---
title: Modular Fusion Categories
url: https://www.emergentmind.com/topics/modular-fusion-category
type: topic
---

# Modular Fusion Categories

Searching arXiv for recent and foundational work on modular fusion categories, modular extensions, and related classification results.
A modular fusion category is, in one common usage, a fusion category over \(\mathbb C\) equipped with a spherical structure and a nondegenerate braiding, with modular data encoded by matrices \(S\) and \(T\). In unitary treatments of \(2+1\)D topological order, the same role is played by a unitary modular tensor category, and in the presence of symmetry one often works relatively: the transparent sector is not required to be trivial, but is fixed to be a symmetric fusion category \(\mathcal E\) [2509.02501; 1602.05936].

## 1. Foundational definitions and variants of modularity

A fusion category is a semisimple, finite, rigid tensor category with finitely many simple objects and simple tensor unit. A premodular category is a braided, balanced fusion category, and it is modular when its Müger center is trivial, \(\mathcal C'=\mathrm{Vec}\). In the braided pivotal setting, nondegeneracy is equivalently expressed by invertibility of the \(S\)-matrix. The Frobenius–Perron dimension is
\[
\operatorname{FPdim}(\mathcal C)=\sum_a d_a^2,
\]
and the pointed subcategory \(\mathcal C_{\mathrm{pt}}\) is generated by the invertible simple objects [1609.04896; 2209.04169; 1203.4180].

Unitary literature refines this picture. A unitary braided fusion category over \(\mathcal E\), abbreviated UBFC/\(\mathcal E\), is a unitary braided fusion category \(\mathcal C\) equipped with a braided full embedding \(\iota_{\mathcal E}:\mathcal E\hookrightarrow \mathcal C'\). If the center is exactly \(\mathcal E\), so that \(\mathcal C'=\mathcal E\), then \(\mathcal C\) is nondegenerate over \(\mathcal E\), or a UMTC/\(\mathcal E\). In this relative sense, modularity is measured after fixing the allowed transparent sector. This terminology matters because some treatments reserve “modular fusion category” for the absolute case \(\mathcal C'=\mathrm{Vec}\), whereas others, especially in the unitary topological-order setting, treat the modular object as a UMTC or as a UMTC/\(\mathcal E\).

The resulting ambiguity is not merely terminological. It separates two regimes: absolute modularity, where all transparent objects are trivial, and relative modularity, where the transparent sector is prescribed by symmetry. The latter is indispensable in symmetry-enriched and fermionic settings.

## 2. Modular data, twists, and numerical invariants

For a modular fusion category, the modular data consist of the matrices \(S\) and \(T\). The matrix \(T\) is diagonal, and its diagonal entries are the twists \(\theta_X\) of simple objects \(X\). These twists are roots of unity. A basic invariant is the Frobenius–Schur exponent,
\[
N=\operatorname{FSexp}(\mathcal C)=\operatorname{ord}(T).
\]
The normalized modular representation factors through a finite quotient
\[
\rho'_{\mathcal C}:\mathrm{SL}(2,\mathbb Z/n\mathbb Z)\to \mathrm{GL}(\cdot),
\]
where \(n=\operatorname{ord}(t)\) and \(N\mid n\). The Gauss sum is
\[
p_+=\sum_i d_i^2\,\theta_i,
\]
and the central charge is defined modulo \(8\) by
\[
p_+ = \sqrt{\mathrm{FPdim}(R)}\, e^{2\pi i c/8}.
\]
The projective modular relations include
\[
(ST)^3=p_+S^2,\qquad S^4=I,\qquad S^2=C
\]
in the integral modular-data setting [2509.02501; 2302.01613].

These invariants are powerful but incomplete. The modular data \((S,T)\) do not determine a modular category in general. A topological refinement is the Borromean tensor \(B\), defined by evaluating the Borromean rings colored by three simple objects:
\[
B_{ijk}:=\operatorname{ptr}\!\left(B\bigl((\sigma_2^{-1}\sigma_1)^3\bigr)\right).
\]
It satisfies symmetries such as \(B_{ijk}=B_{jki}=B_{kij}\) and \(B_{ijk}=B_{jik^*}\). For twisted Drinfeld doubles, \(T\) together with \(B\) distinguishes the \(p\) non-equivalent modular categories of the form \(\mathcal Z(\mathrm{Vec}_G^\omega)\) for \(G=\mathbb Z/q\mathbb Z\rtimes \mathbb Z/p\mathbb Z\), even though the modular data alone do not [1806.03158].

The arithmetic of twists is itself a classification tool. Categories with few distinct twists are highly constrained. If all \(t\)-eigenvalues have pairwise coprime orders, then every twist is in \(\{1,-1\}\), and a nontrivial modular fusion category in that regime must have exactly two distinct twists.

## 3. Drinfeld centers, modular extensions, and relative closure

A foundational source of modularity is the Drinfeld center. If \(\mathcal C\) is a pivotal fusion category over a commutative ring \(K\), then its center \(Z(\mathcal C)\) is modular in the sense of Lyubashenko \(2\)-modularity, and
\[
\dim Z(\mathcal C)=\dim(\mathcal C)^2.
\]
Here modularity is formulated via the coend \(C_B=\int^{Y\in B}Y^*\otimes Y\) and nondegeneracy of the canonical Hopf pairing. Over a field, \(Z(\mathcal C)\) is semisimple iff \(\dim(\mathcal C)\neq 0\); over an algebraically closed field, this is equivalent to \(Z(\mathcal C)\) being a fusion category [1203.4180].

Relative modularity leads to modular extensions. For a UMTC/\(\mathcal E\) \(\mathcal C\), a modular extension is a UMTC \(\mathcal M\) together with a braided full embedding \(\iota_{\mathcal M}:\mathcal C\hookrightarrow \mathcal M\) such that
\[
\mathcal E'|_{\mathcal M}=\mathcal C.
\]
This is the minimal modular closure in which the symmetry \(\mathcal E\) is gauged. The physical proposal is that \(2+1\)D anomaly-free topological/SPT orders with on-site symmetry \(\mathcal E\) are classified, up to stacking with \(E_8\) quantum Hall states, by triples
\[
(\mathcal C,\mathcal M,\iota_{\mathcal M}).
\]

The modular extensions of the symmetry sector itself form a finite abelian group:
\[
\mathcal M_{\mathrm{ext}}(\mathcal E),\ \text{with multiplication } \boxtimes_{\mathcal E}\ \text{and identity } (\mathcal Z(\mathcal E),\iota_0).
\]
For a fixed UMTC/\(\mathcal E\) \(\mathcal C\), if \(\mathcal M_{\mathrm{ext}}(\mathcal C)\neq\varnothing\), then it is a torsor over \(\mathcal M_{\mathrm{ext}}(\mathcal E)\): the action is free and transitive. The relative product \(\mathcal M\boxtimes_{\mathcal E}\mathcal N\) is defined by condensing a canonical algebra \(L_{\mathcal E}\) in \(\mathcal M\boxtimes\mathcal N\). This organizes symmetry gauging, stacking, and symmetry breaking within a single categorical formalism [1602.05936].

## 4. Classification programs and rigidity phenomena

Several classification programs show that modular fusion categories are rigid under rank, twist, and dimension constraints. Rank-finiteness makes classification by rank feasible, and in rank \(5\) the only possible Grothendieck equivalence classes are
\[
SU(2)_4,\qquad SU(2)_9/\mathbb Z_2,\qquad SU(5)_1,\qquad SU(3)_4/\mathbb Z_3.
\]
For integral modular data up to rank \(13\), the classification yields \(19\) non-pointed modular data and \(64\) pointed modular data, with no other integral modular data in that range. In the same range, every perfect integral modular fusion category up to rank \(13\) is trivial [1507.05139; 2302.01613].

A distinct rigidity regime is controlled by the number of twists. If a modular fusion category has fewer than four distinct twists, then for each positive integer \(N\) there are only finitely many such categories up to braided equivalence whose twists form a proper subset of the \(N\)-th roots of unity. For exactly two twists \(1\) and \(\theta\), either \(N=2\), or the category is braided equivalent to one of
\[
\mathcal C(C_2,q),\quad \mathcal C(C_3,q),\quad \mathcal C(\mathfrak{sl}_2,5,q)_{\mathrm{ad}}.
\]
For exactly three twists \(1,\theta,\eta\), either \(N=3\), or the category is braided equivalent to one of the categories listed in Figure \(3\), including examples such as
\[
\mathcal C(C_2,q)\boxtimes \mathcal C(C_2,\overline q),\quad
\mathcal C(C_4,q),\quad
\mathcal I_q,\quad
\mathcal C(C_5,q),\quad
\mathcal C(\mathfrak{sl}_2,7,q)_{\mathrm{ad}},\quad
\mathcal Z(\mathrm{Vec}_{C_2^3}^\omega).
\]

Dimension-based classification yields another sharp dichotomy. Modular categories of dimension \(p^3m\) with \(m\) square-free are pointed whenever \(p\) is odd. For \(p=2\), non-pointed examples are precisely Deligne products built from even metaplectic modular categories of dimension \(8\ell\), pointed odd cyclic modular categories, and Semion factors. The same paper identifies the non-pointed \(p=2\) examples with \(SO(2N)_2\)-type fusion rules when \(N\) is odd [1609.04896].

## 5. Distinguished families and arithmetic models

Modular extensions recover major classification results in topological phases. For \(\mathcal E=\operatorname{Rep}(G)\),
\[
\mathcal M_{\mathrm{ext}}(\operatorname{Rep}(G)) \cong H^3(G,U(1))
\]
as groups, reproducing the group-cohomology classification of \(2+1\)D bosonic SPT orders. For \(\mathcal E=\mathrm{sVec}\),
\[
\mathcal M_{\mathrm{ext}}(\mathrm{sVec}) \cong \mathbb Z_{16},
\]
recovering Kitaev’s \(16\)-fold way; the \(16\) modular extensions consist of \(8\) unitary Ising modular categories and \(8\) pointed modular categories associated to metric groups of order \(4\) [1602.05936].

The Frobenius–Schur exponent \(2\) case is completely rigid. Every modular category of Frobenius–Schur exponent \(2\) is braided monoidally equivalent to
\[
\mathcal C(\mathbb Z_2^{2m},q)
\]
for a non-degenerate quadratic form \(q\), and it decomposes as a Deligne tensor product of the two basic pointed modular categories associated to
\[
q_1(x,y)=(-1)^{xy},\qquad q_2(x,y)=(-1)^{x^2+xy+y^2}.
\]
Its positive Gauss sum \(\tau_+\) is a complete invariant:
\[
\tau_+ = (-1)^{\operatorname{Arf}(q)}2^m.
\]
This is presented as a categorical analog of Arf’s theorem on classification of non-degenerate quadratic forms over fields of characteristic \(2\) [1811.02004].

Metaplectic categories supply a large weakly integral family. Fusion categories Grothendieck equivalent to \(\operatorname{Rep}(so(2p+1)_2)\) admit explicit \(F\)- and \(R\)-symbols, and their monoidal equivalence classes are organized by arithmetic data \((r,\kappa)\) modulo the orbit relation \(r\mapsto rz^2\). Even metaplectic modular categories of dimension \(8N\) with \(N\) odd are \(\mathbb Z_2\)-gaugings of cyclic modular categories of dimension \(2N\), specifically particle-hole symmetry gaugings, and there are exactly
\[
2^{r+2}
\]
inequivalent even metaplectic modular categories of dimension \(8N\) when
\[
2N = 2p_1^{k_1}\cdots p_r^{k_r}
\]
with \(p_i\) distinct odd primes. In the integral metaplectic case, all such categories are group theoretical and therefore have Property \(\mathbf F\). For the special case with fusion rules of \(SO(8)_2\), the center is braided equivalent to \(\operatorname{Rep}(D^\omega G)\) for
\[
G=[32,49].
\]
A parallel dimension-theoretic result states that a non-pointed modular fusion category \(\mathcal C\) with
\[
\dim(\mathcal C)=p\cdot u
\]
for \(p\ge 5\) prime and \(u\) a totally positive algebraic unit is Grothendieck equivalent to \(\mathcal C(\mathfrak{sl}_2,2(p-1))_A^0\) [1608.03762; 1901.04462; 2209.04169].

## 6. Orbifolds, generalized symmetry, and extension problems

Modular fusion categories are also inputs to constructive procedures. An orbifold datum
\[
\mathbb A=(A,T,\alpha,\bar\alpha,\psi,\phi)
\]
in a modular fusion category \(\mathcal C\) defines a new modular fusion category \(\mathcal C_{\mathbb A}\), in a construction described as a generalization of taking the Drinfeld center of a fusion category. In Ising-type modular categories, Fibonacci-type orbifold data yield orbifold modular categories with exactly \(11\) simple objects, and for the parameter choice
\[
h=\exp\!\left(\pi i \frac{19}{24}\right), \qquad \epsilon=-1,
\]
one obtains
\[
(I_{h^3,\epsilon})_{A_{h,\epsilon}} \cong \mathcal C(sl(2),10)
\]
[2010.00932].

Topological approaches recast modular invariance and \(\alpha\)-induction directly inside modular fusion categories. In alterfold theory, a modular fusion category \(\mathcal C\) of non-zero global dimension \(\mu\) over an arbitrary field supports a torus \(Z\)-matrix whose entries are natural numbers and which commutes with the modular \(S\)- and \(T\)-matrices. The trace identities
\[
\operatorname{Tr}(Z)=|Irr(\mathcal M)|,\qquad \operatorname{Tr}(ZZ^t)=|Irr(\mathcal D)|
\]
relate modular invariants to Morita data. The same framework introduces double \(\alpha\)-induction and higher-genus \(Z\)-transformations invariant under mapping class group actions [2412.12702].

The relation between fusion-category symmetry and modularity is especially clear in \(1+1\)D. Boundary symmetry data form a fusion category \(A\), not braided in general, while the bulk lines of the associated Turaev–Viro theory form the Drinfeld center \(Z(A)\), a modular tensor category. The anomaly-free condition is the existence of a module category with one simple object, equivalently a fiber functor \(A\to\mathrm{Vec}_{\mathbb C}\). In parallel, permutation extensions of an arbitrary modular tensor category admit an algorithmic decategorified description: the fusion ring of the \(S_n\)-crossed extension of \(\mathcal C^{\boxtimes n}\) is determined by the base fusion ring, the permutation action, and a \(2\)-cocycle twist [2106.12577; 1909.03003].

A recent extension problem concerns Galois closure. A nondegenerate extension \(\mathcal C\hookrightarrow\mathcal M\) is Galois when its centralizer is integral. Schopieray’s conjecture asks whether every premodular fusion category admits a Galois-modular extension. For pseudounitary braided fusion categories, this has been proved: every such category admits a Galois-modular extension. In this setting, for a modular extension \(\mathcal C\hookrightarrow\mathcal M\), being Galois is equivalent to closure of the simple objects of \(\mathcal C\) under the ambient Galois action [2601.23192].

Source: https://www.emergentmind.com/topics/modular-fusion-category