I just learned compactness today in college analysis course (we are using baby Rudin for textbook). The Rudin's definition is

Open cover: By an open cover of a set $E$, in a metric space $X$, we mean a collection $\{G_a\}$ of open subsets of $X$ such that $E \subset \cup_{\alpha}G_{\alpha}$

Compactness: A subset $K$ of a metric space $X$ is said to be compact if every open cover of $K$ contains a finite subcover.

Professor said that open set like (0, 1) is not compact, but I just thought that why can't we just use (0, 1) itself as a finite subcover because (0, 1) is just one, which I mean here as "finite," set which actually covers (0,1).

I would greatly appreciate it if you could enlighten me so that I can truly understand compactness.

Thank you!

  • 5
    $\begingroup$ Find me a finite subcover of the open cover by (1/3, 1) and (1/n,1/2) for all n. $\endgroup$ – user98602 Oct 12 '18 at 20:31
  • 9
    $\begingroup$ The definition stipulates that any open cover has a finite subcover., so it's not up to you to choose an open cover.since there's nothing to choose $\endgroup$ – Bernard Oct 12 '18 at 20:31

It's important that there must be a finite subcover for every open cover. Otherwise yes, trivially every set has a finite open cover; simply take it to be a single open set consisting of the union of open neighborhoods centered at every point in the set.

One way to think of the finite subcover condition is that, for a compact subset of a metric space, an open cover has to necessarily "overshoot" the set. In doing so, it becomes impossible for the minimal subcover to be infinite.

Example 1. Take $K = [0,1]$. If we have an open cover of $K$ by some open balls, then necessarily there is going to be an open ball that contains the element $1$. But if an open ball contains $1$, then that open ball must actually extend past $1$, in particular that open ball must contain all points greater than $1$ and less than $1 + \delta$ for some $\delta > 0$.

Example 2. Take $A = [0,1)$. This set is not compact, because we could choose the open cover consisting of balls $B_n = (-1, 1 - 1/n)$ for $n = 1, 2, \ldots$. This open cover never "overshoots" $A$ on the right endpoint; rather, the balls $B_n$ get ever and ever closer to $1$. Thus it is impossible for any subcover to be finite; we really do need to take a sequence with $n$ going off to infinity.

  • $\begingroup$ Thank you! I have just two questions more. 1) In example 2, is it right that $B_n=(−1,1−1/n)$ is an open cover but just doesn't have a finite subcover? 2) And we cannot use $B_n=(−1,1−1/n)$ as an open cover for $K=[0, 1]$ in example 1 because $B_n$ does not cover 1. Is this right? $\endgroup$ – Hunnam Oct 12 '18 at 21:39
  • $\begingroup$ (1) Yes, there is no finite subcover. (2) Correct, no $B_n$ contains $1$, so they do not cover $[0,1]$. Whereas it does cover $(0,1)$ because for every number slightly less than $1$, for sufficiently large $n$ there is some $B_n$ containing it. We can already see here the connection between compact sets and the concept of limits. $\endgroup$ – Christopher A. Wong Oct 12 '18 at 21:46
  • $\begingroup$ Thank you so much! Now I feel like I truly understand what compactness is haha. $\endgroup$ – Hunnam Oct 12 '18 at 22:13
  • $\begingroup$ What is $x$ in your example 1? $\endgroup$ – Andrés E. Caicedo Oct 12 '18 at 23:15
  • $\begingroup$ Sorry, that was a typo. $\endgroup$ – Christopher A. Wong Oct 18 '18 at 5:22

An open cover of $(0,1)$ need not have $(0,1)$ among its members; for instance $$ \bigl\{(0,2/3),(1/3,1)\bigr\} $$ is an open cover. Of course this open cover is finite, so it obviously has an open subcover, namely itself.

What about an infinite open cover? Well the simplest choice is $$ \bigl\{(1/2,1),(1/3,1),(1/4,1),\dotsc\bigr\}= \bigl\{(1/n,1):n>1, n\text{ integer}\bigr\} $$ Is this an open cover? Yes, because if $x\in(0,1)$, then there exists $n>1$ such that $1/n<x$. The sets are surely open.

Does it have a finite subcover? Suppose so; then it will be of the form $$ \bigl\{(1/n_1,1),(1/n_2),\dots,(1/n_k,1)\bigr\} $$ for suitable integers $n_1,n_2,\dots,n_k$. But if $m=\max\{n_1,n_2,\dots,n_k\}$, then $$ \bigcup_{i=1}^k\Bigl(\frac{1}{n_i},1\Bigr)= \Bigl(\frac{1}{m},1\Bigr) $$ and this is not the same as $(0,1)$, because $$ \frac{1}{2m}\notin\Bigl(\frac{1}{m},1\Bigr) $$ but clearly $\frac{1}{2m}\in(0,1)$.

Thus there is at least an open cover of $(0,1)$ that doesn't admit finite subcovers. Hence $(0,1)$ is not compact.

You will also learn that a compact subset of a metric space is necessarily closed (not a sufficient condition, though). Then $(0,1)$ is not compact, because it is not closed in $\mathbb{R}$.

  • $\begingroup$ "Does it have an open subcover?" should be "Does it have a finite subcover?" $\endgroup$ – symplectomorphic Oct 12 '18 at 20:58
  • $\begingroup$ @symplectomorphic Lapsus calami, thanks! $\endgroup$ – egreg Oct 12 '18 at 20:59
  • $\begingroup$ Thank you! Your explanation helped me a lot to understand. $\endgroup$ – Hunnam Oct 12 '18 at 22:13

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.