Equicontinuity of Families of Functions

Equicontinuity is the working hypothesis behind Arzelà–Ascoli Theorem, which produces compact subsets of $C(K)$ and is the standard machine for proving existence theorems where a naive limiting process alone is not enough. In differential equations, Peano's existence theorem for $y' = f(t,y)$ constructs a sequence of approximate (Euler/polygonal) solutions that are uniformly bounded and equicontinuous — usually via a uniform Lipschitz-type or derivative bound — and extracts a uniformly convergent subsequence converging to an actual solution. In functional analysis, equicontinuity of families of linear operators underlies the Banach–Steinhaus (uniform boundedness) theorem's function-space analogues, and equicontinuous families of maps appear in the study of normal families in complex analysis (Montel's theorem uses a version of this compactness mechanism for holomorphic functions). In dynamical systems, equicontinuity of the family of iterates $\{f^n\}$ characterizes especially tame ("non-chaotic," rotation-like) dynamics, contrasting with sensitive dependence on initial conditions. In machine learning and approximation theory, controlling the "Lipschitz constant" of a family of neural-network functions — a form of imposing equicontinuity — is used to guarantee generalization bounds and stability of learned function classes. In numerical analysis, equicontinuity estimates justify passing to the limit when constructing weak solutions of PDEs via compactness methods (e.g., the Fréchet–Kolmogorov / Riesz–Fréchet–Kolmogorov compactness criterion in $L^p$, a close relative of Arzelà–Ascoli for integrable rather than continuous functions).

:::quiz[In Example 2, $f_n(x) = n^2x/(1+n^4x^2)$ on $[0,1]$, the family fails equicontinuity at $x_0=0$ because:]

• The functions are not individually continuous
• The family is not uniformly bounded
• Spikes of fixed height ½ occur at points $x_n=1/n^2\to 0$, so no single δ controls all $f_n$ near 0 ✓
• The domain $[0,1]$ is not compact

Ask the tutor: How does equicontinuity combine with pointwise convergence on a countable dense set to prove the full Arzelà–Ascoli theorem?

Picture a family of functions as a bundle of curves drawn over the same domain. Ordinary continuity of each $f \in \mathcal{F}$ only tells you that for that one curve, sufficiently close $x$'s give close $f(x)$'s. Equicontinuity is a promise about the bundle as a whole: no matter which curve you pick from the bundle, the same $\delta$ keeps you within $\varepsilon$. Geometrically, near any point $x_0$, you can draw a "cone" of width $2\delta$ and height $2\varepsilon$ centered at $(x_0, f(x_0))$, and every graph in the family stays inside its own copy of that cone as it passes through the point — the slope-control is shared. The family fails to be equicontinuous when functions in it develop arbitrarily steep transitions localized near some point, with the steepness growing without bound as you move through the family — even though each individual function remains perfectly continuous. The classic offending family is $f_n(x) = \sin(nx)$ on $[0,1]$: each $f_n$ is smooth, but the "wiggle rate" $n$ grows without bound, so no single $\delta$ can control all of them near $x = 0$ once $\varepsilon$ is fixed small.

Example 1 (An equicontinuous family via a uniform Lipschitz bound). Let

Claim: $\mathcal{F}$ is (uniformly) equicontinuous on $[0,5]$. Proof. For any $\alpha \in [-3,3]$ and $x, y \in [0,5]$, the Mean Value Theorem applied to $t \mapsto \sin(\alpha t + 1)$ gives a point $\xi$ between $x$ and $y$ with

Since $|\alpha| \le 3$ and $|\cos(\cdot)| \le 1$ always,

This bound is uniform in $\alpha$ — a single Lipschitz constant $L = 3$ works for every member of the family. Given $\varepsilon > 0$, choose $\delta = \varepsilon/3$. Then for all $\alpha \in [-3,3]$ and all $x,y \in [0,5]$ with $|x-y| < \delta$,

This $\delta$ depends only on $\varepsilon$, not on $\alpha$, $x$, or $y$. Hence $\mathcal{F}$ is uniformly equicontinuous. Numerical check. Take $\varepsilon = 0.3$, so $\delta = 0.1$. Consider $\alpha = 2.7$, $x = 1.05$, $y = 1.00$ (so $|x-y| = 0.05 < 0.1$):

Numerically, $\sin(3.835) \approx -0.6499$ and $\sin(3.7) \approx -0.5298$, so the difference is $\approx -0.1201$, and indeed $|{-0.1201}| \approx 0.12 < 0.3 = \varepsilon$, consistent with (in fact comfortably inside) the bound $3 \times 0.05 = 0.15 < 0.3$. This is one instance confirming (not proving) the general estimate. Any general family $\{f_\alpha\}_{\alpha \in A}$ of differentiable functions with $\sup_{\alpha \in A} \sup_{x} |f_\alpha'(x)| \le L < \infty$ is equicontinuous by the identical MVT argument — this is the standard sufficient criterion in practice.

Example 2 (A pointwise-bounded, non-equicontinuous family — equicontinuity genuinely fails). Let $\mathcal{F} = \{f_n\}_{n=1}^\infty$ on $K = [0,1]$ where

Uniform boundedness check. For each fixed $n$, by AM–GM, $1 + n^4x^2 \ge 2n^2 x$ for $x \ge 0$, so $f_n(x) \le \frac{n^2 x}{2 n^2 x} = \tfrac{1}{2}$ for $x>0$ (and $f_n(0)=0$). So $|f_n(x)| \le 1/2$ for all $n$ and all $x \in [0,1]$: the family is uniformly bounded by $M = 1/2$. Claim: $\mathcal{F}$ is not equicontinuous at $x_0 = 0$. Proof. Evaluate at $x_n := 1/n^2$:

Meanwhile $f_n(0) = 0$ for every $n$. So

while

Fix $\varepsilon_0 = 1/4$. For any proposed $\delta > 0$, choose $n$ large enough that $1/n^2 < \delta$ (possible since $1/n^2 \to 0$). Then $x_n$ satisfies $|x_n - 0| < \delta$, but

So no $\delta > 0$ can simultaneously work for all $f_n$ near $x_0 = 0$ at level $\varepsilon_0 = 1/4$: the family fails equicontinuity at $0$, even though it is uniformly bounded and each $f_n$ individually is smooth (indeed real-analytic) on $[0,1]$. This is the mechanism behind the sharp gap the Arzelà–Ascoli Theorem closes: $(f_n)$ is uniformly bounded on the compact set $[0,1]$, yet — as one can check directly — $f_n \to 0$ pointwise on $(0,1]$ while $f_n(1/n^2) = 1/2 \not\to 0$, so $f_n$ does not converge uniformly, and in fact no subsequence converges uniformly (the same spike argument applies to any subsequence, relocating $n$ within it). The failure of a uniformly convergent subsequence is exactly explained by the failure of equicontinuity.