I know this question has asked to death here on MSE but I have not found a satisfactory solution. A solution found online is extremely elegant but I do not quite understand it!

Given nested sequence of closed nonempty connected subsets of a compact metric space $X$. Prove that $\bigcap_{i=1} X_i$ is nonempty and connected.

It is a standard fact that arbitrary intersection of compact set is nonempty and compact.

I wish to understand the bit of the proof that this is also connected

Proof: https://math.berkeley.edu/sites/default/files/pages/f10solutions.pdf

Suppose that $\bigcap_{i=1} X_i$ is not connected. Let $A$ and $B$ be two disjoint nonempty closed sets so that $\bigcap_{i=1} X_i = A \cup B$. Find disjoint open sets $U$ and $V$ so that $A \subset U$ and $B \subset V$.

Put $F_i = X_i − (U \cup V )$. Then $\{Fi\}$ is a nested sequence of compact sets, whose intersection is empty. Thus $F_i = ∅$ for some $i$. That is, $X_i \subset U \cup V$.

However, $X_i$ intersects both $U$ and $V$ , since $X_i \cap A\neq ∅$ and $X_i \cap B \neq ∅$. This contradicts the assumption that $X_i$ is connected.

Can someone please elaborate on some of the details of the proof?

1) Why bother finding open sets containing $A,B$ and how do we know that they even exist? This move seems sort of unnatural.

2) What is so special about $U \cup V$ especially, why does $F_i = X_i − (U \cup V )$ imply $\{F_i\}$ has empty intersection? Why does intersection being empty imply the existence of $F_i = \varnothing$?

3) Can someone please justify $F_i = \varnothing \Leftrightarrow X_i \subset U \cup V$.

4) Can someone please justify the contradiction?

I know this is the pretty much the entire proof, but I have seriously tried to crack it and failed to understand it. Perhaps the proof is too difficult.


1) We have some fiddley-ness at the start to get $U, V$ disjoint. It's a bit long, so see Disjoint compact sets in a Hausdorff space can be separated

This is possible because the space is Hausdorff (given that it is metrizable). There's a nice phrase by Willard, who says that this is an example of compact sets behaving like points - as $A, B$ are closed in a compact space, they are also compact, and because the space is Hausdorff, there are disjoint open sets containing $A$ and $B$. Neat!

The reason we want such sets will become apparent as we wade on.

2) $U\cup V$ is special because it covers $\bigcap_{i=1}X_i$, and because it's open. So defining $F_i\equiv X_i\setminus (U\cup V)$ is a closed subset of $X$, hence compact, and doesn't contain the intersection $\bigcap_{i=1}X_i$; finally note the $F_i$ also form a nested sequence (draw a diagram!). So if all $F_i$ were non-empty, their intersection would be non-empty - but that's not true! As $$\bigcap_{i=1}F_i=\bigcap_{i=1}(X_i\setminus (U\cup V))=(\bigcap_{i=1}X_i)\setminus (U\cup V)=\emptyset$$ because we already knew $U\cup V$ covers that intersection. So, there is some specific $F_k$ that is empty!

3) Nearly there now. As this $F_k$ is defined as $X_k\setminus (U\cup V)$, then if $F_k=\emptyset$, that implies every element of $X_k$ was also in $U\cup V$. i.e. $X_k\subset (U\cup V)$.

4) So we have that there is a pair of disjoint open sets covering this connected $X_k$. But surely they provide a separation of $X_k$? Just taking the complements of the open sets $U, V$ (namely our original $A$ and $B$...) and then intersecting with $X_k$ to get a separation. The only way this wouldn't be true is if $X_k$ had trivial intersection with one of those complements, i.e. if $X_k$ was entirely contained by one of the open sets. But this can't be true, because we know $$X_k\cap A\supseteq (\bigcap_{i=1}X_i)\cap A\neq\emptyset$$ The same holds for $B$, and so $X_k$ really has a separation. This is a contradiction because it was supposed to be connected!

Phew. I hope that makes sense, please ask for clarifications if anything is too terse. As often seems to be the case, a short looking proof is actually an example of economy of writing, rather than a simple idea. That's not to say this isn't a nice proof, it's just not one you can 'see' in a flash without having spent some time thinking topologically.

  • 1
    $\begingroup$ Hi thank you for this well thought out and complete answer. One point of confusion, if $F_k = \varnothing$, doesn;t this imply that $X_k = U \cup V$ instead of just merely being a subset of $U \cup V$? $\endgroup$ – Carlos - the Mongoose - Danger Jan 29 '16 at 2:17
  • $\begingroup$ Nah, not necessarily. Note that $\bigcap_{i=1}X_i\subseteq U\cup V$, so $U\cup V$ could certainly be bigger than this. The question I linked you too works through this; you might end up with open sets slightly bigger than the closed ones (though you probably don't need to, I feel like Hausdorff could take care of that). Either way the proof works - the crucial fact is that $U\cup V$ covers this $X_k$. $\endgroup$ – Alexander Heyes Jan 29 '16 at 2:26
  • $\begingroup$ Thanks yeah sorry I mixed up the notation. I wish the original proof used different indices $\endgroup$ – Carlos - the Mongoose - Danger Jan 29 '16 at 2:31
  • $\begingroup$ @AlexanderHeyes How does $F_i$ being closed imply it is compact. $\endgroup$ – DLB Mar 10 at 23:41

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.