# Does the Grothendieck construction produce a 2-category or a category?

Let $$F : \mathcal{C} \to \mathbf{Cat}$$ be a lax 2-functor. Then we can form a category $$\int F$$ which is the Grothendieck construction on F. There's a number of resources detailing this construction, but none mentioning if we can get a 2-category, rather than a category this way. It seems like the natural construction would be to generate a 2-category - is this true? I'm looking for a reference.

Namely - it seems like the 2-cells in $$\mathcal{C}$$ in standard definitions don't really play any role at all in $$\int F$$, but it seems like they should correspond to 2-cells in $$\int F$$.

The closest constructions I found were in (https://arxiv.org/abs/2002.06055, Definition 10.7.2.) . They provide a bicategorical Grothendieck construction for a functor $$F : \mathcal{C}^{op} \to \mathbf{Bicat^{ps}}$$ - but of a rather strange kind. It does not use 2-cells in $$\mathcal{C}$$ and it also assumes certain equality of 1-cells in $$\mathcal{C}$$.

Another potential candidate is https://www2.irb.hr/korisnici/ibakovic/sgc.pdf, but it seems to be on a higher level of abstraction than I'm comfortable with. It also seems to talk about the Grothendieck construction of functors whose codomain is $$\mathbf{2-Cat}$$, rather than into $$\mathbf{Cat}$$. It seems like this is extra structure that is not needed for what I'm asking.

So, in short, if there is a lax functor $$F : \mathcal{C} \to \mathbf{Cat}$$, is there a way to make the Grothendieck construction into a 2-category? If so - is there reference with an explicit construction, showing in details what all the 2-cells would look like?

• Bénabou studied this kind of construction: if $\mathcal C=1$ is the terminal category, $F$ consists of a monad in $\bf Cat$, on the category $A=F*$. When $\mathcal C$ is a generic bicategory, $F$ amounts to an "indexed" family of monads, i.e. to a functor $\Sigma_F : \mathcal E \to \mathcal C$ over $\mathcal C$ whose fiber over $C$ is the Kleisli category of the monad determined by $F|_{\{C\}} : \{C\} \to \bf Cat$. [tbh I'm not 100% sure of this statement. But you can try to work out this intuition and see what strictness $\hat F, \mathcal E$ have] – Fosco Apr 5 at 22:01

The usual Grothendieck construction has for $$\mathcal C$$ an ordinary category, so it doesn't have any 2-cells (or at least, doesn't have any non-identity 2-cells). Moreover, we actually get not only a category out of it, but an object of the slice (2-)category $$\mathcal Cat/\mathcal C$$. If $$\mathcal C$$ weren't an ordinary category, this wouldn't make as much sense since $$\mathcal C$$ wouldn't be an object of $$\mathcal Cat$$. (Just like how we can have functors $$\mathcal C \to \mathcal Cat$$, though, it's still possible to talk about functors from 1-categories to 2-categories).

Nonetheless, you can get a meaningful construction when adding in 2-cells. I have no idea if this construction has a name, but once you get the pattern, it's easy to extend this to any arbitrary level. For this construction, we'll assume we have a (lax) 2-functor $$F : \mathcal C \to \mathcal Cat$$.

0-cells of $$\int F$$ are pairs $$(c, x)$$ where $$c$$ is an object of $$\mathcal C$$ and $$x$$ is an object of $$F(c)$$.

1-cells $$(c, x) \to (c', x')$$ are pairs $$(f, g)$$ where $$f : c \to c'$$ and $$g : x \to_f x'$$. $$x \to_f x'$$ is the set of dependent morphisms from $$x$$ to $$x'$$ (terminology adapted from HoTT's dependent paths). Effectively, we use functorality in the types of $$x$$ and $$x'$$ to transport from one type to the other. In this case, the type of $$x$$ is $$F(c)$$ and the type of $$x'$$ is $$F(c')$$ so we can use $$F(f)$$ to map $$x$$ into the type of $$x'$$.

Summing up, $$g$$ should be a morphism $$F(f)(x) \to x'$$, i.e. an element of $$\hom_{F(c')}(F(f)(x), x')$$.

Next, our 2-cells $$(f, g) \to (f', g')$$ should be pairs $$(\alpha, \beta)$$ where $$\alpha$$ is a 2-cell $$f \to f'$$ in $$\mathcal C$$. $$\beta$$ should be a dependent morphism $$g \to_\alpha g'$$.

Now the types of $$g$$ and $$g'$$ are $$\hom_{F(c')}(F(f)(x), x')$$ and $$\hom_{F(c')}(F(f')(x), x')$$ respectively. This time, these type are contravariant in the variable we need to transport ($$f$$ and $$f'$$), so we'll transport $$g'$$ to $$\hom_{F(c')}(F(f)(x), x')$$ via $$\hom_{F(c')}(F(\alpha)(x), x')$$.

Unpacking this, $$g'$$ gets sent to $$g' \circ F(\alpha)(x)$$, so $$\beta$$ is a morphism $$g' \circ F(\alpha)(x) \to g$$. This time, though, we're talking about a morphism between morphisms in $$F(c')$$, which is an ordinary category. So rather than an actual morphism, we'll have an equality $$g' \circ F(\alpha)(x) = g$$.

• Thank you for this very detailed description, I think this might be exactly what I was looking for. Clarifications: I assume $\mathcal{C}$ and $\int F$ are both strict 2-categories here? If this is so, is there a reason why standard references (for the lax 2-functor version, not the 1-functor one) always claim we get a category and completely ignore the 2-categorical part? (people.mpi-sws.org/~dreyer/courses/catlogic/jacobs.pdf, Definition 1.10.1), (arxiv.org/abs/2002.06055, Definition 10.1.2) – Bruno Gavranovic Apr 6 at 1:28
• Furthermore, is it important whether $\mathsf{Cat}$ in your first and second paragraph is considered as a 1-category or a 2-category? I'm a bit confused when we mix, as you say, "functors from 1-categories to 2-categories". – Bruno Gavranovic Apr 6 at 1:31
• @BrunoGavranovic I usually use 2-category for what might be less ambiguously called a bicategory (but I really dislike that naming scheme). For this construction, I don't think I used strictness anywhere. When defining composition and checking all the coherence diagrams, you'll likely need to insert a bunch of structural isomorphisms everywhere. I don't think that would be a problem, though. – SCappella Apr 6 at 6:35
• @BrunoGavranovic A line in your second reference (at the start of chapter 10) I think answers both your other questions. "Throughout this chapter $\mathcal C$ denotes a small category, also regarded as a locally discrete 2-category." So $\mathcal C$ is an ordinary category, but for the purposes of talking about functors $\mathcal C^{op} \to \mathcal Cat$, we treat it as a 2-category whose higher morphisms are just identities. – SCappella Apr 6 at 6:35
• However - I do have one concern. It is known that the Grothendieck construction can be seen as an oplax colimit (ncatlab.org/nlab/show/Grothendieck+construction#AsALaxColimit). But here we have a 2-functor $F : \mathcal{C} \to \mathbf{Cat}$, where $\mathbf{Cat}$ is the 2-category of small categories. A colimit there would give us a category, rather than a 2-category. Does this mean that we need to specify a (?-)functor into $\mathbf{2-Cat}$? – Bruno Gavranovic Apr 6 at 19:27

It produces a functor between categories. In fact, what is called a fibred category. The construction is detailed in Volume 2 of Borceux's Handbook on Categorical Algebra.

Also Angelo Vistolis Notes on Grothendieck topologies, fibered categories and descent theory is worth looking at.