Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Let $\mathcal{E}$ and $\mathcal{F}$ be toposes, $X$ an object of $\mathcal{E}$ and $p: \mathcal{E}/X \rightarrow \mathcal{E}$ the canonical geometric morphism (whose inverse image part is pullback along $p: X \rightarrow 1$.) I am trying to figure out the correspondence between geometric morphisms $\mathcal{F} \rightarrow \mathcal{E}/X$ over $\mathcal{E}$ and "global sections" $1 \rightarrow f^*(X)$. More precisely, I am looking to establish $$\text{Hom}_{\textbf{Topoi}/ \mathcal{E}} ( f: \mathcal{F} \rightarrow \mathcal{E}, \mathcal{E}/X) \cong \text{Hom}_{\mathcal{F}} (1, f^{*} (X))$$

Given a geometric morphism $g$ it is easy to get a section. One simply sends $g$ to the section $$g^* (\Delta) : 1 \rightarrow g^*(\pi_2 : X \times X \rightarrow X) = g^* \circ p^* (X) = f^*(X) $$

where $\pi_2$ is the standard projection and $\Delta$ the diagonal map $X \rightarrow X \times X$. (The idea is that $\Delta$ is the universal section, classifying the identity $\mathcal{E}/X = \mathcal{E}/X$.)

I'm having some difficulty figuring out the converse. Given a section $s : 1 \rightarrow f^*(X)$ we do get a geometric morphism $\mathcal{F} \rightarrow \mathcal{F}/f^*(X)$ and I can see that this ought to be 'sendable' to $\mathcal{E}/X$ but cannot figure out a nice way to do this via a geometric morphism (the wall I keep hitting is that the unit map $\eta_X : X \rightarrow f_* f^*(X)$ induces a geometric morphism in the 'wrong' direction.)

So I guess my first question is:

How does one get the right-to-left correspondence, i.e. from sections to geometric morphisms in this case?

In general, I am also interested to hear how one should generally think of these classifying problems (finding classifying topoi for given structures.) I am new to the game and every case I encounter seems to me now devoid of a general principle of attack. In some cases it seems much easier to go from 'points'/'individual structures' to geometric morphisms. For example, in geometric cases, a point immediately gives the stalk/skyscraper adjunction and hence a geometric morphism. In other cases, like the one above, where the universal structure is clear, it is much easier to go from geometric morphism to a 'point'/'structure'.

So my second question is this:

Is there a good standard way to think/attack such problems? Should one always try and figure out the universal structure first? Should one think geometrically? Any suggestions will be greatly appreciated...

(NOTE: I understand that the second question is not very precise, but neither am I looking for very precise answers.)

Finally, a bonus question: is the correspondence in the problem above (the first equation) a kind of adjunction between functors $\textbf{Topoi}/\mathcal{E} \overset{\leftarrow}{\rightarrow} \mathcal{F}$ (where $\mathcal{F}$ lies over $\mathcal{E}$.) Can one view this as some sort of adjunction in this manner? I suspect not, but I'd like to know if this is a fact made more elegant from a 2-categorical perspective...

share|improve this question
Can you give a source for your claims? I've never seen anything like it. If we take the case $\mathcal{E} = \textbf{Set}$, $\mathcal{F} = \textbf{Set}$, then it's certainly true that an $\mathcal{E}$-morphism $\mathcal{F} \to \mathcal{E} / X$ corresponds to an element of $X$. But it feels like something much more complicated should happen in general... –  Zhen Lin Apr 30 '12 at 8:07
@Zhen Lin This is exercise 6 of chapter VIII of Moerdijk and Maclane. (I didnt put it in the main text so as not to make the exercise searchable) –  Chuck Apr 30 '12 at 12:29
That seems like a silly thing to do. Anyone who wants to search is going to search for keywords rather than an explicit exercise number! –  Zhen Lin Apr 30 '12 at 15:56
@Zhen Yes, you're probably right... –  Chuck Apr 30 '12 at 16:35

1 Answer 1

Allow me to change notation a little. Let $\mathcal{S}$ be a topos, and let $f : \mathcal{E} \to \mathcal{S}$ be a geometric morphism. Let $\mathcal{S}_X = (\mathcal{S} \downarrow X)$. We are trying to look at the hom-category $$\textbf{Geom}_{\mathcal{S}}(\mathcal{E}, \mathcal{S}_X)$$ which is the full subcategory of $\textbf{Geom}(\mathcal{E}, \mathcal{S}_X)$ spanned by geometric morphisms $g : \mathcal{E} \to \mathcal{S}_X$ such that $p \circ g \cong f$, where $p : \mathcal{S}_X \to \mathcal{S}$ is the canonical projection.

The situation we should keep in mind is the case where $\mathcal{E} = \mathcal{S}_Y$ for some other object $Y$. Then, we are just looking at base change along some morphism $Y \to X$ in $\mathcal{S}$. Geometric morphisms like $p : \mathcal{S}_X \to X$ are very special: they come as a triple of adjoint functors $$p_! \dashv p^* \dashv p_* : \mathcal{S}_X \to \mathcal{S}$$ and $p_! 1$ is canonically isomorphic to $X$. Unfortunately, $f : \mathcal{E} \to \mathcal{S}$ is not guaranteed to be of this form, so we must find a surrogate for $f_! 1$.

But, for the moment, let us imagine we had $f_! \dashv f^* \dashv f_*$ of the same form as $p_! \dashv p^* \dashv p_*$. Let $x : 1 \to f^* X$ be a morphism in $\mathcal{E}$. It would correspond to a unique morphism $f_! 1 \to X$ in $\mathcal{S}$. Take $Y$ in $\mathcal{S}_X$. It corresponds to a unique morphism $p_! Y \to X$ in $\mathcal{E}$. We want to find a left exact left adjoint functor $g^* : \mathcal{S}_X \to \mathcal{E}$, so there's only one thing we can do: take the base change of $p_! Y \to X$ along $f_! 1 \to X$. This "is" the required object in $\mathcal{E}$, except it's in $\mathcal{S}$ rather than in $\mathcal{E}$. Now, apply $f^*$ to the pullback diagram; since $f^*$ is left exact, we get a pullback diagram in $\mathcal{E}$. Consider the naturality square of the unit $\textrm{id} \Rightarrow f^* f_!$: if $f_! \dashv f^* \dashv f_*$ comes from base change, then this would be a pullback square. So by the pullback pasting lemma, we can get the base change of $p_! Y \to X$ along $f_! \to X$ as an object in $\mathcal{E}$ by taking the base change of $f^* p_! Y \to f^* X$ along $x : 1 \to f^* X$. Notice that we've erased all mention of $f_!$ by doing this!

But enough handwaving. Let's actually define the functor $g^*$ properly now. Write $x^* : \mathcal{E}_{f^* X} \to \mathcal{E}$ for base change along $x : 1 \to f^* X$. One may check that $f^* p_! : \mathcal{S}_X \to \mathcal{E}$ factors through the forgetful functor $\mathcal{E}_{f^* X} \to \mathcal{E}$; abusing notation slightly, let us also write $f^* p_!$ for the functor $\mathcal{S}_X \to \mathcal{E}_{f^* X}$. This is obviously left exact, so the composite $g^* = x^* f^* p_! : \mathcal{S}_X \to \mathcal{E}$ is also left exact. In less sophisticated terms, $g^*$ maps an object $Y$ of $\mathcal{S}_X$ (i.e. a morphism $p_! Y \to X$ of $\mathcal{S}$) to the object $1 \times_{f^* X} f^* p_! Y$ of $\mathcal{E}$ obtained by the base change of $f^* p_! Y \to f^* X$ along $x : 1 \to f^* X$. One may verify that $g^*$ preserves all colimits, so it has a right adjoint by the special adjoint functor theorem. It is easy to see that $g^* \cong f^* p^*$, so we have the desired geometric morphism.

Your second question about how to find classifying toposes deserves a good answer, but I'm afraid I don't have the time to write one right now. Since it's not quite related to the first question, perhaps it would be best to start a new thread on it and see if anyone else answers.

share|improve this answer
Thanks for this answer. I don't understand what version of the SAFT you are using there (which I associate with the existence of a left adjoint)? I'm assuming what you have in mind is the theorem that says that a functor between Grothendieck toposes has a right adjoint iff it preserves colimits. However we cannot know that the toposes in question here are Grothendieck. Maybe I'm missing something.... –  Chuck May 4 '12 at 17:26
Yes, that's basically a dual form of the special adjoint functor theorem. You're right that we need to assume enough colimits + a "small" generating set, but I imagine an $\mathcal{S}$-indexed version of SAFT should be enough. Or you could construct the right adjoint by hand. shrug –  Zhen Lin May 4 '12 at 17:57

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.