4

Assume all spaces are Tychonoff.

In exercise $12$H of Rings of continuous functions by Gillman and Jerison they ask to prove the following two things:

  1. If $X$ is an extremally disconnected $P$-space of cardinality smaller than first measurable cardinal, then $X$ is discrete.

  2. If a measurable cardinal exists, then there is extremally disconnected $P$-space which is not discrete.

An extremally disconnected space is a space such that for any $U\subseteq X$ its closure $\overline{U}$ is open. All the examples of such spaces are notoriously given by spaces which arise from ultrafilters in some way.

A $P$-space is a space such that if $U_n\subseteq X$ are open, then $\bigcap_n U_n$ is open. In other words, every $G_\delta$-set is open.

There exist extremally disconnected spaces which aren't $P$-spaces, $P$-spaces which aren't extremally disconnected, and both properties imply that a space is basically disconnected, that is if $U\subseteq X$ is a cozero set, then $\overline{U}$ is open. There also exist basically disconnected spaces which are not $P$-spaces and not extremally disconnected.

Existence of measurable cardinals is independent of ZFC, and they are considered to be extremely large, so that point $1.$ says that pretty much all extremally disconnected $P$-spaces are discrete. In the case they exist, point $2.$ says that certain pathological behavior occurs, so that there is a non-discrete extremally disconnected $P$-space. Both points together imply that $$\text{there exists a non-discrete extremally disconnected }P\text{-space}\iff\text{there exists a measurable cardinal.}$$

How can one show the above equivalence?

Jakobian
  • 15,280

1 Answers1

4

In this answer I prove both points $1.$ and $2.$, let's start with point $1.$ The proof of point $1.$ only uses Hausdorff assumption.

Let $X$ be an extremally disconnected space and $p\in X$ be a non-isolated point of $X$. Let $(\mathscr{S}_\alpha)_{\alpha\in A}$ be a chain, where each $\mathscr{S}_\alpha$ is a family consisting of pairwise disjoint clopen subsets of $X$ that don't contain $p$. Their union $\bigcup_{\alpha\in A}\mathscr{S}_\alpha$ consists of pairwise disjoint clopen subsets of $X$ that don't contain $p$ as well. From Zorn's lemma we can take maximal such family $\mathscr{S}$. If $x\notin \overline{\bigcup\mathscr{S}}$ and $x\neq p$, then there exists a clopen set $U\ni x$ disjoint from $\bigcup\mathscr{S}$ and such that $p\notin U$ (Hausdorff extremally disconnected space is totally separated), which is impossible from maximality of $\mathscr{S}$. It follows that $X\setminus\{p\}\subseteq \overline{\bigcup\mathscr{S}}$ and so $\overline{\bigcup\mathscr{S}} = X$ since $X\setminus\{p\}$ is not closed.

Let $\mathfrak{U} = \{\mathscr{T}\subseteq\mathscr{S} : p\in\overline{\bigcup\mathscr{T}} \}$. It follows from the extremally disconnected property of $X$ that $\mathfrak{U}$ is a filter on $\mathscr{S}$. If $\mathscr{T}\notin \mathfrak{U}$, then since $\overline{\bigcup\mathscr{T}}\cup\overline{\bigcup\left(\mathscr{S}\setminus \mathscr{T}\right)} = \overline{\bigcup\mathscr{S}}\ni p$, we have $\mathscr{S}\setminus \mathscr{T}\in \mathfrak{U}$, and so $\mathfrak{U}$ is an ultrafilter. If $U\in\mathscr{S}$ then $p\notin U$ and so $\{U\}\notin\mathfrak{U}$, from which it follows that $\mathfrak{U}$ is a free ultrafilter on $\mathscr{S}$.

Give $\mathscr{S}$ discrete topology, so that $\mathfrak{U}\in\beta \mathscr{S}\setminus\mathscr{S}$. If $|X|$ is smaller than the first measurable cardinal, then $|\mathscr{S}|\leq 2^{|X|}$ is as also, and so $\mathscr{S}$ is realcompact. It follows that there exists bounded continuous function $\varphi:\mathscr{S}\to (0, 1]$ with $\varphi^\beta(\mathfrak{U}) = 0$, where $\varphi^\beta:\beta\mathscr{S}\to [0, 1]$ is the continuous extension of $\varphi$ to $\beta\mathscr{S}$ (see $8.8$ of Rings of continuous functions by Gillman and Jerison). A neighbourhood of $\mathfrak{U}$ in $\mathscr{S}\cup \{\mathfrak{U}\}$ is of the form $\mathscr{T}\cup \{\mathfrak{U}\}$ for some $\mathscr{T}\in\mathfrak{U}$ (see $6E.6$ of Gillman and Jerison), so for $n\neq 0$ there exists $\mathscr{T}_n\in \mathfrak{U}$ with $\varphi(T) < \frac{1}{n}$ for all $T\in\mathscr{T}_n$.

Define $f:\bigcup\mathscr{S}\to\mathbb{R}$ by $f(s) = \varphi(S)$ where $s\in S\in\mathscr{S}$, then $f$ is well-defined, bounded and continuous. Since open sets in an extremally disconnected space are $C^\ast$-embedded, we can extend $f$ continuously to $X$.

Since $|f(t)| < 1/n$ for $t\in \bigcup\mathscr{T}_n$ and $p\in\overline{\bigcup\mathscr{T}_n}$, it follows that $|f(p)|\leq 1/n$ for all $n$, and so $f(p) = 0$. But the $G_\delta$-set $\{x\in X : f(x) = 0\}$ does not intersect $\bigcup\mathscr{S}$, and so is not open. So $X$ is not a $P$-space.


Now let's prove point $2.$

Suppose that $X$ is a discrete space with $|X|$ larger or equal to the first measurable cardinal. Then $X$ is not realcompact, so its Hewitt realcompactification $\nu X$ is not equal to $X$, and so is not discrete. Since $X$ is a $P$-space, $\nu X$ is a $P$-space as well, and because $X$ is extremally disconnected, so is $\beta X$. Since $\nu X$ is a dense subspace of $\beta X$, $\nu X$ is extremally disconnected.

It follows that $\nu X$ is a non-discrete extremally disconnected $P$-space.

Jakobian
  • 15,280
  • 2
    Here is an easier argument for the second half of point $1$: We show directly that, if $X$ is a $P$-space, then $\mathfrak{U}$ is $\sigma$-complete. Indeed, if it were not $\sigma$-complete, then there exists a sequence $\mathscr{T}_n\in\mathfrak{U}$ s.t. $\bigcap_n\mathscr{T}_n=\varnothing$. As $X$ is an extremally connected $P$-space, $p$ belongs to the open set $O=\bigcap_n\overline{\bigcup\mathscr{T}_n}$. Thus, $O$ is nonempty and, as $\overline{\bigcup\mathscr{S}}=X$, there must exist a clopen $U\in\mathscr{S}$ s.t. $O\cap U\neq\varnothing$. But this implies $U\in\mathscr{T}_n$ for… – David Gao Jan 13 '25 at 04:46
  • 2
    … all $n$. Indeed, otherwise $U\cap(\bigcup\mathscr{T}_n)=\varnothing$, so as $U$ is open, $U\cap O\subset U\cap(\overline{\bigcup\mathscr{T}_n})=\varnothing$ as well, a contradiction. But then $U\in\mathscr{T}_n$, contradicting $\bigcap_n\mathscr{T}_n=\varnothing$. This implies $\mathfrak{U}$ is $\sigma$-complete, so $\mathscr{S}$ and thus $X$ must have cardinality at least that of the first measurable cardinal. – David Gao Jan 13 '25 at 04:53
  • @DavidGao I have asked a question about what separation properties does $\nu X$ have here if you have any idea. – Jakobian Jan 14 '25 at 21:19
  • @Jakobian In the first paragraph of part 1, given $x\notin\overline{\bigcup\mathscr{S}}$, to find a clopen nbhd $U$ of $x$ with the stated properties, that relies on $X$ being zero-dimensional, which follows because $X$ assumed regular, right? – PatrickR Jan 15 '25 at 03:39
  • 1
    @PatrickR Hausdorff is enough. You don't need zero-dimensional, just totally separated – Jakobian Jan 15 '25 at 03:59
  • @Jakobian I see it now. $\overline{\bigcup\mathscr{S}}$ is clopen, and so is its complement. And totally separated is used to get a clopen nbhd of $x$ avoiding $p$. Then their intersection works. Thanks. – PatrickR Jan 15 '25 at 04:05