Suppose I have a fusion category $\mathcal{C}$ and an indecomposable module category $\mathcal{M}$ over it. The commutant $\mathcal{C}_\mathcal{M}^*$ is the category of module endofunctors, and gives another fusion category, Morita equivalent to the original $\mathcal{C}$.

Can I bound the rank of $\mathcal{C}_\mathcal{M}^*$?

Recall that the rank is the number of isomorphism classes of irreducible objects. Certainly $\mathcal{C}$ and $\mathcal{C}_\mathcal{M}^*$ have the same global dimension, so easily $\operatorname{rank}(\mathcal{C}_\mathcal{M}^*) \leq \operatorname{dim}(\mathcal{C})$. Are there better upper bounds available?

Update: I'm happy to consider all the 'decategorified' data of $\mathcal{C}$ and $\mathcal{M}$, that is, the Grothendieck groups of both, along with the ring and module structures thereon, when trying to come up with an estimate, not just the rank of $\mathcal{C}$.

As examples:

  • the Haagerup subfactor gives a Morita equivalence between two fusion categories with ranks 4 and 6, and global dimension $\approx 35.725$
  • $\operatorname{Rep}(G)$ and $\text{Vec}_G$ are Morita equivalent, with global dimension $|G|$. Here $\operatorname{rank}(\text{Vec}_G) = |G|$, while when $G$ is non-commutative $\operatorname{rank}(\operatorname{Rep}(G))$ may be much smaller.

2 Answers 2


If $\mathcal{C}$ and $\mathcal{D}$ are Morita equivalent by a pair $({_\mathcal{C}}\mathcal{M}_{\mathcal{D}}, {_\mathcal{D}}\mathcal{N}_{\mathcal{C}})$ of bimodules, then there is a natural map of fusion bimodules ${_D}N \otimes_{C} M_{D} \to {_D}D_D$ that preserves Frobenius-Perron dimension (see section 5.1 of Noah's latest preprint with Pinhas Grossman). So the Frobenius-Perron dimensions of elements of $N \otimes_C M$ (which do not depend on knowing $D$) will give Frobenius-Perron dimensions of elements of $D$. Then I think you may be able to use the known possible small Frobenius-Perron dimensions of objects (from your paper with Noah and Frank Calegari) to determine some nontrivial lower bounds on Frobenius-Perron dimensions of simple objects in $\mathcal{D}$, and thus improve on the global dimension bound.

(Or do arbitrarily small numbers of the form $2 \cos (\pi / n)$ already generate the full ring of real cyclotomic integers? Even if so, this method could at least reduce the bound by 1 for weakly integral categories, although that's not a great improvement.)

  • $\begingroup$ Thanks Evans, this does seem promising at least for some small cases. For example for Haagerup the module has an object of dimension $\alpha = \sqrt{(5+\sqrt{13})/2}$, so $D$ has a (not-necessarily simple) object of dimension $\alpha^2 = (5+\sqrt{13})/2 \simeq 4.30278$. It's easy to see that this can only be decomposed as $1+1+(1+\sqrt{13})/2$, $2+(1+\sqrt{13})/2$ or $2 \cos(\pi/n) + x$, where $x > 76/33$. There's a finite list of such $n$, and it turns out $x$ is never maximal amongst its Galois conjugates except when $n = 2$ or $3$. $\endgroup$ Aug 24, 2012 at 2:31
  • $\begingroup$ Thus we have to split into four cases already, but in each we've already identified the dimensions of some simple objects in $D$, and hence improved the bound on the rank. $\endgroup$ Aug 24, 2012 at 2:32

In the special case of Vec(G) and Rep(G) there's a lot of results in the literature. Usually the phrasing is in terms of conjugacy classes (e.g. the strongest proved bound as far as I know is Keller's "Finite groups have even more conjugacy classes"). I learned about this from Pavel Etingof when Eric Rowell asked him about a similar question for ranks of centers.

Do I understand correctly though that you're happy to have conditions that involve more than just the rank of C? E.g. if I say wanted to have the full list of dimensions of objects in C as input to the bound would that be a problem?

  • $\begingroup$ Thanks Noah for the link. I clarified that I'm happy to use all the information in the fusion ring and its module. $\endgroup$ Aug 22, 2012 at 22:01

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.