Not signed in (Sign In)

Start a new discussion

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.
    • CommentAuthorsherman
    • CommentTimeSep 29th 2016

    The (0,1)-category page says that a (0,1)-category “with the structure of a Grothendieck topos” is a locale, and a (0,1)-category “with the structure of an elementary topos” is a (complete?) Heyting algebra.

    I am trying to understand the formal content of these statements. By the Giraud characterization of Grothendieck toposes, I could believe that (0,1)-categories which happen to be “Giraud toposes” (i.e., Grothendieck toposes) are locales. Running through the Giraud axioms, we have:

    1. “a locally small category with a small generating set” - should be true for any (0,1)-category
    2. “with all finite limits” - this means all finite meets for a (0,1)-category, which locales have
    3. “with all small coproducts, which are disjoint, and pullback stable” - for a (0,1) category, all small coproducts are small/infinitary disjunctions, and pullback stability corresponds to the “frame distributivity” rule which holds for locales, which says that binary meets distribute over joins. I’m not sure about the “disjoint” condition, though.
    4. “where all congruences have effective quotient objects, which are also pullback-stable.” I don’t understand.

    But it does not seem that a (0,1)-category which is also an elementary topos is a complete Heyting algebra. The issue is that there is no subobject classifier. One reasons that the subobject classifier must be the terminal object of the cHa (it would be necessary to have a map into the subobject classifier from any object), but then since the pullback of every iso is an iso, every object must be the terminal object (Top), so the category is made only of terminal objects.

    Is it in fact the case that a (0,1)-category which is a Grothendieck topos is a locale? And is there any formal way to view complete Heyting algebras as decategorified elementary toposes?

    • CommentRowNumber2.
    • CommentAuthorDavidRoberts
    • CommentTimeSep 29th 2016
    • (edited Sep 29th 2016)

    Regarding congruences, these are not very interesting in a poset, all parallel arrows being equal already. So the quotient object is just the codomain.

    At Heyting algebra the qualifier ’almost’ is used when describing the relation.

    Todd Trimble writes on his personal web:

    a topos is a locally cartesian closed Heyting pretopos (in particular an effective regular extensive category with finite colimits, covering all those conditions of Giraud’s theorem that are “elementary”)

    so aside from arbitrary colimits, you can use a lot of what you wrote already.

    It may be that the idea of a subobject classifier is just too impredicative to exist in a poset. I wouldn’t be surprised if there were a fibration over the poset that played the role of the subobject classifier; maybe the codomain fibration does this, since all arrows are monosyllabic.

    • CommentRowNumber3.
    • CommentAuthorTodd_Trimble
    • CommentTimeSep 30th 2016
    • (edited Sep 30th 2016)

    Of course, that “is” (in the sentence from my web page) isn’t an “is” in the same sense as “is the same as” – it signifies satisfaction of a predicate (as in “Todd is blue-eyed”).

    Disjointness of coproducts fails badly in a general poset.

    I’m not sure what was intended on the (0,1)(0, 1)-category page; can someone clarify? I do think of lex totality as coming close to capturing the property of being a Grothendieck topos, and that being a locale could be reasonably construed as a (0,1)(0, 1)-version of lex totality.

    As for the stuff about elementary topos: not sure what was meant there either. We can say that a Heyting algebra is a (0,1)(0, 1)-quasitopos: that’s very different. (Notice for one thing that toposes are balanced categories, whereas posets are balanced iff they are discrete.)

    • CommentRowNumber4.
    • CommentAuthorDavidRoberts
    • CommentTimeSep 30th 2016

    Of course, that “is” (in the sentence from my web page) isn’t an “is” in the same sense as “is the same as”

    yes, I hoped that was clear.

    Disjointness of coproducts fails badly in a general poset.

    I thought that would be a bit tricky to satisfy :-)

    As far as elementary toposes go, I would guess that one could define an elementary (0,1)-topos as a Heyting algebra (not necessarily complete, since that would be a non-elementary notion). Just as a (2,1)-topos is not a (2,1)-category that is a topos, insofar as that makes sense…

    • CommentRowNumber5.
    • CommentAuthorUrs
    • CommentTimeSep 30th 2016

    The statement that “elementary (0,1)(0,1)-toposes are essentially equivalent to Heyting algebras” was added by Toby in revision 2. This was in reaction to me saying “(0,1)-topos” for “Grothendieck (0,1)”-topos in version 1.

    • CommentRowNumber6.
    • CommentAuthorTodd_Trimble
    • CommentTimeSep 30th 2016

    yes, I hoped that was clear

    I knew you knew it, David, but wanted to make sure it was clear to everyone reading. Being a locally cartesian closed Heyting pretopos does come pretty close after all to being a topos. A while back I found a counterexample; I’ll have to recall what it is.

    • CommentRowNumber7.
    • CommentAuthorMike Shulman
    • CommentTimeSep 30th 2016

    Just as a 1-topos (elementary or Grothendieck) is not the same as “a 2-topos that happens to be a 1-category”, a (0,1)-topos is not the same as “a 1-topos that happens to be a poset”.

    In the Grothendieck case, the difference is that in an nn-topos one requires (in addition to existence of all colimits) descent for “quotients of multi-object nn-groupoids”. When n=1n=1, that reduces to coproducts and quotients of equivalence relations. When n=(,1)n=(\infty,1) it’s equivalent to having descent for all (,1)(\infty,1)-colimits. But when n=(0,1)n=(0,1) I think it is a vacuous condition.

    In the elementary case, the difference is that in an nn-topos one requires classifiers for nn-subobjects. When n=1n=1 this is an ordinary subobject classifier. When n=n=\infty it is an classifier for all objects. But when n=(0,1)n=(0,1) I think it is again vacuous.

    • CommentRowNumber8.
    • CommentAuthorTodd_Trimble
    • CommentTimeSep 30th 2016
    • (edited Sep 30th 2016)

    Ah, thank you, Mike! So the statements on the (0,1)-category page are all right after all?

    It seems to me that (0,1)-topos could use more fleshing out, e.g., a definition section. The page (n,1)-topos on the other hand looks useful.

    It may not be important for this discussion, but the example I wanted at the end of #6 is perhaps the ex/lex completion of Set Set^{\bullet \to \bullet}, which if I got things right is a locally cartesian closed pretopos but which has no subobject classifier.

    • CommentRowNumber9.
    • CommentAuthorMike Shulman
    • CommentTimeSep 30th 2016

    Well, I probably wouldn’t say “A (0,1)-category with the structure of a topos is a (0,1)-topos”, especially since topos is linked to a page mostly about 1-toposes. I would say something more like “A (0,1)-topos is a (0,1)-category with extra structure” (in fact, property-like structure).

    • CommentRowNumber10.
    • CommentAuthorTodd_Trimble
    • CommentTimeOct 1st 2016
    • (edited Oct 1st 2016)

    I’m actually a little stuck: what is a (0,1)-topos from first principles? The Idea section of that page reads as if it’s a Heyting algebra by fiat, which of course we don’t want.

    The page (n,1)-topos isn’t as helpful for this as I first imagined, since it only seems to deal with Grothendieck (n,1)-toposes, following Lurie. Please correct me if I’m wrong.

    Edit: looking at Mike’s #5 again, I suppose for a terminal (0,1)(0, 1)-topos the terminal object would be the classifier. But then, if we define a (0,1)(0, 1)-topos to be be a cartesian closed (0,1)(0, 1)-category with such a classifier (which is there vacuously), then we get just a cartesian closed poset, which need not have joins to make it a Heyting algebra. Maybe I don’t understand what an elementary (n,1)(n, 1)-topos is supposed to be.

    • CommentRowNumber11.
    • CommentAuthorMike Shulman
    • CommentTimeOct 1st 2016

    Well, I don’t think the definition of elementary nn-topos is decided in general yet, but it should certainly include finite colimits. I view it as sort of an accident that when n=1n=1 one can construct finite colimits from power objects and so they can be excluded from the definition; in general they should be there.

Add your comments
  • Please log in or leave your comment as a "guest post". If commenting as a "guest", please include your name in the message as a courtesy. Note: only certain categories allow guest posts.
  • To produce a hyperlink to an nLab entry, simply put double square brackets around its name, e.g. [[category]]. To use (La)TeX mathematics in your post, make sure Markdown+Itex is selected below and put your mathematics between dollar signs as usual. Only a subset of the usual TeX math commands are accepted: see here for a list.

  • (Help)