0

Let $A$ and $B$ be disjoint closed subsets of a metric space $(X,d)$. Give a direct proof for the existence of disjoint open subsets $U_a$ and $U_b$ of $X$ such that $A \subset U_a$ and $B \subset U_b$.

My approach: I found this problem a bit trivial, but maybe I was wrong. Since $A$ and $B$ are disjoint closed subsets, there must exist a set of points $\left\{x_1,...,x_n\right\}$ that do not belong to both $A$ and $B$. Take $U_a = A\cup \left\{x_1, x_2, x_3\right\}$ and $U_b = B\cup \left\{x_4,x_5,...x_n\right\}$ such that for each $x_i$ in $U_a$ or $U_b$, exist some $\epsilon>0$ that contains the open ball $B_{\epsilon}(x_i)$. So $U_a$ and $U_b$ are open and disjoint (Q.E.D)

Gerry Myerson
  • 179,216
ghjk
  • 2,859
  • 2
    If this proof were correct, then using $A = [-3,-2]$, $B=[-1,0]$, and $x_1=1$, $x_2=2$, $x_3=3$, … you would have proved that the set $U_a = A \cup {x_1,x_2,x_2} = [-3,-2] \cup {1,2,3}$ is open, which would be quite a dubious accomplishment. – Lee Mosher Oct 01 '14 at 16:59
  • This statement as given isn't true. It's only true of one if $A$ and $B$ is compact, I think.

    Counterexample: the graph of $f(x)=1/x$ for $x>0$ and the line $x=0$ in $\mathbb R^2$.

    – Ian Coley Oct 01 '14 at 17:06
  • @user177196 are you allowed to use the fact that $X$ is hausdorff or regular? – graydad Oct 01 '14 at 17:09
  • @graydad as $X$ is a metric space I'd hope that's allowed. – Ian Coley Oct 01 '14 at 17:14
  • @IanColey The statement is true. Metrizable spaces are normal. – drhab Oct 01 '14 at 17:46

1 Answers1

3

This answer is from Munkres Theorem 32.2 (metrizable spaces are normal)

For any $a\in A$ choose $\epsilon_a>0$ s.t. $B_{\epsilon_a}(a)\cap B=\emptyset$

For any $b\in B$ choose $\epsilon_b>0$ s.t. $B_{\epsilon_b}(b)\cap A=\emptyset$

Define $U=\bigcup_{a\in A}B_{\frac12\epsilon_a}(a)$ and $V=\bigcup_{b\in B}B_{\frac12\epsilon_b}(b)$.

They are open with $A\subseteq U$ and $B\subseteq V$.

It remains to be shown that the sets are disjoint.

Assume that $x\in U\cap V$.

Then $a\in A$ and $b\in B$ exist with $d(a,x)<\frac12\epsilon_a$ and $d(b,x)<\frac12\epsilon_b$ leading to $d(a,b)<\frac12(\epsilon_a+\epsilon_b)$.

If $\epsilon_b\leq\epsilon_a$ then $d(a,b)<\epsilon_a$ contradicting that $B_{\epsilon_a}(a)\cap B=\emptyset$.

If $\epsilon_a\leq\epsilon_b$ then $d(a,b)<\epsilon_b$ contradicting that $B_{\epsilon_b}(b)\cap A=\emptyset$.

Hence no such x exists $\implies U\cap V=\phi$

sonu
  • 712
drhab
  • 151,093
  • For the first 2 lines, you meant $\frac{\epsilon_{a}}{2}$ and $\frac{\epsilon_{b}}{2}$? And all this ball will be open? Also, how do you know U and V, which is the union of open ball, will be sufficient to cover $A$ and $B$? Is it because $A$ and $B$ are closed? – ghjk Oct 02 '14 at 00:54
  • apparently, it seems to me you for the last 2 lines, you switched the correct place for $\epsilon_a\leq \epsilon_b$ and $\epsilon_b\leq \epsilon_a$? – ghjk Oct 02 '14 at 01:31
  • In your second comment you are correct. Thank you, I repaired. With the first two lines nothing is wrong. The balls are meant to be open. For every $a\in A$ you have $a\in B_{\frac{1}{2}e_{a}}\subseteq U$ showing that $U$ covers $A$. It has no connection with the fact that $A$ is closed. – drhab Oct 02 '14 at 07:05
  • @SriKrishnaSahoo Thank you, for repairing. – drhab May 02 '19 at 08:43