6

The image of a convergent sequence plus a limit point is compact. From Is a converging net + limit always compact? the same result does not extend to nets. That is, there are nets whose image unioned with a limit point does not form a compact set.

But what if we are allowed to take a subnet? That is, does every net admit a subnet whose image unioned with a limit point is compact? All the examples in the above link have this property.

I would guess there is an easy proof in first countable spaces but what about in general?

  • 2
    Also, just a reminder: you seem to have never accepted any of the answers to your questions, despite this question and this question having answers that seem to resolve your questions reasonably satisfactorily. You can accept an answer by clicking on the checkmark to the left of the answer. Accepting answers will also incentivize people to continue answering your questions in the future. – David Gao Feb 01 '25 at 05:23

1 Answers1

4

No. For example, let $X = \omega_1 + 1$. Then the net $K$ consisting of all countable successor ordinals is convergent, namely to $\omega_1$. But for any subnet $K’ \subset K$, $K’ \cup \{\omega_1\}$ is not compact. Indeed, if it were compact, then it would be closed, so $K’$ would be closed in $\omega_1$. $K’$, being a subnet of the unbounded $K$, must also be unbounded. But any closed, unbounded subset of $\omega_1$ contains some limit ordinals, while $K’$ does not by definition. This is a contradiction.


Let me also note that the statement is not even true for first countable spaces. The problem here is that, while in a first countable space we can certainly find a sequence of elements in the net which converges to the given limit point, there is no reason the said sequence will be a subnet of the original net. Here is an explicit first countable (in fact, even metrizable) counterexample: Let $X = \mathbb{R}^2$ equipped with the post office metric, i.e.,

$$d(x, y) = \begin{cases} 0 &, \text{ if }x = y\\ \|x\| + \|y\| &, \text{ otherwise} \end{cases}$$

It is not hard to see that compact subsets of $X$ are all countable. Indeed, if $K \subset X$ is compact, then $\{x \in K: \|x\| < \frac{1}{n}\}$ is an open subset of $K$ s.t. every point in $K$ outside of it is isolated. Whence, compactness implies only finitely many elements of $K$ can have norms at least $\frac{1}{n}$. Then, simply take the union over all $n \in \mathbb{N}$ and add the origin to see that $K$ is countable.

Now, let $S^1$ be the unit circle, equipped with a well-ordering $\preceq$ of order type continuum. Let $I = S^1 \times \mathbb{N}$, equipped with the ordering,

$$(x, n) \leq (y, m) \Leftrightarrow x \preceq y \text{ and } n \leq m$$

For each $(x, n) \in I$, let $p_{(x, n)} = \frac{1}{n}x$. Then $p_{(x, n)} \to 0$. But any subnet of $I$ is uncountable, as the cofinality of the continuum is uncountable. Since $I \ni (x, n) \mapsto p_{(x, n)} \in X$ is injective, this implies any subnet must have uncountable image, so no subnet union the limit point $0$ can be compact.


Just for the sake of completeness, let me point out that even $X = [0, 1]$ with its usual topology, which is compact and metrizable (and thus also second countable), does not satisfy the required property. Indeed, pick a subset $A$ of $[\frac{1}{2}, 1]$ of cardinality $\aleph_1$ with the property that any compact subset of $A$ is countable. (This is always possible. Indeed, any uncountable closed subset of the real line has cardinality continuum, so if the continuum hypothesis does not hold, any subset of cardinality $\aleph_1$ would work. In general, any Bernstein set in $[\frac{1}{2}, 1]$ would work.) Equip $A$ with a well-ordering $\preceq$ of order type $\omega_1$. Let $I = A \times \mathbb{N}$, equipped with the ordering,

$$(x, n) \leq (y, m) \Leftrightarrow x \preceq y \text{ and } n \leq m$$

For each $(x, n) \in I$, let $p_{(x, n)} = \frac{1}{3^n}x$. Then $p_{(x, n)} \to 0$. Any subnet of $I$ is uncountable, as the cofinality of $\omega_1$ is uncountable. Since $I \ni (x, n) \mapsto p_{(x, n)} \in X$ is injective, this implies any subnet must have uncountable image. Suppose $K$ is the image of a subnet and $K \cup \{0\}$ is compact. Then $K_n = K \cap [\frac{1}{2 \cdot 3^n}, \frac{1}{3^n}]$ is compact. As it is contained in $\frac{1}{3^n} \cdot A$, by the choice of $A$ we have $K_n$ is countable. But $K = \{0\} \cup (\bigcup_n K_n)$, so $K$ is countable, a contradiction.

David Gao
  • 22,850
  • 9
  • 28
  • 48