Is this proof correct?

Claim: if $(M,d)$ is a separable metric space, then so is any subset $U\subseteq M$


  • let $C=(c_k)_{k\in \mathbb{N}}$ be a countable dense subset of $M$. Fix $n \in \mathbb{N}$. Consider $B_{1/n}(c_k)$ (open ball around $c_k$ of radius $1/n$). If $B_{1/n}(c_k) \cap U$ is non empty, then add to the set $U_n$ a point $u_k^{n}$ in $B_{1/n}(c_k) \cap U$
  • every $U_n$ is non empty by density of $C$, and being every $U_n$ at most countable, so is $U:=\bigcup_n U_n$, which I now claim to be dense in $U$
  • let $u \in U$. Find $c_l\rightarrow_l u$ such that $d(c_l,u)< 1/l$. Thus $B_{1/l}(c_l)\cap U \neq \emptyset$, and by construction there is $u_l^l \in U_l\subseteq U$ with $d(u_l^l,c_l)<1/l$
  • by triangle inequality: $d(u_l^l,u)<1/(2l)$, so that $u_l^l\rightarrow_l u$
  • $\begingroup$ In this answer I show that a separable metric space has a countable base. Then $U$ also has a countable base and we pick a point in each member of a countable base to get a dense subset again. This is in essence your argument too. $\endgroup$ – Henno Brandsma Feb 17 at 22:32
  • $\begingroup$ The proof would be smoother if your definition of dense were not based off sequences, but just used: $D$ is dense iff every ball $B(x,r)$ intersects $D$. $\endgroup$ – Henno Brandsma Feb 17 at 22:42

While you said enough that I can guess with some confidence what you intended, you never really defined the sets $U_n$. Here’s what I think you had in mind:

Fix an arbitrary point $u\in U$. For all $n,k\in\Bbb Z^+$, if $U\cap B_{1/n}(c_k)\ne\varnothing$ let $u_k^n\in U\cap B_{1/n}(c_k)$, and otherwise let $u_k^n=u$. Then for each $n\in\Bbb Z^+$ let $U_n=\{u_k^n:k\in\Bbb Z^+\}$.

(The only reason for $u$ is to allow me to define $u_k^n$ for all $n,k\in\Bbb Z^+$; this makes the later definition of $U_n$ a little simpler.)

It is clear that each $U_n$ is at most countable, and my small trick wth $u$ ensures that each $U_n$ is non-empty. If you define $u_k^n$ only when $U\cap B_{1/n}(c_k)\ne\varnothing$, then you actually have to do a bit of work to show that each $U_n$ is non-empty. You can argue that the fact that $C$ is dense in $M$ ensures that $B_{1/n}(u)\cap C\ne\varnothing$ for each $n\in\Bbb Z^+$, so for each $n\in\Bbb Z^+$ there is a $c_{k(n)}\in C\cap B_{1/n}(u)$. But then $u\in U\cap B_{1/n}(c_{k(n)})$, so $u_{k(n)}^n\in U_n$. (Here $u$ is any fixed point of $U$.) Thus, it is true that $\bigcup_{n\in\Bbb Z^+}U_n$ is a countable subset of $U$, but you cannot name it $U$: in your second bullet point you are using $U$ as a name for two different sets. I’ll call it $D$: $D=\bigcup_{n\in\Bbb Z^+}U_n$.

Your argument to show that $D$ is dense in $U$ would be right if you hadn’t got some of the notation wrong. Let $u\in U$; there is indeed a sequence in $C$ converging to $u$, but you’ve already indexed $C$ as $C=\{c_\ell:\ell\in\Bbb Z^+\}$, so you can’t just say that there is a sequence $\langle c_\ell:\ell\in\Bbb Z^+\rangle$ in $C$ that converges to $u$: you then have two competing (and almost certainly conflicting) indexings of $C$.

Use a different name: there is a sequence $\langle x_\ell:\ell\in\Bbb Z^+\rangle$ in $C$ that converges to $u$ and moreover has the property that $d(u,x_\ell)<\frac1\ell$ for each $\ell\in\Bbb Z^+$. For each $\ell\in\Bbb Z^+$ there is a $k(\ell)\in\Bbb Z^+$ such that $x_\ell=c_{k(\ell)}$, so $u\in U\cap B_{1/\ell}(c_{k(\ell)})$, and therefore $u_{k(\ell)}^\ell\in U\cap B_{1/\ell}(c_{k(\ell)})=U\cap B_{1/\ell}(x_\ell)$. Thus, $d(u_{k(\ell)}^\ell,x_\ell)<\frac1\ell$, and the triangle inequality ensures that

$$d(u,u_{k(\ell)}^\ell)\le d(u,x_\ell)+d(x_\ell,u_{k(\ell)}^\ell)<\frac2\ell\,.$$

(Note that the bound is $\frac2\ell$, not $\frac1{2\ell}$.) It follows that the sequence $\langle u_{k(\ell)}^\ell:\ell\in\Bbb Z^+\rangle$ in $D$ converges to $u$ and hence that $D$ is dense in $U$.

  • $\begingroup$ Terrific answer! Thank YOU Prof. Scott! Your contribution is beyond words. We owe you so much $\endgroup$ – Theoneandonly Feb 19 at 2:05

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.