Now cross-posted to Mathoverflow.

A topological space is a called anticompact if all its compact subsets are finite.

Question 1: Is there a (non-trivial) connected $T_1$ anticompact space?

Obviously the singleton and the empty space satisfy these properties. So let's assume such a space $X$ has at least two points. Here are some observations:

(1) $X$ must be infinite. Because finite $T_1$ spaces are discrete, such a space would not be connected.

(2) $X$ cannot be compact.

(3) $X$ must be totally path disconnected, because anticompact $T_1$ spaces are totally path disconnected. (The image of a continuous map $[0,1]\to X$ is compact, connected and $T_1$, hence finite and $T_1$, so it would be discrete, and connected. Thus it must a single point.)

(4) $X$ cannot be sequential, as a consequence of this. In particular, $X$ cannot be first countable.

Answer (from Ulli): Yes. Any uncountable set with the cocountable topology is connected and anticompact. It is also $T_1$, but not $T_2$.

So updating the question:

Question 2: Is there a (non-trivial) connected Hausdorff anticompact space?

I think the following would be an example. Take $X=[0,1]$. Let $\tau_1$ be the usual Euclidean topology on $X$ and take on $X$ the topology $\tau$ generated by $\tau_1$ and the cocountable topology. By using the Lindelof property for $\mathbb{R}$ it's not too difficult to show that the topology is described by $$\tau=\{U\setminus A: U\in\tau_1 \text{ and } A\subseteq X \text { is countable}\}.$$

Every countable subset of $X$ is closed, and is discrete since all its subsets are also countable, hence closed. So if $A$ is an infinite subset of $X$, it cannot be compact; otherwise a countably infinite subset of it would also be compact (as closed in a compact set) and discrete, which is impossible. This shows that $X$ is anticompact.

$X$ is Hausdorff.

To show that $X$ is connected, suppose $X=Y\cup Z$ with $Y$ and $Z$ nonempty disjoint open sets. Take a point $x\in Y$. There is a Euclidean nbhd $U$ of $x$ such that $Y$ contains all the points of $U$ except countably many of them (where countable = finite or countably infinite). Any point $z\in U\setminus Y$ would have to be contained in $Z$. And since $Z$ is open in $X$, $U\cap Z$ would be a countable nbhd of $z$, which is impossible. This shows that both $Y$ and $Z$ are open in the Euclidean topology. Again that is impossible since $[0,1]$ is connected in its usual topology.

I would appreciate if you could confirm that the example above is correct.

This leads me to the revised question:

Question 3: Is there a countably infinite connected $T_1$ (or $T_2$) anticompact space?

I can't find an example. The Arens-Fort space or the Appert space are anticompact (and not first countable) but not connected. Countably infinite connected examples that I know of, like the irrational slope topology or the Golomb space, are first countable, hence not anticompact.

  • $\begingroup$ Your uncountable example $X$ is NOT Hausdorff but it IS $T_1.$ $\endgroup$ Commented Nov 20, 2022 at 6:29
  • $\begingroup$ Just taking the cocountable topology on $\mathbb{R}$ (or $\omega_1$) is also anticompact, connected and T1 (but not Hausdorff). $\endgroup$
    – Ulli
    Commented Nov 20, 2022 at 6:34
  • $\begingroup$ @DanielWainfleet: the example of the OP is finer than the euclidean topology, hence it is Hausdorff. Just taking the cocountable topology wouldn't be Hausdorff. $\endgroup$
    – Ulli
    Commented Nov 20, 2022 at 6:38
  • $\begingroup$ @Ulli I kind of knew that, but thanks for reminding me. I guess the nice thing is it's a Hausdorff example (which should have been a better original question). I'll update the post. So the only thing not obvious is to find a countably infinite example (T_1 or T_2). $\endgroup$
    – PatrickR
    Commented Nov 20, 2022 at 7:27
  • 1
    $\begingroup$ According to space 63, item 9 in Counterexamples your X in Question 2 is connected. I think your proof for anticompact is also correct (item 7 in Counterexamples is exactly this as well, though they didn't track "anticompact") and will contribute it to pi-Base. $\endgroup$ Commented Nov 22, 2022 at 17:36

2 Answers 2


Here is a partial answer to question 3, namely an example of a connected, anticompact T1-topology on the (positive) integers:

Let $\mathfrak{F}$ be a free ultrafilter on $\mathbb{N}$ and $\tau := \mathfrak{F} \cup \{\emptyset\}$ the ultrafilter topology on $\mathbb{N}$. Since $\mathfrak{F}$ is free, for each $n \in \mathbb{N}$, $\mathbb{N} \setminus \{n\} \in \mathfrak{F}$, hence $\{n\}$ is closed. Thus, the topology is T1.

The intersection of any two non-empty, open sets is non-empty again. Hence the topology is connected.

It is anticompact:
Let $A$ be an infinite subset of $\mathbb{N}$. Then there exists $U \in \mathfrak{F}$ such that $A \setminus U$ is infinite.
[$A$ is the disjoint union of two infinite subsets $S, T$. If $S \in \mathfrak{F}$ or $T \in \mathfrak{F}$, we are done. Otherwise, $\mathbb{N} \setminus S \in \mathfrak{F}$ and $\mathbb{N} \setminus T \in \mathfrak{F}$. Hence $\mathbb{N} \setminus A = \mathbb{N} \setminus (S \cup T) = (\mathbb{N} \setminus S) \cap (\mathbb{N} \setminus S) \in \mathfrak{F}$.]
Now, $A \setminus U$ is discrete.
[For any $x \in A \setminus U$, $\{x\} = (U \cup \{x\}) \cap (A \setminus U)$ is open in $A \setminus U$.]
Furthermore, $A \setminus U \notin \mathfrak{F}$. Hence its complement is open, hence $A \setminus U$ is closed in $\mathbb{N}$ and thus in $A$. Therefore $A$ cannot be compact.

Note: Of course, this topology is far from being Hausdorff.

  • $\begingroup$ Very nice example for the $T_1$ case. $\endgroup$
    – PatrickR
    Commented Nov 22, 2022 at 7:43
  • $\begingroup$ Talking about ultrafilters, there is the Single ultrafilter topology = example #114 in "Counterexamples in Topology". It is countable and $T_2$ and it seems the same arguments that you used can show it is anticompact. But it is far from connected, as it's zero-dimensional. $\endgroup$
    – PatrickR
    Commented Nov 22, 2022 at 7:55
  • $\begingroup$ Gustin's sequence space seems to be a candidate: it's countable, T2, and connected. Anti-compact is open in the pi-Base. EDIT: or not, it's first-countable, that's just not tracked. $\endgroup$ Commented Nov 23, 2022 at 4:43

As answered in the cross-post to mathoverflow, a paper by T. Banakh and Y. Stelmakh provides examples of countably infinite connected Hausdorff anticompact spaces. Their examples are even strongly rigid.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .