Let $X$ be a separable compact Hausdorff topological space. Then every closed subset of $X$ is a $G_{\delta}$ set (i.e. $X$ is perfectly normal).

I've been skimming through some topology textbooks recently. As we know, every closed subset of a metric space is a $G_{\delta}$ set. But I am not sure whether the above proposition is correct because I did not find any counterexample. Can anyone give a proof or counterexample? Thanks.


Let $X = \{0,1\}^{\mathbb{R}}$. Since $\{0,1\}$ is separable and $\operatorname{card} (\mathbb{R}) \leqslant 2^{\aleph_0}$, $X$ is separable. Let

$$F = \{0\} = \{ x \in X : t \in \mathbb{R} \implies x(t) = 0\}.$$

$F$ is a closed subset of $X$ that is not a $G_{\delta}$-set. For if $U$ is an open subset of $X$ containing $F$, then there is a finite subset $I = I(U)$ of $\mathbb{R}$ such that

$$V_I = \{ x \in X : t \in I \implies x(t) = 0\} \subset U.$$

Then if we have a countable family $\{ U_n : n \in \mathbb{N}\}$ of open subsets containing $F$, the union

$$K := \bigcup_{n\in \mathbb{N}} I(U_n)$$

is a countable, thus proper, subset of $\mathbb{R}$, and hence

$$\bigcap_{n\in \mathbb{N}} U_n \supset \bigcap_{n \in \mathbb{N}} V_{I(U_n)} = V_K \supsetneqq F.$$

  • $\begingroup$ Why if $U$ is an open subset of $X$ containing $F$, then there is a finite subset $I = I(U)$ of $\mathbb{R}$ such that $$V_I = \{ x \in X : t \in I \implies x(t) = 0\} \subset U?$$ $\endgroup$ – Jay Apr 2 '16 at 13:47
  • $\begingroup$ I've simplified the example, it was needlessly complicated, and now the existence of such a $V_I \subset U$ follows easily from the definition of the product topology. I'm not sure it actually did in the original formulation. (Though the original $F$ was not a $G_\delta$, the proof may have needed to be different.) $\endgroup$ – Daniel Fischer Apr 2 '16 at 13:49
  • $\begingroup$ @JacobZhang Not sure whether you have seen the updated answer. In the updated version, note that a base of the product topology is formed by "open rectangles" $R = \prod_{t\in \mathbb{R}} W_t$, where $\bigl\{ t\in \mathbb{R} : W_t \neq \{0,1\}\bigr\}$ is finite, and $W_t \neq \varnothing$ for all $t$, by definition of the product topology. For such a rectangle, we have $0 \in R$ if and only if there is a finite $I$ with $R = V_I$. $\endgroup$ – Daniel Fischer Apr 2 '16 at 13:53
  • $\begingroup$ I have seen the undated answer. Your construction is skillful. Thanks. $\endgroup$ – Jay Apr 2 '16 at 13:57

Although Henno’s general result renders specific examples a bit superfluous, I’ll note that $\beta\Bbb\omega$ is another relatively easy example. It is certainly compact, Hausdorff, and separable. The ultrafilter construction of $\beta\Bbb\omega$ makes it obvious that it has a base of cardinality $2^\omega=\mathfrak{c}$: if for each $A\subseteq\omega$ we set

$$\widehat A=\{p\in\beta\omega:A\in p\}\;,$$

then $\mathscr{B}=\left\{\widehat A:A\subseteq\omega\right\}$ is a base for $\beta\omega$.

If $p\in\beta\omega$, and $\{p\}$ is a $G_\delta$, there is a countable $\mathscr{B}_p\subseteq\mathscr{B}$ such that $\bigcap\mathscr{B}_p=\{p\}$. $\mathscr{B}$ has only $\left(2^\omega\right)^\omega=2^\omega$ countable subsets, so there can be at most $2^\omega$ points $p\in\beta\omega$ such that $\{p\}$ is a $G_\delta$. But $|\beta\omega|=2^{\mathfrak{c}}$, so most singletons in $\beta\omega$ are closed sets that are not $G_\delta$ sets.


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.