# Canonical topology for infinity topoi revisited.

A while ago I asked this quetion: Canonical topology for big infinity topoi

and this question: How to resolve size issues with the regular epimorphism topology

Let me first summarize some of what I learned from these:

If $U$ is a fixed ambient Grothendieck universe, then one can define a $U$-site to be a (not necessarily $U$-small) Grothendieck site $(C,J)$ such that there exists a $U$-small set $G$ of objects, called topological generators, such that every object of $C$ admits a covering family all of whose sources are in $G$.

$U$-sites are useful to have around, because they allow you to deal with "large" Grothendieck sites whose topos of sheaves are equivalent to the topos of sheaves of a small site. For example, if $E$ is a $U$-topos, i.e. a category equivalent to sheaves of $U$-small sets on some $U$-small site, then it is certainly not $U$-small itself. However, it does carry a Grothendieck topology, the canonical Grothendieck topology which is generated by jointly surjective epimorphisms. If we choose a $U$-small site of definition for $E$ such that $K$ is subcanonical, so that $E\cong Sh_K\left(D\right)$, then the objects of $D$, considered as representable sheaves, form a $U$-small set of topological generators for $E$, showing that $E$ with the canonical topology is in fact a $U$-site. Now, the category of $U$-small presheaves on $E$ is not $U$-small, but it's locally $U$-small, and the inclusion of the full subcategory of sheaves admits a left-exact left-adjoint (expose ii, theorem 3.4 of SGA 4). Moreover, this category of sheaves, is equivalent to $E$ itself.

I'm pretty sure that I can prove that all of this goes through for $n$-topoi when $n$ is finite. The complication arises when $E$ is a genuine infinty topos. Indeed, one can still equip $E$ with the canonical topology, and by careful use of Grothendieck universes as above, construct the infinity topos $Sh_{\infty}\left(E,can\right)$. However, its possible that $E$ is not equivalent to infinity sheaves on some site (e.g. it could be hypersheaves), so in this case, $Sh_{\infty}\left(E,can\right)$ can not be equivalent to $E$, correct? Even worse, $E$ could land somewhere between sheaves and hypersheaves, so we can't just say $E$ is the hypercompletion of $Sh_{\infty}\left(E,can\right)$.

My quetsion is, what is the relationship between $E$ and $Sh_{\infty}\left(E,can\right)$ when $E$ is an infinity topos? When $E$ is equivalent to infinty sheaves on a site, are these the same?

• I think that the keyword that you're looking for is cotopological localization (page 531 of Higher Topos Theory: xxx.lanl.gov/pdf/math/0608040v4.pdf) – André Henriques Apr 25 '12 at 19:03
• Thanks Andre. This looks like a nice approach. I am guessing you'd like me to show that $E$ is always a cotopological localization of $Sh_{\infty}\left(E\right)$. I can show the Yoneda embedding has a left-adjoint, but I'm having trouble (so far) in proving that it's left-exact. – David Carchedi Apr 25 '12 at 22:37
• If $E$ is hypersheaves on one site, does that preclude its being equivalent to sheaves on some other site (like itself)? – Mike Shulman Apr 26 '12 at 8:44
• @Mike: I am not sure. This is sort of implicit in the question. (hence my saying "correct?") – David Carchedi Apr 26 '12 at 12:31

I will write what I think is a proof that in fact every infinity topos is equivalent to sheaves over itself. Please let me know if I am making any errors. I am basically adapting a proof from SGA4 of the classical statement.

Let $U$ a Grothendieck universe and suppose that $E$ is a left-exact localization of presheaves of $U$-small infinity groupoids on some $U$-small site. Then $E$ possesses a $U$-small set of generators, $X_\alpha$, $\alpha \in A$. By HTT 6.3.5.17, the Yoneda embedding of $E$ into sheaves of $V$-small infinity groupoids on $E$, with $U \in V$ a larger Grothendieck universe, preserves $U$-small colimits.

Lemma: If $i:F \hookrightarrow G$ is a mono with $F$ and $G$ infinity sheaves on $E$, with $G$ representable, then $F$ is representable.

Pf: Consider the family $\left(f:X_\alpha \to F, f \in F\left(X_\alpha\right)_0\right), \alpha \in A$, which is jointly epimorphic. Hence, the corresponding Cech-nerve is effective. Each iterative fiber product in this nerve, say $$X_{\alpha_1} \times_F ...\times_{F} X_{\alpha_n}\simeq X_{\alpha_1} \times_G ...\times_{G} X_{\alpha_n},$$ since $i$ is mono. But the right-hand side is representable. Hence, this is actually an effective groupoid object in $E$, so it has a colimit $C$ in $E$. Since the inclusion of $E$ into $V$-sheaves (the Yoneda embedding) preserves colimits, we conclude that $C$ represents $F$.

Now consider $H$ to be an arbitrary sheaf of $U$-small infinity groupoids on $E$. Consider the family $\left(g:X_\beta \to H, f \in H\left(X_\beta\right)_0\right), \beta \in B$. Each iterative fiber product in the associated nerve, say $X_{\alpha_1} \times_H ...\times_{H} X_{\alpha_n}$ is a subobject of the representable sheaf $X_{\alpha_1} \times...\times X_{\alpha_n}$, but also an infinity sheaf, hence it is representable by an object of $E$ by the lemma. So we again have an effective groupoid object which actually lies entirely in $E$, and as in the lemma, we conclude $H$ is representable.

• In fact, we don't need to appeal to 6.3.5.17, since a groupoid object coming from an epi in an infinity topos, corresponds to a cover in the canonical topology, so, the Yoneda embedding will preserve this colimit automatically. – David Carchedi Apr 27 '12 at 1:06
• I think I believe it. – Mike Shulman Apr 27 '12 at 16:02
• It still seems fishy to me. Because, the full subcategory on the small set of generators should also be a site for $E$. But, if $E$ is a left-exact localization of presheaves on $C$, the the Yoneda embedded image of $C$ is a small set of generators. So this would seem to imply that $E$ is necessarily sheaves on $C$, but we know that need not be the case... – David Carchedi Apr 27 '12 at 22:58
• Oh, of course: in an $(\infty,1)$-category the pullback is not a subobject of the product. – Mike Shulman Apr 29 '12 at 9:10
• For the record: When trying to think about this question I find it very confusing that 6.3.5.17 overloads the notation "Shv(C)" to mean colimit-preserving functors, rather than sheaves for a topology. – Mike Shulman Apr 29 '12 at 9:56

I've convinced myself the answer is the following CLAIM however, I have not finished proving it yet, but it is too long to leave as a comment. Hopefully soon I will update this with a proof of the claim as well (but feel free to beat me to it).

ClAIM: For any $n$, possibly infinite, if $E$ is an $n$-topos, and $C$ is a small $n$-category from which $E$ can be obtained by a left-exact localization $$a:Psh_n(C) \to E$$ then the canonical topology on $E$ restricts to a Grothenieck topology on $C$ via the composite of $a$ with the Yoneda embedding (this is clear), and this induces an equivalence of $n$-topoi between $Sh_n(C)$ and $Sh_n(E)$ (this needs proof).

Why do I think this is true?

1.) For finite $n$, this implies that the functor $Sh_\infty:n-Top \to \infty-Top$ which assigns an $n$-topos its infinity topos of infinity sheaves over itself with respect to the canonical topology is a fully faithful right-adjoint to the functor $\tau_{\le n-1}$ which assigns an infinity topos its $n$-topos of $\left(n-1\right)$-truncated objects. (See HTT 6.4.5.7).

2.) For $n=\infty$, by using HTT 6.5.2.19, we know that if $E$ is an infinity topos and $$L:Psh_\infty(C) \to E$$ is a left-exact localization, then it factors uniquely as localizations $$Psh_\infty(C) \to Sh_\infty(C,J) \to E,$$ for some Grothendieck topology on $C$, where the localization $Sh_\infty(C,J) \to E$ is cotopological. However, it is not hard to see that this $J$ is the topology induced from the canonical topology on $E$. Combining this with the claim would show that not only is this factorization unique, but the infinity topos $Sh_\infty(C,J)$ does not depend on $C$. In particular, it would imply that if $E$ is a non-sheaf topos for some site, than it is not a sheaf topos for any site.