Skip to main content

Section 2.6 Equicontinuous Family of Functions

Definition 2.6.1. Equicontinuous Family.

A family \(\mathcal F\) of functions \(f\) defined on a set \(E\) in a metric space \((X, d)\) is said to be equicontinuous on \(E\) if for every \(\epsilon > 0\) there exists some \(\delta > 0\) such that
\begin{equation} |f(x)-f(y)| < \epsilon \text{ whenever } d(x, y) < \delta, x, y \in E, \text{ and } f \in \mathcal F.\tag{2.6.1} \end{equation}

Remark 2.6.2. On the notion of equicontinuous family of functions.

A fundamental property of the set of real numbers \(\mathbb R\) is the Bolzano-Weierstrass Theorem: any bounded sequence in \(\mathbb R\) has a convergent subsequence. One would like to find an appropriate extension of this property on function spaces such as the set \(C(E)\) of continuous functions on a metric space \(E\text{.}\) Unfortunately, the direct extension does not hold, as demonstrated by the sequence of functions \(\left\{f_{n}(x)=\sin n x\right\}\) in \(C([0,2\pi])\text{.}\)
Using \(\int_{0}^{2\pi} |\sin nx -\sin m x|^{2}\, dx =2\pi\) when \(n\ne m\text{,}\) it’s clear that \(\left\{\sin n x\right\}\) can’t have a subsequence converging uniformly on \([0, 2\pi]\text{.}\) In fact it can’t have a subsequence converging pointwise on \([0, 2\pi]\text{.}\)
In the situation of the example, what one can directly extend is the following property: there exists a countable dense subset \(F\) of \([0,2\pi]\) and a subsequence \(\left\{ f_{n_{k}}(x) \right\}\) such that it converges at every \(x\in F\text{.}\) To be able to say that one can choose a subsequence \(\left\{ f_{n_{k}}(x) \right\}\) such that it converges at every \(x\in [0,2\pi]\text{,}\) one needs to control the behavior of \(\left\{ f_{n_{k}}(x) \right\}\) for \(x\in [0,2\pi]\setminus F\text{.}\) The condition of being equicontinuous is the one what would give us the desired property.
Heuristically, the equicontinuous condition guarantees that the oscillation of \(f_{n}(x)\) to be smaller than \(\epsilon > 0\) for \(x\) over any neighborhood \(V\) of suitably small diameter, uniformly in \(n\text{.}\) This allows us to propagate the property
\begin{equation*} |f_{n_{i}}(p)-f_{n_{j}}(p)| < \epsilon \text{ for one point } p \in V \end{equation*}
to
\begin{equation*} |f_{n_{i}}(x)-f_{n_{j}}(x)| < 3\epsilon \text{ for all } x \in V. \end{equation*}
Using the total boundedness of \(E=[a, b]\text{,}\) this shows that \(\{ f_{n_{i}}(x)\}\) is uniformly Cauchy on \(E\text{.}\)

Definition 2.6.3.

A sequence of functions \(\{f_{n}\} \) defined on \(E\) is said to be pointwise bounded on \(E\) if for every \(x\in E\) the sequence of scalars \(\{f_{n}(x)\} \) is bounded. \(\{f_{n}\} \) is said to be uniformly bounded on \(E\) if there exists \(M > 0\) such that
\begin{equation*} |f_{n}(x)| < M \text{ for all } n \text{ and } x\in E. \end{equation*}

Proof.

Let \(\{p_{i}\}\text{,}\) \(i=1,\cdots\text{,}\) be the points of \(C\) arranged in a sequence. Then \(\{f_{n}(p_{1})\} \) is a bounded sequence, therefore, has a convergent subsequence, say \(\{f_{n_{1,k}}(p_{1})\} \text{.}\) Next we pick a convergent subsequence \(\{f_{n_{2, k}}(p_{2})\} \) from the bounded sequence \(\{f_{n_{1,k}}(p_{2})\} \text{.}\) Continuing in this fashion for each \(p_{i}\text{,}\) we obtain \(\{f_{n_{i, k}}\}\) such that for each \(i\text{,}\) \(\{f_{n_{i, k}}(p_{j})\}\) converges for each \(j\le i\) as \(k\to \infty\text{.}\)
Finally, \(\{f_{n_{i, i}}\}\) is a subsequence of \(\{f_{n}\} \) which converges at each \(x_{j}\text{.}\)

Proof.

For any \(\epsilon > 0\text{,}\) let \(\delta > 0\) be such that (2.6.1) holds. Since \(K\) is compact, it can be covered by a finite number of sets of diameter \(< \delta\text{,}\) say, \(V_{1}, \cdots, V_{m}\text{.}\)
Pick \(p_{i} \in V_{i}\) for each \(i=1,\cdots, m\text{.}\) Then for each \(i=1,\cdots,m\text{,}\) there exists some \(M_{i} > 0\) such that \(|f_{n}(p_{i})|\le M_{i}\) for all \(n\text{.}\) Any \(x\in V_{i}\) satisfies
\begin{equation*} |f_{n}(x)-f_{n}(p_{i})| < \epsilon, \text{ therefore } |f_{n}(x)|\le M_{i}+\epsilon. \end{equation*}
Let \(M=\max_{1\le i\le m}M_{i}\text{.}\) Then any \(x\in K\) satisfies \(|f_{n}(x)|\le M+\epsilon\) for all \(n\text{,}\) which shows (a).
The compactness of \(K\) implies that it has a dense countable subset \(\{q_{i}\}\text{.}\) It then follows from PropositionΒ 2.6.4 that we can pick a subsequence \(\{f_{n_{k}}\}\) such that \(\{f_{n_{k}}(q_{i})\}\) converges for each \(q_{i}\text{,}\) \(i=1,\cdots\text{.}\)
Each \(V_{i}\) contains some \(q_{j_{i}}\text{.}\) It follows that there exists some \(N\) such that \(|f_{n_{k}}(q_{j_i})-f_{n_{l}}(q_{j_i})| < \epsilon\) for all \(k, l\ge N\) and \(i=1,\cdots, m\text{.}\)
Using the finite cover \(\{V_{i}\}_{i=1}^{m}\) of \(K\) and (2.6.1) on \(\{f_{n_{k}}\}\text{,}\) we find that for any \(x\in K\text{,}\) \(d(x, q_{j_i}) < \delta \) for some \(i=1,\cdots, m\text{,}\) therefore,
\begin{align*} \amp |f_{n_{k}}(x)-f_{n_{l}}(x)|\\ \le \amp |f_{n_{k}}(x)-f_{n_{k}}(q_{j_i})|+| f_{n_{k}}(q_{j_i})-f_{n_{l}}(q_{j_i})| + |f_{n_{l}}(q_{j_i})-f_{n_{l}}(x)| \le 3\epsilon \end{align*}
for all \(k, l\ge N\text{.}\) This show that \(\{f_{n_{k}}\}\) is uniformly Cauchy on \(K\text{,}\) therefore proving (b).

Remark 2.6.6.

The Bolzano-Weierstrass/Ascolli-Arzel property has extensions to function spaces where the convergence is not uniform but in integral norms. We first describe a generalization in the context of the space \(l^{p}\text{,}\) defined for \(p < \infty\) as the space of infinite sequences \(\mathbf x:=\{\mathbf x(k)\}_{k=1}^{\infty}\) such that
\begin{equation*} \Vert \mathbf x\Vert_{p} :=\left(\sum_{k=1}^{\infty} |\mathbf x(k)|^{p}\right)^{1/p} < \infty, \end{equation*}
we note that, for \(p < \infty\text{,}\) if a sequence \(\left\{{\mathbf x}_{n}\right\} \subset l^{p}\) converges to some \({\mathbf x}_{\infty}\) in \(l^{p} \text{,}\) then for any \(\epsilon >0\text{,}\) there exists \(L\) such that
\begin{equation*} \left(\sum_{k=L}^{\infty} |{\mathbf x}_{\infty}(k)|^{p} \right)^{1/p} < \epsilon, \end{equation*}
and there exists some \(N\) such that for \(n\ge N\text{,}\)
\begin{equation*} ||{\mathbf x}_{n}-{\mathbf x}_{\infty}||_{p}= \left(\sum_{k=1}^{\infty} |{\mathbf x}_{n}(k)-{\mathbf x}_{\infty}(k)|^{p} \right)^{1/p} < \epsilon. \end{equation*}
This then implies, by Minkowski’s inequality, that
\begin{equation*} \left(\sum_{k=L}^{\infty} |{\mathbf x}_{n}(k)|^{p} \right)^{1/p} \le \left(\sum_{k=L}^{\infty} |{\mathbf x}_{n}(k)-{\mathbf x}_{\infty}(k)|^{p} \right)^{1/p} + \left(\sum_{k=L}^{\infty} |{\mathbf x}_{\infty}(k)|^{p} \right)^{1/p} < 2\epsilon. \end{equation*}
For the finite number of elements \(\left\{{\mathbf x}_{n}\right\}_{n=1}^{N-1}\text{,}\) we can certainly find some \(L' \ge L\) such that, for \(n=1, \ldots, N-1\text{,}\)
\begin{equation*} \left(\sum_{k=L'}^{\infty} |{\mathbf x}_{n}(k)|^{p} \right)^{1/p} < \epsilon, \end{equation*}
so
\begin{equation*} \left(\sum_{k=L'}^{\infty} |{\mathbf x}_{n}(k)|^{p} \right)^{1/p} < 2\epsilon, \forall \;n. \end{equation*}
This turns out to be also a sufficient condition in order to extract a convergent subsequence in \(l^{p}\text{.}\) We formulate the condition more precisely in the following.

Definition 2.6.7. Equi-summable Family in \(l^{p}\).

A family \(\left\{{\mathbf x}_{s}\right\}_{s\in I}\) of elements in \(l^{p}\) is said to be equi-summable in \(l^{p}\text{,}\) if for any \(\epsilon >0\text{,}\) there exists \(L\) such that
\begin{equation} \left(\sum_{k=L}^{\infty} |{\mathbf x}_{s}(k)|^{p} \right)^{1/p} < \epsilon \; \forall \; s\in I.\tag{2.6.2} \end{equation}

Exercise 2.6.9.

Exercise 2.6.10.