6

I just read a proof that adds the assumption of T1 to conclude that a first countable, limit compact and T1 space must be sequentially compact, but I didn't understand what happens if we drop the T1 assumption.

If $X$ is limit point compact and first countable, and $\{a_n\}$ is a sequence in $X$, then it has a limit point $x$.

By first countability, we can find a basis of neighborhoods of $x$ and then build a subsequence that converges to $x$. Isn't that enough?

Reference: The proof is from relationship among different kinds of compactness by rm50 (Theorem 2). As you can see, $q$ appears from nowhere.

hardmath
  • 37,015
Rodrigo
  • 7,646
  • Pobably to avoid confusion in the definition of sequential compactness. If a space is not $T_1$ then a single sequence may have multiple limits. – Justpassingby Dec 30 '15 at 18:19
  • Thought the same thing, but by definitions a converging sequence doesn't need to converge uniquely. – Rodrigo Dec 30 '15 at 18:22
  • This post came to my attention because of the broken link to PlanetMath.org. Although the original PDF no longer appears (even in the Wayback archive), thanks to your succinct description of what was to be found there I located this topic page at PlanetMath. Please have a look and judge whether you find it to be a suitable replacement link. Indeed $q$ does "appear from nowhere". – hardmath Jan 21 '22 at 17:52

1 Answers1

8

Consider the right topology on $\mathbb{R}$: the topology generated by the sets of the form $$(a,\infty) = \{ x \in \mathbb{R} : x > a \}$$ for $a \in \mathbb{R}$.

  • It is clearly first- (even second-) countable. (Consider only rational $a$ for the basic open sets.)
  • If $ A\subseteq \mathbb {R} $ is nonempty, then any $x \in \mathbb{R}$ which is (strictly) less than an element of $A$ is a limit point of $A$, hence it is limit point compact.
  • The sequence $\langle -n \rangle_{n \in \mathbb {N}}$ has no convergent subsequence, hence it is not sequentially compact.

Note that in T1-spaces, to be a limit point of a set every neighbourhood must contain infinitely many points of that set. This is what allows for the construction of the subsequence.

If $\langle x_n \rangle_{n \in \mathbb {N}}$ is a sequence in $X$, then either it has a constant subsequence (which obviously converges), or the set $A= \{ x_n : n \in \mathbb {N} \}$ is infinite, and hence has a limit point, $ x $.

Fix a decreasing countable base $\{ V_k \}_{k \in \mathbb{N}}$ for $x$.

Recursively pick $n_k \in \mathbb{N}$ ($k \in \mathbb{N}$) so that $x_{n_k} \in V_k$ and $x_{n_k} > x_{n_{k-1}}$. This can be done because $V_k \cap A$ is infinite, and thus so is the set $\{ n \in \mathbb {N} : x_n \in V_k\}$. (Or, to go a bit more slowly, since $X$ is T1 finite sets are closed, and so $V_k \setminus \{ x_n : n \leq n_{k-1}, x_n \neq x \}$ is an open neighbourhood of $x$, and so contains an element $z$ of $A$ distinct from $x$. Since $z\in A$ there must be an $n$ such that $z=x_n$, and by choice of neighbourhood of $x$ it must be that $n > n_{k-1}$, so set $n_k = n$.)

The subsequence $\langle x_{n_k} \rangle_{k \in \mathbb {N}}$ then converges to $x$.

However for non-T1-spaces a neighbourhood of a limit point may only contain a single point of the set, which makes the construction of a subsequence as in the third paragraph of the above demonstration problematic.

In the above space, the only open set which contains infinitely many points of the sequence $\langle -n \rangle_{n \in \mathbb {N}}$ is the entire space $\mathbb {R}$, although $\overline { \{ -n : n \in \mathbb {N}\}} = (-\infty,-1]$.

user642796
  • 52,188
  • What an horrible topology! That's a great counter example. – Zach Stone Dec 30 '15 at 20:11
  • @ZachStone This topology isn't too bad. It's sort of "half" the usual topology on $\mathbb {R} $, and can be used to characterise "lower semicontinuous functions". – user642796 Dec 30 '15 at 20:39