5
$\begingroup$

I’m wondering if there is a version of Arzelà-Ascoli for continuous functions on not-necessarily compact metric/Hausdorff spaces $X$, i.e. a characterization of the compact subsets of $C_b(X)$ (under the sup norm) for $X$ not-necessarily compact, e.g. the titular $X=(0,1)$, using a neat set of conditions.

I assume there is not such a characterization, which brings me to perhaps the more important question: is there a philosophical reason or guiding intuition/heuristic that tells us that we “should not expect” such a theorem for non-compact spaces $X$?

Or dually is there a philosophical reason or guiding intuition/heuristic that tells us that we “should expect” such a theorem for compact spaces $X$? I know that continuous functions on compact sets enjoy a wide variety of nice properties (boundedness, attaining extrema, uniformity of continuity and convergence, etc.) but how do I know that these properties are “the best properties” and therefore indicative that continuous functions “really should belong on compact sets”/“live most naturally on compact sets”?

EDIT: there is a version of Arzelà-Ascoli for LCH spaces under compact subset convergence https://math.stackexchange.com/questions/20670/theorem-of-arzel%C3%A0-ascoli, but (1) that's not exactly what I am asking about and (2) compactness still permeates that entire result, i.e. it is clearly just a slight augmentation of the standard compact Arzelà-Ascoli. Some comments bring up inheriting an Arzelà-Ascoli theorem from a compactification, but that still leaves the question of why it is "more natural" to first do the compactness case and then try to extend from there.

(Cross-post of https://math.stackexchange.com/questions/4339552/arzel%c3%a0-ascoli-for-c-b0-1-or-more-generally-why-is-that-continuous-function).

EDIT: https://math.stackexchange.com/questions/35115/classifying-the-compact-subsets-of-lp tells us there is a nice characterization of compact subsets of $L^p(\mathbb R^n)$, and the Hanche-Olsen & Holden paper the answer links (https://arxiv.org/pdf/0906.4883.pdf) references an article of Phillips (Thm. 3.7 of https://www.ams.org/journals/tran/1940-048-03/S0002-9947-1940-0004094-3/) providing a characterization of compact subsets of any Banach space, including of course $C_b(X)$ for any topological space $X$. I am unsure as to the usefulness of Phillips' characterization.

$\endgroup$
2
  • 1
    $\begingroup$ What about Bourbaki, General Topology (1989), Theorem 2 in X.2.5 with necessary and partially sufficient assumptions? $\endgroup$ Dec 27, 2021 at 22:51
  • 4
    $\begingroup$ The algebra of bounded continuous functions on a locally compact Hausdorff space $X$ is a closed subspace of the algebra of continuous functions on the Srone-Cech compactification $\hat{X}$, so $X$ does inherit an Arzela-Ascoli theorem from $\hat{X}$. The Stone-Cech compactification is usually a disaster, though, so that explains your observations. You get a nicer result for $C_0(X)$ since the relevant compactification is the much nicer one-point compactification. $\endgroup$ Dec 27, 2021 at 22:55

2 Answers 2

3
$\begingroup$

A possible proof of the Arzelà-Ascoli theorem for $C(X)$ with compact $X$ is to show that equicontinuity of a subset $\mathcal F\subseteq C(X)$ implies that the topology of pointwise convergence coincides on $\mathcal F$ with the topology of the sup-norm. That $\mathcal F$ is relatively compact for the pointwise topology follows from Tychonov's theorem. For a general topological space and the space $C_b(X)$ this proof gives an abstract characterization:

A subset $\mathcal F$ of $C_b(X)$ is compact for the sup-norm $\|\cdot\|$ if and only if it is closed and bounded and, for all $f\in\mathcal F$ and $\varepsilon>0$, there are $\delta>0$ and a finite subset $E\subseteq X$ such that, for every $g\in \mathcal F$ with $\max\{|f(x)-g(x)|:x\in E\}<\delta$ we have $\|f-g\|<\varepsilon$.

It depends very much on the concrete situation whether this is useful. As an example, the general result implies that a closed and bounded subset $\mathcal F\subseteq C_b(X)$ for a precompact metric space $X$ is compact if it is equi-uniformly continuous, i.e., for every $\varepsilon>0$ there is $\delta>0$ such that, for all $f\in\mathcal F$ and $x,y\in X$ with $d(x,y)<\delta$, one has $|f(x)-f(y)|<\varepsilon$.

Indeed, given $f$ and $\varepsilon$ and the corresponding $\delta$ there is a finite set $E\subseteq X$ such that every $x\in X$ is $\delta$-close to some $x'\in E$. If then $\max\{|f(y)-g(y)|:y\in E\}<\varepsilon$ one gets, for every $x\in X$, that $|f(x)-g(x)|\le |f(x)-f(x')|+|f(x')-g(x')|+|g(x')-g(x)|< 3\varepsilon$. This gives $\|f-g\|\le 3\varepsilon$ where, of course, the factor $3$ is inessential.

I know that this "application" can also be reduced to the classical Arzelà-Ascoli theorem by extending all uniformly continuous functions to the completion of $X$ which is compact.

$\endgroup$
1
  • 1
    $\begingroup$ To state this as a ``philosophical reason or guiding intuition/heuristic'', we can say that the Arzelà-Ascoli theorem is really about uniformly continuous functions. Theorem 37 of Chapter III in "Uniform Spaces" by J.R.Isbell: An equiuniformly continuous, pointwise precompact family of functions on a precompact domain is precompact. On every space which is not precompact there is an equiuniformly continuous family of functions into $[0,1]$ which is not precompact. $\endgroup$
    – user95282
    Mar 20, 2022 at 11:13
0
$\begingroup$

The Azrela Ascoli theorem is true on any space $C^0(Y,Z)$ with $Z$ metric and $Y$ an arbitrary topological space, as soon as $C^0(Y,Z)$ is endowed with the compact open topology. See e.g. Dugundgi, Topology, chap. XII Theorem 6.4.

$\endgroup$
1
  • $\begingroup$ I addressed this in my first "EDIT". $\endgroup$
    – D.R.
    Feb 10, 2022 at 8:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.