Consider an extension of the 1-category of vector spaces and linear maps down to a 2-category $\mathcal{C}$ whose objects are $k$-linear categories. What is the symmetric monoidal structure on the objects, and how does one interpret a morphism of unit $k$-linear categories as a vector space?

In other language, consider a d-dimensional once-extended TFT $Z:\text{Bord}_d^2\rightarrow\mathcal{C}$ that assigns $k$-linear categories to $(d-2)$-manifolds and etcetera. A bordism of of closed $(d-1)$-manifolds maps under $Z$ to a linear map between vector spaces. In particular, a bordism of empty sets is a closed $d$-manifold and maps to an endomorphism of the unit vector space $k$, i.e. an element of $k^\times$, as expected. I wish to understand the lower dimensional analog: a bordism $B$ of closed $(d-2)$-manifolds $M$ and $N$ maps to a functor $Z(B):Z(M)\rightarrow Z(N)$ of $k$-linear categories. If $M$ and $N$ are empty, $B$ is a closed $(d-1)$-manifold. How can I see that $Z(B)$ is a vector space?


Najib Idrissi's answer has a subtle error which I'd like to address. I'd also like to give some context.

It's easiest to restrict our attention to the nicest $k$-linear categories, namely those which arise as the categories $\text{Mod}(A)$ of (right) $A$-modules for $A$ a $k$-algebra. In this context one attractive candidate for the tensor product of $\text{Mod}(A)$ and $\text{Mod}(B)$ is $\text{Mod}(A \otimes_k B)$. In fact there is a lovely symmetric monoidal $2$-category, the Morita $2$-category over $k$, which can be described in the following equivalent ways: it has

  • objects the $k$-linear categories $\text{Mod}(A)$, equipped with the above tensor product,
  • morphisms the cocontinuous $k$-linear functors $\text{Mod}(A) \to \text{Mod}(B)$,
  • $2$-morphisms the ($k$-linear? This might be automatic) natural transformations of such functors.

Equivalently, it has

  • objects the $k$-algebras $A$, equipped with the usual tensor product,
  • morphisms the $k$-bimodules, or equivalently the right modules over $A^{op} \otimes_k B$ (which compose under the tensor product of bimodules),
  • $2$-morphisms the morphisms of $k$-bimodules.

These $2$-categories are equivalent by the Eilenberg-Watts theorem, which identifies the category of cocontinuous $k$-linear functors $\text{Mod}(A) \to \text{Mod}(B)$ with the category of $k$-bimodules over $A$ and $B$. Given a bimodule $M$ the corresponding cocontinuous functor is $(-) \otimes_A M$, and conversely given a cocontinuous $k$-linear functor $F$ the corresponding bimodule is $F(A)$. Now:

The unit object is $\text{Mod}(k)$ in the first description of this $2$-category and $k$ in the second description, and by Eilenberg-Watts we have an identification between the category of cocontinuous $k$-linear functors $\text{Mod}(k) \to \text{Mod}(k)$ and $\text{Mod}(k)$ itself.

You should think of this result as a categorification of the Yoneda lemma result that endomorphisms of a ring $R$ as a right $R$-module is the same as $R$ itself, thinking of $\text{Mod}(k)$ as a categorified ring with multiplication the tensor product and thinking of the Morita $2$-category as a $2$-category of (nice) module categories over $\text{Mod}(k)$. (The module structure is given not by the enrichment but by the tensoring of $\text{Mod}(A)$ over $\text{Mod}(k)$.)

That $k$-linearity condition is essential (and this is the error in Najib Idrissi's answer): without it, the category of cocontinuous (this implies $\mathbb{Z}$-linear but not necessarily $k$-linear) functors $\text{Mod}(k) \to \text{Mod}(k)$ is more complicated, and is instead identified with the category of $(k, k)$-bimodules, or equivalently modules over $k \otimes_{\mathbb{Z}} k$. In general this category contains many interesting bimodules which are not equivalent to direct sums of copies of $k$ as a bimodule. For example, any automorphism of $k$ lets you twist the usual bimodule structure on $k$, and leads to an interesting automorphism of $\text{Mod}(k)$ that you can think of as acting by that automorphism on matrices.

In any case, one of the many attractive things about this symmetric monoidal $2$-category is that every object $A$ is automatically dualizable, with dual $A^{op}$, hence automatically defines a $1$-dimensional field theory. The value of this field theory on a circle turns out to be a version of Hochschild homology, although to get Hochschild homology in the ordinary sense we need to work in a derived version of the Morita $2$-category. It's a nice exercise to work out which $A$ are $2$-dualizable, so can define a $2$-dimensional field theory.

| cite | improve this answer | |
  • $\begingroup$ Thanks for the correction. But unless I'm mistaken, I did include the linearity condition in my answer, when I wrote "additive"? I was under the impression that this was enough, please correct me if I'm wrong. In any case, don't we generally want these functors to be $k$-linear in the context of such a TQFT? $\endgroup$ – Najib Idrissi Nov 4 '14 at 8:32
  • $\begingroup$ @Najib: additive only means $\mathbb{Z}$-linear; at least that's its standard meaning. $\endgroup$ – Qiaochu Yuan Nov 4 '14 at 17:21
  • $\begingroup$ You're right.${}$ $\endgroup$ – Najib Idrissi Nov 4 '14 at 17:41

When $M=N=\emptyset$, then $Z(\emptyset) = \mathsf{Vec}_f$ is the category of finite dimensional vector spaces, aka the unit for the tensor product. In the end the question becomes: how does one identify functors $\mathsf{Vec}_f \to \mathsf{Vec}_f$ with finite dimensional vector spaces? (We need finite dimensionality here I think, and vector spaces are finite dimensional in this context anyway.)

Define a functor $\operatorname{ev} : \mathsf{Fun}(\mathsf{Vec}_f, \mathsf{Vec}_f) \to \mathsf{Vec}_f$ by $\operatorname{ev}(F) = F(k)$, where $k$ is the ground field. Then $\operatorname{ev}$ is an equivalence of categories. Indeed, by this MO answer of Todd Trimble, additive covariant functors preserve direct sums (this is the kind of functors we're studying here). This means that every $F \in \mathsf{Fun}(\mathsf{Vec}_f, \mathsf{Vec}_f)$ is uniquely determined by its value on $k$, because every $V \in \mathsf{Vec}_f$ is a (finite) direct sum of copies of $k$.

As for the actual symmetric monoidal structure, I think the best you can do is read Remark 1.2.7 in Lurie's article On the classification of topological field theories. Roughly speaking, if $\mathsf{C,D}$ are cocomplete $k$-linear categories, then $\mathsf{C} \otimes \mathsf{D}$ is the category representing "bilinear" functors $\mathsf{C} \times \mathsf{D} \to \mathsf{E}$.

| cite | improve this answer | |
  • $\begingroup$ This isn't quite right. You've uniquely determined what $F$ does on objects but not on morphisms. I'm about to post an answer explaining the correct statement. $\endgroup$ – Qiaochu Yuan Nov 4 '14 at 4:52

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.