5
$\begingroup$

Let $X$ be an infinite set and let $C(X)$ denote the collection of connected Hausdorff topologies on $X$. Suppose $N\subseteq C(X)$ has the property that whenever $\tau\neq\sigma \in N$ then $(X,\tau)$ and $(X,\sigma)$ are not homeomorphic. In terms of $|X|$, how large can $|N|$ be at most?

$\endgroup$
9
  • 1
    $\begingroup$ +1 for this very interesting question. BTW what about if we add "compactness", too.? for example, up to homeomorphism, how many topologies on [0 1] make it a compact connected Hausdorff space? $\endgroup$ Apr 17, 2016 at 8:36
  • 1
    $\begingroup$ when the cardinality of $X$ is $c$, then the cardinality of $N$ is at least $c$. Because there are uncountable number of open contractible subsets of $\mathbb{R}^{3}$, mutually non homeomorphics. $\endgroup$ Apr 17, 2016 at 8:50
  • 2
    $\begingroup$ @Jeremy: No, it doesn't. Given a set $X$, there is a well defined set of topologies on $X$, and a well defined set of connected Hausdorff topologies on $X$, and homeomorphism is an equivalence relation. So far, no choice. The OP formulated this using a set of representatives which is the tour de force of choice. I, on the other hand, ask what is the cardinality of the set of equivalence classes, which again involves no choice. But of course it wasn't intentional or anything... :-) $\endgroup$
    – Asaf Karagila
    Apr 17, 2016 at 10:35
  • 1
    $\begingroup$ @Asaf I don't deny that both questions make sense without choice. It was your statement "So you're asking ...", claiming that they're the same question, that I was teasing you about. $\endgroup$ Apr 17, 2016 at 10:54
  • 1
    $\begingroup$ @Jeremy: And how does that work for thus far? :-P $\endgroup$
    – Asaf Karagila
    Apr 17, 2016 at 10:58

1 Answer 1

10
$\begingroup$

Let $\kappa:=|X|$. Then $2^{2^\kappa}$ is an obvious upper bound for the number of topologies on $X$.

Every ultrafilter on $\kappa$ will give you a Hausdorff topological space on $\kappa+1$; these are $2^{2^\kappa}$ many spaces. Some of them might be homeomorphic, but there are only $2^\kappa$ many bijections, so you get $2^{2^\kappa}$ many non-homeomorphic spaces.

Assuming that $\kappa$ is at least continuum, you can add one more point, and copies of the unit interval to each point of your original space; this will now give you a connected (even path-connected) space.

If $\kappa$ is smaller than continuum, then replace the unit interval by some countable connected Hausdorff space.

$\endgroup$
5
  • 1
    $\begingroup$ So if two spaces are not homeomorphic, after your "suspension" adding a point and lots of intervals, they are still hot homeomorphic? $\endgroup$ Apr 17, 2016 at 13:00
  • $\begingroup$ @Gerald: Indeed, if I understand the "suspension" procedure correctly, doing it with a one-point space and a two-point space will yield homeomorphic spaces. $\endgroup$ Apr 17, 2016 at 14:14
  • 1
    $\begingroup$ @GeraldEdgar That doesn't matter if you do the "up to homeomorphism" bit at the end. Goldstern's argument shows that for any infinite $X$ there is a set with the same cardinality $\kappa$ as $X$ and with $2^{2^\kappa}$ connected Hausdorff topologies. Since the number of permutations of this set is only $2^\kappa$, there must be $2^{2^\kappa}$ homeomorphism classes. $\endgroup$ Apr 17, 2016 at 14:35
  • $\begingroup$ Call the original space $A$, the "strings" (copies of the open unit interval or connected space) $B$, and let $c$ be the point where are strings are glued together. As Jeremy Rickard points out, the very many topologies on $A$ induce very many topologies on $A\cup B\cup \{c\}$, and the number of homeomorphism classes stays "very large" (i.e., $2^{2^\kappa}$). -- However, at least in the case $\kappa\ge 2^{\aleph_0}$, it is clear that each autohomeomorphism must map $B$ onto $B$ (all neighborhoods are $\simeq (0,1)$), and $c$ to $c$, hence $A$ onto $A$. $\endgroup$
    – Goldstern
    Apr 17, 2016 at 16:20
  • $\begingroup$ My answer suggests that a more interesting question might have been "how many locally connected Hausdorff spaces...". But I am pretty sure that the answer would stay the same. $\endgroup$
    – Goldstern
    Apr 17, 2016 at 16:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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