Solving Function-Space Compactness Problems

A Practical Framework for Function-Space Compactness

Compactness questions about a family of continuous functions involve two different kinds of control. At each point of the domain, the functions must not take values that escape to infinity; across the domain, their changes must be controlled uniformly over the family. Arzelà–Ascoli combines these requirements into a useful test. The main task in an application is to choose the topology and then verify each hypothesis for the particular family.

Throughout, \(K\) is a compact metric space with metric \(d\), and \(C(K,\mathbb{R})\) is the set of continuous functions from \(K\) to \(\mathbb{R}\). We use the uniform metric \(d_\infty(f,g)=\sup_{x\in K}|f(x)-g(x)|\). Since \(K\) is compact, every continuous real-valued function on \(K\) is bounded, so this supremum is finite.

Definition: A family \(F\subseteq C(K,\mathbb{R})\) is equicontinuous if for every \(x\in K\) and every \(\varepsilon>0\), there is a \(\delta>0\) such that, for every \(f\in F\) and every \(y\in K\), \(d(x,y)<\delta\) implies \(|f(x)-f(y)|<\varepsilon\). It is pointwise relatively compact if, for each \(x\in K\), the closure in \(\mathbb{R}\) of \(\{f(x):f\in F\}\) is compact.

The quantifiers in equicontinuity matter: the same \(\delta\) must work for every function \(f\in F\), once \(x\) and \(\varepsilon\) have been fixed. Pointwise relative compactness is a separate condition at each \(x\); it does not by itself control how quickly functions can change between nearby points.

Which Function-Space Topology?

For compact metric \(K\), the compact-open topology on \(C(K,\mathbb{R})\) agrees with the topology of the uniform metric \(d_\infty\). This lets us use the uniform version of Arzelà–Ascoli for questions stated in the compact-open topology. Here is a brief verification of the agreement. A compact-open subbasic set has the form \([C,U]=\{g:g(C)\subseteq U\}\), where \(C\subseteq K\) is compact and \(U\subseteq\mathbb{R}\) is open. If \(f\in[C,U]\), the compact set \(f(C)\) lies inside \(U\), so it has positive distance from \(\mathbb{R}\setminus U\) when that complement is nonempty. A sufficiently small uniform ball about \(f\) therefore lies in \([C,U]\); if \(U=\mathbb{R}\), the condition imposes no restriction.

In the other direction, fix a uniform ball of radius \(\varepsilon>0\) about \(f\). By continuity of \(f\) and compactness of \(K\), choose finitely many compact sets \(C_i\) covering \(K\), with points \(x_i\in K\), such that \(|f(y)-f(x_i)|<\varepsilon/3\) for \(y\in C_i\). The compact-open neighborhood \(\bigcap_i[C_i,(f(x_i)-2\varepsilon/3,f(x_i)+2\varepsilon/3)]\) lies in that ball: for \(y\in C_i\), its defining condition gives \(|g(y)-f(x_i)|<2\varepsilon/3\), and hence \(|g(y)-f(y)|<\varepsilon\). Thus the two topologies agree.

Theorem (Arzelà–Ascoli for a compact metric domain): For \(F\subseteq C(K,\mathbb{R})\), the closure of \(F\) in the uniform topology is compact if and only if \(F\) is equicontinuous and pointwise relatively compact.

We use this standard compact-domain form of Arzelà–Ascoli. It gives a two-stage strategy: first verify equicontinuity and pointwise relative compactness to obtain relative compactness; then, if compactness of \(F\) itself is required, also show that \(F\) is closed. Indeed, a closed subset of the compact closure \(\overline{F}\) is compact.

Worked Example: A Compact Family of Quadratic Functions

Let \(K=[0,1]\) with its usual metric and let \(F=\{f_a:a\in[-1,2]\}\), where \(f_a(x)=ax^2\). We check the Arzelà–Ascoli hypotheses and then show that \(F\) is closed.

Equicontinuity. For \(x,y\in[0,1]\) and \(a\in[-1,2]\), \(|a|\le2\), and \(x+y\le2\). Therefore \(|f_a(x)-f_a(y)|=|a||x-y|(x+y)\le4|x-y|\). Given \(\varepsilon>0\), choose \(\delta=\varepsilon/4\). If \(|x-y|<\delta\), the displayed bound is strictly less than \(\varepsilon\), for every \(a\). Thus the family is equicontinuous.

Pointwise relative compactness. Fix \(x\in[0,1]\). Since \(a\) ranges over \([-1,2]\), the value set is \(\{ax^2:a\in[-1,2]\}=[-x^2,2x^2]\), a compact interval. For example, at \(x=1/2\), this interval is \([-1/4,1/2]\), since \((1/2)^2=1/4\).

Arzelà–Ascoli now gives relative compactness. To prove compactness of \(F\), suppose a sequence \(f_{a_j}\) converges uniformly to \(g\). Evaluation at \(1\) gives \(a_j=f_{a_j}(1)\to g(1)=a\). Since every \(a_j\in[-1,2]\) and this interval is closed, \(a\in[-1,2]\). For every \(x\in[0,1]\), \(f_{a_j}(x)=a_jx^2\to ax^2\); uniform convergence also implies pointwise convergence to \(g(x)\). Uniqueness of real limits gives \(g(x)=ax^2=f_a(x)\), so \(g\in F\). Thus \(F\) is closed, and hence compact. As a check, \(f_a(1)=a\) recovers the parameter from the function, so the limit parameter found above is consistent with the family’s definition.

When Pointwise Control Is Not Enough

Worked Example: The Powers \(x^n\)

Consider \(F=\{f_n:n\in\mathbb{N}\}\subseteq C([0,1],\mathbb{R})\), where \(f_n(x)=x^n\). At each fixed \(x\in[0,1]\), all values lie in \([0,1]\). The closure of the value set is a closed subset of the compact interval \([0,1]\), so the family is pointwise relatively compact.

But \(F\) is not equicontinuous at \(1\). For each \(n\in\mathbb{N}\), set \(x_n=2^{-1/n}\). Then \(x_n\in[0,1]\), \(x_n\to1\), and \(|1-x_n|\to0\), while \(|f_n(1)-f_n(x_n)|=|1-(2^{-1/n})^n|=|1-1/2|=1/2\). Thus, with \(\varepsilon=1/2\), no single \(\delta>0\) can work for every \(n\): choose \(n\) large enough that \(|1-x_n|<\delta\), and the required strict inequality fails because the function-value difference equals \(\varepsilon\). Arzelà–Ascoli shows that \(F\) is not relatively compact in the uniform topology.

There is another way to check the obstruction. Pointwise on \([0,1]\), \(x^n\) tends to the function that is \(0\) for \(0\le x<1\) and \(1\) at \(x=1\), which is discontinuous at \(1\). A uniformly convergent sequence of continuous functions has a continuous limit, so no subsequence of \((f_n)\) can converge uniformly: any such subsequence would have the same pointwise limit. This confirms the conclusion from the failed equicontinuity test.

The example separates the two hypotheses. Bounded values at every individual point do not prevent sharper and sharper changes near a point. Conversely, a common modulus of continuity does not prevent functions from taking values that escape to infinity at a fixed point. Both checks are needed.

A Common Lipschitz Bound Gives Compactness

Worked Example: Functions with a Common Lipschitz Bound

Let \(F\) be the family of all \(f\in C([0,1],\mathbb{R})\) satisfying \(|f(0)|\le1\) and \(|f(x)-f(y)|\le3|x-y|\) for all \(x,y\in[0,1]\). We show that \(F\) is compact in the uniform topology.

Equicontinuity. Given \(\varepsilon>0\), take \(\delta=\varepsilon/3\). If \(|x-y|<\delta\), the defining Lipschitz inequality gives \(|f(x)-f(y)|\le3|x-y|<\varepsilon\), for every \(f\in F\).

Pointwise relative compactness. For any \(x\in[0,1]\), \(|f(x)|\le |f(0)|+|f(x)-f(0)|\le1+3x\le4\). Thus at each \(x\), the value set is bounded in \(\mathbb{R}\), and its closure is a closed subset of \([-4,4]\), hence compact. More precisely, at \(x=1/2\), \(|f(1/2)|\le1+3(1/2)=5/2\).

Arzelà–Ascoli gives relative compactness. To see that \(F\) is closed, let \((f_j)\) in \(F\) converge uniformly to \(g\). Then \(|g(0)|=\lim_{j\to\infty}|f_j(0)|\le1\). For any \(x,y\in[0,1]\), pointwise convergence and continuity of absolute value give \(|g(x)-g(y)|=\lim_{j\to\infty}|f_j(x)-f_j(y)|\le3|x-y|\). So \(g\in F\). In a metric space, a set containing the limits of all its convergent sequences is closed; therefore \(F\) is closed. Since it is both closed and relatively compact, it is compact.

Check: the bound at \(x=1/2\) uses \(|f(0)|\le1\) and \(3|1/2-0|=3/2\), giving \(|f(1/2)|\le1+3/2=5/2\). The bound is uniform over the entire family, as required for pointwise relative compactness.

A Pitfall and a Working Checklist

Common mistake. A tempting but incorrect calculation for the powers example is to use the derivative \(f_n'(x)=nx^{n-1}\) and claim a common Lipschitz bound. The derivative depends on \(n\), and at \(x=1\) its value is \(n\), not a constant independent of the family. A bound for each separate function is not the uniform bound required for equicontinuity. The explicit test points \(x_n=2^{-1/n}\) show the failure directly: they approach \(1\), but \(|f_n(1)-f_n(x_n)|=1/2\) for every \(n\).

For a new family \(F\subseteq C(K,\mathbb{R})\), with \(K\) compact metric, use this sequence of checks:

Identify the topology. In this setting, compact-open and uniform topologies agree.
For each fixed \(x\), bound or otherwise identify \(\{f(x):f\in F\}\). In \(\mathbb{R}\), boundedness suffices for relative compactness.
Find a common continuity estimate. A uniform Lipschitz bound is one option; test possible failures using nearby points and functions that may vary with those points.
Apply Arzelà–Ascoli to conclude relative compactness only after both hypotheses hold.
If the question asks whether \(F\) itself is compact, verify that \(F\) is closed as well.

Check Your Understanding

For each question, specify whether you are testing relative compactness or compactness in the uniform topology.

For \(f_a(x)=a x\) on \([0,1]\), where \(a\in[0,1]\), verify equicontinuity and pointwise relative compactness.
For \(f_n(x)=x^n\), identify which Arzelà–Ascoli hypothesis fails and at which point equicontinuity fails.
Explain why a common Lipschitz bound together with a uniform bound on \(f(0)\) gives pointwise boundedness on \([0,1]\).
Suppose a family is equicontinuous and pointwise relatively compact. What does Arzelà–Ascoli imply, and what additional property would show that the family itself is compact?
Why does pointwise convergence of a sequence in \(C([0,1],\mathbb{R})\) not by itself establish convergence in the uniform topology?