Is there a classification of categories in which the inverse of a bijective morphism is automatically a morphism?

89 Views Asked by At

This question arises when looking at ways to describe isomorphism in certain categories.

In groups, rings, vector spaces: two objects $A,B$ are isomorphic if there exists a bijective morphism from $A$ to $B$.

In graphs, topological spaces, measure spaces: two objects $A,B$ are isomorphic if there exists a bijective morphism from $A$ to $B$, whose inverse is also a morphism.

Now, the second definition is a bit more general, it is easy to verify for the examples I have mentioned that the second definition reduces to the first because of the homomorphism structure.

My question is, is there a way to classify all such categories in which this reduction is possible? Or are homomorphisms simply that special when it comes to imbuing structure?

Edit: Let us restrict the discussion to categories where the objects are built upon sets and the morphisms are special set maps on these underlying sets.

1

There are 1 best solutions below

1
On BEST ANSWER

As was already pointed out, the question doesn't make sense for arbitrary categories, because in general the notion of bijection doesn't make sense. A non-example would be the free walking arrow category $\bullet \rightarrow \bullet$ consisting of two objects and one nontrivial morphism between them. It has nothing to do with sets and functions.

What we can do is to restrict the question to categories $\mathcal{C}$, which are equipped with a functor $\mathcal{U:C}\longrightarrow\mathsf{Set}$ (e.g. many categories of sets with structure have such a forgetful functor). We can then say that a morphism $f:X\rightarrow Y$ in $\mathcal{C}$ is $\mathcal{U}$-bijective, if $\mathcal{U}(f)$ is a bijective function. Since in the category $\mathsf{Set}$ of sets being bijective and being an isomorphism is the same, $f$ is $\mathcal{U}$-bijective if and only if $\mathcal{U}(f)$ is an isomorphism. The condition that in $\mathcal{C}$ the notions of $\mathcal{U}$-bijective and isomorphism coincide is then precisely the requirement that $\mathcal{U}$ reflects isomorphisms i.e. that $\mathcal{U}$ is a conservative functor.

So far we have just rephrased the question, but I guess you would also like to know some conditions for when a functor $\mathcal{F:C}\longrightarrow \mathcal{D}$ is actually conservative. Two instances are:

  1. the functor $\mathcal{F}$ is full and faithful (this is a nice short exercise to get comfortable with definitions)
  2. the functor $\mathcal{U}$ is monadic (a really interesting condition, but a bit further down the category rabbit hole. For all purely algebraic notions (monoids, groups, abelian groups, modules, rings etc.) the forgetful functor to sets satisfies this condition*)

I am unaware of a complete characterization of conservative functors and don't really think it is feasible, based on how different the sufficient conditions 1. and 2. are.

Fun fact: It is well known that the forgetful functor $\mathsf{Top}\longrightarrow\mathsf{Set}$ is not conservative (in other words: that a continuous bijection need not be a homeomorphism). However you can prove a variation of the compact-Hausdorff-trick by purely categorical means: The forgetful functor from the category $\mathsf{CHaus}$ of compact Hausdorff spaces and continuous maps to the category of sets is monadic. By 2. it is conservative, in other words a continuous bijection between compact Hausdorff spaces is a homeomorphism!

Remark 1: Why call it $\mathcal{U}$-bijective instead of just bijective? Well, it really depends on the functor. For example take the category $\mathsf{Set}\times\mathsf{Set}$ of pairs of sets. $\mathcal{U}$ could forget the second set, forget the first set or take a pair $(X,Y)$ to, say, their product set $X\times Y$. Now whether $\mathcal{U}$-bijections are isomorphism or not obviously depends on what $\mathcal{U}$ you choose...

Remark 2: One could also have a completely different take on this question and ask, what conditions you could put on your category $\mathcal{C}$ to even be able to formulate a definition of bijective morphism (e.g. as injective + surjective morphism, whatever this means now) inside that category. Then one could ask, under which conditions this notion coincides with isomorphism. Topos theory might be a good starting point to look for an answer to that question.