(i) Let $X$ be locally compact, and $K$ compact. Show that, given $\delta > 0$, there exists finitely many compact closed balls of radius at most $\delta$ that cover $K$.

Proof: Let $\delta >0$. Since $X$ is locally compact, to each $x \in X$ there exists $\epsilon(x) \leq \delta$ such that $C_{\epsilon(x)} (x) : = \{y \in X \, : \, d(x,y) \leq \epsilon(x)\}$ is closed and compact. Now, consider the covering $$K \subset \bigcup_{x \in K} B_{\epsilon(x)} (x) $$ so that $$K \subset \bigcup_{i=1}^{n} B_{\epsilon_i}(x_i)$$ and therefore, $$K \subset \bigcup_{i=1}^{n} C_{\epsilon_i}(x_i).$$

(ii) Deduce that if $\epsilon >0$ is small enough, then the set $S:= \{x \in X : d(x,K) \leq \epsilon\}$ with $$d(x,K) = \inf_{y \in K} d(x,y)$$ is compact.

Proof: The above argument works for any compact subset of $X$, in particular, $\partial K \subset K$, the boundary, is closed and therefore compact. Therefore, we get finitely many closed compact balls that cover $\partial K$, and given $\epsilon < \delta$, cover $S$. Thus it is enough to show that the set $S':= \{x \in X, x \notin K^\circ : d(x,K) \leq \epsilon \}$ is closed, since $$S' \subset S \subset \bigcup_{i=1}^{n} C_\delta (x_i)$$ with $x_i \in \partial K$, and since $d(x,K) = 0$ for any $x \in K$ anyways and thus we can take $S = S' \cup K$, to show $S$ is compact. Since $K$ is compact and metrics are continuous, there is $y \in \partial K$ such that $d(x,K) = d(x,y)$ for each $x \in X$. Now, $S'$ is clearly closed, since if $S' \ni x_n \to x$, then $d(x,y) \leq d(x,x_n) + \epsilon$, and this implies $d(x,y) \leq \epsilon$ as well. Thus, $S'$ is a closed subset of a compact set and therefore compact and so $S$ is compact as the union of two compact sets.

Note: ($K^\circ$ is the interior of $K$)

Any issues here?? I've been doing measure theory and funct analysis for a little while, so I wanted to do an exercise to keep up on my analysis I/II type skills.


Part $(i)$ is correct.

For part $(ii)$, you obviously made the argument much complicated than it should be. Geometrically you might want to consider $\partial K$, but in a general metric space and general $K$, you basically knows nothing about $\partial K$ (Yes, intuitions fails here). Working with $K$ will be more or less the same as working with $\partial K$.

More seriously, you haven't quite shown that $S'$ (or $S$) is compact. Thye problem is that even if $\epsilon <\delta$, $$S\subset \bigcup_{i=1}^n C_\delta (x_i)$$ might not be true (Counterexample: let $K$ be a line with length $1$ sit inside $X = \mathbb R^2$. Let $x_1, x_2$ be two endpoints of $K$ and $\delta$ a bit bigger then $1/2$. Then $S$ is not contained in $C_{\delta} (x_1) \cup C_\delta(x_2)$ whenever $\epsilon$ is not small enough, in particular when $\epsilon =1/2$).

Another serious problem is that even if you have the inclusion, $C_\delta (x_i)$ might not be compact, only that $C_{\epsilon _i}(x_i)$ is compact.

To remedy this, you have to be more careful in part $(i)$. Instead of $B_{\epsilon(x)} (x)$, you use $B_{\epsilon(x)/2} (x)$ as the open cover. Then as $K$ is compact, there are $x_1, \cdots x_n\in K$ so that $$K \subset \bigcup_{i=1}^n C_{\epsilon_i/2} (x_i),$$ where $0<\epsilon_i \le \delta$ and each ball $C_{\epsilon_i} (x_i)$ is compact. Now let $\epsilon >0$ be chosen so that $2 \epsilon <\min\{\epsilon_1, \cdots, \epsilon_n\}$. Thus if $y\in S= \{ y\in X : d(y, K)\le \epsilon\}$, there is $x\in K$ so that $d(x, y)\le \epsilon$. As $x\in K$, there is $x_i$ so that $d(x, x_i) \le \epsilon_i/2$. Hence $$d(y, x_i)\le d(y, x) + d(x, x_i) \le \epsilon + \epsilon_i/2 <\epsilon_i. $$ Thus $y\in C_{\epsilon_i} (x_i)$ and so

$$S\subset \bigcup_{i=1}^n C_{\epsilon_i} (x_i).$$ As the right hand side is a compact set and $S$ is a closed subset, $S$ is also a compact set.

  • $\begingroup$ I've still actually a tad confused as to why I can't remedy my proof... If I just replace $\delta$ by $\epsilon_i$ and $\epsilon$ by $\frac{\min{\epsilon_i}}{2}$ wouldn't I be alright? $\endgroup$ – Anthony Peter Nov 11 '15 at 11:28
  • 1
    $\begingroup$ I thought about that too, but it doesn't work. For example again if $K$ is the line of length one again, with only two open balls $B_{\epsilon_1}(x_1)$, $B_{\epsilon_2} (x_2)$ covering $K$, where $x_1, x_2\in K$ are the end points and $\epsilon_1, \epsilon_2$ are just slightly larger than $1/2$. Then $\epsilon < \min\{\epsilon_1, \epsilon_2\}/2$ don't work. (Try to draw a picture) @AnthonyPeter $\endgroup$ – user99914 Nov 11 '15 at 11:38

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.