Let us recall that a symmetric on a set $X$ is any function $d:X\times X\to[0,\infty)$ such that for every $x,y\in X$ the following two conditions are satisfied:

$\bullet$ $d(x,y)=0$ if and only if $x=y$;

$\bullet$ $d(x,y)=d(y,x)$.

A topological space $X$ is called symmetrizable if there exists a symmetric $d$ on $X$ such that a subset $U\subseteq X$ is open if and only if for every $x\in X$ there exists $\varepsilon>0$ such that $B_d(x,y)\subseteq U$, where $B_d(x,\varepsilon)=\{y\in X:d(x,y)<\varepsilon\}$.

By the Urysohn Metrization Theorem, every regular second-countable space is metrizable and hence symmetrizable.

Question 1. Is each Hausdorff second-countable space symmetrizable?

A weaker version of Question 1 is also of interest:

Question 2. Is each countable first-countable Hausdorff space symmetrizable?

Added in Edit. I have just realized that Question 2 has a simple affirmative answer (which I present below), so only Question 1 remains open.

Added in Edit 31.05.2022. Ups: Question 1 has a negative answer!

  • 2
    $\begingroup$ "A symmetric"? this sounds (to me — not native) as phoney English. $\endgroup$
    – YCor
    May 30, 2022 at 17:55
  • 3
    $\begingroup$ @YCor Nonetheless this is a standard terminology, see e.g. Section 9 of Gruenhages' survey paper "Generalized metric spaces" in Handbook of Set-Theoretic Topology. $\endgroup$ May 30, 2022 at 18:01
  • 2
    $\begingroup$ Arhangelskii's 1966 paper Mappings and spaces (this is the Russian Mathematical Surveys translation) might be useful. Gruenhage has a more recent survey, although I suppose you know of it. FYI, I know very little about this topic. I simply looked up the term in this book (p. 237, right column), where I saw the Arhangelskii reference, which I surprisingly have a photocopy of, so I very briefly glanced at it. $\endgroup$ May 30, 2022 at 19:40
  • $\begingroup$ @DaveLRenfro You are right. A more general result was proved by Akhangelski in his paper in Russian Math Surveys: according to his Theorem 2.9, every first-countable $T_1$-space with $\sigma$-discrete network is symmetrizable. Thank you for this reference to Arhangelski. I had to look at it before asking this question. $\endgroup$ May 30, 2022 at 19:59
  • 1
    $\begingroup$ @Ycor Perhaps it's a pun on the word "metric", so one should read it as "a sym-metric"? $\endgroup$ Jun 1, 2022 at 8:38

1 Answer 1


I have just realized that the first question has a simple affirmative answer.

Theorem 1. Every countable first-countable $T_1$ space $X$ is symmetrizable.

Proof. For every $x\in X$, fix a neighborhood base $(U_n(x))_{n\in\omega}$ such that $U_{n+1}(x)\subseteq U_n(x)$ for all $n\in\omega$. Since $X$ is countable, there exists a linear order $\le$ on $X$ such that for every $x\in X$ the initial interval $\{y\in X:y\le x\}$ is finite.

It is easy to see that the topology of $X$ is generated by the symmetric $$d(x,y)=\inf\big\{2^{-n}:\max\{x,y\}\in U_n(\min\{x,y\})\big\}.\quad\square$$

Important Remark. After looking at a reference Dave L. Renfro suggested in a comment as possibly being relevant, I discovered that affirmative answers to both my questions follow from

Theorem 2.9 (Arhangelski, 1966): Every first-countable $T_1$ space with a $\sigma$-discrete (closed) network is symmetrizable.

But reading the proof of this theorem I discovered that it works only for regular spaces. Arhangelski writes that one loses no generality assuming that the $\sigma$-discrete network is closed, i.e., consists of closed sets. But after taking the closure of elements of a network in a $T_1$-space (even in a Hausdorff space), the network property can be destroyed. So, Question 2 seems to stay open and not answered even for Hausdorff (not mention $T_1$) spaces.

Added in Edit 31.05.2022. To my big surprise (and contrary to what was claimed by Arhangelski in his Theorem 2.9), I have just discovered that Question 1 has negative answer!

First let us prove that the symmetrizability of second-countable spaces is equivalent to the perfectness. Recall that a topological space $X$ is perfect if every closed subset of $X$ is of type $G_\delta$.

Theorem 2. A second-countable (Hausdorff) $T_1$-space $X$ is symmetrizable if (and only if) it is perfect.

Proof. If $X$ is symmetrizable and Hausdorff, then the first-countability of $X$ implies that $X$ is semi-metrizable and perfect, see the paragraph before Theorem 9.8 in Gruenhage's "Generalized metric spaces".

Conversely, if a second-countable $T_1$-space $X$ is perfect, then each open subset of $X$ is an $F_\sigma$ in $X$, which implies that $X$ has a countable closed network. Now we can apply Arhangelski's Theorem 2.9 to conclude that $X$ is symmetrizable. $\square$

Let $\mathrm{non}(\mathcal M)$ denote the smallest cardinality of a nonmeager set in the real line.

Example 1. There exists a second-countable Hausdorff space of cardinality $\mathrm{non}(\mathcal M)$ which is not perfect and hence not symmetrizable.

Proof. Take any nonmeager linear subspace $L$ of cardinality $\mathrm{non}(\mathcal M)$ in $\mathbb R^\omega$ such that for every $n\in\omega$ the intersection $L_n=L\cap(\{0\}^n\times\mathbb R^{\omega\setminus n})$ is dense in $\{0\}^n\times\mathbb R^{\omega\setminus n}$. Consider the quotient space $X=L_\circ/_\sim$ of $L_\circ=L\setminus\{0\}$ by the equivalence relation $\sim$ defined by $x\sim y$ iff $\mathbb R x=\mathbb Ry$. Since the space $L_\circ$ is Baire and the quotient map $q:L_\circ\to X$ is open, the space $X$ is second-countable and Baire. It is easy to check that the closure of every nonempty set in $X$ contains the set $q[L_n\setminus\{0\}]$ for some $n\in\omega$. This implies that the space $X$ is superconnected in the sense that for every nonempty open sets $U_1,\dots,U_n$ in $X$ the intersection of their closures $\overline U_1\cap\dots\cap\overline U_n$ is not empty.

Now take any disjoint nonempty open sets $U,V$ in $X$. Assuming that $V$ is of type $F_\sigma$, we can apply the Baire Theorem and find a nonempty open set $W\subseteq V$ whose closure in $X$ is contained in $V$. Then $\overline{U}\cap\overline{W}=\emptyset$, which contradicts the superconnectedness of $X$. $\square$

The cardinality $\mathrm{non}(\mathcal M)$ is the above example can be lowered to $\mathfrak q_0$, where $\mathfrak q_0$ is the smallest cardinality of a second-countable metrizable space which is not a $Q$-space (= contains a subset which is not of type $G_\delta$).

A topological space is submetrizable it it admits a continuous metric. Each submetrizable space is functionally Hausdorff in the sense that for any distinct elements $x,y\in X$ there exists a continuous function $f:X\to\mathbb R$ such that $f(x)\ne f(y)$.

Example 2. There exists a submetrizable second-countable space $X$ of cardinality $\mathfrak q_0$, which is not symmetrizable.

Proof. By the definition of the cardinal $\mathfrak q_0$, there exists a second-countable metrizable space $Y$, which is not a $Q$-space and hence contains a subset $A$ which is not of type $G_\delta$ in $X$. Let $\tau'$ be the topology on $X$, generated by the subbase $\tau\cup\{X\setminus A\}$ where $\tau$ is the topology of the metrizable space $Y$. It is clear that $X=(Y,\tau')$ is a second-countable space containing $A$ as a closed subset. Since $\tau\subseteq\tau'$, the space $X$ is submetrizable. Assuming that $X$ is symmetrizable and applying Theorem 2, we conclude that $X$ is perfect and hence the closed set $A$ is equal to the intersection $\bigcap_{n\in\omega}W_n$ of some open sets $W_n\in\tau'$. By the choice of the topology $\tau'$, for every $n\in\omega$ there exists open sets $U_n,V_n\in \tau$ such that $W_n=U_n\cup(V_n\setminus A)$. It follows from $A\subseteq W_n=U_n\cup(V_n\setminus A)$ that $A=A\cap W_n=A\cap U_n\subseteq U_n$. $$A=\bigcap_{n\in\omega}W_n=A\cap\bigcap_{n\in\omega}W_n=\bigcap_{n\in\omega}(A\cap W_n)=\bigcap_{n\in\omega}(A\cap U_n)\subseteq \bigcap_{n\in\omega}U_n\subseteq \bigcap_{n\in\omega}W_n=A$$ and hence $A=\bigcap_{n\in\omega}U_n$ is a $G_\delta$-set in $X$, which contradicts the choice of $A$. This contradiction shows that the submetrizable second-countable space $X$ is not symmetrizable. $\square$

On the other hand we have the following partial affirmative answer to Question 1.

Theorem 3. Martin's Axiom implies that every second-countable $T_1$ space of cardinality $<\mathfrak c$ is perfect and hence symmetrizable.

Proof. It is known that Martin's Axiom implies that every second-countable $T_1$-space $X$ of cardinality $\mathfrak c$ is a $Q$-space, which means that every subset of $X$ is of type $G_\delta$. In particular, $X$ is perfect and by Theorem 2 is symmetrizable. $\square$

  • 2
    $\begingroup$ For the record, I didn't know that either or both theorems were consequences of a more general result in Arhangelski's paper -- I didn't look at the paper much, other than to see that the symmetrizable property appears non-trivially in the paper. More accurate would be to say something like: "After looking at a reference Dave L. Renfro suggested in a comment as possibly being relevant, I discovered that both theorems follow from the known ..." $\endgroup$ May 30, 2022 at 21:02
  • $\begingroup$ @DaveLRenfro Thank you for the suggestion of editing my answer. What is surprising that reading the proof of Arhangelski I found that it works only for Hausdorff spaces. So, the case of $T_1$ spaces seems to be open. $\endgroup$ May 30, 2022 at 21:53
  • $\begingroup$ Correct me if I'm wrong, but symmetrisability and semimetrisability are equivalent in Fréchet-Urysohn spaces, while every semimetrisable space with a point-countable base is developable (I don't thing anything here requires regularity, but I didn't check). B. Scott gives here an example of a second-countable $T_2$ space which is not developable. By the above, it cannot be symmetrisable. $\endgroup$
    – Tyrone
    May 31, 2022 at 12:50
  • $\begingroup$ @Tyrone Thank you for your comment. In fact, besides your or my example there are even more simple examples of submetrizable second-countable submetizable spaces which are not symmetrizable. Now I will add a suitable example to my answer. $\endgroup$ May 31, 2022 at 13:02

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.