18

How to prove for every open cover of the real numbers $\mathbb{R}$ there is a countable subcover? Without using more sophisticated results from topology, assuming only a real analysis background.

I've found a proof using second-countable space characterization, but since i never studied general topology before, it's hard to associate a countable base on the real line. My intuition says to transform the open cover into disjoint open subsets, but how to achieve that?

hardmath
  • 37,015
juliohm
  • 615

2 Answers2

32

I’m going to assume that you want to prove that $\mathbb{R}$ is Lindelöf. You definitely do not want to try to transform the open cover into disjoint open sets, because it can’t be done: no family of two or more pairwise disjoint non-empty open sets covers $\mathbb{R}$.

Getting a countable base for $\mathbb{R}$ isn’t at all hard, provided that you know that $\mathbb{Q}$, the set of rational numbers, is countable. Just let $\mathscr{B}$ be the set of open intervals with rational endpoints: each pair $\{p,q\}$ of distinct rational numbers determines exactly one such interval, $(p,q)$ if $p<q$, and $(q,p)$ if $p>q$, and there are only countably many pairs of rational numbers, so $\mathscr{B}$ is countable. It only remains to show that $\mathscr{B}$ is a base for the topology of $\mathbb{R}$, which just means showing that every open set in $\mathbb{R}$ is a union of members of $\mathscr{B}$.

Every non-empty open set in $\mathbb{R}$ is a union of open intervals. If we can show that every open interval in $\mathbb{R}$ is a union of members of $\mathscr{B}$, i.e., of open intervals with rational endpoints, it will immediately follow that every non-empty open subset of $\mathbb{R}$ is also such a union. To this end let $(a,b)$ be any non-empty open interval in $\mathbb{R}$. Then there are sequences $\langle p_n:n\in\mathbb{N}\rangle$ and $\langle q_n:n\in\mathbb{N}\rangle$ of rational numbers such that:

  • $p_0>p_1>p_2>\dots\;$;
  • $q_0<q_1<q_2<\dots\;$;
  • $\lim\limits_{n\to\infty}p_n = a\;$;
  • $\lim\limits_{n\to\infty}q_n = b\;$; and
  • $p_0<q_0$.

In other words, $\langle p_n:n\in\mathbb{N}\rangle$ is a decreasing sequence converging to $a$, $\langle q_n:n\in\mathbb{N}\rangle$ is an increasing sequence converging to $b$, and $p_0<q_0$. It easily follows that $$(a,b) = \bigcup_{n\ge 0}(p_n,q_n)\;,$$ and each interval $(p_n,q_n)$ obviously has rational endpoints. Thus, every non-empty open interval in $\mathbb{R}$ is a union of members of $\mathscr{B}$, so every non-empty open set of any kind in $\mathbb{R}$ is such a union, and $\mathscr{B}$ is therefore a countable base for the topology of $\mathbb{R}$.

It’s now trivial to see that $\mathbb{R}$ is Lindelöf: if $\mathscr{U}$ is any open cover of $\mathbb{R}$, just let $$\mathscr{B}_\mathscr{U}=\{B\in\mathscr{B}:\exists U\in\mathscr{U}\big(B\subseteq U\big)\}.$$ Each $U\in\mathscr{U}$ is the union of the members of $\mathscr{B}_\mathscr{U}$ contained in it, so $\mathscr{B}_\mathscr{U}$ covers $\mathbb{R}$. It’s also countable, since it’s a subset of the countable set $\mathscr{B}$. Now for each $B\in\mathscr{B}_\mathscr{U}$ choose some $U(B)\in\mathscr{U}$ such that $B\subseteq U$; the definition of $\mathscr{B}_\mathscr{U}$ guarantees that there is one. Let $\mathscr{U}_0=\{U(B):B\in\mathscr{B}_\mathscr{U}\}$. Then $\mathscr{U}_0$ is countable, since it’s no bigger than $\mathscr{B}_\mathscr{U}$, and $$\bigcup\mathscr{U}_0 = \bigcup_{B\in\mathscr{B}_\mathscr{U}}U(B)\supseteq \bigcup_{B\in\mathscr{B}_\mathscr{U}}B=\mathbb{R},$$ so $\mathscr{U}_0$ is indeed a countable subcover of $\mathscr{U}$.

Added: This idea can be extended to $\mathbb{R}^n$. Instead of open intervals with rational endpoints, you take for your countable base the set of Cartesian products of such intervals. In other words, you take as a base for $\mathbb{R}^n$ the set of open boxes of the form $B_1\times\dots\times B_n$, where $B_1,\dots,B_n\in\mathscr{B}$. Then you show that each open box in $\mathbb{R}^n$ is a union of these ‘rational boxes’. Since there are only countably many rational boxes, and every non-empty open set in $\mathbb{R}^n$ is a union of them, it follows that $\mathbb{R}^n$ is Lindelöf: the remainder of the argument is just like that for $\mathbb{R}$.

There is another way to proceed, if you know that closed, bounded subsets of $\mathbb{R}^n$ are compact, meaning that every open cover of such a set has a finite subcover. I’ll do it for $\mathbb{R}$; the generalization to $\mathbb{R}^n$ is pretty straightforward. $\mathbb{R}$ is the union of the closed intervals $[n,n+1]$ for $n\in\mathbb{Z}$. There are only countably many such intervals, and each of them is compact. Now let $\mathscr{U}$ be any open cover of $\mathbb{R}$. For each $n\in\mathbb{Z}$ let $\mathscr{U}_n = \{U\in\mathscr{U}:U\cap [n,n+1]\ne\varnothing\}$. Then $\mathscr{U}_n$ is an open cover of $[n,n+1]$, so it has a finite subcover, $\mathscr{V}_n$. Finally, let $$\mathscr{V}=\bigcup_{n\in\mathbb{Z}}\mathscr{V}_n\;;$$ $\mathscr{V}$ is the union of countably many finite sets, so it’s a countable subset of $\mathscr{U}$, and it clearly covers $\mathbb{R}$.

Brian M. Scott
  • 616,228
  • Thank you for your answer, it's really helpful. I'll take a better look to investigate all the details and then back to probably mark the topic as solved. – juliohm Nov 11 '11 at 22:21
  • Can't you just, given an uncountably infinite cover of the real numbers, just choose one open set containing each rational? – Carl Nov 12 '11 at 00:26
  • That is, unless you have an objection to the axiom of choice I suppose. – Carl Nov 12 '11 at 00:32
  • 2
    @Carl: No, you can’t: the union of the resulting sets will be a dense open set, but it won’t necessarily be all of $\mathbb{R}$. For a very simple example, the union might be $\mathbb{R}\setminus{n\sqrt{2}:n\in\mathbb{Z}}$. In that case you could just throw in one open set for each $n\sqrt{2}$, but the uncovered bit could be uncountable, e.g., a Cantor set contained in the irrationals. (They do exist.) – Brian M. Scott Nov 12 '11 at 04:58
  • @BrianM.Scott: For construct monotone sequences $(p_n)$ and $(q_n)$ converging to $\inf p_n = a$ and $\sup q_n = b$, respectively, we use the fact that $\mathbb{Q}$ is dense and is possible to pick $p_0 < q_0 \in (a,b)$. We continue doing this now with the reduced intervals $(a,p_0)$ and $(q_0,b)$ and the sequences will show up, correct? I'm thinking now in the core proof ($\mathbb{R}$ is Lindelöf)... – juliohm Nov 12 '11 at 09:52
  • @juliohm: Almost. To make sure that the sequences actually do converge to $a$ and $b$, you should pick the rational $p_{n+1}\in\left(a,\frac{a+p_n}2\right)$ and $q_{n+1}\in\left(\frac{q_n+b}2,b\right)$, say. That way $|p_{n+1}-a|<|p_n-a|/2$ and $|b-q_{n+1}|<|b-q_n|/2$, so that $\lim_n|p_n-a|=0$ and $\lim_n|b-q_n|=0$. Otherwise you might find the $p_n$’s converging to something larger than $a$ or the $q_n$’s to something smaller than $b$. – Brian M. Scott Nov 12 '11 at 10:18
  • @BrianM.Scott: That's true! Nice notion of proximity. Please give me time to think better on the rest of the proof. I'm too busy, but i won't forget the topic. Thank you. – juliohm Nov 12 '11 at 18:58
  • @BrianM.Scott: I have some final questions, what you mean by the $\mathscr{U}(B\subset U)$ notation? Is just to emphasize the objects dependency, correct? Like saying in calculus given $\varepsilon$, find $\delta(\varepsilon)$ such that... The other question refers to the statement: "Then $\mathscr{U}0$ is countable, since it's no bigger than $\mathscr{B}{\mathscr{U}}$". How you're comparing this sets which are not subsets each other? – juliohm Nov 15 '11 at 20:33
  • @juliohm: It’s not $\mathscr{U}(B\subseteq U)$; it’s $\exists U\in\mathscr{U}(B\subseteq U)$, i.e., there is a $U\in\mathscr{U}$ such that $B\subseteq U$. – Brian M. Scott Nov 15 '11 at 20:47
  • @juliohm: For your other question, the map $B\mapsto U(B)$ is a surjection from $\mathscr{B}\mathscr{U}$ to $\mathscr{U}_0$, so $\mathscr{U}_0$ must be the same size as $\mathscr{B}\mathscr{U}$ or smaller. – Brian M. Scott Nov 15 '11 at 20:50
  • @BrianM.Scott: Ok, it was what i had in mind. And the second question? If i understood correctly, we can substitute the $\mathcal{R}$ set by any $X\subset\mathcal{R}$ and the proof will still valid. – juliohm Nov 15 '11 at 20:55
  • @juliohm: I’m not sure what you mean by the $\mathcal{R}$ set. Do you mean that the proof works for any subset $X$ of $\mathbb{R}$, the reals? Yes, that’s true. Once you have a countable base for $\mathbb{R}$, you can just intersect its members with $X$ to get a countable base for $X$. – Brian M. Scott Nov 15 '11 at 20:57
  • @BrianM.Scott: Oh, really. This is a surjection, again the notation got me. Thank you for your attention. – juliohm Nov 15 '11 at 20:59
  • R-Lindelof depends of AC? – Gaston Burrull Apr 28 '13 at 02:05
  • So it seems. At least, the given proof seems to. Precisely in the «choose some $U(B)$» part. – MickG Apr 23 '15 at 19:45
  • @MickG: What is required is specifically $\mathsf{CC}(\Bbb R)$, the axiom of countable choice for subsets of $\Bbb R$. – Brian M. Scott Apr 23 '15 at 19:54
  • And the same in $\mathbb{R}^n$, I'd say. Or can that be deduced from the $\mathbb R$ case? And proof two also requires the AC in choosing the finite subcovers, right? The same $\mathsf{CC}(\mathbb R)$ I'd say. – MickG Apr 23 '15 at 21:28
  • 1
    @MickG: I just checked: $\mathsf{CC}(\Bbb R)$ is equivalent to every second countable space being Lindelöf. – Brian M. Scott Apr 23 '15 at 21:33
  • That is interesting and sounds incredible... I guess I'll ask a new question about that. – MickG Apr 24 '15 at 08:16
  • @BrianM.Scott Done it. – MickG Apr 24 '15 at 08:43
  • @BrianM.Scott Seems to me even "$\mathbb N$ is Lindelöf" is equivalent to $\mathsf{CC}(\mathbb R)$, is that right? – bof Apr 24 '15 at 09:50
  • @bof: Yes: Theorem $\mathbf{3.8}$ in Herrlich, Axiom of Choice, Lecture Notes in Mathematics 1876. – Brian M. Scott Apr 24 '15 at 17:08
4

NOTE: I don't know if this is what the user would call "sophisticated results". It uses some ostensibly highfalutin language, but the ideas are very simple. I hope it suffices.

I think what the user is asking is to prove that every second countable space is Lindelof (in more common notation).

Really, the user is asking to prove that "If $X$ is second countable and $A$ is a subset of $X$, the any open cover of $A$ admits a countable subcover." To see how $\text{Second Countable}\implies \text{Lindelof}$ gives us this merely note that if $\Omega$ is a $X$-open cover of $A$ then $\Omega$ induces a $A$-open cover of $A$ and since $A$ is second countable (since second countability is hereditary) our result $\text{Second Countable}\implies\text{Lindelof}$ gives us what we want.

So, let's prove $\text{Second Countable}\implies\text{Lindelof}$. So, let $X$ be second countable with countable basis $\mathscr{B}$, and let $\Omega=\left\{U_\alpha\right\}_{\alpha\in\mathcal{A}}$ be an open cover for $X$. By assumption, for each $\alpha\in\mathcal{A}$ we can cover $U_\alpha$ with some collection $B_\alpha$ of elements of $\mathscr{B}$. Note then that $\displaystyle \Sigma=\bigcup_{\alpha\in\mathcal{A}}B_\alpha$ is a countable open cover for $X$. So, for each element $O$ of $\Sigma$ choose an element $U$ of $\Omega$ containing it. Then, this subset, call it $\Gamma$, of $\Omega$ is an open cover of $X$ (since its union contains the union over all the elements of $\Sigma$ which is $X$) and is countable since there is a surjection $\Sigma\to\Gamma$ and $\Sigma$ is countable. Thus, $\Gamma$ is our desired countable subcover of $\Omega$.

Alex Youcis
  • 54,059
  • Thank you for your answer, it's very elucidative. Unfortunely, i'm asking a more simple question. – juliohm Nov 11 '11 at 22:06
  • Ah, I see. I'm sorry that it was overly complicated! – Alex Youcis Nov 11 '11 at 22:08
  • @AlexYoucis I suppose by $\bigcup_{\alpha\in\frak A}B_{\alpha}$ you mean the union of the sets in that collection. However, I would read that as a new collection of sets containing all the open sets in the various $B_\alpha$. I suggest you add an extra $\bigcup$ before the one already present in that $\Sigma$. – MickG Apr 23 '15 at 19:38
  • @juliohm in fact this answer looks like just a generalization to a generic topological space of the above answer, which was specific to $\mathbb{R}$ first and $\mathbb{R}^n$ then. The arguments are much the same. – MickG Apr 23 '15 at 19:39
  • And the answer's start is the generalisation of this comment :). – MickG Apr 23 '15 at 19:43