Is every (second-countable) topological manifold completely metrizable?

It is known that every smooth manifold possess a complete Riemannian metric, hence in particular it is completely metrizable, however there are non smoothable manifolds.


2 Answers 2



The proof I know is a little roundabout. Let $M$ be your manifold. It is locally compact and Hausdorff, so it has a one-point compactification $M^*$ which is compact Hausdorff. Now $M^*$ is again second countable (see One point compactification is second contable), and (locally) compact Hausdorff spaces are regular, so by the Urysohn metrization theorem, $M^*$ is metrizable with some metric $d^*$. Since $M^*$ is compact, then $(M^*, d^*)$ is of course complete. Now $M$ is an open subset of $M^*$, and every open (or even $G_\delta$) subset of a complete metric space is completely metrizable (with a different metric). See Theorem 1.2 of this note for a proof; it's also in Kechris's Classical Descriptive Set Theory and probably many other standard texts.

In fact, unless I am mistaken, we just showed any locally compact Hausdorff second countable space is completely metrizable.

If there is a more direct proof, I would be interested to see it!

  • 3
    $\begingroup$ A more direct proof might show that every manifold has a proper embedding into Euclidean space. For $n \neq 4$-manifolds, every topological manifold has a handlebody structure, so you should be able to use this to construct a proper embedding by hand. Or one could use dimension theory to show that $M^*$ embeds into a Euclidean space, then enbed it into $S^N$ such that $\infty$ maps to $\infty$; then this gives a proper embedding into $\Bbb R^N$. $\endgroup$
    – user98602
    Aug 5, 2015 at 16:43
  • $\begingroup$ @MikeMiller: This question is relevant - it seems that this is possible but difficult. It also only solves part of the question - not every subset of $\mathbb{R}^n$ is completely metrizable, so we would have to verify that the image of the embedding is $G_\delta$ and then apply the result I mention above. $\endgroup$ Aug 10, 2015 at 18:17
  • 1
    $\begingroup$ I was suggesting a closed embedding. If you can embed the one-point compactification into $S^n$ with the point at infinity mapping to $\infty$, then this restricts to a closed embedding of the manifold into $\Bbb R^n$.. Then the restriction of the Euclidean metric provides a complete metric. $\endgroup$
    – user98602
    Aug 10, 2015 at 18:19
  • $\begingroup$ @MikeMiller: Hmm. Is it a problem that the one-point compactification of a topological manifold need not be a topological manifold? Or is this covered by the "dimension theory" argument (I don't know much about dimension theory)? $\endgroup$ Aug 10, 2015 at 18:26
  • $\begingroup$ I have no idea! I, too, am a dimension theory ignoramus. Probably it will work? $\endgroup$
    – user98602
    Aug 10, 2015 at 18:29

Express $M$ as a union of open sets $\{U_n\}$ with $clos(U_n)$ compact and contained within $U_{n+1}$ (always possible if $M$ is not itself compact).

Put a Riemannian metric $g_0$ on $M$. Revise $g_0$ to $g_1$ as follows: $g_1 = f_1 g_0$ for $f_1$ a positive scalar function obeying:
(a) $f_1 = 1$ on $U_1$ and
(b) on $U_3 - U_2$, $f_1 < 1/\sqrt{d_0(U_2, M - U_3)}$ (where $d_0$ = distance function generated by $g_0$).

Note that for $d_1$ the distance function generated by $g_1$, $d_1(U_2, M - U_3)$ is at least $1$; in particular, any curve starting in $U_1$ and escaping to infinity must have at least length $1$, because it crosses from the boundary of $U_2$ to the boundary of $U_3$, a distance of at least $1$.

Now continue inductively: treating $U_3$, $U_4$, and $U_5$ as we just did $U_1$, $U_2$, and $U_3$ and creating $g_2 = f_2 g_1$, with any curve from $U_1$ escaping to infinity having length at least $2$, as it crosses both from boundary of $U_2$ to $U_3$ (same metric as $g_1$) as well as from boundary of $U_4$ to boundary of $U_5$ (again with distance at least $1$); and so on.

We end with a well-defined metric $g_{\infty} = \lim g_n$. Any curve escaping to infinity must cross an infinite number of bands of width at least $1$ in $g_{\infty}$, so it must have infinite length in $g_{\infty}$. That means $(M, g_{\infty})$ is complete.

  • $\begingroup$ Note: $M$ here is only assumed to be a topological manifold. Thus, you cannot put a Riemannian metric on it (is n general). Also, OP explicitly stated that he knows a proof in the smooth setting. $\endgroup$ Feb 6 at 19:19

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.