I'm trying to prove that if $X$ is a metric space, then for a set $A\subset X$, the closure $\bar A$ is compact iff for every sequence $(x_n)_{n=1}^\infty\subset A$ there exists a subsequence $(x_{n_k})_{k=1}^\infty$ that converges to some $x\in X$.

The forward direction is pretty straightforward as a particular case of the definition of compactness by B-W property.

For the reverse, let $a\equiv(x_n)_{n=1}^\infty\subset \bar A$ be some sequence. Then, if $a$ has any subsequence in $A$ we are done by assumption. Otherwise, there are at most finitely many points in $a$ that are in $A$. So there exists some subsequence which resides entirely in $\bar A \backslash A$, and also for some $N\in \mathbb N$ the $N$-tail of $a$ is in $\bar A\backslash A$.

But I don't see where this gets me. What guarantees that I can come up with a convergent subsequence? The closure of a set contains all its limit points, but why shouldn't I be able to choose a non-convergent sequence this subset? Or in $\bar A$ in general?

Other approaches such as finite open covers or being a closed subset of a compact set don't seem to be relevant here.


  • $\begingroup$ You might get some mileage out of a diagonal trick: each $x_n\in\bar A$ has its own sequence $\{x_{n,i}\}\subset A$ that converges to $x_n$. Now consider $\{x_{n,n}\}$.... $\endgroup$ – Greg Martin Mar 29 '17 at 21:28
  • $\begingroup$ @Greg Martin Could you please explain a little further? I didn't quite get it. Specifically, who's ${x_n,n}$? Do you mean that If I have a sequence, then each point is a limit point of some other sequence, so I can pick only points from the original sequence that coincide with the corresponding point in the ${x_n}$ sequence? I'm not sure I am guaranteed to get a non-empty set out of it, nor a convergent one. $\endgroup$ – Yoni Mar 30 '17 at 7:07

PART 1. Let $\bar A$ be compact and let $(x_n)_{n\geq 0}$ be a sequence in $A.$

Let $C_0$ be a finite cover of $\bar A$ by open balls of $X$ of radius $1.$ Let $\beta_0\in C_0$ such that $\{n:x_n\in \beta_0\}$ is infinite. Let $\gamma_0=\bar {\beta_0}\cap \bar A$ .

Now recursively suppose that $\gamma_m$ is a compact subset of $\bar A$ and that $\{n:x_n\in \gamma_m\}$ is infinite. Let $C_{m+1}$ be a finite cover of $\gamma_m$ by open balls of $X$ of radius $2^{-(m+1)}$ and let $\beta_{m+1}\in C_{m+1}$ such that $\{n:x_n\in \beta_{m+1}\}$ is infinite. And let $\gamma_{m+1}=\bar \beta_{m+1}\cap \gamma_m.$

Now we may take a strictly increasing $f:\mathbb N_0\to \mathbb N_0$ such that $x_{f(n)}\in \gamma_n$ for every $n.$ Now $d(x_{f(n)},x_{f(n')})<2^{-n}$ when $n<n'$, so the sequence $(x_{f(n)})_{n\geq 0}$ is a Cauchy sequence in the closed set $\gamma_0$ , so it converges to a limit point $x$.

PART 2.(Converse). Let $\bar A$ be non-compact. Let $C$ be an open cover of $\bar A$ with no finite subcover.

CASE 1. $C$ has no countable subcover.

For each $p\in \bar A$ take $q_p\in \mathbb Q^+$ and $c_p\in C$ such that $B(p,q_p)\subset c_p.$ For $q\in \mathbb Q^+$ let $$S(q)=\{p\in \bar A: q_p=q\}.$$ Suppose that for every $q\in \mathbb Q^+$ there is a finite $T(q)\subset S(q)$ such that $\cup_{p\in T(q)}B(p,q)\supset S(q).$ Then $\cup_{q\in \mathbb Q^+}\{c_p:p\in T(q)\}$ is a countable sub-cover of $C$, contrary to hypothesis.

Therefore there exists $q_0\in \mathbb Q^+$ such that no finite subset of $\{B(p,q_0):p\in S(q_0)\}$ covers $S(q).$

So take $p_1\in S(q_0)$ and take $p_{n+1}\in S(q_0)$ \ $\cup_{j\leq n}B(p_j,q_0).$ We now have $d(p_n,p_m)\geq q_0$ whenever $n\ne m$.

Finally take $x_n\in A$ such that $d(x_n,p_n)<q_0/4.$ Then $(x_n)_n$ is a sequence in $A$ with $d(x_n,x_m)\geq q_0/2$ whenever $n\ne m$, so it has no convergent subsequence.

CASE 2. $C$ has a countable subcover. Let $\{c_n:n\in \mathbb N\}$ be a countable subcover of $C.$ Since $C$ has no finite subcover, let $$x_n\in \bar A \ \cup_{j<n}c_j.$$ Suppose $(x_n)_n$ had a subsequence $(x_{f(n)})_n$ converging to $x$ (with $f :\mathbb N\to \mathbb N$ strictly increasing).

We have $x\in \bar A$ so $x\in c_{n_0}$ for some $n_0.$

But then $B(x,r)\subset c_{n_0}$ for some $r>0,$ and for all but finitely many $n$ we have $x_{f(n)}\in B(x,r)$ (because $(x_{f(n)})_n$ converges to $x$).... So there exists $n$ such that $f(n)>n_0$ and $x_{f(n)}\in c_{n_0}$, contrary to $x_{f(n)}\not \in \cup_{j<f(n)}c_j.$

Therefore $(x_n)_n$ has no convergent subsequence. Finally let $y_n\in A$ with $d(x_n,y_n)<2^{-n}.$ Then $(y_n)_n$ has no convergent subsequences for if $(y_{f(n)})_n$ converged to $y$ then $(x_{f(n)})_n$ would also converge to $y$.

  • $\begingroup$ Thanks, but isn't this the forward direction? I want to prove $\bar A$ is compact given the what is known about $A$. $\endgroup$ – Yoni Mar 30 '17 at 7:02
  • $\begingroup$ I added the converse, which can be done more easily using the theory of ordinals and transfinite recursion, but I dk whether you know that stuff, and it's basically the same proof. An equivalent to this result is that a metric space is non-compact iff it has an infinite closed discrete subspace (which is not the case for all topological spaces). $\endgroup$ – DanielWainfleet Mar 30 '17 at 9:02
  • $\begingroup$ I'm familiar with ordinals etc., but the above is sufficient! Thank you $\endgroup$ – Yoni Mar 30 '17 at 9:28

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.