In analysis, one frequently encounters sequences of functions and needs to extract convergent subsequences. For sequences of real numbers, the Bolzano--Weierstrass theorem guarantees that every bounded sequence has a convergent subsequence. For sequences of continuous functions, however, boundedness alone is not enough. The family $\mathcal{F} = \{f_n\}_{n=1}^\infty$ defined by

\begin{align*} f_n: [0, 1] &\to \mathbb{R} \\ x &\mapsto \sin(n\pi x) \end{align*}

is uniformly bounded --- $\|f_n\|_\infty \le 1$ for all $n \in \mathbb{N}$ --- yet no subsequence converges uniformly. The oscillations of $f_n$ become increasingly rapid as $n$ grows, and no uniform limit can tame them. What additional condition on the family $\mathcal{F}$ would prevent this pathological oscillation and guarantee that bounded sequences have uniformly convergent subsequences?

The answer is **equicontinuity**: a condition that controls how much functions in a family can oscillate, uniformly across the entire family. Whereas [uniform continuity](/page/Uniform%20Continuity) asks for a single function's modulus of continuity to be independent of the point, equicontinuity asks for the modulus to be independent of both the point and the choice of function in the family. This notion, introduced by Ascoli (1884) and refined by Arzelà (1895), provides the precise characterisation of relatively compact subsets of $C(K)$ and lies at the heart of compactness arguments throughout analysis, PDE theory, and functional analysis.

[example: Oscillation Obstruction] Consider the family $\mathcal{F} = \{f_n\}_{n=1}^\infty$ where $f_n: [0,1] \to \mathbb{R}$ is defined by $f_n(x) = \sin(n\pi x)$. Each $f_n$ is continuous and satisfies $|f_n(x)| \le 1$ for all $x \in [0,1]$, so the family is uniformly bounded. However, fix $x_0 = 0$ and consider the point $x = 1/(2n)$. Then \begin{align*} |f_n(x) - f_n(x_0)| &= |\sin(n\pi \cdot \tfrac{1}{2n}) - \sin(0)| \\ &= |\sin(\tfrac{\pi}{2})| \\ &= 1. \end{align*} The distance $|x - x_0| = 1/(2n)$ can be made arbitrarily small by choosing $n$ large, yet the function values differ by $1$. For any $\delta > 0$, the function $f_n$ with $n > 1/(2\delta)$ satisfies $|x - x_0| < \delta$ but $|f_n(x) - f_n(x_0)| = 1$. No single $\delta$ works for all $f_n$ simultaneously, so the family fails to be equicontinuous at $x_0 = 0$. This oscillation is precisely what prevents the extraction of a uniformly convergent subsequence. [/example]

## Definition

The central question is: when can we choose a modulus of continuity that works for every function in a family simultaneously? This leads to two notions, depending on whether the uniformity is local or global.

[definition: Equicontinuous Family] Let $(X, d_X)$ and $(Y, d_Y)$ be metric spaces, and let $\mathcal{F}$ be a family of functions from $X$ to $Y$. **Pointwise equicontinuity.** The family $\mathcal{F}$ is *equicontinuous at a point* $x_0 \in X$ if for every $\varepsilon > 0$ there exists $\delta > 0$ such that \begin{align*} d_X(x, x_0) < \delta \implies d_Y(f(x), f(x_0)) < \varepsilon \quad \text{for all } f \in \mathcal{F}. \end{align*} The family $\mathcal{F}$ is *pointwise equicontinuous* (or simply *equicontinuous*) if it is equicontinuous at every point $x_0 \in X$. **Uniform equicontinuity.** The family $\mathcal{F}$ is *uniformly equicontinuous* if for every $\varepsilon > 0$ there exists $\delta > 0$ such that \begin{align*} d_X(x, y) < \delta \implies d_Y(f(x), f(y)) < \varepsilon \quad \text{for all } f \in \mathcal{F} \text{ and all } x, y \in X. \end{align*} [/definition]

The distinction between these two notions mirrors the distinction between continuity and [uniform continuity](/page/Uniform%20Continuity): in pointwise equicontinuity, the $\delta$ may depend on $x_0$ (but not on $f$); in uniform equicontinuity, $\delta$ depends only on $\varepsilon$. For a single function, pointwise equicontinuity reduces to ordinary continuity, and uniform equicontinuity reduces to uniform continuity. The power of equicontinuity is that the modulus is shared across the entire family.

When $X$ is a [compact metric space](/page/Compact%20Space), the two notions coincide for families of functions into $\mathbb{R}$ (or any metric space), just as continuity and uniform continuity coincide for a single function on a compact domain. On non-compact domains, however, uniform equicontinuity is strictly stronger.

[remark: Equicontinuity vs Uniform Continuity] Consider the hierarchy of uniformity conditions for a family $\mathcal{F}$ of functions $f: X \to Y$: 1. Each $f \in \mathcal{F}$ is continuous: $\delta$ depends on $f$, $x_0$, and $\varepsilon$. 2. The family $\mathcal{F}$ is equicontinuous: $\delta$ depends on $x_0$ and $\varepsilon$, but not on $f$. 3. Each $f \in \mathcal{F}$ is uniformly continuous: $\delta$ depends on $f$ and $\varepsilon$, but not on $x_0$. 4. The family $\mathcal{F}$ is uniformly equicontinuous: $\delta$ depends only on $\varepsilon$. Equicontinuity and uniform continuity are independent conditions --- neither implies the other for a general family on a non-compact domain. Equicontinuity makes $\delta$ uniform over functions; uniform continuity makes it uniform over points. Uniform equicontinuity achieves both. [/remark]

## The Arzelà--Ascoli Theorem

The most fundamental question about equicontinuity is: *what does it buy us?* The Bolzano--Weierstrass theorem says that bounded sets in $\mathbb{R}^n$ are relatively compact --- every sequence has a convergent subsequence. In the infinite-dimensional space $C(K)$ of continuous functions on a [compact space](/page/Compact%20Space) $K$, bounded sets are *not* relatively compact. The closed unit ball in $C([0,1])$, for instance, contains the sequence $f_n(x) = x^n$, which converges pointwise to a discontinuous function and has no uniformly convergent subsequence. Something beyond boundedness is needed.

The Arzelà--Ascoli theorem identifies equicontinuity as exactly the missing ingredient. Together with pointwise boundedness, equicontinuity characterises the relatively compact subsets of $C(K)$.

### Failure of Compactness Without Equicontinuity

To appreciate why equicontinuity is necessary, consider two families that are bounded but not equicontinuous, each illustrating a different mechanism of failure.

[example: Power Functions on the Unit Interval] Define $f_n: [0,1] \to \mathbb{R}$ by $f_n(x) = x^n$ for $n \in \mathbb{N}$. Each $f_n$ is continuous and $\|f_n\|_\infty = 1$, so the family $\{f_n\}$ is uniformly bounded. The pointwise limit is \begin{align*} f(x) = \lim_{n \to \infty} x^n = \begin{cases} 0 & \text{if } 0 \le x < 1, \\ 1 & \text{if } x = 1. \end{cases} \end{align*} This limit is discontinuous, so the convergence cannot be uniform. To verify that equicontinuity fails at $x_0 = 1$, fix $\varepsilon = 1/2$ and let $\delta > 0$. Choose $x_\delta = 1 - \delta/2 \in (1-\delta, 1)$ and $n_\delta = \lceil 2\log 2 / \delta \rceil$. Then $|x_\delta - 1| = \delta/2 < \delta$, but \begin{align*} |f_{n_\delta}(x_\delta) - f_{n_\delta}(1)| &= |x_\delta^{n_\delta} - 1| = 1 - (1 - \tfrac{\delta}{2})^{n_\delta}. \end{align*} Using the inequality $(1 - t)^n \le e^{-nt}$ for $t \in (0,1)$, we obtain \begin{align*} (1 - \tfrac{\delta}{2})^{n_\delta} \le e^{-n_\delta \cdot \delta/2} \le e^{-\log 2} = \tfrac{1}{2}, \end{align*} so $|f_{n_\delta}(x_\delta) - f_{n_\delta}(1)| \ge 1/2 = \varepsilon$. No $\delta$ works for all $n$ simultaneously, confirming that the family is not equicontinuous at $x_0 = 1$. The concentration of the graphs near the corner $(1,1)$ --- where the functions transition from near $0$ to exactly $1$ in an interval of length $O(1/n)$ --- is the geometric manifestation of this failure. [/example]

The power function example fails because all oscillation concentrates at a single boundary point. A fundamentally different failure mechanism arises when the family "drifts" spatially --- the functions are individually well-behaved, but their supports escape every compact set.

[example: Translating Bump Functions] Let $\varphi \in C_c^\infty(\mathbb{R})$ satisfy $\operatorname{supp} \varphi \subset [-1, 1]$ and $\|\varphi\|_\infty = 1$. Define $f_n: \mathbb{R} \to \mathbb{R}$ by $f_n(x) = \varphi(x - n)$. Each $f_n$ is smooth with $\|f_n\|_\infty = 1$, and the family is uniformly bounded. The supports $\operatorname{supp} f_n = [n-1, n+1]$ are pairwise disjoint for $n \ge 3$, so \begin{align*} \|f_n - f_m\|_\infty = 1 \quad \text{for all } n \ne m \text{ with } |n-m| \ge 2. \end{align*} No subsequence can be Cauchy in $C_b(\mathbb{R})$. The family is equicontinuous --- since all functions are translates of the same smooth bump, they share the same modulus of continuity --- but it is not *pointwise bounded* in the sense needed for Arzelà--Ascoli. At each point $x_0$, eventually $f_n(x_0) = 0$ for all large $n$, so the sequence converges pointwise to $0$, but the convergence is not uniform. This example shows that equicontinuity alone, without a compactness condition on the domain or a total boundedness condition, does not suffice. [/example]

### Statement and Discussion