The Day convolution monoidal category structure $(\mathsf{PSh}(\mathcal{C}),\circledast,\mathsf{h}_{\mathbf{1}_{\mathcal{C}}})$ on the category of presheaves of a monoidal category $(\mathcal{C},\otimes,\mathbf{1}_{\mathcal{C}})$ satisfies the following universal properties:

  • In A universal property of the convolution monoidal structure, Im–Kelly prove that it is the free monoidal cocompletion of $\mathcal{C}$. Here, a monoidal category is monoidally cocomplete if it is cocomplete and the functors $A\otimes-$ and $-\otimes B$ preserve colimits for all $A,B\in\mathrm{Obj}(\mathcal{C})$. Im–Kelly then prove:

The monoidal category $(\mathsf{PSh}(\mathcal{C}),\circledast,\mathsf{h}_{\mathbf{1}_{\mathcal{C}}})$ is the universal monoidally cocomplete category on $\mathcal{C}$ in that, given any monoidally cocomplete monoidal category $(\mathcal{D},\otimes,\mathbf{1}_{\mathcal{D}})$, precomposition with $よ\colon\mathcal{C}\hookrightarrow\mathsf{PSh}(\mathcal{C})$ defines an equivalence of categories $$よ^*\colon\mathsf{Fun}^{\otimes,\mathsf{strong}}_{\mathsf{cocont.}}(\mathsf{PSh}(\mathcal{C}),\mathcal{D})\longrightarrow\mathsf{Fun}^{\otimes,\mathsf{strong}}(\mathcal{C},\mathcal{D}).$$ That is, $(\mathsf{PSh}(\mathcal{C}),\circledast,\mathsf{h}_{\mathbf{1}_{\mathcal{C}}})$ is uniquely determined by the following requirements:

  1. The Yoneda embedding $よ\colon\mathcal{C}\hookrightarrow\mathsf{PSh}(\mathcal{C})$ is strong monoidal.
  2. $\circledast$ is cocontinuous in each variable.

The analogue of the first of these for bimonoidal categories, however, doesn't work.

Question. Given a bimonoidal category $(\mathcal{C},\otimes,\oplus,\mathbf{1}_{\mathcal{C}},\mathbf{0}_{\mathcal{C}})$, is there a universal way to put a bimonoidal category structure on $\mathsf{PSh}(\mathcal{C})$?

In particular, is there a bimonoidal category structure on $\mathsf{PSh}(\mathcal{C})$ such that, given another bimonoidal category $(\mathcal{D},\otimes_{\mathcal{D}},\oplus_{\mathcal{D}},\mathbf{1}_{\mathcal{D}},\mathbf{0}_{\mathcal{D}})$, we have an equivalence of categories $$ \left\{ \begin{gathered} \text{symmetric strong }\color{red}{\text{bi}}\text{monoidal}\\ \text{functors $\mathcal{C}\times\mathcal{D}\to\mathsf{Sets}$} \end{gathered} \right\} \cong \left\{ \begin{gathered} \text{symmetric strong }\color{red}{\text{bi}}\text{monoidal}\\ \text{functors $\mathcal{D}\to\mathsf{PSh}(\mathcal{C})$} \end{gathered} \right\}, $$ natural in $\mathcal{D}$?


$\def\C{\mathcal{C}}\def\Set{\mathsf{Set}}$ (What follows comes from a private chat with Todd Trimble)

Notation. If $(\C,\otimes,\oplus)$ is a bimonoidal category, I will call $\otimes$ the multiplicative structure and $\oplus$ the additive structure; if $\oplus$ is the cocartesian monoidal structure, I will call $\C$ a 2-rig, following https://arxiv.org/abs/2103.00938.

I am convinced that in general (=for a general bimonoidal category) little can be said, because one needs some compatibility between the multiplicative structure and co/products. Even when $\C$ is a 2-rig the most I can formulate until now is a

Conjecture. When $\C$ is a 2-rig, the category $[\C^o,\Set]_\times$ of functors $\C^o \to \Set$ that are product-preserving (=sending coproducts in $\C$ to products in $\Set$) is the free 2-rig on $\C$.

In order for this freeness property to be legitimate, the least we can ask is that

  1. $F\C=[\C^o,\Set]_\times$ is a 2-rig if the multiplicative structure is Day convolution;
  2. The Yoneda embedding $y : \C \to F\C$ is a morphism of 2-rigs.

Unfortunately, I am still unable to prove that the Day convolution restricts to "models" of the "theory" $\C$ (it is a fruitful intuition to think of $\C$ like it was a Lawvere theory even if it's not, were it only because it's easier to query google with questions ;-) )

Update: I couldn't because it's not true, but it falls very short from being true, in the sense that the conjecture is "true up to reflecting the monoidal structure": observe that $[\C^o,\Set]_\times$ has many desirable properties for the free cocomplete 2-rig on $\C$:

  1. $[\C^o,\Set]_\times$ is a cocomplete[¹], reflective subcategory of the entire $[\C^o,\Set]$.
  2. the yoneda embedding $y : \C \to [\C^o,\Set]$ clearly factors through $[\C^o,\Set]_\times$.
  3. $F\C=[\C^o,\Set]_\times$ has the following universal property, if $\mathcal D$ is cocomplete: $$\{\text{cocontinuous } g : F\C \to\mathcal D\}\cong \{\text{coproduct preserving } h : \C \to \mathcal D\}$$
  4. Another useful universal property for $[\C^o,\Set]_\times$ is that it is the cocompletion of $\C$ under sifted colimit.

All these facts turn out to be useful to establish that $[\C^o,\Set]_\times$ is the free 2-rig over $\C$: the coend that expresses the usual Day convolution formula is to be interpreted in a way that we first take the coend in the usual category of presheaves, but then to that apply the reflection functor $r$ that is left adjoint to the full inclusion. So, it's not true that the ordinary Day convolution takes a pair of product-preserving functors to a product-preserving functor; you have to sheafify. After you do that, everything falls into place.

Look how neat everything becomes!

Suppose we have a functor $F$ in $[\C^o,\Set]_\times$; by 4 above, it's a sifted colimit of representables $\C(-, c)$. Since we assume $\C$ is a 2-rig, we have a composite of (finite) coproduct-preserving functors

$$Y : \C \stackrel{C \otimes -}{\to} \C \stackrel{y}{\to} [\C^o,\Set]_\times$$

under the equivalence of point 3, this becomes a colimit preserving functor $[\C^o,\Set]_\times \to [\C^o,\Set]_\times$, exactly the reflected convolution $\C(-,C)\ast \_$.

But since the convolution we want to build must be cocontinuous, it is uniquely determined by this construction!

The missing detail is some result ensuring that $\ast$ is a monoidal structure. All in all I expect this to be a consequence of a theorem about transport of monoidal structures $\otimes$ (Day convolution of presheaves) into $\ast$ ("reflected" Day convolution) given some lax monoidality assumptions on $r$, so I won't enter the details.

Let me just add a final neat detail: the coend formula expressing the Day convolution product is a reflexive coequalizer; a reflexive coequalizer is a sifted colimit, so it can be interpreted as the usual, pointwise colimit in Set. Moreover, the sums = coproducts involved in the coend formula are filtered colimits of finite coproducts; again, filtered colimits are sifted colimits. So the only "re-interpretation" involved with the reflector resides in the way you compute finite coproducts in models, i.e.: you don't compute finite coproducts set-wise, but at the level of models, and every other colimit as usual.

[¹] But colimits in $[\C^o,\Set]_\times$ are not computed as colimits in the presheaf category (thikn again to the case of Lawvere theories and coproducts of monoids...), and this will be crucial, because for example, in $[\C^o,\Set]_\times$ the object $\C(-,A+B)$ has the universal property of the coproduct $\C(-,A)+\C(-,B)$.


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.