Not signed in (Sign In)

Not signed in

Want to take part in these discussions? Sign in if you have an account, or apply for one below

  • Sign in using OpenID

Discussion Tag Cloud

Vanilla 1.1.10 is a product of Lussumo. More Information: Documentation, Community Support.

Welcome to nForum
If you want to take part in these discussions either sign in now (if you have an account), apply for one now (if you don't).
    • CommentRowNumber1.
    • CommentAuthorDavidCarchedi
    • CommentTimeSep 28th 2012
    • (edited Sep 28th 2012)
    I have noticed the following error in the categorification of a local homeomorphism of spaces into a concept of a local homeomorphism of n-topoi. I have a suspicion that it must have been noticed before (at least for 1-topoi), and if this is true, please view this discussion, in particular, as a reference request.

    **Background**

    For a topological space X, if F is a sheaf, it has an \'etal\'e space Y_F \to X which is a local homeomorphism over X, and sections of it are exactly the sheaf F. The \'etal\'e space construction yields an equivalence of categories
    Sh\left(X\right) \simeq Et/X between the category of sheaves on X and the category of local homeomorphisms over X. Moreover, assuming X is sober, since sober topological spaces embed fully faithfully into topoi, for a sheaf F, Y_F \to X corresponds to a geometric morphism Sh(Y_F) \to Sh(X) and there turns out to be an equivalence of topoi Sh(Y_F) \simeq Sh(X)/F under which this geometric morphism is equivalent to the one induced by slicing: Sh(X)/F \to Sh(X).

    Using this as an example, topos theorists said that a geometric morphism of 1-topoi \mathcal{F} \to \mathcal{E} is *\'etale* if it is equivalent to one of the form \mathcal{E}/E \to \mathcal{E}, and this was to be a ``local homeomorphism'' of 1-topoi.

    **The Problem**

    Define an \'etale geometric morphism of n-topoi the same way, i.e.

    \mathcal{F} \to \mathcal{E} is *\'etale* if it is equivalent to one of the form \mathcal{E}/E \to \mathcal{E}.

    Now let X be a (sober) topological space. In particular, it is a locale, hence a 0-topos in the sense of Lurie. Lets write this 0-topos by O(X) (as it is in fact the lattice of opens of X.) The maps of locales that are induced by slicing are exactly those which correspond to inclusions of open sublocales. So, an \'etale map of 0-topoi does not correspond to a local homeomorphism (but it is a particular kind of local homeomorphism).

    You might ask: So what? Maybe this is just a defect for n=0.

    No, this persists. An \infty-topos \mathcal{X} is *n-localic* if it is equivalent to \infty-sheaves on an n-site. There is a functor from n-topoi to \infty-topoi which is fully faithful and whose essential image is n-localic \infty-topoi, and if \mathcal{X} is an n-localic, the n-topos which is corresponds to is the \left(n,1\right)-category of \left(n-1\right)-truncated objects of \mathcal{X}.
    Moreover, if X is an object of an \mathcal{X}, then \mathcal{X}/X is also n-localic *if and only if* X is n-truncated. So let X be an object of \mathcal{X} which is n-truncated but not \left(n-1\right)-truncated. Then the \'etale map of \infty-topoi \mathcal{X}/X \to \mathcal{X} is a morphism between n-localic \infty-topoi. Let \mathcal{F} denote the n-topos associated to \mathcal{X}/X and \mathcal{E} the one associated to \mathcal{X}. Since n-topoi embed fully faithfully into \infty-topoi as n-localic \infty-topoi, this \'etale map induced by X
    must correspond to a geometric morphism \mathcal{F} \to \mathcal{E}. However, it cannot be an \'etale map of n-topoi, since the object X is not \left(n-1\right)-truncated. Nonetheless, such a geometric morphism is the "correct" notion of a local homeomorphism since if we let n=0, this is a local homeomorphism of locales, in the usual sense.

    Any comments? Has this been noticed for 1-topoi and is there a name for such geometric morphisms (and is there a characterization of them internal to 1-topoi?)
    • CommentRowNumber2.
    • CommentAuthorMike Shulman
    • CommentTimeSep 28th 2012

    That’s interesting! In summary, the point is that you can’t define local homeomorphisms between n-toposes to be slice categories unless you move up to their associated (n+1)(n+1)-toposes. This is clear when n=0, which is an essential case because it’s where the terminology ’local homeomorphism’ comes from, but it also implies that the usual definition of local homeomorphism for 1-toposes (as found, for instance, in the Elephant), is wrong. Is that an accurate summary?

    I don’t think I’ve heard this mentioned before, and I’d probably remember if I had. Terminology-wise, I think I’d be in favor of using the term ’local homeomorphism’ correctly, even if it doesn’t agree with previous usage.

    • CommentRowNumber3.
    • CommentAuthorMike Shulman
    • CommentTimeSep 28th 2012

    On the other hand, one might argue that we should define an “nn-local-homeomorphism” to be a morphism of \infty-topoi that is equivalent to slicing over an (n1)(n-1)-truncated object. Then the existing definitions of local homeomorphism for 0-topoi and 1-topoi would both be special cases of 1-local-homeomorphisms, while the “correct” notion of local homeomorphism between nn-topoi that you are advocating would be an (n+1)(n+1)-local-homeomorphism. That way, the existing terminology could be maintained, according to the general principle that an unprefixed “foo” is equivalent to a “1-foo”.

    As for a characterization internal to 1-topoi of the 2-local-homeomorphisms, shouldn’t they be the geometric morphisms FEF\to E where FF is equivalent to the topos of internal diagrams on some internal groupoid in EE?

    • CommentRowNumber4.
    • CommentAuthorDavidCarchedi
    • CommentTimeSep 28th 2012
    • (edited Sep 28th 2012)
    As for your first comment: Yes, that's about sums it up.

    As far as your second comment, that's *exactly* how I have already decided to remedy this in a paper I am writing; I define n-\'etale maps, etc. It's nice to hear that you thought of the same solution, so I'm reassured that it was a good call.

    As for your characterization, whereas I do agree that is a characterization, I was looking for one more akin to the "slice free" characterization of etale maps (f^* has an additional left adjoint f_{!} with special properties...).
    • CommentRowNumber5.
    • CommentAuthorDavidCarchedi
    • CommentTimeSep 28th 2012
    By the way, although I won't prove it here, another way at getting at local homeomorphisms is that for a given n-topos \mathcal{E}, it has a canonical topology, so we have the \left(n+1,1\right) category of sheaves of n-groupoids over \mathcal{E} (i.e. it's associated (n+1)-topos. We also have the (n+1,1)-category \mathfrak{Top}_n/\mathcal{E}. There is a canonical functor \chi:\mathcal{E} \to \mathfrak{Top}_n/\mathcal{E} given by slicing. The left Kan extension \mathbf{ET}:=\mathbf{Lan}_{y_\mathcal{E}} \chi:\mathbf{Sh}_{n}\left(\mathcal{E}\right) \to\mathfrak{Top}_n/\mathcal{E} of \chi along the yoneda embedding y_\mathcal{E}:\mathcal{E} \hookrightarrow \mathbf{Sh}_{n}\left(\mathcal{E}\right) exists as we write it as \mathbf{Lan}_{y_\mathcal{E} \circ y} \left(\chi \circ y\right) where y:\mathscr{C} \to \mathcal{E} is the Yoneda embedding associated to any n-site of definition for \mathcal{E}. Local homeomorphsms are precisely those in the essential image of \mathbf{ET}. (This is like a topos-theoretic \'etal\'e space construction.)
    • CommentRowNumber6.
    • CommentAuthorDavidCarchedi
    • CommentTimeSep 29th 2012
    Better yet, if you use presheaves instead of sheaves (with some size issues to deal with), you get an adjunction between presheaves and the slice category, restricting to an equivalence. (Oh yeah, \mathbf{ET} is fully faithful, I should've said that).
    • CommentRowNumber7.
    • CommentAuthorMike Shulman
    • CommentTimeSep 29th 2012

    I was looking for one more akin to the “slice free” characterization of etale maps

    Ah. Well, I can’t think of any. (-:

    BTW, you’ve probably noticed this by now, but the TeX syntax \'e for é doesn’t work here. You can write a literal é if you have a way to input Unicode, or (if you use the “Markdown+Itex” format option, which you really should), you can write the HTML é.

    • CommentRowNumber8.
    • CommentAuthorDavidCarchedi
    • CommentTimeSep 29th 2012
    • (edited Sep 29th 2012)

    Thanks for the comment about how to typeset é. Anyhow, it should be mentioned that classically defined 'etale maps are still an important class of morphisms, because after all:

    For any nn-topos ,\mathcal{E}, 𝔗𝔬𝔭 n et/.\mathcal{E} \simeq \mathfrak{Top}^{et}_n/\mathcal{E}.

    However, the theory of nn-étale morphisms does give a nice explanation as to why 𝔗𝔬𝔭 n et/\mathfrak{Top}^{et}_n/\mathcal{E} is an (n,1)(n,1)-category (since apriori one would expect it to be an (n+1,1)(n+1,1)-category, as it is the slice of one). The real statement is that

    Sh n()𝔗𝔬𝔭 n net/Sh_n\left(\mathcal{E}\right) \simeq \mathfrak{Top}^{n-et}_n/\mathcal{E}

    where netn-et refers to taking only nn-étale morphisms, and whereby Sh nSh_n is the functor Sh n:𝔗𝔬𝔭 n𝔗𝔬𝔭 n+1Sh_n: \mathfrak{Top}_n \hookrightarrow \mathfrak{Top}_{n+1} which promotes an nn-topos to an (n+1)(n+1)-topos. (The notation is suggestive since, at least for n,n \ne \infty, this is equivalent to taking sheaves on \mathcal{E} regarded as an nn-site with its canonical topology. For n=,n=\infty, this is problematic however (since an \infty-topos is not always sheaves over itself) so by convention the functor Sh Sh_\infty is the identify, but I digress…) Now, when n=n= \infty there is no difference between nn-étale and étale, and Sh =idSh_\infty=id so this reduces to the classical fact that etale maps over an \infty-topos is equivalent to the \infty-topos in question. When n,n \ne \infty, then we have that Sh n1()\mathcal{E} \simeq Sh_{\mathbf{n-1}}\left(\mathcal{E}\right) which in turn is equivalent to the full subcategory of Sh n()𝔗𝔬𝔭 n net/Sh_n\left(\mathcal{E}\right) \simeq \mathfrak{Top}^{n-et}_n/\mathcal{E} spanned by (n1)(n-1)-truncated objects, hence an (n,1)(n,1)-category, and the (n1)(n-1)-truncated objects in 𝔗𝔬𝔭 n net/\mathfrak{Top}^{n-et}_n/\mathcal{E} are precisely the étale ones! Phew! That was a bit more complicated to spell out than I thought!

    • CommentRowNumber9.
    • CommentAuthorMike Shulman
    • CommentTimeSep 29th 2012

    That might be the first time I’ve ever heard a theorem about \infty-toposes referred to as “classical”. Progress marches on, I suppose. (-;

    • CommentRowNumber10.
    • CommentAuthorDavidCarchedi
    • CommentTimeSep 29th 2012

    Well, I meant more that this was a special case of 𝔗𝔬𝔭 n et/,\mathcal{E} \simeq \mathfrak{Top}^{et}_n/\mathcal{E}, (which is the “n-ification” of the classical result for ordinary topoi) when n=.n=\infty.

    • CommentRowNumber11.
    • CommentAuthorMike Shulman
    • CommentTimeSep 29th 2012

    (-:

    • CommentRowNumber12.
    • CommentAuthorDavidCarchedi
    • CommentTimeSep 29th 2012

    I may be getting philosophical, but I think that maybe 𝔗𝔬𝔭 n et/\mathcal{E} \simeq \mathfrak{Top}^{et}_n/\mathcal{E} is the fundamental statement, and it is this equivalence which justifies the importance of étale maps. It seems that in some sense that it’s an accident that historically the fundamental statement was Sh(L)Et/LSh\left(L\right) \simeq Et/L for LL a locale, since this has a mismatch between viewing LL as a 11-topos and a 00-topos at the same time. Indeed, at n=0n=0, the statement L𝔗𝔬𝔭 0 et/LL \simeq \mathfrak{Top}^{et}_0/L for a locale LL is an interesting and true, and the fact that this holds true for n=1n=1 allows us to make the claim that Sh(L)𝔗𝔬𝔭 1 et/Sh(L)Sh(L) \simeq \mathfrak{Top}^{et}_1/Sh(L)- and then notice that an étale morphism into a localic topos must have a localic domain and correspond to a local homeomorphisms between them. Does anyone have any thoughts?

    • CommentRowNumber13.
    • CommentAuthorUrs
    • CommentTimeSep 29th 2012

    Hi Dave,

    I am certainly looking forward to seeing this stuff written up cleanly. That’ll be foundational.

    • CommentRowNumber14.
    • CommentAuthorDavidCarchedi
    • CommentTimeSep 29th 2012

    @Urs: Thanks! Do you have any opinion about whether one should view étale maps being the same as the topos as foundational, or the “étalé-space construction” as foundational, as discussed in my last comment?

    • CommentRowNumber15.
    • CommentAuthorDavidCarchedi
    • CommentTimeSep 29th 2012

    (Also, I am started to question as to whether or not I need to discuss all this in the paper I am writing after all, and was wondering if perhaps it should just be a separate paper entirely? Suggestions/advice? Would an OK tactic be to touch upon it, but then leave further developments for another paper?)

    • CommentRowNumber16.
    • CommentAuthorMike Shulman
    • CommentTimeSep 29th 2012

    You could write an nLab page about it. (-;

    • CommentRowNumber17.
    • CommentAuthorUrs
    • CommentTimeSep 29th 2012
    • (edited Sep 29th 2012)

    What you say in #12 seems to me indeed to be a most useful way of thinking about the situation.

    I don’t know of course exactly what paper you are about to write, but I’d think in any case in the introduction section it would be great to just sketch transparantly the situation discussed here, maybe in a kind of crisp summary as in Mike’s #2 above, followed by a paragraph with more details as in your #1.

    • CommentRowNumber18.
    • CommentAuthorDavidCarchedi
    • CommentTimeSep 29th 2012

    @Urs: Thanks again, and fyi: The paper I am writing at the moment is the one establishing a general theory of higher etale stacks.

    • CommentRowNumber19.
    • CommentAuthorDavidCarchedi
    • CommentTimeOct 13th 2012

    I have a hunch that the following is a characterization of local homeomorphisms (in the way I defined them) of topoi:

    f:f:\mathcal{E} \to \mathcal{F}

    is a local homeomorphism if and only if ff is open and its diagonal

    × \mathcal{E} \to \mathcal{E} \times_\mathcal{F} \mathcal{E}

    is. Moreover, I believe that if we define a geometric morphism

    f:f:\mathcal{E} \to \mathcal{F}

    of infinity topoi to be open if

    f:τ 1τ 1f:\tau_{\le -1}\mathcal{E} \to \tau_{\le -1}\mathcal{F}

    is an open map of locales (this agrees with the ordinary definition of an open geometric morphism for 11-topoi), then ff is 'etale if and only if ff and its diagonal are open. However, I haven’t thought of how to prove this yet.