Fusion Categories, Module Categories, and Bimodule Defects

A fusion category provides finite, semisimple, rigid fusion data for topological lines. A left module category describes boundary conditions on which those lines may end, and a bimodule category describes a wall between two fusion theories. The relative tensor product composes walls. These statements require finiteness, semisimplicity, rigidity, and coherent module associators; the same Grothendieck matrices without that structure do not determine a defect theory.

Required background. Defects on Stratified Spacetimes supplies the higher-morphism interpretation. DHR Sectors and Modular Tensor Categories of Nets supplies a rigorous QFT source of fusion categories. Categories, Functors, Natural Transformations, and Universal Properties supplies categorical composition and universal properties. Helpful background. Non-Invertible Topological Defects and Fusion gives the physical meaning of the fusion data.

Fusion and module data

Over $\mathbb C$, a fusion category $\mathcal C$ is a $\mathbb C$-linear semisimple rigid monoidal category with finite-dimensional hom-spaces, finitely many simple isomorphism classes, and simple tensor unit $\mathbf1$. If $a,b,c$ are simple, then

$$a\otimes b\cong\bigoplus_c N_{ab}^{\ c}\,c, \qquad N_{ab}^{\ c}\in\mathbb Z_{\ge0}.$$

The integers are necessary but not sufficient. One must also give associators satisfying the pentagon, unit constraints, and left and right duality maps satisfying the snake identities. Ostrik states the monoidal, rigid, and module-category structures in Ostrik 2003, §§2.2–2.3, printed pp. 3–7 (PDF).

A left $\mathcal C$-module category $\mathcal M$ has an action $\triangleright:\mathcal C\times\mathcal M\to\mathcal M$ and coherent isomorphisms

$$(a\otimes b)\triangleright m\cong a\triangleright(b\triangleright m), \qquad \mathbf1\triangleright m\cong m.$$

Its objects may label boundary conditions, while a morphism in $\mathcal M$ labels a boundary-changing point operator. A $(\mathcal C,\mathcal D)$-bimodule category carries commuting left $\mathcal C$- and right $\mathcal D$-actions and represents an interface. Given compatible walls $\mathcal M$ and $\mathcal N$, their composite is the balanced relative tensor product

$$\mathcal M\boxtimes_{\mathcal D}\mathcal N,$$

characterized by functors that identify $(m\triangleleft d,n)$ with $(m,d\triangleright n)$. Existence and good finiteness properties rely on the finite semisimple setting. Etingof, Nikshych, and Ostrik construct this tensor product and the Brauer–Picard higher groupoid in Etingof, Nikshych, and Ostrik 2010, §§3–4, printed pp. 14–26 (PDF).

Morita equivalence means the existence of an invertible bimodule category, not equality of tensor categories. It preserves the bulk Drinfeld center but can forget a chosen boundary realization, pivotal structure, or symmetry enrichment.

A subgroup module and its fusion matrices

Let $G=S_3$, $H=\langle(12)\rangle\cong\mathbb Z_2$, and $\mathcal C=\operatorname{Rep}(S_3)$. The category $\mathcal M=\operatorname{Rep}(H)$ is a $\mathcal C$-module category through restriction:

$$V\triangleright M=\operatorname{Res}^{S_3}_{H}(V)\otimes M.$$

Write $m_+$ and $m_-$ for the trivial and sign representations of $H$. The three simple $S_3$ representations are $\mathbf1$, $\varepsilon$, and the two-dimensional standard representation $v$. Their restrictions give

$$N_{\mathbf1}=\begin{pmatrix}1&0\\0&1\end{pmatrix}, \quad N_{\varepsilon}=\begin{pmatrix}0&1\\1&0\end{pmatrix}, \quad N_v=\begin{pmatrix}1&1\\1&1\end{pmatrix}.$$

These action matrices reproduce the fusion ring:

$$N_v^2=N_{\mathbf1}+N_{\varepsilon}+N_v,$$

because $v\otimes v\cong\mathbf1\oplus\varepsilon\oplus v$. The two module simples are boundary sectors, and the entries of $N_a$ count the possible boundary transitions generated by bringing line $a$ to the boundary. This is the exact first application returned to Non-Abelian Topological Orders: the categorical calculation supplies boundary-defect fusion data, while the physical phase, anyon observables, and measurements remain there.

The matrix identity is an independent decategorified check, but it does not verify the module associator. Distinct module categories can have the same based module, and a based module need not admit a categorification at all. Ostrik emphasizes this distinction between fusion-rule modules and module categories in Ostrik 2003, §2.1 and Definition 6, printed pp. 2–6 (PDF).

Where the fusion formulas stop

Drop rigidity. Then an object may have no evaluation and coevaluation maps, so an orientation-reversed line need not exist. Drop semisimplicity. Then an object need not split into simples, the fusion matrices no longer capture extensions, and formulas based on positive Frobenius–Perron dimensions can lose their intended meaning. Drop finiteness. Then the relative tensor product may require completions and analytic control absent from the finite theorem.

The adversarial case is therefore a monoidal category with the advertised multiplication table but no compatible duals. Its integer coefficients still define a based ring. What must be withdrawn are the statements that every defect has an orientation reverse, that categorical dimensions obey the fusion eigenvector formulas, and that bimodule composition remains inside a finite semisimple category. The strongest surviving conclusion is algebraic fusion data, not a fusion category or a QFT realization.

There is also a useful two-sided consistency check. Rigidity implies the Frobenius reciprocity relation

$$\dim\operatorname{Hom}(a\otimes m,n) =\dim\operatorname{Hom}(m,a^\vee\otimes n),$$

so the action matrix of $a^\vee$ is the transpose of that of $a$ in an orthonormal simple basis. In the subgroup example all three $S_3$ simples are self-dual, and the displayed matrices are symmetric. If a proposed module action violates this identity, either its dual labels, its action multiplicities, or its choice of simple basis is wrong. Passing the check still does not determine the associator, but failing it decisively rules out the claimed rigid module category.

Exercises

Verify the $S_3$ module relation for $v\otimes\varepsilon\cong v$.

Solution

Matrix multiplication gives $N_vN_\varepsilon=\left(\begin{smallmatrix}1&1\\1&1\end{smallmatrix}\right)=N_v$. This agrees with restricting $v\otimes\varepsilon\cong v$ to $H$. The check concerns the based module; the coherent module associator is additional data.

References

• Etingof, Pavel, Dmitri Nikshych, and Victor Ostrik. “Fusion Categories and Homotopy Theory.” Quantum Topology 1 (2010): 209–273. DOI; Open PDF.
• Ostrik, Victor. “Module Categories, Weak Hopf Algebras and Modular Invariants.” Transformation Groups 8 (2003): 177–206. DOI; Open PDF.