Want to take part in these discussions? Sign in if you have an account, or apply for one below
Vanilla 1.1.10 is a product of Lussumo. More Information: Documentation, Community Support.
Stubby start to topological topos. Will be adding material by and by.
Could we call this something else? Johnstone’s article was just called “on A topological topos”; I think there could be many different toposes that deserve to be thought of as “topological” for various reasons. Johnstone remarked slightly tongue-in-cheek that the objects of this topos might be called “consequential spaces,” so if we wanted to go with that we could write about them on a page called consequential space. Or we could think of another name.
There could be, but there isn’t. :-) Seriously: if someone wanted to look this up or google this, they’d type “topological topos”, wouldn’t they? At least that’s what I tried. I doubt they would try consequential space.
I could go along with something like “Johnstone’s topological topos”.
What about the topos of sheaves on some small subcategory of Top, or the topos of sheaves on a category of (ultra)filters? In the article Johnstone lists a few other contenders like that. I’ve never heard anyone refer to this as “the topological topos.” And that name isn’t really correct anyway; it’s not the topos that’s topological, its the objects of the topos.
Okay, let’s focus less on my first sentence of #3 (which seems to be leading to an argument I didn’t intend), and more on what might be the most useful title.
You’ll notice I didn’t say “the topological topos”. On the other hand, despite your last sentence, it was Johnstone’s phrase, not mine. So, what say you to “Johnstone’s topological topos”? I’m thinking of the person trying to find out what this thing is that’s being referred to in, for example, the page on subsequential space.
Edit: and it might not be just from reading that page on the nLab – it could be some poor slob like me who’s heard of the darned thing, googles it, and comes up with little except bibliographic references and maybe a link to the paper for a price. I wanted to give it a title which is more likely to attract hits.
Okay, fair enough. I can certainly go along with “Johnstone’s topological topos.”
I added to Johnstone’s topological topos a proof that the Cauchy real numbers object therein also has the usual topology. (Johnstone’s paper only discusses the Dedekind real numbers.) It would be nice to have a more abstract proof — e.g. does this topos possess some general property that is sufficient for the Cauchy and Dedekind to coincide? — but I can’t think of one.
Nice. Could you include a reference that it does not support countable choice?
Hmm… no, I don’t think I could. Actually I’m not completely sure, but my instinct is that it doesn’t. (Though it does satisfy “crisp countable choice”, i.e. is externally projective.)
Fourman - continuous truth II shows that the gross topos of sheaves over separable locales has local choice. Both this gross topos and Johnstone’s topological topos are ways to cut down the Giraud gross topos. It would be nice to understand what breaks countable choice in Johnstone’s topos.
What is “local choice”?
It’s defined in Continuous truth I below Prop 4.3. Locally, we have a choice function, i.e. there is an open cover U_i on which we can define a choice function on each U_i.
This does not seem enough to derive countable choice. However, it is suggested that CAC holds in this model. I am surprised it is not clearly stated. Perhaps, I am overlooking something.
A similar model is described by van der Hoeven, Moerdijk - sheaf models for choice sequences. Countable choice holds in this topos (2.2.6) and the real number object is computed on p81. The topological monoid of endomorphism of Baire space seems to be a topological site as in continuous truth II, but I haven’t checked it formally yet. Similarly, for Johnstone’s monoid of endomorphisms on . I expect this to be folklore.
After some conversation with Martin and trying to build a counterexample, I’m no longer sure that the topological topos fails countable choice. In fact, here is an argument claiming that it satisfies countable choice. Please poke a hole in it.
We have to show that is internally projective, i.e. that dependent product along preserves epis. Since is a countable coproduct of copies of 1, is a countable coproduct of copies of , which means that this dependent product can be identified with the countably infinite product functor on the slice category .
Now, Johnstone’s explicit description of the Grothendieck topology defining this topos implies that for a map to be epi means that (1) it is surjective on points and (2) for any convergent sequence in , and any infinite subsequence , there is a further subsequence and a convergent sequence in that maps via to .
Thus, suppose we have epis for natural numbers ; we want to show that is epi, where is the pullback of all the ’s over . It’s certainly surjective on points. Let be a convergent sequence in , and an infinite subsequence. Since is epi, there is a subsequence of and a convergent sequence mapping via to . Now since is epi, there is a subsequence of and a convergent sequence mapping via to . And so on. (Of course, in choosing particular such subsequences we are using countable/dependent choice in Set.) Now define , and a sequence in the pullback by . Since convergence in is defined in each factor, where , and it maps via to , with a subsequence of the given . Thus, is also epi.
Also, a (un?)related question. In Johnstone’s paper he cites an abstract by Isbell as showing that the pair of maps and are not universally effective-epimorphic in the monoid of continuous endomorphisms of the topological unit interval, and claims that this is related to the fact that the interval is locally connected. Can anyone reproduce (or know a citation for) such a proof? I’m guessing it has something to do with pulling back along some continuous map that behaves very badly near , but I haven’t been able to get any further, or get any intuition for what it has to do with local connectedness.
Moerdijk-Reyes have some remarks about the failure of countable choice in the Euclidean topos. For them a topological topos is a topos of sheaves on a topological site.
This should connect with Johnstone’s topological topos, but I haven’t seen the details spelled out.
Interesting; they specifically say it is because of the “connectedness” of the euclidean site.
I added a remark to topological topos that while LPO fails, LLPO holds.
1 to 18 of 18