8
$\begingroup$

If $C$ is a monoidal category, we can define the category $Mon(C)$ of monoids in $C$; call $U_C : Mon(C) \to C$ the forgetful functor. I'm interested in functors between categories of monoids:

Suppose $(C, I, \otimes)$, $(D, J, \boxtimes)$ are (strict) monoidal categories, and call $Mon(C)$, $Mon(D)$ the corresponding monoid categories. It is well known that any lax monoidal functor $F : C \to D$ preserves monoids; in other words we can define another functor $\tilde{F} : Mon(C) \to Mon(D)$ such that $U_D \circ \tilde{F} = F \circ U_C$.

My question: Suppose we have a functor $\tilde{G} : Mon(C) \to Mon(D)$: is it possible to produce a functor $G : C \to D$ such that the above equality holds? If not, are there some results in the literature providing conditions under which it exists (or might exist)?

My failed attempts: At first I thought that such an "extension" would always exist (and be a lax monoidal functor), but when I tried to compute it in some special cases, my strategy failed.

All the cases where I failed use the existence of a coproduct in some other category $B$; let me elaborate on one of such cases:

Example of failure: Suppose $B$ has finite coproducts $+$, as well as a terminal object $\ast$. Consider the following:

$$\Phi : [B, B] \to [B, B]$$ $$F(-) \mapsto F(- + \ast)$$

In other words, $\Phi(F)(a) = F(a + \ast)$; it acts in the obvious way on morphisms (both in $B$ and in $[B, B]$). It can be shown that if $M : [B, B]$ is a monoid (that is, a monad on $B$) then so is $\Phi(M)$:

The unit (of the monoid structure on $\Phi(M)$) is given by the composition: $$ (\epsilon_{\Phi(M)})_a =: a \xrightarrow{\iota_1} a + \ast \xrightarrow{\epsilon_{M(a + \ast)}} M(a + \ast) $$

The multiplication is quite more involved to define: $$(\mu_{\Phi(M)})_a =: M(M(a + \ast) + \ast) \xrightarrow{M(\psi)} M(M(a + \ast)) \xrightarrow{\mu_{M(a + \ast)}} M(a + \ast)$$ It is defined as long as we can define $\psi : M(a + \ast) + \ast \to M(a + \ast)$.

But such a $\psi$ can be defined: consider the universal property of $+$. By that universal property, given any couple of morphisms $f : a \to c$ and $g : b \to c$, we can produce a morphism $<f, g> : a + b \to c$.

Now, consider the case where $b = \ast$ and $c = a + \ast$: by the monoid structure on $M$ we have a map $$\phi : \ast \xrightarrow{\iota_2} a + \ast \xrightarrow{\epsilon_{a + \ast}} M(a + \ast)$$ apply the construction given before to $1_{M(a + \ast)} : M(a + \ast) \to M(a + \ast)$ and $\phi : \ast \to M(a + \ast)$ to produce a morphism $$\psi := <1_{M(a + \ast)}, \phi> : M(a + \ast) + \ast \to M(a + \ast)$$ This concludes the construction of the monoid structure on $\Phi(M)$ (since the naturality of $\psi$ is quite straightforward to prove).

As some of the readers will have noticed, the morphism $M(\psi)$ does exactly what the "laxness of $\Phi$" would have done, did it exist: $M(\psi) : \Phi(M) \circ \Phi(M) \to \Phi(M \circ M)$. However, we had to assume $M$ to be a monoid in the first place: were it not, $\psi$ couldn't be constructed.

This does not mean that $\Phi$ is not a lax monoidal functor: it could well have different laxness structural morphisms, that "reduce" to the above construction when the argument $M$ is in fact a monoid in $[B, B]$ (monad on $B$).

The other two cases where this happened are:

  • $\Phi(F)(a) = F(a^*)$, where $a^*$ is the "Kleene Star" (or, in category-theory speech, the free monoid on $a$); this requires the further assumption that finite product exist, countable coproduct exist (in addition to finite ones), and finite products distribute over countable coproducts, in $B$

  • $\Phi(F)(a) = F(b + a)$ for some object $b : B$; this does not even require the existence of a terminal object.

The above construction carries over to these other two cases, with slight modifications.

Motivation: I am particularly interested in the case where the two categories (and their monoidal structures) are the same, being an endofunctor category with the obvious monoidal structure (as in the "failed attempt")

$$(C, I, \otimes) = (D, J, \boxtimes) = ([B, B], 1_B, \circ)$$

where $Mon([B, B]) = Monads(B)$ the category of monoids in $[B, B]$ is the category of monads on $B$: as many will have guessed, this question arised from haskell programming and monad transformers.

Rephrasing the question: Given a functor between categories of monoids (internal to some monoidal categories), when can it be extended as a functor between the underlying monoidal categories? What properties should I expect such an extension to have (if such an extension does indeed exist)?

Lastly, is the "failed attempt" really a counter-example? If so, is there a counter-example that does not rely on functor categories? In the last few days, I searched for a counter-example in the category of non-commutative rings (as many will know, non-commutative rings are monoid objects in the category of abelian groups) without success.

$\endgroup$
8
$\begingroup$

Here is a simple counterexample: take $C=D=\mathrm{Set}$, and let $\tilde{G}$ be the abelianization of a monoid. Then $\tilde{G}$ cannot arise from any $G:\mathrm{Set}\to\mathrm{Set}$, because the cardinality of the underlying set of $\tilde{G}(M)$ depends on the monoid structure of $M$, not just on its cardinality. For instance, take an abelian monoid and a nonabelian monoid of the same cardinality, say the symmetric group $S_3$ and the cyclic group $C_6$: then $U(S_3) \cong U(C_6)$, but $U \tilde{G}(S_3) \not\cong U\tilde{G}(C_6)$ since the latter still has 6 elements but the former has fewer; thus $U \tilde{G} \not\cong G U$ for any functor $G$.

$\endgroup$
  • $\begingroup$ Very interesting: does this mean, by the Eckmann-Hilton argument, that the free monoid functor on monoid categories (which should be the abelianization functor, if I'm not mistaken) does not generally extend to the underlying category? $\endgroup$ – LorenzoPerticone Jun 3 '18 at 11:21
  • $\begingroup$ @LorenzoPerticone I guess so. $\endgroup$ – Mike Shulman Jun 3 '18 at 18:51
0
$\begingroup$

@Mike Shulman's answer is already quite interesting, and answers one of my main perplexities. Here I want to add a little note on why my "proposed counterexample" really is a counterexample (again, this carries over to the other two cases with little modifications):

Set $\Phi(F) (a) := F(\ast + a)$ as in the question.

We want to show that there are $C$ a category, $F, G : C \to C$ functors such that there is no natural transformation $\Phi(F) \circ \Phi(G) \to \Phi(F \circ G)$.

Take $C = Set$, so that $\ast$ is any singleton, $F = 1_C$, and $G=\Delta_{\emptyset}$ to be the functor constant on $\emptyset$ the empty set; obviously $$\Phi(F) \circ \Phi(G) = \Delta_\ast$$ $$\Phi(F \circ G) = \Delta_\emptyset$$ and we all know that there is no function from any set to the empty set.

$\endgroup$

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.