Remark 167.1.
Recall that a set is compact iff it is complete and totally bounded. Compactness is a topological property, but completeness and totally boundedness are not; they depend on the metric. Let and be metrics on , as defined here. We have seen that while and are the same topological space (they have the same open sets), the former is complete and the latter is not. Also, is not totally bounded, but is!
We now have the tools to prove Theorem 142.3 more succinctly:
Theorem 167.2.
Let be compact, be continuous. Then, is uniformly continuous.
Proof.
Let. For , let . is open since is continuous. is an open cover of . By the Lebesgue covering lemma, there exists such that for any , there exists such that . This works: If , then , so .□
Example 167.3.
bd2833Closed balls in are not compact (we have shown that any ball centered at the origin in these spaces is not totally bounded; any ball in an NLS can be obtained by scaling and translating the unit ball).
We will now show that balls in are not compact. Consider the sequence . clearly does not have a converging subsequence: if it did, the subsequence would have to converge pointwise to the limit, and the pointwise limit of any subsequence is the pointwise limit of , which is not continuous.
In fact, unit balls at not compact in any infinite dimensional NLS. Proving this requires some functional analysis.
The Arzelà–Ascoli Theorem
Recall that for compact , is a Banach space. The motivating question for this section is: For compact , when is a subset of compact? If were finite dimensional, the answer is trivial by the Heine Borel Theorem. The Arzelà–Ascoli Theorem provides a characterization of compact subsets of when is infinite dimensional: In addition to being closed and bounded, compact subsets of are equicontinuous.
Equicontinuity is the natural generalization of uniform continuity to families of functions:
Definition 167.4.
Let be a compact metric space. is a family of equicontinuous functions if for every there exists such that for all and for all .
For example, a family of Lipschitz mappings with the same Lipschitz constant is equicontinuous (more generally, their it suffices for their Lipschitz constants to be bounded above).
Example 167.5.
Let , be metric spaces. Let . There exists an equivalent metric on such that each is Lipschitz continuous.
Proof.
First, note that for a single function, the metric on makes Lipschitz with . We can use the same idea when working with a countably family of functions. First, define the equivalent bounded metric on .
Next, define
Then, .
You’ll have to show and are equivalent.□
Note, however, that this does not make equicontinuous: blows up as .
Example 167.6.
Let be a compact metric space. Let be continuous. Let . Show that is equicontinuous.
Proof.
Let. Since is compact, is uniformly continuous. If we consider the metric on , there exists such that for all . In particular, for all , .□
Example 167.7.
Let be convex and open. Let be a family of differentiable functions from to . Assume that there exists such that for all and for all . Then, by Theorem 187.2, we have
for all and . Thus, is equicontinuous.
Theorem 167.8(Arzelà–Ascoli, Kumaresan (2005) 4.4.8).
Let be a compact metric space. Then a set is compact iff is bounded1, closed, and equicontinuous.
Proof.
If is compact, it is closed. It is also complete and totally bounded. It only remains to show equicontinuity. Let . There exist for such that . Let be chosen by the uniform continuity of for the given . Let be the minimum of the ‘s. Then, for such that , for and such that , we have
Suppose is closed, bounded, and equicontinuous. Since is closed, it is complete. It remains to show that is totally bounded. Let . Let be chosen by the equicontinuity of . Since is compact, there exist such that . Suppose for all . Then, we can restrict the codomain of each to be . is compact in , so there exist such that . Define
. For , define
We now claim that
E1Indeed, for , there exists such that if lands in , hence .
It now remains to show that the diameter of the ‘s are bounded by some constant multiple of . Suppose and for some . Then,
by the equicontinuity of and the definition of , so . We’re done: for every , there exists an open ball of diameter which contains , and it follows trivially from (E1) that
□
Corollary 167.9.
If is infinite dimensional, then any compact subset of must be nowhere dense.
All we need to show is that closed balls are not compact in ; this can be done by adding triangle functions with increasing slopes to the function at the center of ball, violating equicontinuity.
This holds in general for any infinite dimensional NLS, since closed balls in infinite dimensional normed linear spaces are not compact.
Footnotes
-
As a subset of ! ↩