# Why do we care about locally small categories (as opposed to large categories)?

I'm going through an algebra course that starts with category theory, and the definition of a category given is actually that of a locally small category, i.e. we assume that $$\text{Mor}_{\mathcal{C}}(X,Y)$$ is a set.

The issue arises when we start considering a category that has as its objects functors between two categories, and as its morphisms the natural transformations between said functors. The author said that this is "morally" not a category (hinting at 2-categories). I don't really understand what is meant by "morally" not a category. Is the only problem that the morphisms form a class? If so, why are we more interested in locally small categories specifically? It seems like allowing the objects to be a class is fine, but having too many morphisms could be a problem. Is it because morphisms are "more important" than the objects themselves (following Grothendieck)?

More generally, what are striking differences between large and locally small categories?

• I don't know much set theory but I think there is a problem when you have a class ($Ob(\mathcal C) \times Ob(\mathcal C)$) indexing classes (homs), as opposed to a class indexing sets in the locally small case. That is, collecting a class of classes into a single object called a category seems much dicier to me than collecting a class of sets together. Additionally, this article by Mike Shulman entitled "Set Theory for Category Theory" may be of use to you, but it focuses a lot of foundations. Jun 20 '21 at 19:59

This is a duplicate question, but I cannot find the duplicate right now.

It has been already explained why locally small categories are important (tons of examples, and the Yoneda Lemma). Examples of categories which are not locally small have been given at SE/219539 and MO/3278. Here is an attempt to answer the set-theoretic concerns.

First of all, Grothendieck universes are very useful to manage size-isszes in category theory. For example, locally small categories can be considered to be small in this way. In detail:

Let $$U$$ be a Grothendieck universe. A $$U$$-small (or just: small) category is a category such that both morphisms and objects form sets in $$U$$, also called $$U$$-small sets.

Many categories in practice are not small, but they are locally small. But in order to formalize this, it is best to assume another Grothendieck universe $$U \in V$$. Then we define* a locally small category to be a $$V$$-small category such that all hom-sets are $$U$$-small. Notice that such a category is still small, just with respect to a larger Grothendieck universe. This is very useful, since you can apply all of the known constructions of small categories also to all categories, when you keep track that you are working with $$V$$-small categories. In particular, we can consider functor categories for $$V$$-small categories. So from this perspective, there are no "large categories".

If $$\mathbf{Set}$$ denotes the category of $$U$$-small sets, for every locally small category $$\mathcal{C}$$ we have the Yoneda embedding $$\mathcal{C} \to \mathrm{Hom}(\mathcal{C}^{\mathrm{op}},\mathbf{Set})$$, $$A \mapsto \mathrm{Hom}(-,A)$$. Notice that both sides here are $$V$$-small. The Yoneda embedding is very important, since it reduces many claims from an arbitrary locally small category to the category of sets.

*Unfortunately, this is not the only definition in the literature. See MO/3409.

• Thank you very much for your answer. I did not see a duplicate so I went ahead and asked. Are large categories studied extensively like locally small categories are? Jun 21 '21 at 15:54
• For a $V$-small category you don't necessarily have a Yoneda embedding in the same universe. Jun 21 '21 at 18:48

I'm not sure about the "morally" statement, but: (1) we might want to restrict to locally small categories so that we can apply the Yoneda lemma; and (2) I think that the categories people care about in practice are locally small (although functor categories are an exception, as you say).

• That's definitely a good explanation, thank you. Do you know why people care about large categories them (or what large categories are interesting in the first place)? Jun 21 '21 at 8:39
• @Saegusa Exactly because they arise naturally as functor categories (and from some other constructions). Jun 21 '21 at 12:15
• When we work with universes, then we can always apply a version Yoneda Lemma relative to a category of sets which is sufficiently large. Jul 8 '21 at 16:46

I believe that functors and natural transformation morally form a 2-category is independent from the issue with smallness. There are also 1-categories which are not locally small. For example the localisation of a locally small category can have large hom sets. Localisations are morally 1-categories. Whenever you have a 2-category, consising of 0,1,and 2-cells, then you can two 1-cells and consider the 2-morphisms between them, and this is a 2-category.

In my opinion the author is saying that the functor category is morally not a 1-category not because of the size of the hom-sets, but because you loose a lot of information when you can not compose natural transformations horizontally.

If you feel confused about classes/sets/smallness, then I suggest you read somewhere online about Grothendieck universes. Alternatively you can read this text, which I have written for myself last year, when I got confused about the same questions:

• You may find people who think localisations are morally $\infty$-categories… Jun 21 '21 at 15:15