Universes are used in a category theory to handle size-issues. Assuming Grothendieck's axiom UA that

every sets is contained in some universe

there are two approaches to $U$-smallness given a universe $U$.

1) A set is $U$-small if it is isomorphic to an element of $U$ (reference: SGA),

2) A set is $U$-small if it is an element of $U$ (references: Murfet, Low, Dwyer-Hirschhorn-Kan-Smith).

Of course, two main things that differs in these two approaches are definitions of a small $U$-category (resp., a $U$-small category) and of a $U$-category (resp., a locally $U$-small category).

In the second approach it's customary to say a $U$-category for a category whose set of objects is a subset of $U$ (some may want to drop this condition) and, more importantly, whose $\mathsf{Hom}$-sets are $U$-small with respect to 2), that is, are elements of $U$.

An important notion in category theory is a category of functors $[\mathsf{C},\mathsf{D}]$. The problem with the second approach to $U$-smallness is that $[\mathsf{C},\mathsf{D}]$ is not necessarily a $U$-category even if $\mathsf{C}$ is $U$-small and $\mathsf{D}$ is a $U$-category, unlike if we work with classes or with the first approach to $U$-smallness.

The definitions of a functor and a natural transformations that I use:

  • A functor $F\colon\mathsf{C}\to\mathsf{D}$ is a special kind of quadruple $(\mathsf{C},\mathsf{D},F_0\colon\mathsf{Ob(C)}\to\mathsf{Ob(D)}, F_1\colon\mathsf{Mor(C)}\to\mathsf{Mor(D)}),$
  • A natural transformation $\alpha\colon F\Rightarrow G$ is a special kind of triple $(F,G,(\alpha_X\colon F(X)\to G(X))_{X \in \mathsf{Ob(C)}})$.

These definitions are motivated by the assumption that every functor (resp, every natural transformation) needs to have a well-defined domain and a well-defined codomain. However, they get in the way of a category of functors between a $U$-small category and a $U$-category being a $U$-category. But what if we redefine a functor category in terms of sets?

Say, given categories $\mathsf{C}$ and $\mathsf{D}$, the functor category $[\mathsf{C},\mathsf{D}]$ is not a category of functors and of natural transformations per se, but a category

  • whose objects are pairs $(F_0,F_1)$ of surjective functions which define a functor $F\colon\mathsf{C}\to\mathsf{D}$,

  • whose morphisms between two such pairs $(F_0,F_1)$ and $(G_0,G_1)$ are triples $((F_0,F_1),(G_0,G_1),A)$ where $A = \{ (X,\alpha_X) \mid X \in \mathsf{Ob(C)} \}$ which defines a natural transformations between respective functors which are defined by $(F_0,F_1)$ and $(G_0,G_1)$

  • the composition law and the identities are evident.

Of course, it is only a technicality, and given two categories $\mathsf{C}$ and $\mathsf{D}$ we can identify objects and morphisms of our newly defined $[\mathsf{C},\mathsf{D}]$ with "real" functors and natural transformations.

Zhen Lin Low seems to do something similar is his aforementioned notes, that is, he says that such a category is isomorphic to a category of functors and Daniel Murfet seems to also do it implicitly (or maybe he uses a different definition of functors and natural transformations).

My question is this: is redefining a functor category such as this safe? Wouldn't it lead to any problems with other constructions?

  • 1
    $\begingroup$ What is your definition of function? $\endgroup$ – Fred Rohrer Nov 26 '18 at 15:51
  • 1
    $\begingroup$ Why do you want the maps $F_0$ and $F_1$ to be surjective? $\endgroup$ – Fred Rohrer Nov 27 '18 at 8:03
  • 1
    $\begingroup$ The codomain of $F_0$ or $F_1$ is the set of objects or morphisms of ${\sf D}$, which need be neither an element of $U$ nor $U$-small. But anyway, not every functor (in the usual sense) can be described by two surjective maps to the set of objects or morphisms of its target category. $\endgroup$ – Fred Rohrer Nov 27 '18 at 9:38
  • 1
    $\begingroup$ @FredRohrer No, I don't request a set of objects to be an element of $U$ in a $U$-category, merely a subset of $U$. Hence $U$-$\mathsf{Set}$, $U$-$\mathsf{Grp}$ etc.are $U$-categories. For the proof why my $[\mathsf{C},\mathsf{D}]$ is a $U$-category you can see linked notes of Murfet or Low. As for considering surjective versions of $F_0\colon\mathsf{Ob(C)}\to\mathsf{Ob(D)}$ and $F_1\colon\mathsf{Mor(C)}\to\mathsf{Mor(D)}$ instead these functions themselves, you may instead consider relations $F_0$ and $F_1$ as sets of ordered pairs, it's essentially the same as $\endgroup$ – Jxt921 Nov 27 '18 at 10:26
  • 1
    $\begingroup$ functional relations and surjective functions are in bijection. We only need to define functions as pairs $(f,Y)$ where $f$ is functional relations with $\mathrm{ran}(f) \subseteq Y$ because we want to consider nonsurjective functions. $\endgroup$ – Jxt921 Nov 27 '18 at 10:27

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.