→ {e} (3.28)

where ea is the set of unital vector charges, namely (ea )i = δi,a (i, a ∈ {1, . . . 4}). Such
operators are the minimal gauge-invariant operators, both in the untwisted and twisted
Hilbert space of the ungauged theory, which become genuine operators after the gauging19 .
A trial version of the subgroup G is parametrized by five integers {k, a, b, c, d} such that
its action on a vertex operator reads
2πi
′ (an1 +bn2 +cw1 +dw2 )
G : Vn,w → Vn,w =e k Vn,w . (3.29)

Being invarinat under the action of G, the operators in S σ satisfy the condition

Vn,w ∈ S σ ⇒ Vn,w = Vn,w



. (3.30)

Hence, given a matrix σ ∈ O(2, 2, Q), determining the set S σ allows to easily derive
the action of the symmetry G, i.e. determining k, a, b, c, d.20 In order to illustrate this
procedure, we apply it to the following example
   1 1

1 1 0 0 2 2 0 0
 1 −1 0 0   1 −1 0 0 
  −1 T  2 2 
σ=  , (σ ) =   (3.31)
 0 0 12 1 
2
 0 0 1 1 
0 0 12 − 21 0 0 1 −1

for which (3.28) corresponds to


 
σ
S = V1,1,0,0 , V1,−1,0,0 , V 1 1,V 1 1 . (3.32)
0,0, 2 , 2 0,0, 2 ,− 2

Now, imposing the invariance condition given in (3.30), it is straightforward to verify that
integer solutions for {a, b, c, d} only exist for k ∈ 2Z. Finally, the maximal non-trivial
solution (see footnote 20) corresponds to

k = 2 , a = b = 1 , c = d = 0 ⇒ G ≡ ZD
2 ⊂ U (1)n1 × U (1)n2 (3.33)

We will encounter the matrix (3.31) again when studying the quadri-critical point in the
next section.
19
For instance, in the c = 1 case a gauging of Zk ⊂ U (1)n acts on the charges as (n, w) → ( nk , kw). There-
fore primaries with charge (k, 0), (0, k1 ) are mapped to (1, 0), (0, 1) and they correspond to the invariant
operators under the action of Zk .
20
Generically, the solution is not unique. More precisely, the condition (3.30) is equally satisfied by any
subgroup H ⊂ G. However, there is always a solution for which gcd(k, a, b, c, d) = 1, which we identify
with the maximal group.

18
We want now to discuss the existence of non-invertible duality symmetries on the
toroidal branch. Even if the condition (3.25) in principle generates all the possible
D ∈ O(2, 2, Q) implying a duality symmetry defect D, explicit solutions of such equa-
tion are generically hard to find. To this aim, we proceed to describe a particular method
by which some solutions can be easily obtained. In particular by performing some topolog-
ical manipulation T , namely a combination of gaugings and dualities, we can connect the
τ , ρb) in the toroidal branch for which some duality sym-
theory at (τ, ρ) to another point (b
metries are manifest. We denote the latter symmetries by D b and their associated matrix
b Therefore the original theory will automatically enjoy a non-invertible symmetry
by D.
defect D corresponding to a matrix D constructed as

b T−1 ,
D = TD (3.34)

where T ∈ O(2, 2, Q) is the matrix representation of the topological manipulation T used


to connect the points (τ, ρ) and (b τ , ρb) in the conformal manifold. For instance, a very
τ , ρb) is one for which the theory factorizes in terms of two c = 1
convenient choice of (b
RCFT’s, namely τb1 = ρb1 = 0. In particular, one can show that such a factorized theory
enjoy duality symmetries coming from the separate gaugings of subgroups of U (1)n1 ×
U (1)w1 and U (1)n2 × U (1)w2 .21
Consider a generic c = 2 RCFT with

p p′
τ= + iτ2 , ρ= + iρ2 , p, q, p′ , q ′ ∈ Z . (3.35)
q q′
where τ2 and ρ2 are determined in terms of the rationality condition reviewed previously in
this section. For any p, q ∈ Z, we can always perform a gauging σ ∗ that makes τ1 , ρ1 ∈ Z.
Indeed this can be achieved by the gauging of Zni ,wi ⊂ U (1)ni ,wi with

w1 pq ′ w2
= ′ , = qq ′ . (3.36)
n1 qp n2

This gauging brings the theory at (τ, ρ) to a theory with

q 2 p′ q 2 p′
τ ′ = qp′ + i τ2 , ρ′ = pq ′ + i ρ2 (3.37)
p p
′ ′
which, by applying the dualities Tρpq Tτqp , is equivalent to the factorized point

q 2 p′ q 2 p′
τb = i τ2 , ρb = i ρ2 . (3.38)
p p
21
This is a trivial consequence of the existence of duality symmetries in any c = 1 RCFT as shown in
section 2.

19
Duality symmetry defects D b of the factorized theory above are easily detected by the logic
′ ′
exposed in section 2 and the topological manipulation just described reads T = Tρpq Tτqp σ ∗ .
This simple computation shows that, because of (3.34), any c = 2 RCFT enjoy duality
symmetries which are more manifest in a factorzied point connected to the original RCFT
via topological manipulations. We will illustrate this procedure when studying some par-
ticular examples in the next section. Before concluding, we emphasize that the duality
symmetries which are manifest in the factorized point generically do not span the entire
set of duality symmetries in a given RCFT. For instance there can be symmetries coming
from non-diagonal gauging in the factorized point which are, however, more manifest if
we look at different points connected by topological manipulations. We will encounter an
example of this kind when we look at the bi-critical point in section 4.2.1.

4 Duality symmetries at multicriticality

As mentioned in the previous section, the intricate structure of the bosonic c = 2 conformal
manifold allows for several special loci where various orbifold branches intersect. Moreover,
the theories sitting at these multicritical loci are generically rational. Examples of this
kind will be the focus of this section, putting particular emphasis on the interplay between
non-invertible duality symmetries and several deformations. Regarding the latter, we will
study the additional exactly marginal deformations arising at multicriticality, together
with some examples of RG flows triggered by duality preserving relevant operators.
We begin from the points on the conformal manifold of c = 2 theories where the global
symmetry is enhanced. These loci are given in (3.16) and we focus on the following:
2
(τ, ρ) = (i, i) : SU (2)2 × SU (2) ,

(τ, ρ) = (ei2π/3 , ei2π/3 ) : SU (3) × SU (3) . (4.1)

Orbifolding by the Z2 and Z3 symmetries of the SU (2)2 point and the SU (3) point respec-
tively, one obtains two multicritical points, where four and two branches of the conformal
manifold interesect. See figure 2 for a pictorial representation of the conformal mani-
fold. Therefore, these are respectively a quadri-critical and a bi-critical points. These two
multicritical points are respectively located in the toroidal branch at (τ, ρ) = (i, 2i) and
2πi √
(τ, ρ) = (ω, α) with ω = e 3 , α = ω + i 3. The reason behind this choice is the fact
that they exemplify many of the properties of c = 2 RCFTs with non-invertible duality
symmetries. At the quadri-critical, point the theory is factorized into two copies of c = 1,
while at the bi-critical point the B-field takes non-zero value and the theory is genuinely
c = 2.

20
4.1 The quadri-critical point
2
We begin by looking at the point (τ, ρ) = (i, i), featuring an SU (2)2 × SU (2) enhanced
global symmetry. We review some features of this point in Appendix A.1. The theory
is factorized into two copies of c = 1, as it can be seen already in the holomorphic
decomposition of the fields
1 a 1 a
φa = √ X a + X , φea = √ X a − X . (4.2)
2 2
Each factor enjoys two Z2 symmetries, i.e. the shift φa → φa + π and the reflection

φa → −φa . On the chiral fields, these act as a shift by 2π/2 and a sign flip. One can
then identify four (non-anomalous) Z2 symmetries at (τ, ρ) = (i, i), which in the chiral
basis take the following form
 √ √ 
a a
Zs2 : X a , X → X a + 2π/2, X + 2π/2 ,
a a
Zorb
2 : X a , X → −X a , −X ,
   √ √ 
1 1
ZR
2
1
: X 1, X → X 1 + 2π/2, X + 2π/2 (4.3)
   
2 2
X 2, X → −X 2 , −X ,
   √ √ 
1 1
ZR
2
2
: X 1, X −X 1 + 2π/2, −X + 2π/2 ,

   √ √ 
2 2
X 2, X → X 2 + 2π/2, X + 2π/2 .

The Zs2 shift symmetry corresponds to a simultaneous translation in field space. The Zorb2
implements a reflection on both fields, analogous to the c = 1 case (see appendix A.1).
The remaining Z2 symmetries implement combinations of the previous actions. For more
details, see [55]22 . As usual, the orbifold by the Zs2 shift symmetry leads to another
theory in the toroidal branch. The remaining Z2 actions define three independent orbifold
branches. Furthermore, the four actions just described are mapped to each other by the
2
SU (2)2 × SU (2) global symmetry. Therefore, performing the quotient by any of them
leads to four different descriptions of the same point of the conformal manifold, where the
toroidal and the three orbifold branches meet.
Note that the orbifold by the Zs2 symmetry is implemented precisely by the matrix
in Eq. (3.31). The resulting theory sits on the toroidal branch at (τ, ρ) = (i, i/2), and
by acting with Sρ : ρ → −1/ρ, the theory is mapped to (τ, ρ) = (i, 2i). One therefore
identifies the theory at (τ, ρ) = (i, 2i) with the quadri-critical point. The overall map of
the charges read
n1 + n2 n1 − n2
n′1 = , n′2 = , w1′ = w1 + w2 , w2′ = w1 − w2 , (4.4)
2 2
22
The reader should be aware that we are using a slightly different notation than in [55].

21
where the non-primed symbols denote the original charges at the enhanced symmetry
point.
The theory at the quadri-critical point is itself rational and, moreover, corresponds to
a diagonal modular invariant since ρ = 2τ [58]. This property is also manifest by the fact
that this theory can be regarded as a product of two rational c = 1 theories at radius equal

to 2. Even if many of the properties of the global symmetries featured at (τ, ρ) = (i, 2i)
naturally descend from its product structure, we will see that there are still some salient
features pertaining to the c = 2 theory, such as the non-Abelian fusion satisfied by the
(non-invertible) duality symmetries. This is just a consequence of the fact that the latter
symmetries descend from a non-Abelian subgroup of the c = 2 duality group (3.17) which
is not a product of two c = 1 dualities.

4.1.1 Duality symmetry and fusion category

As we have already discussed in the general case, RCFT points in the toroidal branch
enjoy duality symmetries, obtained by composing gaugings and dualities in different ways.
We now discuss some of the duality symmetries present in the quadri-critical point (τ, ρ) =
(i, 2i). Since it is a factorized point (i.e. it can be decomposed as a product of two c = 1
theories) with ρ2 , τ2 ∈ Z, it generically hosts duality symmetries coming from diagonal
gaugings, of the form23
   
0 0 21 0 0 0 0 − 21
0 1 0 0 1 0 0 0 
   
DA = M σ2,1 → DA =   , DB = Sρ M σ̃2,1 → DB =  (4.5)
2 0 0 0 0 −2 0 0 
0 0 0 1 0 0 1 0

together with the invertible duality symmetries Sτ and I that do not involve any gauging.
e In
The matrices above represent the action of the associated operators over the (φ, φ).
(4.5) we introduced the following notation for the diagonal gauging

σN,M ≡ gauging ZN × ZM ⊂ U (1)n1 × U (1)w1 (4.6)


eN,M ≡ gauging ZN × ZM ⊂ U (1)n2 × U (1)w2 .
σ

and their corresponding matrix representation reads


N   
M 0 0 0 1 0 0 0
 0 1 0 0 0 N
0 0
   M 
σN,M →   , σ eN,M →  , (4.7)
0 0 M N 0  0 0 1 0
M
0 0 0 1 0 0 0 N

23
For sake of notational simplicity, from now on we we will omit the symbol ◦ denoting composition
dualities and topological manipulations. For instance M σ2,1 ≡ M ◦ σ2,1 .

22
As emphasized before, this set of duality symmetries does not capture the full set
of symmetries of this point; for instance duality symmetries coming from non-diagonal
gaugings can be present24 . It would be nice to develop a systematic method capable of
identifying all duality symmetries, along the lines of [63].
Given a set of duality symmetries closed under fusion, we can discuss the fusion alge-
bra and the underlying fusion category (see appendix C for some details on this type of
categorical symmetries). Since they are symmetries coming from self-dualities, the fusion
algebra will be graded by (a subgroup of) the underlying O(2, 2, Z) duality group at c = 2.
This grading fixes the generic structure of the fusion algebra to be

Dα × Dβ = Dαβ × C0 , (4.8)

where Dαβ is the composed duality symmetry whose matrix representation is obtained
by multiplying the matrix representation of Dα and Dβ and C0 is some combination of
topological lines generating a subgroup of the invertible U (1)2n × U (1)2w symmetry. The
occurrence of the gauging σα in the definition of Dα ensures that this topological defect
absorbs all the lines generating the gauged symmetry Gα ⊂ U (1)2n × U (1)2w , namely

Dα × ηα = ηα × Dα = Dα ∀α ∈ Gα ⊂ U (1)2n × U (1)2w . (4.9)

The fusion of Dα with its orientation reversal can be uniquely fixed by imposing the
consistency with the above property of absorption, finding
X
Dα × D α = ηα Gα ⊂ U (1)2n × U (1)2w . (4.10)
α∈G

Let us now delve into the specific examples presented in (4.5). From the gauging matrices
used to define those defects we find

ηa,0 ηeb,0 × DA = ηeb,0 × DA , ηa,0 ηeb,0 × DB = ηa,0 × DB , (4.11)

where ηa,0 (η0,b ) generate Z2 ⊂ U (1)n1 (Z2 ⊂ U (1)w1 ) while ηea,0 (e


η0,b ) generate Z2 ⊂ U (1)n2
(Z2 ⊂ U (1)w2 ). Here a, b ∈ Z2 . Because of the grading structure we also need to include
additional duality defects

e2,1
DC = Sρ σ2,1 σ , e2,1
Dτ C = Sτ Sρ σ2,1 σ , C, (4.12)

where C is the topological defect implementing charge conjugation and the new duality
defects satisfy
ηa,0 ηeb,0 × DC = DC , ηa,0 ηeb,0 × Dτ C = Dτ C . (4.13)
24
It was also recently noticed in [31] that duality defects coming from gauging non-invertible symmetries
can be present in 2d CFT.

23
Applying the generic rules described above we find the fusion between defects and their
inverse to be
X X X
DC × D C = ηa,0 ηeb,0 , DA × D A = ηa,0 , DB × D B = ηeb,0 (4.14)
a,b=0,1 a=0,1 b=0,1

while some fusion rules between different duality symmetries read


X X
DC × DA = DB ηa,0 , DA × DC = Sτ DA ηeb,0
a=0,1 b=0,1
X (4.15)
DA × DB = Sτ ηa,0 ηeb,0 , DB × DA = DC ,
a,b

together with other ones that can be easily derived. We notice that the category is non-
commutative, as as a consequence of the non-Abelian O(2, 2, Z) grading25 . In order to
describe the non-commutative grading more explicitly, we introduce the group with five
generators {s, dc , da , i, c} satisfying the following multiplication law

s2 = d2c = c , d2a = c2 = i2 = 1 , isi = s3 , idc i = d3c


asa = dc , adc a = s . (4.16)

It can be easily verified that the fusion of the category described before is graded by
the non-commutative group (4.16) by making the following assignment of (non-)invertible
defects to the generators

Sτ → s , DC → dc , DA → da , I → i , C → c , (4.17)

with the remaining defects corresponding to products of the ones above.


As already emphasized, the quadri-critical point is the product of two c = 1 CFT at

R = 2. Therefore the partition function can be written as
X √ √
Zc=2 [τ = i, ρ = 2i] = χm χn χ−m χ−n = Zc=1 [R = 2]Zc=1 [R = 2] (4.18)
m,n∈Z4

and we can trivially prove that the partition function of the c = 2 theory at the quadri-
critical point is self-dual under gauging of Z2 ⊂ U (1)n1 and Z2 ⊂ U (1)n2 , implying the
existence of the defects described above (see appendix C).

4.1.2 Modular bootstrap at the quadri-critical point

As a further check that the above structure is actually consistent, we compute the twisted
partition functions and make use of modular covariance. We find that all the elements
25
Non-commutative categorical symmetries appeared before in the literature in 4d (see e.g. [29, 40]) and
also in 2d c = 2 theories (see e.g. [24]).

24
introduced above lead to well defined defect Hilbert spaces, hence complying with (2.11).
For sake of the extension of this exposition, we do not include all the examples but just a
few illustrative ones and comment on the rest.
Regarding the defect DA , it can be explicitly checked that its action emulates the
action of T -duality in one of the sectors of the factorized theory, namely p1 → −p1 , hence

the computation works out exactly as at c = 1 and R = 2 [20]. Similarly, target space
inversion I implements a reflection in one of the two sectors, hence clearly leading to a
well defined defect on the product theory.
We will then proceed to illustrate the computation for the non-invertible defects DB
and DC . Making use of (3.27) and (3.7), the action of DB maps
! !
0 1 0 1
p → RB p , p → RB p , RB = , RB = . (4.19)
−1 0 1 0

In addition, it turns out that the phase αB (n, w) determined by equation (A.9) is trivial
once we impose the transformed operator to be local26 . Due to (4.19), when computing
the partition function twisted by the insertion of DB along the spatial cycle, the sum gets
contributions only from states with p1 = p2 = 0, p1 = p2 = 2n (n ∈ Z). Finally, in order
to perform the sum over oscillator modes, we consider a basis over which RB (RB ) act
diagonally with eigenvalues ±i (±1). Putting all together we get
1 1 P 2
√ q − 12 q − 12 n∈Z q 4n √ ϑ4 (4τ )ϑ4 (2τ̄ )ϑ3 (8τ̄ )
Z(1,DB ) = 2 Q∞ m m m m
= 2
m=1 (1 + iq )(1 − iq )(1 + q̄ )(1 − q̄ ) η(2τ )η(τ̄ )2
τ →− 1 ϑ2 (τ /4)ϑ2 (τ̄ /2)ϑ3 (τ̄ /8) 1  3 1 3 1

τ
−−−−→ Z(DB ,1) = = q 32 q 16 + 2q 32 q 8 + . . . . (4.20)
4η(τ /2)η(τ̄ )2 |η(τ )|4

We therefore verify that Z(DB ,1) has a consistent interpretation as a trace over the defect
Hilbert space.
Now we turn to the defect DC implementing the following action
!
0 1
p → RC p , p → RC p , RC = −RC = . (4.21)
−1 0

In addition, the states get multiplied by the Z2 phase αC = n·w. Given the transformation
(4.21), the twisted partition function gets contribution from states in the identity multiplet
26
In fact, the action of DB leads to the map of charges {n1 , n2 , w1 , w2 } → {−2w2 , n1 , − n22 , w1 }. There-
fore, the resulting operator corresponds to a genuine local operator only for n2 = 2Z. Moreover, plugging
the previous map of charges into equation (A.9) leads to the solution αB (n, w) = n2 w2 mod 2 for the
phase entailing mutual locality. The latter clearly becomes trivial for n2 even.

25
pa = pa = 0 (a = 1, 2). Following similar steps as the ones described above, we get
1 1
q − 12 q − 12 ϑ4 (4τ )ϑ4 (4τ̄ )
Z(1,DC ) = 2 Q∞ m m m m
=2
m=1 (1 + iq )(1 − iq )(1 + iq̄ )(1 − iq̄ ) |η(2τ )|2
(4.22)
   
τ →− 1
τ ϑ2 (τ /4)ϑ2 (τ̄ /4) 4 3 3 3 3 1 1
−−−−→ Z(DC ,1) = = q 32 q 32 + q 32 q 32 q 4 + q 4 + ... ,
|η(τ /2)|2 |η(τ )|4
hence corroborating that DC leads to a consistent defect Hilbert space.
We end this subsection by considering the case of Dτ C whose action maps pa → −pa
(a = 1, 2) while leaving the right movers invariant. The Z2 phase again obtains ατ C = n·w.
By comparison with the c = 1 case reviewed in section 2, one notices that this action can
be regarded as a composition of two (left) T -dualities on each c = 1 factor (composed with
a gauging of the corresponding Z2 subgroups), hence leading to
ϑ4 (2τ )2 ϑ3 (4τ )2
Z(1,Dτ C ) = 2 (4.23)
|η(τ )|4
τ →− 1
τ ϑ2 (τ /2)2 ϑ3 (τ /4)2 1  1 1 1 1 1

−−−−→ Z(Dτ C ,1) = = q 8 + 4q 8 q 8 + 4q 8 q 4 . . .
4|η(τ )|4 |η(τ )|4
Note the important role played by the quantum dimension here. If it was absent, the final
result would be
ϑ2 (τ /2)2 ϑ3 (τ /4)2 1 1
4
∼ q8 + ... (4.24)
8|η(τ )| 2
clearly spoiling the interpretation of the twisted partition function as a trace over a Hilbert
space. Of course, the positive outcome of the modular bootstrap analysis for DC and Dτ C
naturally implies the consistency of the invertible defect implementing Sτ .

4.1.3 Marginal and relevant deformations at the quadri-critical point

We proceed to study the interplay between the symmetry structure presented in the pre-
vious sections and the various marginal and relevant deformations of the theory. As a
multicritical point, one expects finding additional independent exactly marginal deforma-
tions which span the orbifold branches, besides the ones of the form ∂µ φa ∂ν φb which span
the four dimensional toroidal branch. The strategy consists in constructing the set of
b
marginal operators of the form Jia J j , where a, b = 1, 2 and i, j = +, −, and in terms of
the chiral fields
√ √ 1
1 2X 1 1
J± = V±1,0,±1,0 = e±i , J ± = V±1,0,∓1,0 = e±i 2X
,
√ √ 2
2 2X 2 2
J± = V0,±1,0,±1 = e±i , J ± = V0,±1,0,∓1 = e±i 2X
,
i a i a
J3a = √ ∂X a , J3 = √ ∂ X . (4.25)
2 2

26
b 2
At the enhanced symmetry point, they are all equivalent to J3a J 3 by the SU (2)2 × SU (2)
global symmetry. We then construct invariant combinations under the Z2 automorphisms
and charge conjugation. The action on the currents is
 
Zs2 : J3a , J±a
→ J3a , −J±
a
,
 
Zorb
2 : J3a , J±a
→ −J3a , J∓
a
,
 
ZR
2
1
: J31 , J±1
→ J31 , −J±
1
,
2 2
 2 2

J3 , J± → −J3 , J∓ ,
 
ZR
2
2
: J31 , J±1
→ −J31 , J∓
1
,
 
J32 , J±
2
→ J32 , −J±
2
. (4.26)

Given the Z2 -invariant operators, the exactly marginal operators are further combinations
of these, whose OPE with themselves has no simple poles [49]. Finally, the exactly marginal
operators are mapped to the quadri-critical point by application of the map (4.4).
For the case of Zorb
2 , the four exactly marginal operators

a b a b a b a b
Oab = J+ J + + J− J − + J+ J − + J− J + , a b = 1, 2 , (4.27)

are mapped to the quadri-critical points as




 O′ 11 = V0,0,+1,−1 + V+2,−2,0,0 + V0,0,−1,+1 + V−2,+2,0,0 ,



O ′
12 = V2,0,0,−1 + V0,−2,+1,0 + V−2,0,0,+1 + V0,+2,−1,0 ,
Zorb
2 branch

 O′ 21 = V−2,0,0,−1 + V0,−2,−1,0 + V+2,0,0,+1 + V0,+2,+1,0 ,



 ′
O 22 = V0,0,−1,−1 + V−2,−2,0,0 + V0,0,+1,+1 + V+2,+2,0,0 ,

where they span the 4-dimensional Zorb 2 orbifold branch.


Similarly, and considering the action in Eq. (4.3), the deformations into the ZR 1
2 branch
are generated by the following ZR 1
2 -invariant exactly marginal operators


O ′ R 1
= V+2,0,0,−1 − V0,−2,+1,0 + V−2,0,0,+1 − V0,+2,−1,0 ,
R1 A
Z2 branch

O′ R1 = V−2,0,0,−1 − V0,+2,+1,0 + V+2,0,0,+1 − V0,−2,−1,0 ,
B

while for ZR2


2 , we get the following


O ′ R 2
= V0,0,+1,−1 − V+2,−2,0,0 + V0,0,−1,+1 − V−2,+2,0,0 ,
R2 A
Z2 branch

O′ R2 = V0,0,−1,−1 − V−2,−2,0,0 + V0,0,+1,+1 − V+2,+2,0,0 .
B

Both the ZR1 R2


2 and Z2 orbifold branches are 2-dimensional.

27
The duality defects discussed in the previous subsection are preserved by some com-
binations of the exactly marginal operators at the quadri-critical point27 . For the four
operators O′ ab parametrizing the 4-dimensional Zorb
2 orbifold branch, we get

O′ 11 + O′ 12 , O′ 21 + O′ 22 preserved by DA ,

O′ 11 + O′ 21 , O′ 12 + O′ 22 preserved by DB ,

O′ 11 + O′ 22 , O′ 12 + O′ 21 preserved by Sτ , I , DC ,

O′ 11 , O′ 12 , O′ 21 , O′ 22 preserved by Dτ C . (4.28)

In particular Dτ C is preserved along the full Zorb


2 orbifold branch. In addition, there are
two extra non-invertible symmetries DA,B which are preserved along certain 2-dimensional
submanifolds of this particular orbifold branch.
Regarding the ZR 1 R2
2 , Z2 branches the invariant combinations are

R R
O′ A1 + O′ B1 preserved by I , DC ,
R R
O′ A1 − O′ B1 preserved by Sτ ,
R R
O′ A2 + O′ B2 preserved by Sτ , I ,
R R
O′ A2 − O′ B2 preserved by DC . (4.29)

Interestingly, we find that DA and Dτ B = Sτ−1 M Sτ ◦ σ


e2,1 act as
R R
DA : O ′ A1 → O ′ A2 ,
R R
O′ B1 → O′ B2 ,
R R
Dτ B : O′ A1 → −O′ A2 ,
R R
O′ B1 → −O′ B2 , (4.30)

thus they effectively exchange the two orbifold branches.

Relevant deformations

We now proceed with the analysis of the relevant deformations of the quadri-critical point.
Classifying them by their dimension, we have
27
Note that, given their charges, none of these vertex operators acquires a phase under the action of the
duality symmetries described in the previous subsection.

28
(h, h) (n, w)
1 1

,
8 8 (±1, 0, 0, 0) , (0, ±1, 0, 0)
1 1

4, 4 (±1, ±1, 0, 0) , (±1, ∓1, 0, 0)
1 1

2, 2 (±2, 0, 0, 0) , (0, ±2, 0, 0) , (0, 0, ±1, 0) , (0, 0, 0, ±1)

5 5
 (0, ±1, ±1, 0) , (±2, ∓1, 0, 0) , (±1, 0, 0, ±1) , (±1, 0, 0, ∓1)
8 , 8 (±2, ±1, 0, 0) , (±2, ∓1, 0, 0) , (±1, ±2, 0, 0) , (±1, ∓2, 0, 0)
Some non-invertible defects at the quadri-critical point commute with particular com-
binations of these operators. We thus proceed to analyse some representative cases of
duality-preserving RG flows, triggered by some of the relevant deformations presented in
the above table. Given the factorized structure quadri-critical theory, these RG flows can
be studied by building up on the intuition acquired at c = 1 (see [20] for some exam-
ples of the latter). This enables to check some non-trivial constraints that are a direct
consequence of the presence of non-invertible duality symmetries.
We begin by considering the case of the most relevant operators, that is with conformal

weights (h, h) = 18 , 18 . There are two possible independent charge conjugation invariant
combinations, namely
1
R1/8 = V1,0,0,0 + V−1,0,0,0 ∼ cos φ1 , 2
R1/8 = V0,1,0,0 + V0,−1,0,0 ∼ cos φ2 (4.31)

Each of them preserve a U (1)m × U (1)w invertible symmetry participating in a mixed ’t


Hooft anomaly of the form
Z
1
A3d = AdB A ∈ U (1)m , B ∈ U (1)w . (4.32)

Due to this anomaly the IR fixed point must be gapless. Moreover both R1 and R2
preserve respectively Dτ B = Sτ−1 Sρ M σ
e2,1 and DA = M σ2,1 .28 Therefore the IR gapless
theory must enjoy such non-invertible defect29 . In this case, it is straightforward to show

that the IR theory corresponds to the c = 1 compact boson at R = 2, hence exactly
satisfying all the symmetry constraints.
Let us now comment on slightly more involved flows, triggered by the relevant operators

with (h, h) = 21 , 12 of the form
1
R1/2 ∼ cos(φe1 ) + cos(2φ2 ) , 2
R1/2 ∼ cos(φe2 ) + cos(2φ1 ) . (4.33)
1,2
Both R1/2 break U (1)4 to a subgroup U (1) × U (1) × Z2 with a Z2 anomaly
Z
A3d = πi a ∪ β(b) (4.34)

28
Note that Dτ B = Sτ−1 DB and therefore is not an independent defect.
29
The non-invertible symmetry cannot decouple from the gapless sector since its invertible subgroup
partecipates in the ’t Hooft anomalies generating the gapless mode.

29
where a, b ∈ H 1 (X3 , Z2 ) are discrete gauge fields for Z2 × Z2 ⊂ U (1) × U (1) × Z2 and β
is the Bockstein homomorphism (see e.g. [64]). In the case of R1/2 1 , the discrete subgroup

corresponds to a Z2 × Z2 momentum and winding symmetries acting on (φ2 , φe2 ), and


2 . On the contrary, the relevant combination R1 + R2
vice versa for R1/2 1/2 1/2 preserves an
anomaly free Z2 × Z2 . Therefore there are no continuous anomalies and the theory can
1,2
be non-trivially gapped if we deform with R1/2 or even trivially gapped if we consider
the combination R1/2 1 2 . However both deformations in (4.33) preserve the non-
+ R1/2
invertible symmetry generated by DC and the sum R1/2 1 + R2
1/2 preserves all the duality
defects DA , DC , Dτ C . Since both DA and Dτ B have a non-invertible anomaly, while DC is
anomaly free (see appendix C and references therein)30 , the IR theory must match such
anomalies. We proceed now to study the RG flows
R 1
S = Sc=2 [(i, 2i)] + µ R1/2 (4.35)
R 1 2
S = Sc=2 [(i, 2i)] + µ (R1/2 + R1/2 ) (4.36)

and check explicitly the non trivial constraints coming from the non-invertible ’t Hooft
anomalies. In order to do that it is instructive to recall that a c = 1 compact boson at

radius 2 corresponds to the bosonization of two free Majorana fermions (equivalently
a Dirac fermion). From this perspective, the relevant deformations cos 2φ and cos φe are
mapped to mass terms for the Majorana fermions. More precisely, in terms of the two
Majora components χ1,2 , the bosonization map reads

χ1 + iχ2 = eiX , χ1 + iχ2 = eiX (4.37)

hence, recalling that 2φ = X + X and φ̃ = X − X, one finds

cos 2φ ∼ χ1 χ1 − χ2 χ2 , cos φe ∼ χ1 χ1 + χ2 χ2 . (4.38)

These considerations, when applied to the theory at hand, imply

cos(φe1 ) + cos(2φ2 ) → χ1 χ1 + χ2 χ2 + χ3 χ3 − χ4 χ4 (4.39)


cos(φe1 ) + cos(2φ2 ) + cos(φe2 ) + cos(2φ1 ) → χ1 χ1 + χ3 χ3 . (4.40)

Therefore the IR theory of the RG flow triggered by R1/2 1 is the bosonized version of a
trivial theory, i.e. a Z2 × Z2 gauge theory, compatible with the anomaly free T Y (Z2 × Z2 )
symmetry and an anomalous Z2 invertible symmetry. On the other hand, the RG triggered
1 +R2 leads in the IR to two Majorana fermions in the fermionized theory.
by the sum R1/2 1/2
Upon bosonizing each of them separately, we land on the c = 1 Ising2 theory. In the latter
theory, the two preserved non-invertible defects are naturally realized by the Kramers-
Wannier duality defect preserved by each factor.
30
In particular, we are referring to the anomalies coming from the TY(Z2 ) subcategory spanned by one
of the three non-invertible defects together with the its corresponding Z2 symmetry.

30
4.2 The bi-critical point

Let us now delve into the more √


interesting case of the bi-critical point (τ, ρ) = (ω, α) where
2πi/3 1 3
ω=e and α = − 2 + i3 2 . Note that this theory cannot be regarded as a product
of two c = 1 theories. As discussed previously, we take the theory (τ, ρ) = (ω, ω) with
enahnced global symmetry SU (3) × SU (3) as the starting point for the analysis. In this
theory, the holomorphic decomposition of the fields takes the form
r 
1 1  1 1
 1  2 2

2 2 2

φ = √ X +X + √ X +X , φ = X2 + X , (4.41)
2 6 3
and enjoys the shift and rotational Z3 symmetries determined by the following action on
the chiral fields
√ !
 2
Zs3 : X 1, X 2 → X 1 + 2 π, X 2 ,
3
  √ r !
1 2 1 2 2 2
X , X → X + π, X − π ,
3 3
 
orb 1 2
 2π 1 2π 2 2π 1 2π 2
Z3 : X , X → cos X + sin X , − sin X + cos X , (4.42)
3 3 3 3
   
1 2 2π 1 2π 2 2π 1 2π 2
X , X → cos X + sin X , − sin X + cos X .
3 3 3 3
and similarly for the antiholomorphic fields. Again, these two Z3 actions can be mapped to
each other by means of global SU (3) × SU (3) rotations. Consequently, the corresponding
orbifolds lead to two equivalent descriptions of the same theory, leading to a bi-critical
point.
In order to move to the bi-critical point, we gauge the Zs3 symmetry. This operation
implements the following map of charges
2 1 1 1
n′1 = n1 + n2 , n′2 = − n1 + n2 , w1′ = w1 + w2 , w2′ = −w1 + 2w2 (4.43)
3 3 3 3
and (τ, ρ) = (ω, ω) → (ω, ω/3), subsequently performing the SL(2, Z) transformation
Tρ−2 Sρ , finally landing at the point (τ, ρ) = (ω, α). For these two operations together, the
overall map of the charges results
4 2 2 2
n′1 = n1 + n2 + w1 − 2w2 , n′2 = − n1 + n2 + w1 + w2 ,
3 3 3 3
1 1 2 1
w1′ = n1 − n2 , w2′ = n1 + n2 . (4.44)
3 3 3 3
as it can be easily verified by conjugating the generalized metric with the corresponding
matrix31 .
31
Recall that, as described in section 3, for a map implementing (n, w) → (M −1 )⊺ (n, w), the generalized
metric transforms accordingly as E → (M −1 )⊺ E M −1 .

31
4.2.1 Duality symmetry and fusion category

As shown in section 3, a generic RCFT can always be connected to a factorized point via
the composition of dualities and gaugings on each factor. In this case, we choose to move
√ √
to the factorized point at (τ, ρ) = (i 3, i3 3) by means of the following naminpulation
√ √
f=2 : (ω, α) → (i 3, i3 3) .
Tρ Tτ σM (4.45)

Duality symmetries pertaining to the the bi-critical theory can be found by composing
the above map with the duality symmetries of the factorized point. Indeed, following
the procedure described in section 3, we find the following defects and their associated
matrices in O(2, 2, Q)
 1 1
  
0 3 3 0 −1 2 −4 −2
0 1 0 0 1 −2 −4
   −2 1 
D1 → D1 =   , D2 → D2 =  . (4.46)
 3 −1 0 0 9 −28 14 −1 −2
−1 31 13 1 14 −28 2 1

Another useful point at which additional duality symmetries become more manifest is the
diagonal RCFT (τ = w, ρ = 3w), which is connected to the bi-critical point via Tρ−1

Tρ−1 : (w, α) → (w, 3w) . (4.47)

At this point we have the duality symmetry coming from composing mirror symmetry M
and a gauging of Z3 shift symmetry. However, this can be checked to coincide with D1
when written in the bi-critical point. In addition, there is a new duality symmetry coming
from the duality matrix IM Sρ Sτ . When mapped back to the bi-critical point, the action
of such a symmetry becomes implemented by the following matrix
 
1 0 0 0
 1 0 0 − 13 
 
D3 → D3 =  31 1  ∈ O(2, 2, Q) , (4.48)
− 3 1 1 3 
1 −3 0 0

Accordingly, it can be easily verified that these three operations leave the generalized
metric invariant, that is

(D−1 ⊺ −1
a ) EDa = E , a = 1, 2, 3 . (4.49)

Note that we are omitting the description of the actions implemented by the above defects
in terms of a string of dualities and topological manipulations, just because these are quite
involved and barely illuminating.

32
We will focus on the three duality symmetries just described. Of course, we will identify
new ones when consider their fusion products below. Before delving into those aspects, let
us first identify the discrete gaugings leading to the self-dualitites implied by these non-
invertible defects. We do that by following the arguments in Sec. 3.1. We can interpret
D1,2,3 as performing the gauging of the following subgroups of the global symmetry
(1) 2πik
(n1 +w2 )
(D1 ) Z3 : Vn,w → e 3 Vn,w
2πik1 2πik2
(2) (n1 −n2 +4(w1 +w2 )) (n2 −w1 )
(D2 ) Z3 × Z9 : Vn,w → e 9 e 3 Vn,w , (4.50)
2πik2
(2) (n2 −w1 )
(D3 ) Z3 : Vn,w → e 3 Vn,w .

These expressions should not be confused with the action of the corresponding non-
invertible defects on the vertex operators. The latter will be implemented by the cor-
responding map of charges implied by the matrices in (4.46), (4.48), together with the
associated quantum dimensions and Z2 phases, as discussed around (3.26). We will be
more explicit about these action when testing the defects against the modular bootstrap
in the next subsection.
We can now discuss the fusion between those defects. As matrices acting on the
generalized metric, all D1,2,3 square to the identity matrix. Therefore we get
X X X
D1 × D1 = ηa , D2 × D2 = ηb , D3 × D3 = ηc , (4.51)
(1) (2) (2)
a∈Z3 b∈Z3 ×Z9 c∈Z3

so that they separately generate a Tambara-Yamagami category with a Z2 grading. More-


over fusion between different defects generates new duality defects, arising from self-duality
(1) (2)
under gauging subgroups of Z3 × Z3 × Z9 . For instance
(1,2)
D1 × D2 = C0 D4 (4.52)
(1,2)D
where D4 is the duality defect constructed by gauging on half-space the symmetry Z3 ×
Z9 , where
(1,2)D 2πik
Z3 : Vn,w → e 3 (n1 −n2 +w1 +w2 ) Vn,w . (4.53)
Comparing the symmetries on the l.h.s. and on the r.h.s. of the fusion (4.52) we can
(1,2)
derive the form of C0

X (1) (2)
(1,2) (1,2)AD Z3 × Z3
C0 = ηa , Z3 = (1,2)D
. (4.54)
(1,2)AD Z3
a∈Z3

Similarly one can compute the other fusion rules, like


X
D2 × D3 = D5 ηb ,
(4.55)
b∈ZAD
3

33
where D5 comes from the half-gauging of
(2) 2πik
Z3 × Z9 ⊃ ZD
9 : Vn,w → e 9
(−n1 +4n2 +2w1 +5w2 )
Vn,w , (4.56)

and
(2)
Z3 × Z9
ZAD
3 = . (4.57)
ZD
9

Self-duality under gauging

The full set of non-invertible symmetries together imply that the RCFT at the bi-critical
point is self-dual under the gauging of
(1) 2πik
(n1 +w2 )
Z3 : Vn,w → e 3 Vn,w ,
2πik2
(2) (n2 −w1 )
Z3 : Vn,w → e 3 Vn,w , (4.58)
2πik1
(n1 −n2 +4(w1 +w2 ))
Z9 : Vn,w → e 9 Vn,w .
By using the closed form of the torus partition function for RCFTs, we proceed to explicitly
check such self-dualities, thus confirming the above findings.
Following the general construction reviewed in App. B, the chiral algebra at the bi-
critical point can be written as
!
6 −3
u(1)2K × u(1)2K , K = . (4.59)
−3 6

Furthermore, the partition function for this RCFT is given by a (diagonal) modular in-
variant combination of the chiral algebra characters
X
Z= χλ χλ , (4.60)
λ∈D

where
1 X 21 |λ+Kl|2 −1 1 X 12 |λ+Kr|2 −1
χλ = q K , χλ = q K . (4.61)
η(τ )2 2
η(τ )2 2
l∈Z r∈Z
The vector λ labelling each representation belongs to the lattice

L ≡ λ ∈ Z2 , λ ∼ λ + Kv , v ∈ Z2 . (4.62)

The fact that the theory corresponds to a diagonal RCFT becomes evident by noticing
that there is a duality frame in which ρ = 3τ (see equation (4.47)).
(1)
It turns to be useful to rewrite the action of the Z3 over the elements comprised in the
different characters. This can be done straightforwardly by using the expressions (B.14)
to write the character labels in terms of the global charges. More precisely
2π T
χλ χλ = e2πi(λ+λ) K −1 v3
(1)
Z3 : χλ χλ → ei 3
(λ1 +λ1 )
χλ χλ , (4.63)

34
with the generator of the Z3 sublattice v3 ≡ (2, −1)T . Armed with these tools, we can
write the partition function twisted by (η a , η b ) along the temporal and spatial cycles
(1)
respectively, where η denotes the generator of Z3
1 X i4πaλT K −1v3 i2πb(µ+µ)T K −1 v3 −i2πλT K −1 (µ−µ)
Z(ηa ,ηb ) = e e e χµ χµ . (4.64)
|L|
λ,µ,µ∈L

Summing over λ and using the orthogonality relation (B.12) leads to


X T −1
Z(ηa ,ηb ) = ei4πb(µ−av3 ) K v3 χµ χµ−2av3 . (4.65)
µ∈L

The partition function of the gauged theory then obtains


1 X
Z(ηa ,ηb ) . (4.66)
3
a,b∈Z3

Now, the sum over b imposes the constraint a = 2µ1 mod 3, leading to
!
1 X X X −1 0
Z(ηa ,ηb ) = χµ χCµ = χµ χµ , C= , (4.67)
3 1 1
a,b∈Z3 µ∈L µ∈L

(1)
hence establishing the self-duality of the theory under gauging Z3 . In the last equality,
we used that the matrix C preserves the lattice pairing, namely

C T K −1 C = K −1 . (4.68)

A completely analogous computation shows that the theory at (τ, ρ) = (ω, α) is self-dual
(2)
upon gauging Z3 in (4.58).
For sake of completeness, let us briefly comment on the computation that shows the
self-duality of the theory under gauging the Z9 in (4.58). Its action on the characters is
 T

2π 2πi λT K −1 v9 −λ K −1 v9
Z9 : χλ χλ → ei 9
(2λ1 +λ2 −λ1 −2λ2 )
χλ χλ = e χλ χλ , (4.69)

with v9 ≡ (1, 0)T and v9 ≡ (0, 1)T . Therefore the twisted partition function formally
reads
1 X i2πaλT K −1(v9 −v9 ) i2πb(µT K −1 v9 −µT K −1 v9 ) −i2πλT K −1 (µ−µ)
Z(ηa ,ηb ) = e e e χµ χµ .
|L|
λ,µ,µ∈L
(4.70)
The sum over λ now sets µ = µ − a(v9 − v9 ). Finally, taking the orbifold and summing
over b ∈ Z9 fixes a = µ1 − µ2 mod 9 hence proving the claimed self-duality
!
1 X X X
′ 0 1
Z(ηa ,ηb ) = χµ χC ′ µ = χµ χµ , C = . (4.71)
9 1 0
a,b∈Z3 µ∈L µ∈L

where we made use of the fact that the matrix C ′ also preserves the lattice pairing.

35
4.2.2 Modular bootstrap at the bi-critical point

As with the previous example, we now proceed to run the modular bootstrap analysis
to confirm that the symmetry operations introduced above define consistent topological
defects.
Let us begin with D1 which, by plugging its matrix representation D1 into (3.27), leads
to the following map of the U (1) charges
n1 w2 n1 + w2
n1 → 3w1 − w2 , n2 → + n 2 − w1 + , w1 → , w2 → w2 . (4.72)
3 3 3
Before continuing, let us determine the phase associated to the action of D1 on vertex
operators. As explained in Appendix A, mutual locality of correlation functions under the
action (4.72) demands

α1 (n + n′ , w + w′ )+α1 (n, w) + α1 (n′ , w′ ) = (4.73)


n1 + w2 + n′1 w2′
(3w1′ − w2′ ) + (3w1 − w2 ) − w1 w2′ − w1′ w2 mod 2 .
3 3
Note that, in order to recast the phase to the above form, we imposed n1 + w2 ∈ 3Z and
n′1 + w2′ ∈ 3Z, hence rendering the resulting vertex operators genuine. As such, there are
multiple solutions, each of them differing by stacking D1 with a particular Z2 subgroup
of the invertible symmetry. For a reason that will be clear momentarily, we choose the
following solution
n1 + w2
α1 (n, w) = (3w1 − w2 ) − w1 (w1 + w2 ) (4.74)
3
Having fixed the phase in the action of the D1 defect, we proceed to run the modular
bootstrap. The map (4.72) translates in the momentum basis to

(p1 , p2 , p1 , p2 ) → (p1 , p2 , −p1 , p2 ) (4.75)

hence implying that only states with p1 = 0 contribute to the twisted partition function
Z(1,D1 ) . In particular, this implies n1 = 3w1 − w2 . Moreover, (4.74) trivializes in Z2 for
these values of the charges, which is the main reason for choosing this particular solution.
In terms of the chiral algebra characters, and using the conversion formulas (B.14), the
states contributing to the twisted partition function have λ = (0, λ2 )T and r = (r, 2r)T
with r ∈ Z. The basis of independent right moving characters then reads

λ + K · r = (0 , λ2 + 9r)T . (4.76)

Therefore, the independent values of λ2 span a Z9 . Turning to the holomorphic blocks,


the scaling dimensions are given by (l = (l1 , l2 )T )
   
1 3 λ2 2 9 λ2 2
(λ + Kl)T K −1 (λ + Kl) = l+ + + l− + . (4.77)
2 4 3 4 9

36
where we diagonalized the matrix K in terms of l+ ≡ l1 +l2 and l− ≡ −l1 +l2 . Consistently
with the fact that the theory at the bi-critical point (τ, ρ) = (ω, α) does not factorize, the
l± variables are not free within Z. In particular, they must satisfy either (l+ , l− ) = (2l, 2l′ )
or (l+ , l− ) = (2l + 1, 2l′ + 1) for l, l′ ∈ Z 32 . We then need to sum over both the even and
odd sectors. In addition, states with an odd number of right-moving oscillator modes come
with a (−1) phase, as usual. Summing over oscillator modes and plugging the solution
described above, the twisted partition function results
√ ( !
3 X X 3l+ λ2 2 X 9l′ + λ2 2 X 3l+ λ2 +3 2 X 9l′ + λ2 +9 2
Z(1,D1 ) = q 6
q 18
+ q 6
q 18
|η(τ )|4
λ2 ∈Z9 l∈Z l′ ∈Z l∈Z l′ ∈Z
 2 )
X 9 r+ λ2
× q 9
ϑ4 (2τ ) , (4.78)
r
√ ( " # " # " # " # !
3 X λ2 λ2 λ2 +3 λ2 +9
= ϑ 6
(6τ ) ϑ 18 (18τ ) + ϑ 6 (6τ ) ϑ 18 (18τ )
|η(τ )|4 0 0 0 0
λ2 ∈Z9
" # )
λ2
×ϑ 9 (18τ ) ϑ4 (2τ ) . (4.79)
0

Performing the modular S transformation τ → −1/τ and using the transformation prop-
erties listed in appendix D one gets
1 1 X   0    0  

0



0



τ τ τ τ
Z(D1 ,1) = ϑ 6 ϑ − λ2 18 + ϑ − λ2 +3 6 ϑ − λ2 +9
36 |η(τ )|4 − λ62 18 6 18
18
λ2 ∈Z9
  
0 τ
 
×ϑ λ2 18 ϑ2 τ2 , (4.80)
−9
1  1 1 13 1 25 1 1

≈ q 16 + 4q 9 q 144 + 2q 9 q 144 + 2q 3 q 16 + . . . .
|η(τ )|4
In going to the last line, we performed the sum over λ2 . As it becomes evident from the
first orders in the expansion above, the defect Hilbert space has a consistent expansion in
terms of Virasoro characters.
We now proceed to bootstrap the defect D2 . We first notice that the right hand side
of the equation (A.9) determining the phase α2 is already in 2Z once restricted to genuine
operators33 . We can therefore set α2 (n, w) = 0. In addition, the action of D2 flips the
32 l+ −l− l+ −l−
This can be easily verified by noticing that l1 = 2
and l2 = 2
have integer solutions only for
l± =even or l± =odd.
33 (1)
More precisely, D2 implements an orbifold by the Z3 × Z9 symmetry in (4.58). Hence, by restricting
to genuine local operators we mean

n1 − n2 + 4(w1 + w2 ) ∈ 9Z , n2 − w1 ∈ 3Z .

37
sign of both p1 and p2 . The subset of states contributing to Z(1,D2 ) thus have p1 = p2 = 0,
which in terms of the characters reads λ1 = λ2 = l1 = l2 = 0. By manipulations analogous
to the ones described above, one obtains

27
Z(1,D2 ) = ϑ4 (2τ )2 (ϑ3 (6τ )ϑ3 (18τ ) + ϑ2 (6τ )ϑ2 (18τ ))
|η(τ )|4
τ →− τ1 1  τ 2   τ   τ     
τ τ
−−−−→ Z(D2 ,1) = 4
ϑ2 ϑ3 ϑ3 + ϑ4 ϑ4
4|η(τ )| 2 6 18 6 18
1  1 1 1 1 1

≈ 2q 8 + 12q 8 q 9 + 12q 8 q 3 + . . . . (4.81)
|η(τ )|4

As a final example, consider the defect D5 which, in particular, implements the orbifold
by the Z9 gauging in (4.58). For the phase required by mutual locality, we choose the
following solution
α5 (n, w) = n2 (w1 + w2 ) + w12 + w22 (4.82)

which turns out to be trivial when evaluated on the states contributing to the twisted
partition function (recall that all possible solutions differ just by stacking with an invertible
Z2 symmetry defect). The action of D5 amounts to the following
√ !
1 1 3
p→R·p , R=− √ (4.83)
2 3 −1

Note that R2 = 1, hence defining a Z2 action in momentum space (equivalently, D5 : X →


R · X). In terms of the characters, the states contributing to the trace have λ2 = −2λ1
l1 = 0 and their holomorphic scaling dimension read
 
1 λ =−2λ1 λ1 2
|(λ + Kl)|2K −1 −−2−−−−→ 3 l− . (4.84)
2 l1 =0 , l2 =l 3

It becomes evident from the expression above that the independent representations con-
tributing to the trace are labelled by λ1 taking values in Z3 . On the other hand, the right
moving sector can be dealt with by means of analogous manipulations as in (4.77).
In order to perform the sum over oscillator modes, we notice that the matrix R is no
more than a rotation in field space, followed by flip of one of the momenta. This can be
verified by performing the following change of basis on the holomorphic fields
√ !
3 1
′ 2 −
√2
X → X = BX , B = 1 3
(4.85)
2 2

for which the action of R reduces to X1′ → −X1′ . Hence, by going to the appropriate basis,
the summand acquire the usual sign factor depending on one of the oscillator numbers.

38
Putting all together we end up with
√ ( " # " # " # " # !
9 X − λ61 − λ61 −λ1 +3
6
−λ1 +3
6
Z(1,D5 ) = ϑ (6τ ) ϑ (18τ ) + ϑ (6τ ) ϑ (18τ )
|η(τ )|4 0 0 0 0
λ1 ∈Z3
" # )
− λ31
×ϑ (6τ ) ϑ4 (2τ ) (4.86)
0

Finally, upon performing a modular S transformation we obtain the trace over the twisted
Hilbert space
1 X   0   τ   0   τ  
0
  
τ 0
  
τ
Z(1,D5 ) = ϑ λ1 ϑ λ1 + ϑ λ1 −3 ϑ λ1 −3
12|η(τ )|4 6 6 6 18 6 6 6 18
λ1 ∈Z3
     
0 τ τ
×ϑ ϑ2 (4.87)
− −λ3
1
6 2
1  1 1 1 7 1 7 1

≈ q 16 + 2q 16 q 9 + 4q 48 q 9 + 6q 48 q 3 + . . . .
|η(τ )|4

Categorical Data

As explained in [20] (see also appendix C) the computation of the twisted Hilbert space
enables to determine the extra categorical data of the Tambara-Yamagami categorical
symmetry, i.e. the non-degenerate bi-character χ(a, b) and the Frobenius-Shur indicator
ǫ. As in the analysis of the modular bootstrap, we focus on the duality defects D1,2 and
D5 . The data corresponding to the remaining defects can be derived in a similar manner.
Following the analysis reviewed in appendix C, the F-symbols of the categorical sym-
metry imply selection rules for the spins of the states in the defect Hilbert space. In
particular we get
ǫ X
e4πis |ψi = √ ηba |ψi , (4.88)
A a∈G
where ηba are the invertible symmetry operators generating G ⊂ U (1)2n × U (1)2w acting on
the Hilbert space twisted by D, with a particular resolution of the 4-valent junction (see
the discussion around Eq. (C.12))34 and |ψi are eigenstates of ηba in such twisted Hilbert
space. By performing F-moves, it is straightforward to check that

ηba ηbb = χ(a, b)ηc


ab (4.89)

so that the group generated by ηba is generically an extension of G. Let us now start by
(1)
applying these constraints on D1 . In this case the abelian group G = Z3 and the possible
bi-characters are
2πi
χr (a, b) = e 3 rab r = ±1 . (4.90)
34
The two possible resolutions are related by the bi-character.

39
Therefore we get
iπ 2 ∓ πi
e4πis = ǫe± 3 n 4 r = ±1 , (4.91)

which is compatible with the expansion (4.80) only for the values r = 1, ǫ = 1.
In the case of D5 , the abelian group is Z9 and the possible non-degenerate bi-characters
are
2πi
χr (a, b) = e 9 rab gcd(9, r) = 1 , (4.92)

which imply
πin2 πi
e4πis = ǫer 9 e−r 4 . (4.93)

Such spin selection rules are compatible with (4.87) if ǫ = 1, r = −1.


The combined duality defect D2 , coming from the self-duality under Z3 × Z9 gauging,
will therefore have a diagonal bi-character
" !# !
  b1 1/3 0
χ(a, b) = exp 2πi a1 a2 χ , χ= , (4.94)
b2 0 −1/9

where a, b ∈ Z3 × Z9 and trivial Frobenius-Shur indicator.

4.2.3 Marginal and relevant deformations at the bi-critical point

As in the easier case of the quadri-critical point, in order to construct the marginal defor-
mations spanning the orbifold branch in the bi-critical point, it is instructive to analyze
symmetries and marginal operators of the SU (3) enhanced symmetry point (w, w). Fol-
lowing the discussion done in appendix A.2, we find the following holomorphic currents
√ √ 1
2X 1
I± = ±V±1,∓1,±1,0 = ±e±i , I ± = ±V±1,0,∓1,0 = ±e±i 2X

q q
1 3 2
±i √1 X 1 ±i 3 2
X ±i √1 X ±i X
K± = ∓V±1,0,±1,±1 = ∓e 2 2 , K ± = ∓V0,±1,∓1,∓1 = ∓e 2 2 ,
q q
1 3 2
∓i √1 X 1 ±i 3 2
X ∓i √1 X ±i X
U± = ±V0,±1,0,±1 = ±e 2 2 , U ± = ±V∓1,±1,0,∓1 = ±e 2 2 ,
√ 3 √ 1
J 3 = i 2∂X 1 , J = i 2 ∂ X
√ 8 √ 2
J 8 = i 2∂X 2 , J = i 2 ∂ X , (4.95)

which span the su(3) algebra, along with a similar set of antiholomorphic currents for
su(3). The two relevant Z3 symmetries are [55]:

Zs3 : I± → ω ±2 I± , K± → ω ±1 K± , U± → ω ±2 U± ,

Zorb
3 : I ± → U± , K± → I∓ , U± → K∓ . (4.96)

40
At this point, the various marginal operators are all related to {J3 J 3 , J3 J 8 , J8 J 3 , J8 J 8 } by
means of the enhanced su(3) chiral algebra, and the invariant combinations under both
Zs3 and Zorb
3 are

O1 = OA + OB + OC ,

O2 = OAω + OBω + OCω (4.97)

where

OA = I+ I + + U+ U + + K− K − + h.c. = V2,−1,0,0 + V−1,2,0,0 + V1,1,0,0 + h.c. ,

OB = I+ K − + U+ I + + K− U + + h.c. = V1,−2,2,1 + V1,1,−1,1 + V2,−1,1,2 + h.c. ,

OC = I+ U + + U+ K − + K− I + + h.c. = V0,0,1,−1 + V0,0,1,2 + V2,0,0,1 + h.c. ,

OAω = ω 2 I+ I + + ωU+ U + + K− K − + h.c. ,

OBω = ωI+ K − + U+ I + + ω 2 K− U + + h.c. ,

OCω = I+ U + + ω 2 U+ K − + ωK− I + + h.c. . (4.98)

Finally, at the bi-critical point this set of operators is mapped to

O′ A = V2,−2,1,1 + V0,2,−1,0 + V2,0,0,1 + h.c. ,

O′ B = V0,1,1,0 + V1,0,0,−1 + V−1,1,1,1 + h.c. ,

O′ C = V3,0,0,0 + V3,−3,0,0 + V0,3,0,0 + h.c. , (4.99)

and similarly for O′ Aω , O′ Bω , O′ Cω . Therefore we have the two exactly marginal operators
O′ 1 , O′ 2 that span the Zorb
3 branch. However in this case all the non-invertible duality
symmetries are broken by all these marginal deformations.

Relevant deformations

The relevant operators at the bi-critical point are the following:

(h, h) (n, w)
1 1

9, 9 (±1, 0, 0, 0) , (0, ±1, 0, 0) , (±1, ∓1, 0, 0)
1 1

3, 3 (±1, ±1, 0, 0) , (±2, ∓1, 0, 0) , (±1, ∓2, 0, 0)
4 4

9, 9 (±2, 0, 0, 0) , (0, ±2, 0, 0) , (±2, ∓2, 0, 0)
(0, 0, 0, ±1) , (±1, 0, 0, ±1) , (0, ±1, ∓1, 0) , (0, 0, ±1, ±1)
7 7

9 , 9 (0, 0, ±1, 0) , (±1, ∓1, ±1, ±1) , (±2, ±1, 0, 0) , (±1, ±2, 0, 0)
(±3, ∓1, 0, 0) , (±1, ∓3, 0, 0) , (±3, ∓2, 0, 0) , (±2, ∓3, 0, 0)

41
As in the quadri-critical point, we can find combinations that preserve some non-invertible
duality symmetry, therefore leading to interesting constraints on the corresponding RG
flow. In this case however, it is more challenging to determine the IR theory resulting
from these duality preserving RG flows. Therefore all the possible symmetry constraints
which we will describe in the following may lead to useful predictions for the low energy
dynamics.
As before, let us give some representative examples which will serve to highlight some
interesting features regarding these RG flows. A simple relevant deformation which pre-
serves D1 is
R1/9 ∼ cos(φ2 ) . (4.100)
Such a deformation preserves U (1)n1 × U (1)w1 × U (1)w2 , with the first two factors partic-
ipating in a continuous mixed ’t Hooft anomaly; therefore we conclude that the IR theory
is gapless. Because of the presence of D1 we also deduce that such gapless theory must
enjoy a non-invertible TY(Z3 ) symmetry. A very natural candidate for this IR theory is

therefore the R = 3 c = 1 CFT. However, as opposite to the quadri-critical case, since
the bi-critical point is not factorized, we cannot immediately deduce the IR fate of this
process. In order to determine it, we can however perform a topological manipulation
which brings the theory to a factorized point, perturbed by a genuine relevant operator.
As explained around (4.45) we have
√ √
f=2 : (ω, α) → (i 3, i3 3) ,
Tρ Tτ σM (4.101)

and, by this manipulation, the relevant deformation is mapped to

Tρ Tτ σM √ √
f=2 : cos(φ2 )|(ω,α) → cos(2φ2 )|(i 3,i3 3) . (4.102)
√ √
Therefore, at (i 3, i3 3), we find that the sector described by φ2 flows to a gapped phase
with two degenerate vacua, i.e. the Z2 ∈ U (1)n2 is spontaneously broken. The reason for
this is that the Z2 shift symmetry which is preserved by the deformation (4.102) at the
factorized point is itself involved in a mixed anomaly with a Z2 subgroup of the winding
symmetry. We then conclude that the IR theory is described by
√ √  √ 
RG flow
(i 3, i3 3) + cos(2φ2 ) −−−−−→ c = 1 R = 3 × ( 2 vacua) (4.103)

In order to understand the IR of the bi-critical point we can now perform the inverse
topological manipulation
√ √
σNe =2 Tρ−1 Tτ−1 (i 3, i3 3) = (ω, α) (4.104)

which maps the two vacua mentioned above to a single trivial ground-state. Therefore we
get 
RG flow √ 
(ω, α) + cos(φ2 ) −−−−−→ c = 1 R = 3 (4.105)

42
in agreement with the expectations inferred from the symmetries.
We now discuss a slightly more involved example, constructed by considering the com-
bination of relevant deformations35

R1/9,7/9 ∼ cos(φ2 ) + cos(3φ1 − φ2 ) + cos(φe1 ) . (4.106)

While cos(φ2 ) preserves the non-invertible symmetry D1 , the relevant operator cos(3φ1 −
(w )
φ2 ) + cos(φe2 ) preserves only the diagonal symmetry generated by D e1 ≡ ηe1 D1 (e
η1 ∈ Z2 2 )
due to the phase (4.74) produced by the action of D1 . The full set of symmetries preserved
by R1/9,7/9 is then
(1) e1 =⇒ T Y (Z(1) ) × U (1)
Z3 × U (1)w2 × D 3 (4.107)

By looking just at the invertible part of the symmetry structure, this RG flow is compatible
with a trivially gapped phase. However the presence of an anomalous TY(Z3 ) implies
that the theory is either gapples or gapped with a spontaneoulsy broken non-invertible
symmetry [35, 38, 39].

5 Duality symmetries of SU(3)1 WZW

We now want to comment about the (non-invertible) symmetry structure of the c = 2


theory at (τ = w, ρ = w), where the chiral algebra enhances to SU (3) × SU (3) (see ap-
pendix A.2) and the theory is equivalent to SU (3)1 Wess-Zumino-Witten model (WZW).
As opposed to the c = 1 case, this theory also enjoys non-invertible duality defects exactly
as the other rational points of the toroidal branch36 . An easy way to express such sym-
metries is by combining the ones discussed at the bi-critical point in section 4.2.1 with the
duality and gauging operations which connect the bi-critical point with the SU (3)1 WZW
(see the discussion around (4.43)). We then find that the non-invertible defect D1 leads
to the following one at the SU (3) point
 
−1 2 −4 −2
SU (3) SU (3) 1−2 1 −2 −4

D1 → D1 =   , (5.1)
3 −4 2 −1 −2
2 −4 2 1
35
Because these three deformation have no the same scaling dimensions, there is a range in the parameter
space where cos(φ2 ) dominates and the IR theory is approximately the one described before. However in
a generic point of the phase diagram, both deformations need to be taken into account.
36
Also the SU (2) × SU (2) point enjoys non-invertible duality defects as already emphasized in [47]. Here
we focus on the SU (3) point since it cannot be expressed as product of c = 1 theories. We also emphasize
that these non-invertible symmetries are not part of the Verlinde lines of WZW since for SU (N )1 all the
Verlinde lines are invertible.

43
where, for sake of simplicity, we chose to display only its representation in terms of an
O(2, 2, Q) matrix. It is straightforward to verify that such matrix leaves invariant the
generalized metric at (τ, ρ) = (ω, ω).
By similar methods, we can further evaluate the duality symmetries descending from
the non-invertible defects D2 and D3 at the bi-critical point. However, we find that, at
the SU (3)1 WZW theory, these defects reduce to the one in (5.1) stacked with invertible
defects. Explicitly we find
SU (3) SU (3) SU (3) SU (3)
D2 = −M D1 , D3 = M3 D1 , M3 = Tτ−1 Sτ Tρ−1 Sρ (5.2)

where M is mirror symmetry and both M and M3 are invertible symmetries of this point37 .
SU (3)
Therefore we have just one independent non-invertible defect, say D1 . Therefore, we
have found that the SU (3)1 enhanced symmetry point at (w, w) has a TY(Z3 )χ,ǫ non-
invertible symmetry.
The presence of this defect implies that the theory at (τ = w, ρ = w) is self-dual under
gauging the following Z3 subgroup of the global symmetry
2πi
k(2n1 +n2 +2w1 +2w2 )
Z3 : Vn,w → e 3 Vn,w . (5.3)

Now we proceed to prove the self-duality under gauging the Z3 symmetry in (5.3) by
explicitly computing the orbifold partition function. The chiral algebra at the SU (3) point
is characterized by a K-matrix of the form
!
2 −1
K= , (5.4)
−1 2

and the partition function is given by the diagonal modular invariant


X
Z(1,1) = χλ χλ , (5.5)
λ∈L

where the characters and the lattice L are defined as before, but now with the K-matrix
introduced in (5.4). Furthermore, note that
T
K −1 v3
Z3 : χλ χλ → e2πiλ χλ χλ (5.6)

with v3 ≡ (0, −1)T . Denoting by η to the generator of the above Z3 action, the twisted
partition function results
1 X 2πiaλT K −1 v3 2πibµT K −1v3 −i2πλT K −1(µ−µ)
Z(ηa ,ηb ) = e e e χµ χµ (5.7)
|L|
λ,µ,µ∈L

37
Another invertible symmetries of the SU (3) point are Sρ Tρ and Sτ Tτ . It is easy to check that they do
note generate other non-trivial non-invertible symmetries when translated to the bi-critical point. Indeed
Sτ Tτ is still an invertible symmetry of the bi-critical point while Sτ Tτ Sρ Tρ = M32 .

44
Following the same steps as in the previous sections, summing over λ and subsequently
over a, b ∈ Z3 , we obtain
!
1 X X 1 0
Z(ηa ,ηb ) = χµ χCµ , C = . (5.8)
3 −1 −1
a,b∈Z 3 µ∈L

T
Again, it can be easily verified that C preserves the pairing, namely CKC = K, hence
K Cµ ∼ K µ , leading to the claimed self-duality.
SU (3)
Just for completeness, let us remark that the topological defect D1 induces a
consistent defect Hilbert space, as it complies with the conditions imposed by the modular
bootstrap. First, the right hand side of equation (A.9) trivializes when evaluated on
invariant states under (5.3) and consequently we set α(n, w) = 0. Furthermore, the
SU (3)
action of D1 amounts to take p1 → −p1 and p2 → −p2 , hence the twisted partition
function only accounts for the contribution of states with p1 = p2 = 0. The computation
is almost identical to the one depicted around (4.81) and we will not include the details
here. Performing the modular S transformation, the trace over the defect Hilbert space is
then of the form
1 2    
Z(DSU (3) ,1) = 4
ϑ2 τ2 ϑ3 τ̄6 ϑ3 τ̄2 + ϑ4 τ̄6 ϑ4 τ̄2
1 4|η(τ )|
 
2 1 1 1 5 5 1
= q + 6q q̄ + 2q + 12q q̄ + . . .
8 8 3 8 8 3 (5.9)
|η(τ )|4

Regarding the set of relevant deformations at this point, we get

(h, h) (n, w)
(±1, 0, 0, ±1) , (±1, 0, 0, 0) , (±1, ∓1, ±1, ±1) , (±1, ∓1, 0, 0)
1 1

3 , 3 (0, ±1, 0, 0) , (0, ±1, ∓1, 0) , (0, 0, ±1, ±1) , (0, 0, ±1, 0)
(0, 0, 0, ±1)
Contrary to the multicritical points, now there is no relevant deformation which preserves
the non-invertible duality symmetry since any of the above operators is send to a non
SU (3)
genuine one under the action of D1,2,3 .

Acknowledgments

We would like to thank Riccardo Argurio, Luigi Tizzano, Christian Copetti, Andrea Antin-
ucci and Giovanni Rizi for a careful reading of the manuscript and for giving us precious
comments. The work of S.M. is supported by ”Fondazione Angelo Della Riccia” and by
funds from the Solvay Family. J.A.D. is a Postdoctoral Researcher of the F.R.S.-FNRS
(Belgium). G.G. is partially supported by funds from the Solvay Family. The research

45
of J.A.D. is supported by IISN-Belgium (convention 4.4503.15) and through an ARC ad-
vanced project.

A Details on compact bosons

In this section we expose some conventions and useful formulas that are used in the main
text. We adopt the conventions in which a periodic free holomorphic field X satisfies

X(z) ∼ X(z) + 2 2πR , hX(z)X(w)i = − log(z − w) (A.1)

Along the toroidal branch of a c = D theory of free periodic scalar fields, a generic vertex
operator is written as
e
Vn,w (z, z) ≡ ein.φ(z,z)+iw.φ(z,z) = eip.X(z)+ip.X(z) (A.2)

where we defined the D-dimensional vectors

φ = (φ1 , . . . , φD ) , e = (φe1 , . . . , φeD ) ,


φ
X = (X1 , . . . , XD ) , X = (X 1 , . . . , X D ) , (A.3)
n = (n1 , . . . , nD ) , w = (w1 , . . . , wD ) ,

where {n1 }, {wi } denote the charges under the U (1)D D


n ×U (1)w global symmetry. Of course,
operators along the orbifold branch can be written in a similar fashion, though arbitrary
e
charges are not generically gauge invariant. The exact relation between the fields (φ, φ)
and (X, X) (or equivalently between (n, w) and (X, X)) depends on the details of the
theory in consideration and we will be explicit in the examples considered in this paper.
By omitting the antiholomorphic part for notational simplicity, the operator product
expansion of two vertex operators reads
′ ′
P∞ m m ′
Vn,w (z, z)Vn′ ,w′ (w, w) ∼ (−1)n·w (z − w)p.p : ei m (z−w) (p·∂X(w))
:: ei(p+p )·X (w) :
(A.4)
n·w ′
Let us briefly comment on the (−1) phase arising in the above OPE. The reason for
it is due to imposing locality on the correlation functions of the theory. Take for instance
a correlation function of the form

hVn,w (z, z)Vn′ ,w′ (w, w) . . .i (A.5)

where the . . . denote possible additional insertions at points away from z and w. By
locality, we mean that the above correlation function should be single valued, i.e. invariant
under taking a point, say w, along a loop around z and back to its original position. This
turns out to be non-trivial by the fact that momentum and winding modes are not mutually

46
local, due to the mixed anomaly. The operation just described is achieved by exchanging
the insertions and then analytically continuing to a function of (z − w) ((z − w)). For
sake of simplicity, let us omit the additional insertions in (A.5) and focus on the two-point
function. At the level of the OPE, one obtains
′ ′ ′
Vn,w (w, w)Vn′ ,w′ (z, z) ∼ (−1)n ·w (w − z)p.p (w − z)p.p (· · · )
′ ′ ′ ′ ′
∼ (−1)n ·w (−1)n·w +n ·w (z − w)p.p (z − w)p.p (· · · ) (A.6)

where (· · · ) denote normal ordered operators which do not affect the present argument.
In going to the second line, we analytically continued by taking (w − z) → eiπ (z − w)
((w − z) → e−iπ (z − w)) and then expressed the resulting phase in terms of the mutual
spin38 . The correlation function is then invariant

hVn,w (w, w)Vn′ ,w′ (z, z)i = hVn,w (z, z)Vn′ ,w′ (w, w)i , (A.7)

consistently with locality. More generally, it can be proven that the overall phase in (A.4)
is necessary and sufficient to ensure the locality of a generic correlation function (see [14]
and [65] for a recent discussion).
Moreover, the presence of this phase also affects the action of some topological opera-
tors implementing dualities that exchange momentum and winding modes. We illustrate
this for the case of the T -duality defect at c = 1 in Figure 3, leading to the following
constraint

α(n + n′ , w + w′ ) + α(n + n′ , w + w′ ) + α(n + n′ , w + w′ ) = nw′ + n′ w mod 2 (A.8)

A possible solution of the above is α(n, w) = nw, hence obtaining the action (2.8).
The above argument can be generalized for any duality defect D at c = 2. For this
purpose, let us denote by Dn and Dw respectively to momentum and winding charges
transformed by the action of D. These can be easily read from resorting to the matrix
representation of the duality symmetry action, as we explain in the main text. Again, we
denote by α(n, w) to the potential phase generated by the action of D on Vn,w . Now, the
argument depicted in Figure 3 leads to

α(n + n′ , w + w′ ) + α(n, w) + α(n′ , w′ ) = n · w′ + (Dn) · (Dw′ ) mod 2 (A.9)

A.1 The self-dual point at c = 1

As the dynamics of one of the c = 2 multicritical theories studied in this paper, namely
the quadri-critical point, keeps a close relation with the self-dual point at c = 1, we briefly
review some features about the latter theory in this subsection.
38
In fact, regardless of the particular relation between (p, p) and (n, w), the 2D-dimensional momentum
lattice corresponding to this kind of theories always satisfies that the mutual spin p.p′ −p.p′ = n·w′ +n′ ·w.
Notice that we are assuming that both insertions are genuine local bosonic operators, hence n·w′ , n′ ·w ∈ Z

47
D

Vn,w Vn′ ,w′


z ω

Vn+n′ ,w+w′ VN w, Nn VN w′ , n′

= (−1)α(n,w)+α(n′ ,w′ )
N
(−1)nw z z ω

VN (w+w′ ), n+n′ VN (w+w′ ), n+n′


′ ′ ′ N ′ ′ ′ N
(−1)α(n+n ,w+w ) (−1)nw z
= (−1)α(n,w)+α(n ,w )+n w z

Figure 3: Constraint on the phase factor α coming from mutual locality, combined with
the action of the non-invertible symmetry.

The c = 1 theory at radius R = 1 is characterized by the presence of additional spin


one conserved currents, that enhance the chiral algebra to su(2) × su(2). More precisely,
the holomorphic currents
1  i√2X(z) √  1  i√2X(z) √  i
J1 (z) = e − e−i 2X(z) , J2 (z) = e + e−i 2X(z) , J3 (z) = √ ∂X(z)
2 2i 2
satisfy the su(2)1 chiral algebra

1 δij iǫijk Jk (w)


Ji (z)Jj (w) = 2
+ + ... (A.10)
2 (z − w) z−w

with the totally antisymmetric tensor satisfying ǫ123 = 1. Equivalently, in the complexified
basis √ 1 2iJ3 (w)
J± (z) = e±i 2X(z) ⇒ J+ (z)J− (w) = + + ... (A.11)
(z − w)2 z−w
The key observation is that at R = 1 there are two Z2 symmetries, the shift Zs2 and
the orbifold Zorb
2 which act on the fields and currents as
√ π √ π
Zs2 : X→X+ 2 , X→X+ 2 , (J3 , J± ) → (J3 , −J± )
2 2
Zorb
2 : X → −X , X → −X , (J3 , J± ) → (−J3 , J∓ ) , (A.12)

48
with a similar action on the anti-holomorphic part. The quotient of Zs2 , combined with
T -duality39 , maps the theory to the KT point (R = 2). On the other hand, the quotient
by Zorb
2 defines the origin of the orbifold branch (Rorb = 1). Since the global symmetry
exchanges the two Z2 , the two theories must be equivalent, leading to a multicritical
point. In [20] it is showed that the KT point hosts a non-invertible duality symmetry,
that satisfies the structure of a Z4 Tambara-Tamagami category symmetry. Moreover, the
duality symmetry is preserved on the orbifold branch.
Taking the product of two copies of the self-dual theory described above describes the
SU (2)2 point (τ, ρ) = (i, i) on the toridal branch of the c = 2 conformal manifold. The
symmetry algebra of the SU (2)2 point is then simply accounted for by two copies of the
above currents, constructed in terms of (φ1 , φe1 ) and (φ2 , φe2 ) respectively.

A.2 The SU (3) point at c = 2

The point with maximal enhanced symmetry within the c = 2 toroidal branch corresponds
2π √
to (τ, ρ) = (ω, ω), with ω = ei 3 . Given that τ ,ρ are in Q( −3), the corresponding theory
is rational and, moreover, diagonal since τ = ρ. The condition pi = 0 has eight solutions
with h = 1, which we assemble in the following currents
√ √ 1
2X 1
I± = ±V±1,∓1,±1,0 = ±e±i , I ± = ±V±1,0,∓1,0 = ±e±i 2X

q q
1 3 2
±i √1 X 1 ±i 3 2
X ±i √1 X ±i X
K± = ∓V±1,0,±1,±1 = ∓e 2 2 , K ± = ∓V0,±1,∓1,∓1 = ∓e 2 2 ,
q q
1 3 2
∓i √1 X 1 ±i 3
X2 ∓i √1 X ±i X
U± = ±V0,±1,0,±1 = ±e 2 2 , U ± = ±V∓1,±1,0,∓1 = ±e 2 2 ,
√ 3 √ 1
J 3 = i 2∂X 1 , J = i 2 ∂ X
√ 8 √ 2
J 8 = i 2∂X 2 , J = i 2 ∂ X , (A.13)

which respect the su(3) algebra


   3   3 
J 3 , I± = ±2I± , J , U± = ∓U± , J , K± = ±K± ,
 8   8  √  8  √
J , I± = 0 , J , U± = ± 3U± , J , K± = ± 3K± ,

[I+ , I− ] = J 3 , [I+ , U+ ] = K+ [I+ , K− ] = −U− ,


1 √ 8  1 √ 8 
[U+ , K− ] = I− , [U+ , U− ] = 3J − J 3 , [K+ , K− ] = 3J + J 3 , (A.14)
2 2
39
Equivalently, one may quotient by a Z2 ⊂ Uw (1).

49
and similarly for the antiholomorphic currents. The remaining commutators are zero. In
terms of the basis
1 1 
I± = J ± iJ 2 ,
2
1 4 
K± = J ± iJ 5 ,
2
1 6 
U± = J ± iJ 7 , (A.15)
2
the su(3) chiral algebra reads

1 δij ifijk Jk (w)


Ji (z)Jj (w) = 2
+ (A.16)
2 (z − w) (z − w)

with the fully antisymmetric structure constants given in the Gell-Mann presentation

1 3
f123 = 1 , f147 = −f156 = f246 = f257 = f345 = −f367 = , f458 = f678 =
2 2
(A.17)
Note that there are three distinguished SU (2) subalgebras generated by
√ √
{J 1 , J 2 , J 3 } , {J 4 , J 5 , J 3 + 3J 8 } , {J 6 , J 7 , + 3J 8 − J 3 } . (A.18)

B Partition function for RCFTs at c = 2

In this section, we briefly review the construction of the partition function in terms of
representations of the extended chiral algebra at rational points along the toroidal branch.
This construction is well known and we will follow the recent presentation in [16].
At a generic rational point in the toroidal branch, the theory features an extended
chiral algebra of the form
u(1)2KL × u(1)2KR (B.1)

where KL and KR are positive symmetric even integer matrices. In the following, we
describe how these matrices are defined and, more importantly, how to construct the
representations of (B.1).
We start by writing the scaling dimensions as40
1 2 1 2
h= ~n + M w
~ G−1
, h̄ = ~n − M T w
~ G−1
(B.2)
4 4
where M = G + B, with G and B the metric and B-field respectively introduced in
section 3. The charge vectors ~n, w
~ are two dimensional in this case, but one may apply
40
We denote |v|2M = v i Mij v j .

50
this construction to any c ∈ Z. The condition for an operator to be chiral then reads
~n = M T w.~ Notice that for generic M ∈ GL(2, Q), then M T w ~ is not necessarily in Z2
for all w,
~ hence not corresponding to any genuine primary operator. In order to avoid
that, we pick a sublattice ΛL such that for any w~ ∈ ΛL then M T w ~ ∈ Z2 . We assemble a
particular basis for ΛL into the columns of a matrix PL . Of course, the choice of basis is
completely arbitrary, hence PL is itself defined up to conjugation by unimodular integer
matrices. The same procedure is applied to the anti-chiral sector, where the condition
instead reads ~n = −M w,
~ hence defining PR in a similar fashion.
In terms of the objects just introduced, we proceed to define the KL , KR matrices in
(B.1), namely
KL = 2PLT GPL , KR = 2PRT GPR . (B.3)

The (finite) set of representations of (B.1) are in one-to-one correspondence with the
elements of the following lattices

LL,R ≡ v ∈ Z2 , v ∼ v + KL,R v′ , v′ ∈ Z2 . (B.4)

Notice that there is a total of |LL | = |LR | = detKL = detKR non-equivalent elements
in these lattices. A given (anti-)chiral primary of (B.1) is labeled by a vector λ ∈ LL
(λR ∈ LR ) and the characters of its associated representation then read

1 X 2 |λL +KL l|2K −1 1 X 2 |λR +KR r|2K −1


1 1

χλL (τ ) = q L , χ̄λR (τ̄ ) = q̄ R (B.5)


η(τ )2 2
η(τ̄ )2 2
l∈Z r∈Z

Notice that the dimensions associated to the primaries are


1 1
h = |λL |2K −1 , h̄ = |λR |2K −1 (B.6)
2 L 2 R

whereas the integer vectors l and r parametrize their descendants under the action of the
higher spin currents of the chiral algebra (B.1).
The modular invariant partition function is determined by a particular pairing between
the left- and right-moving sectors. Such a pairing is given by a group isomorphism ω̂ :
LL → LR defined as follows. First, it is easy to verify that, for a given M = G + B, there
exist integer 2 × 2 matrices N1,2 , such that
!
PLT PLT M
S= ∈ SL(4, Z) (B.7)
N1 N2

Its inverse is also a unimodular integer matrix and reads


!
−1 R0 M N3
S =S= (B.8)
−S0 −N3

51
for some R0 , S0 and N3 (see [16] for a proof that these matrices always exist). The
left-right pairing ω̂ is therefore written as
ω̂ = PRT R0 + PRT M T S0 (B.9)
and the partition function results
X
Z= χλL χω̂λL (B.10)
~
λL ∈DL

Finally, the modular S and T matrices acting on the holomorphic characters are of the
form
1 −1 −i π6 +iπ|λ|2 −1
−2πiλT KL µ K
Sλ,µ = e , Tλ,µ = δλ,µ e L , (B.11)
|LL |1/2
with analogous counterparts for the antiholomorphic blocks. A useful identity is the
following X T −1
e−i2πλ K µ = |D|δµ,0 . (B.12)
~
λ∈L
with 0 denoting the trivial vector in L. In all these expressions, the constraints imposed
by the delta functions are intended to hold up to lattice identifications. In particular, the
above properties ensure the modular invariance of the partition function.
Given the representations introduced above, it is instructive to write the corresponding
charges n, w under U (1)n ×U (1)w symmetry in terms of λL , λR , l and r. To this purpose,
we define the orthogonal matrix O and such that
K = O T diag(k1 , k2 )O (B.13)
for some integers ki . Since K is symmetric, the orthogonal matrix O is guaranteed to
exist. Comparing the expressions for the scaling dimensions in terms of either variables
one arrives to the following
1
pi = √ Oij (λj + (KL l)j ) (B.14)
ki
1 
pi = √ Oij λ̄j + (KR l)j (B.15)
ki
where we redefined λ ≡ λL (λ ≡ λR ), and pi (pi ) are the usual left (right) chiral weights.
The above equivalence defines a set of four linear equations for ni , wi .

C Tambara-Yamagami categories

In this appendix we review some of the properties of the Tambara Yamagami categories [66]
which describe categorical symmetries arising from self-dualities under gauging some dis-
crete symmetry. Given a discrete abelian group A, a symmetric non-degenerate bicharacter
χ : A × A −→ U (1) (C.1)

52
and a class ǫ ∈ H 3 (Z2 , U (1)) = Z2 called Frobenius Shur indicator, the Tambara Yam-
agami category, also dubbed as TY(A)ǫ, χ , is a graded fusion category:
C = C0 ⊕ C1 , C0 = VecA , C1 = D (C.2)
with fusion rules
M
a × b = ab , a×D = D×a = D, D×D = a (C.3)
a∈A

and non-trivial F-symbols41


h i h i h i ǫ
FDa, D, b = FbD, a, D = χ(a, b) , FDD, D, D =p χ(a, b)−1 . (C.5)
D, D D, D a, b |A|
The category VecA , i.e. the category of A graded vector spaces, describes an anomaly free
invertible 0-form symmetry A while D is the non-invertible defect which grades A and
p
with quantum dimension dim(D) = |A|, as evident from the last fusion rule in (C.3).
From the structure of TY(A)ǫ, χ follows that a theory with this categorical symmetry is
automatically self-dual under gauging of A. Indeed the gauging of A can be performed by
L
inserting a fine enough mesh of the algebra object A = a∈A a inside the 2d space-time.
However, because of (C.3) a mesh of A can always be reduced to contractible insertions
of D
D
A
=
A D

A D
(C.6)
automatically implying the claimed self-duality.
Even more interestingly, also the opposite is true: any 2d theory self-dual under the
gauging of an abelian symmetry A enjoys an additional topological line D which follows
the structure described above [26, 67]. More concretely, such topological line can be con-
structed by gauging A on half-space and imposing Dirichlet boundary conditions for the
gauge field

T T /A

D (C.7)
41
The F-symbols are a base-dependent representation (i.e. after choosing a set of simple objects) of the
natural isomorphism
Fx,y,z : (x ⊗ y) ⊗ z → x ⊗ (y ⊗ z) (C.4)
where x, y, z are some objects inside the category. In the physical language of topological lines, it relates
to in-equivalent line configurations

53
Because T /A ≃ T , the domain-wall D actually describes a topological defect within the
theory T . From this point of view, the choice of the bicharacter χ has a nice physical
interpretation. While on the l.h.s. of D we have a theory with global symmetry A, on
the r.h.s. the gauged theory has global symmetry A∨ ≡ Hom(A, U (1)) which is non-
canonically isomorphic to A. In order to identify the two symmetries, we actually need to
choose a group isomorphism φ : A → A∨ and the bicharacter is then defined as

χ(a, b) = φ(a)b ∈ U (1) . (C.8)

Such categorical structure can be generalized by grading A with any group G (possibly
also non-Abelian). In this case the full category C can be decomposed as direct sum of
abelian categories M
C= Cg (C.9)
g∈G

where C1 = A while Cg contain different non-invertible duality defects as their set of objects.
These kind of symmetries are marginal in our considerations and we refer to [31, 68] for
more details. We just mention that theories which are self-dual under gauging of A with
some non-trivial discrete torsion enjoy this kind of categorical symmetries.
Duality symmetries following TY(A)ǫ, χ can also have anomalies which constrain the
possible IR behaviors of symmetry preserving RG flows [20, 38]. Such anomalies can be
defined as obstruction to have a trivially gapped phase or as obstruction to gauge the
categorical symmetry. Even if the two concepts coincides for invertible symmetries, it
was shown in [37] that this two notion are generically inequivalent for fusion category
symmetries. In the particular case of TY(A)χ,ǫ with A ≡ ZN , a symmetric RG flow is
compatible with a symmetry preserving gapped phase only if N = r 2 for some integer r
while the IR theory must either break spontaneously the duality symmetry or must be
gapples when N 6= r 2 . In the former case, the symmetry can be preserved in the IR if and
only if ǫ is trivial. Similar conditions for generic groups A can be found in e.g. [38].
We conclude this section with a brief overview on how the topological data correspond-
ing to the type of categories described above can be extracted from the modular bootstrap.
We refer to [20] for a more detailed treatment. Resorting to modular covariance, applying
the modular T transformation (i.e. τ → τ + 2) to the twisted Hilbert space, we end up
with the configuration depicted on the right of

54
τ →τ +2
−−−−−→ (C.10)

D D
Alternatively, the action τ → τ + 2 generates a phase proportional to the spin, namely
ˆ X
Z(D,1) (τ + 2) = TrHD e−2πτ2 ∆ e2πi(τ1 +2)ŝ = e4πis q h q h (C.11)
HD

where ∆ˆ = L0 + L0 and ŝ = L0 − L0 are the usual dilatation and spin operators.


Now, the trick amounts to map the configuration on the right of (C.10) to a trace over
the defect Hilbert space HD . This can be done by combining different F-moves, as shown
in (C.12).

ǫ X ǫ X a
=p =p a
e |A| a a |A| a

D D D
(C.12)
We insert an identity line e running from two consecutive pieces of the defect D. Then
we perform an F -move to map it to a configuration on which D runs parallel to the
time direction, subsequently making use of χ(e, a) = 1 for any a. The resulting diagram
computes the the trace over the defect Hilbert space with a sum over the elements of A
inserted. In order to determine the action of the symmetry A on the states in HD , it is
required to resolve
h the ifour-way junction. The two possibilities are related by an F -move,
more precisely FDa,D,a d
= χ(a, a). Following [], we denote by (η a ) to the resolution

D,D
in the right hand side of (C.12), namely the one in which the left-insertion is placed above
the right-insertion.
Finally, modular covariance implies the following constraints over the spin of states in
the defect Hilbert space
ǫ X da
e4πiŝ =p (η) − (C.13)
HD |A| a∈A HD

Note the above equality is written as an operator equation. In practice, one decomposes

55
HD in terms of common eigenstates for both operators42 and demands for the eigenvalues
to match. Note that the spin selection rule (C.13) involves both χ and ǫ (the bi-character
is implicit in the choice of resolution for the junction). Therefore, this piece of topological
data can be a priori determined by inspection of the spectrum of spins arising in HD . We
show how it works in several examples in the main body of this article. Let us conclude
by mentioning that there may be cases for which (C.13) is satisfied by more than one
inequivalent combination of (χ, ǫ). In such a case, a more refined analysis is required to
fully fix the categorical data.

D Modular functions

Let us give a brief overview of the modular functions employed in the calculations exposed
in the main text. We will mainly follow [69] for the conventions and transformation
properties.
First, the Dedekind eta function is defined as

Y
1
η(τ ) = q 24 1 − qm , (D.1)
m=1

with q (q̄) defined as usual q = e2πiτ


(q̄ = e2πiτ̄ ). Under the modular S transformation,
the function above transforms as
 √  √
η − τ1 = −iτ η(τ ) , η − τ̄1 = iτ̄ η(τ̄ ) . (D.2)

In addition, we introduce the elliptic Jacobi theta function with characteristics defined as
follows43  
α X 2
ϑ (τ ) = eπiτ (n+α) e2πi(n+α)β , (D.3)
β
n∈Z
where the real parameters α, β take values on the interval [0, 1). These functions have
well defined transformations under the action of the modular group, in particular
   
α 1
 √ 2πiαβ β
ϑ − τ = −iτ e ϑ (τ ) . (D.4)
β −α
A predominant role is usually played by these functions when α ∼ −α, β ∼ −β, that is
α, β ∈ {0, 12 }, hence they are given given special names
" # 1    
1
2 2 0 0
ϑ1 (τ ) ≡ −ϑ 1 (τ ) , ϑ2 (τ ) ≡ ϑ , ϑ3 (τ ) ≡ ϑ , ϑ4 (τ ) ≡ ϑ 1 (τ ) .
2
0 0 2
(D.5)
42
This is always possible as these are commuting operators.
43
More generally, these are functions of two complex variables τ and z, where z is usually interpreted as
a fugacity. Throughout this work we will always set z = 0, as it does not play any role in for our purposes.

56
From (D.4), one can readily obtain their transformation under τ → −1/τ , namely
 √  √
ϑ1 − τ1 = −i −iτ ϑ1 (τ ) , ϑ2 − τ1 = −iτ ϑ4 (τ )
 √  √
ϑ3 − τ1 = −iτ ϑ3 (τ ) , ϑ4 − τ1 = −iτ ϑ2 (τ ) . (D.6)

Finally, there is also a twisted version of the Dedekind eta function which, in the
context of the present paper, often appears when the oscillator modes are odd under a
particular symmetry operation, as it is the case of T -duality at c = 1. In terms of the
elliptic functions just introduced, such a distribution reads

Y
1 η(τ )
q 24 1 + qm = . (D.7)
ϑ4 (2τ )
m=1

References

[1] P. Di Francesco, P. Mathieu, and D. Senechal, Conformal Field Theory. Graduate


Texts in Contemporary Physics. Springer-Verlag, New York, 1997. 3, 7

[2] P. H. Ginsparg, “APPLIED CONFORMAL FIELD THEORY,” in Les Houches


Summer School in Theoretical Physics: Fields, Strings, Critical Phenomena. 9,
1988. arXiv:hep-th/9108028. 3

[3] R. Dijkgraaf, E. P. Verlinde, and H. L. Verlinde, “C = 1 Conformal Field Theories


on Riemann Surfaces,” Commun. Math. Phys. 115 (1988) 649–690. 3

[4] S. Hellerman, “A Universal Inequality for CFT and Quantum Gravity,”


JHEP 08 (2011) 130, arXiv:0902.2790 [hep-th]. 3

[5] J. L. Cardy, “Operator Content of Two-Dimensional Conformally Invariant


Theories,” Nucl. Phys. B 270 (1986) 186–204. 3

[6] N. Benjamin, J. Lee, H. Ooguri, and D. Simmons-Duffin, “Universal Asymptotics


for High Energy CFT Data,” arXiv:2306.08031 [hep-th]. 3

[7] S. Ferrara, A. F. Grillo, G. Parisi, and R. Gatto, “Covariant expansion of the


conformal four-point function,” Nucl. Phys. B 49 (1972) 77–98. [Erratum:
Nucl.Phys.B 53, 643–643 (1973)]. 3

[8] A. M. Polyakov, “Nonhamiltonian approach to conformal quantum field theory,”


Zh. Eksp. Teor. Fiz. 66 (1974) 23–42. 3

[9] A. A. Belavin, A. M. Polyakov, and A. B. Zamolodchikov, “Infinite Conformal


Symmetry in Two-Dimensional Quantum Field Theory,”
Nucl. Phys. B 241 (1984) 333–380. 3

[10] R. Rattazzi, V. S. Rychkov, E. Tonni, and A. Vichi, “Bounding scalar operator


dimensions in 4D CFT,” JHEP 12 (2008) 031, arXiv:0807.0004 [hep-th]. 3

57
[11] D. Poland, S. Rychkov, and A. Vichi, “The Conformal Bootstrap: Theory,
Numerical Techniques, and Applications,” Rev. Mod. Phys. 91 (2019) 015002,
arXiv:1805.04405 [hep-th]. 3

[12] E. P. Verlinde, “Fusion Rules and Modular Transformations in 2D Conformal Field


Theory,” Nucl. Phys. B 300 (1988) 360–376. 3, 4

[13] J. Fuchs, I. Runkel, and C. Schweigert, “TFT construction of RCFT correlators 1.


Partition functions,” Nucl. Phys. B 646 (2002) 353–497, arXiv:hep-th/0204148.
3, 4

[14] J. Fuchs, M. R. Gaberdiel, I. Runkel, and C. Schweigert, “Topological defects for the
free boson CFT,” J. Phys. A 40 (2007) 11403, arXiv:0705.3129 [hep-th]. 3, 47

[15] D. S. Freed, G. W. Moore, and C. Teleman, “Topological symmetry in quantum


field theory,” arXiv:2209.07471 [hep-th]. 3

[16] F. Benini, C. Copetti, and L. Di Pietro, “Factorization and global symmetries in


holography,” SciPost Phys. 14 no. 2, (2023) 019, arXiv:2203.09537 [hep-th]. 3,
8, 50, 52

[17] L. Bhardwaj and S. Schafer-Nameki, “Generalized Charges, Part II: Non-Invertible


Symmetries and the Symmetry TFT,” arXiv:2305.17159 [hep-th]. 3

[18] C.-M. Chang, Y.-H. Lin, S.-H. Shao, Y. Wang, and X. Yin, “Topological Defect
Lines and Renormalization Group Flows in Two Dimensions,” JHEP 01 (2019) 026,
arXiv:1802.04445 [hep-th]. 3, 4, 6, 8, 9

[19] R. Thorngren and Y. Wang, “Fusion Category Symmetry I: Anomaly In-Flow and
Gapped Phases,” arXiv:1912.02817 [hep-th]. 3, 4, 9

[20] R. Thorngren and Y. Wang, “Fusion Category Symmetry II: Categoriosities at c =


1 and Beyond,” arXiv:2106.12577 [hep-th]. 3, 4, 6, 8, 9, 25, 29, 39, 49, 54

[21] C. Cordova and G. Rizi, “Non-Invertible Symmetry in Calabi-Yau Conformal Field


Theories,” arXiv:2312.17308 [hep-th]. 3

[22] L. Bhardwaj, L. E. Bottini, D. Pajer, and S. Schafer-Nameki, “The Club Sandwich:


Gapless Phases and Phase Transitions with Non-Invertible Symmetries,”
arXiv:2312.17322 [hep-th]. 3

[23] L. Bhardwaj, L. E. Bottini, D. Pajer, and S. Schafer-Nameki, “Gapped Phases with


Non-Invertible Symmetries: (1+1)d,” arXiv:2310.03784 [hep-th]. 3

[24] Y. Nagoya and S. Shimamori, “Non-invertible duality defect and non-commutative


fusion algebra,” arXiv:2309.05294 [hep-th]. 3, 24

[25] A. Giveon, M. Porrati, and E. Rabinovici, “Target space duality in string theory,”
Phys. Rept. 244 (1994) 77–202, arXiv:hep-th/9401139. 4, 10

58
[26] Y. Choi, C. Cordova, P.-S. Hsin, H. T. Lam, and S.-H. Shao, “Noninvertible duality
defects in 3+1 dimensions,” Phys. Rev. D 105 no. 12, (2022) 125016,
arXiv:2111.01139 [hep-th]. 4, 53

[27] Y. Choi, C. Cordova, P.-S. Hsin, H. T. Lam, and S.-H. Shao, “Non-invertible
Condensation, Duality, and Triality Defects in 3+1 Dimensions,”
Commun. Math. Phys. 402 no. 1, (2023) 489–542, arXiv:2204.09025 [hep-th]. 4

[28] A. Antinucci, F. Benini, C. Copetti, G. Galati, and G. Rizi, “The holography of


non-invertible self-duality symmetries,” arXiv:2210.09146 [hep-th]. 4

[29] V. Bashmakov, M. Del Zotto, A. Hasan, and J. Kaidi, “Non-invertible symmetries


of class S theories,” JHEP 05 (2023) 225, arXiv:2211.05138 [hep-th]. 4, 5, 24

[30] J. A. Damia, R. Argurio, F. Benini, S. Benvenuti, C. Copetti, and L. Tizzano,


“Non-invertible symmetries along 4d RG flows,” arXiv:2305.17084 [hep-th]. 4

[31] Y. Choi, D.-C. Lu, and Z. Sun, “Self-duality under gauging a non-invertible
symmetry,” arXiv:2310.19867 [hep-th]. 4, 23, 54

[32] O. Diatlyk, C. Luo, Y. Wang, and Q. Weller, “Gauging Non-Invertible Symmetries:


Topological Interfaces and Generalized Orbifold Groupoid in 2d QFT,”
arXiv:2311.17044 [hep-th]. 4

[33] V. B. Petkova and J. B. Zuber, “Generalized twisted partition functions,”


Phys. Lett. B 504 (2001) 157–164, arXiv:hep-th/0011021. 4

[34] V. Ostrik, “Fusion categories of rank 2,” arXiv:math/0203255 [math.QA]. 4

[35] A. Apte, C. Cordova, and H. T. Lam, “Obstructions to gapped phases from


noninvertible symmetries,” Phys. Rev. B 108 no. 4, (2023) 045134,
arXiv:2212.14605 [hep-th]. 4, 43

[36] C. Zhang and C. Córdova, “Anomalies of (1 + 1)D categorical symmetries,”


arXiv:2304.01262 [cond-mat.str-el]. 4

[37] Y. Choi, B. C. Rayhaun, Y. Sanghavi, and S.-H. Shao, “Remarks on boundaries,


anomalies, and noninvertible symmetries,” Phys. Rev. D 108 no. 12, (2023) 125005,
arXiv:2305.09713 [hep-th]. 4, 54

[38] A. Antinucci, F. Benini, C. Copetti, G. Galati, and G. Rizi, “Anomalies of


non-invertible self-duality symmetries: fractionalization and gauging,”
arXiv:2308.11707 [hep-th]. 4, 43, 54

[39] C. Cordova, P.-S. Hsin, and C. Zhang, “Anomalies of Non-Invertible Symmetries in


(3+1)d,” arXiv:2308.11706 [hep-th]. 4, 43

[40] A. Antinucci, C. Copetti, G. Galati, and G. Rizi, “”Zoology” of non-invertible


duality defects: the view from class S,” arXiv:2212.09549 [hep-th]. 5, 24

59
[41] J. L. Cardy, “Boundary Conditions, Fusion Rules and the Verlinde Formula,”
Nucl. Phys. B 324 (1989) 581–596. 5

[42] N. Ishibashi, “The Boundary and Crosscap States in Conformal Field Theories,”
Mod. Phys. Lett. A 4 (1989) 251. 5

[43] J. L. Cardy, “Boundary conformal field theory,” arXiv:hep-th/0411189. 5

[44] J. Dai, R. G. Leigh, and J. Polchinski, “New Connections Between String Theories,”
Mod. Phys. Lett. A 4 (1989) 2073–2083. 5

[45] J. Polchinski, “Dirichlet Branes and Ramond-Ramond charges,”


Phys. Rev. Lett. 75 (1995) 4724–4727, arXiv:hep-th/9510017. 5

[46] C.-T. Hsieh, Y. Nakayama, and Y. Tachikawa, “Fermionic minimal models,”


Phys. Rev. Lett. 126 no. 19, (2021) 195701,
arXiv:2002.12283 [cond-mat.str-el]. 5

[47] W. Ji, S.-H. Shao, and X.-G. Wen, “Topological Transition on the Conformal
Manifold,” Phys. Rev. Res. 2 no. 3, (2020) 033317,
arXiv:1909.01425 [cond-mat.str-el]. 5, 43

[48] J. D. Sau, B. I. Halperin, K. Flensberg, and S. Das Sarma, “Number conserving


theory for topologically protected degeneracy in one-dimensional fermions,”
Phys. Rev. B 84 (Oct, 2011) 144509. 5

[49] P. H. Ginsparg, “Curiosities at c = 1,” Nucl. Phys. B 295 (1988) 153–170. 6, 27

[50] J. A. Harvey and G. W. Moore, “An Uplifting Discussion of T-Duality,”


JHEP 05 (2018) 145, arXiv:1707.08888 [hep-th]. 7, 10

[51] Y.-H. Lin and S.-H. Shao, “Bootstrapping noninvertible symmetries,”


Phys. Rev. D 107 no. 12, (2023) 125025, arXiv:2302.13900 [hep-th]. 9

[52] Y.-H. Lin and S.-H. Shao, “Anomalies and Bounds on Charged Operators,”
Phys. Rev. D 100 no. 2, (2019) 025013, arXiv:1904.04833 [hep-th]. 9

[53] Y.-H. Lin and S.-H. Shao, “Duality Defect of the Monster CFT,”
J. Phys. A 54 no. 6, (2021) 065201, arXiv:1911.00042 [hep-th]. 9

[54] Y.-H. Lin and S.-H. Shao, “ZN symmetries, anomalies, and the modular bootstrap,”
Phys. Rev. D 103 no. 12, (2021) 125001, arXiv:2101.08343 [hep-th]. 9

[55] S. Dulat and K. Wendland, “Crystallographic orbifolds: Towards a classification of


unitary conformal field theories with central charge c = 2,” JHEP 06 (2000) 012,
arXiv:hep-th/0002227. 10, 14, 15, 21, 40

[56] W. Nahm, “On quasi-rational conformal field theories,”


Nucl. Phys. B Proc. Suppl. 49 (1996) 107–114. 12

60
[57] T. Gannon, “U(1)-m modular invariants, N=2 minimal models, and the quantum
Hall effect,” Nucl. Phys. B 491 (1997) 659–688, arXiv:hep-th/9608063. 12

[58] S. Gukov and C. Vafa, “Rational conformal field theories and complex
multiplication,” Commun. Math. Phys. 246 (2004) 181–210,
arXiv:hep-th/0203213. 12, 13, 22

[59] B. Rostand, “On multicritical lattice / orbifold models,”


Phys. Lett. B 248 (1990) 89–94. 14

[60] B. Rostand, “A Classification of multicritical toroidal conformal field theories,”


Phys. Lett. B 262 (1991) 240–248. 14

[61] O. Hohm, C. Hull, and B. Zwiebach, “Generalized metric formulation of double field
theory,” JHEP 08 (2010) 008, arXiv:1006.4823 [hep-th]. 15

[62] D. S. Berman and C. D. A. Blair, “The Geometry, Branes and Applications of


Exceptional Field Theory,” Int. J. Mod. Phys. A 35 no. 30, (2020) 2030014,
arXiv:2006.09777 [hep-th]. 15

[63] P. Niro, K. Roumpedakis, and O. Sela, “Exploring non-invertible symmetries in free


theories,” JHEP 03 (2023) 005, arXiv:2209.11166 [hep-th]. 23

[64] A. Hatcher, Algebraic Topology. Cambridge University Press, 2002.


https://pi.math.cornell.edu/~hatcher/AT/AT.pdf. 30

[65] I. Runkel, L. Szegedy, and G. M. T. Watts, “Parity and Spin CFT with boundaries
and defects,” arXiv:2210.01057 [hep-th]. 47

[66] D. Tambara and S. Yamagami, “Tensor categories with fusion rules of self-duality
for finite abelian groups,” Journal of Algebra 209 no. 2, (1998) 692–707. 52

[67] J. Kaidi, K. Ohmori, and Y. Zheng, “Kramers-Wannier-like Duality Defects in


(3+1)D Gauge Theories,” Phys. Rev. Lett. 128 no. 11, (2022) 111601,
arXiv:2111.01141 [hep-th]. 53

[68] S. Gelaki, D. Naidu, and D. Nikshych, “Centers of graded fusion categories,”


Algebra & Number Theory 3 no. 8, (2009) 959–990. 54

[69] E. D’Hoker and J. Kaidi, “Lectures on modular forms and strings,”


arXiv:2208.07242 [hep-th]. 56

61

You might also like