I'm studying some things related to ultrafilters on metric and topological spaces and trying to prove theorem in a general setting, so the following question came to my mind.

Let $S$ be a topological space. Does there exist a natural directed set $(M,\leq)$ for which the nets $\left\{x_\lambda\right\}_{\lambda\in M}$ describe the topology of $S$?

By "natural", I mean that the directed set associated to a metric, or first countable, space is $\mathbb{N}$, for example. If we let $\mathcal{U}_x$ denote the collection of open neighbourhoods of a point $x\in S$ ordered by reverse inclusion ($U\leq V\iff V\subseteq U$), then $M=\prod_{x\in S}\mathcal{U}_x$ with the product order is sufficient to describe the topology of $S$ with nets (I believe), but it is not exactly "natural".

Most of the proofs I know of theorems which describe the topology of $S$ using nets usually consider, at some point, a product directed set. When we are dealing with first countable spaces, we usually consider not the product directed set $\mathbb{N}\times\mathbb{N}$, but the diagonal $\left\{(n,n):n\in\mathbb{N}\right\}$, which is in fact cofinal in $\mathbb{N}\times\mathbb{N}$ (see here, for example).

  • $\begingroup$ Did you realize that the space itself is much much more natural in some sense than the natural numbers - the natural numbers have nothing to do with the space a priori. It is just a bad incident that these occur so natural to us as they provide a simple example of a directed set. But apart from a simple example that's it. $\endgroup$ – C-Star-W-Star Dec 20 '14 at 17:24
  • $\begingroup$ @Freeze_S Indeed, I'm starting to realize that the natural numbers are probably too good to serve as examples. I found a 1978 paper by Saks which characterizes compact spaces as the spaces $S$ for which all nets indexed by cardinals $\mathfrak{m}\leq|S|$ converge relatively to all/some tail ultrafilter (using the generalized notion of convergence along an ultrafilter). I was hoping that the character of the space would have a bigger role when dealing to convergence, but apparently that's not the case. Thank you for your comment. $\endgroup$ – Questioner Dec 22 '14 at 18:04

Note that nets can be seen as special filters.

Hausdorff's approach to topology uses neighborhood filters: $\mathcal{N}_x$

Most commonly these are constructed by neighborhood bases: $\mathcal{B}_x$

So one has: $\mathcal{N}_x:=\uparrow\mathcal{B}_x:=\{N:\exists B\in\mathcal{B}_x:B\subseteq N\}$

Now, the topology of metric spaces is easily defined in this way: $B_\varepsilon(x)$

(Precisely, it is defined via a uniform structere constructed via a basis.)


Let's consider the space $(ω_1 + 1)$, i.e. the transfinite convergent sequence of length $ω_1$. What would your natural directed set be here? You need both $ω$ and $ω_1$ types. So I don't see anything better than $ω × ω_1$ here.

  • 1
    $\begingroup$ I understand your point. We'd need $\omega$ to take care of nets converging to points before $\omega_1$ and $\omega_1$ to take care of $\omega_1$ itself. $\endgroup$ – Questioner Dec 20 '14 at 15:28

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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