The Arzela-Ascoli Theorem

⚡ Key Takeaways

The Arzelà–Ascoli Theorem characterizes the relatively compact subsets of C(K)C(K), the space of continuous functions on a compact space: a family is relatively compact (in the uniform norm) if and only if it is uniformly bounded and equicontinuous. It is the principal tool for extracting uniformly convergent subsequences. Applies in: ODE/PDE existence theory, functional analysis, harmonic analysis, dynamical systems. Originated with Cesare Arzelà and Giulio Ascoli, Italy, 1880s–1890s.

💡 The Core Insight

Boundedness gives you control at each point; equicontinuity transfers that control uniformly across points. The marriage of the two lets a diagonal argument pick out a subsequence that converges first on a countable dense set, and then equicontinuity spreads that pointwise convergence into uniform convergence on the whole compact domain. Compactness = (pointwise pre-compactness on a dense set) + (uniform regularity that interpolates the rest).

Formal Statement

DefinitionEquicontinuity

Let (X,d)(X,d) and (Y,ρ)(Y,\rho) be metric spaces and F⊆C(X,Y)\mathcal{F} \subseteq C(X,Y) a family of continuous functions.

  • F\mathcal{F} is equicontinuous at a point x0∈Xx_0 \in X if
1
∀ε>0    ∃δ>0    ∀f∈F:  d(x,x0)<δ  ⟹  ρ(f(x),f(x0))<ε.\quad \forall \varepsilon>0 \;\; \exists \delta>0 \;\; \forall f\in\mathcal{F}: \; d(x,x_0)<\delta \implies \rho\big(f(x),f(x_0)\big)<\varepsilon.

Derivation hint: This is ordinary continuity at x0x_0, but with a single δ\delta valid for ALL ff simultaneously — the quantifier ∀f\forall f sits inside, not outside.

  • F\mathcal{F} is (pointwise) equicontinuous if it is equicontinuous at every x0∈Xx_0\in X.
  • F\mathcal{F} is uniformly equicontinuous if δ\delta may be chosen independent of x0x_0:
2
∀ε>0    ∃δ>0    ∀f∈F    ∀x,x′∈X:  d(x,x′)<δ  ⟹  ρ(f(x),f(x′))<ε.\quad \forall \varepsilon>0 \;\; \exists \delta>0 \;\; \forall f\in\mathcal{F}\;\;\forall x,x'\in X: \; d(x,x')<\delta \implies \rho\big(f(x),f(x')\big)<\varepsilon.

Derivation hint: Move the ∀x0\forall x_0 out from after ∃δ\exists\delta to before it; on a compact XX the two notions coincide (proved below).

DefinitionUniform Boundedness and Pointwise Boundedness

F⊆C(X,Rn)\mathcal{F} \subseteq C(X,\mathbb{R}^n) is:

  • pointwise bounded if for each xx, sup⁡f∈F∥f(x)∥<∞\sup_{f\in\mathcal{F}}\|f(x)\| < \infty;
  • uniformly bounded if sup⁡f∈Fsup⁡x∈X∥f(x)∥=sup⁡f∈F∥f∥∞<∞\sup_{f\in\mathcal{F}}\sup_{x\in X}\|f(x)\| = \sup_{f\in\mathcal F}\|f\|_\infty < \infty.

On a compact XX, pointwise boundedness + equicontinuity automatically upgrades to uniform boundedness.

TheoremArzelà–Ascoli, Classical Form

Let KK be a compact metric space and let {fn}n≥1⊆C(K,Rm)\{f_n\}_{n\ge 1} \subseteq C(K,\mathbb{R}^m). If 1. {fn}\{f_n\} is uniformly bounded: sup⁡n∥fn∥∞<∞\sup_n \|f_n\|_\infty < \infty, and 2. {fn}\{f_n\} is equicontinuous, then {fn}\{f_n\} has a subsequence converging uniformly on KK (i.e. in the norm ∥⋅∥∞\|\cdot\|_\infty).

TheoremArzelà–Ascoli, Compactness Form

Let KK be a compact Hausdorff space. A subset F⊆C(K,Rm)\mathcal{F} \subseteq C(K,\mathbb{R}^m) is relatively compact (its closure is compact) in (C(K),∥⋅∥∞)(C(K),\|\cdot\|_\infty) if and only if F\mathcal{F} is pointwise bounded and equicontinuous. Equivalently: F\mathcal{F} is relatively compact   ⟺  \iff it is uniformly bounded and equicontinuous. Since C(K)C(K) is a complete metric space, relative compactness   ⟺  \iff total boundedness of F\mathcal F.

TheoremEquicontinuity on Compact Domains

If KK is compact and F⊆C(K,Y)\mathcal F\subseteq C(K,Y) is pointwise equicontinuous, then F\mathcal F is uniformly equicontinuous.

Why It Holds — Derivation

ProofProof of the Classical Form: Diagonal + Spreading

Step 1 — A countable dense set. Since KK is a compact metric space, it is separable. Choose a countable dense subset D={x1,x2,x3,… }⊆KD=\{x_1,x_2,x_3,\dots\}\subseteq K.

Step 2 — Diagonal extraction of pointwise convergence on DD. At x1x_1, the sequence {fn(x1)}\{f_n(x_1)\} lies in a bounded subset of Rm\mathbb{R}^m (by uniform boundedness), hence by Bolzano–Weierstrass has a convergent subsequence {fn(1)}\{f_{n}^{(1)}\}. From it, the values at x2x_2 are bounded, giving a further subsequence {fn(2)}\{f_{n}^{(2)}\} converging at x2x_2 (and still at x1x_1). Iterate. Define the diagonal sequence

3
gn:=fn(n).\quad g_n := f_{n}^{(n)}.

Derivation hint: {gn}n≥k\{g_n\}_{n\ge k} is, after the kk-th term, a subsequence of {fn(k)}\{f_n^{(k)}\}, so it converges at xkx_k for every kk — the classic Cantor diagonal trick. Thus gn(x)g_n(x) converges for every x∈Dx\in D.

Step 3 — Cauchy estimate via equicontinuity (the spreading). Fix ε>0\varepsilon>0. By equicontinuity (uniform, since KK compact) pick δ>0\delta>0 so that

4
d(x,x′)<δ  ⟹  ∥gn(x)−gn(x′)∥<ε3for all n.\quad d(x,x')<\delta \;\Longrightarrow\; \|g_n(x)-g_n(x')\| < \tfrac{\varepsilon}{3}\quad\text{for all }n.

Derivation hint: Use Definition [2] with the single δ\delta valid for the whole family {gn}⊆F\{g_n\}\subseteq\mathcal F. By compactness, KK is covered by finitely many balls B(yj,δ)B(y_j,\delta), j=1,…,Nj=1,\dots,N, with yj∈Dy_j\in D (density lets us put centers in DD).

Step 4 — Uniform Cauchy. For each jj, {gn(yj)}\{g_n(y_j)\} converges, hence is Cauchy: choose MM such that n,p≥M  ⟹  ∥gn(yj)−gp(yj)∥<ε/3n,p\ge M \implies \|g_n(y_j)-g_p(y_j)\| < \varepsilon/3 for all j=1,…,Nj=1,\dots,N. For arbitrary x∈Kx\in K, pick yjy_j with d(x,yj)<δd(x,y_j)<\delta. Then for n,p≥Mn,p\ge M:

5
∥gn(x)−gp(x)∥≤∥gn(x)−gn(yj)∥⏟<ε/3+∥gn(yj)−gp(yj)∥⏟<ε/3+∥gp(yj)−gp(x)∥⏟<ε/3<ε.\quad \|g_n(x)-g_p(x)\| \le \underbrace{\|g_n(x)-g_n(y_j)\|}_{<\varepsilon/3}+\underbrace{\|g_n(y_j)-g_p(y_j)\|}_{<\varepsilon/3}+\underbrace{\|g_p(y_j)-g_p(x)\|}_{<\varepsilon/3} < \varepsilon.

Derivation hint: Triangle inequality; first and third terms controlled by [4], middle by Step 4's pointwise Cauchy choice — and jj depends on xx but the bound MM does not. Therefore {gn}\{g_n\} is uniformly Cauchy. Since (C(K),∥⋅∥∞)(C(K),\|\cdot\|_\infty) is complete, gn→gg_n \to g uniformly, and g∈C(K)g\in C(K) (uniform limit of continuous functions is continuous). ■\qquad\blacksquare

∎
ProofProof that Compactness ⇒\Rightarrow the Two Conditions (Necessity)

Suppose F‾\overline{\mathcal{F}} is compact in C(K)C(K).

Bounded. A compact subset of a metric space is bounded, so sup⁡f∈F∥f∥∞<∞\sup_{f\in\mathcal F}\|f\|_\infty<\infty — uniform boundedness.

Equicontinuous. Compactness ⇒\Rightarrow total boundedness: given ε>0\varepsilon>0 there exist h1,…,hp∈Fh_1,\dots,h_p\in\mathcal F forming an ε/3\varepsilon/3-net, i.e. every f∈Ff\in\mathcal F satisfies ∥f−hk∥∞<ε/3\|f-h_k\|_\infty<\varepsilon/3 for some kk. Each hkh_k is uniformly continuous on compact KK, so there is δk\delta_k with d(x,x′)<δk⇒∥hk(x)−hk(x′)∥<ε/3d(x,x')<\delta_k \Rightarrow \|h_k(x)-h_k(x')\|<\varepsilon/3. Let δ=min⁡kδk\delta=\min_k\delta_k. Then for any ff and d(x,x′)<δd(x,x')<\delta:

6
∥f(x)−f(x′)∥≤∥f(x)−hk(x)∥+∥hk(x)−hk(x′)∥+∥hk(x′)−f(x′)∥<ε.\quad \|f(x)-f(x')\| \le \|f(x)-h_k(x)\| + \|h_k(x)-h_k(x')\| + \|h_k(x')-f(x')\| < \varepsilon.

Derivation hint: The finitely many hkh_k "anchor" the whole family; total boundedness converts a continuity statement for each anchor into equicontinuity for all of F\mathcal F. ■\qquad\blacksquare

∎
ProofProof of Equicontinuity on Compact Domains

Assume F\mathcal F pointwise equicontinuous on compact KK; fix ε>0\varepsilon>0. For each x∈Kx\in K pick δx>0\delta_x>0 so that d(z,x)<δx⇒∥f(z)−f(x)∥<ε/2d(z,x)<\delta_x \Rightarrow \|f(z)-f(x)\|<\varepsilon/2 for all ff. The balls {B(x,δx/2)}\{B(x,\delta_x/2)\} cover KK; extract a finite subcover with centers x1,…,xNx_1,\dots,x_N and set δ=12min⁡iδxi\delta=\tfrac12\min_i\delta_{x_i}. If d(x,x′)<δd(x,x')<\delta, then x∈B(xi,δxi/2)x\in B(x_i,\delta_{x_i}/2) for some ii, and d(x′,xi)≤d(x′,x)+d(x,xi)<δ+δxi/2≤δxid(x',x_i)\le d(x',x)+d(x,x_i)<\delta+\delta_{x_i}/2\le\delta_{x_i}, so both lie within δxi\delta_{x_i} of xix_i; hence

7
∥f(x)−f(x′)∥≤∥f(x)−f(xi)∥+∥f(xi)−f(x′)∥<ε∀f∈F.\quad \|f(x)-f(x')\| \le \|f(x)-f(x_i)\| + \|f(x_i)-f(x')\| < \varepsilon \quad \forall f\in\mathcal F.

Derivation hint: A Lebesgue-number / finite-subcover argument; the same trick that promotes continuity to uniform continuity on a compact set. ■\qquad\blacksquare

∎
CA
Cesare Arzelà & Giulio Ascoli
1847–1912, 1843–1896

Giulio Ascoli (Trieste, Milan) introduced the notion of equicontinuity in 1883–84 while studying the question of when a set of curves admits a "limit curve," giving the sufficiency of equicontinuity for extracting convergent sequences. Cesare Arzelà (Bologna) sharpened and completed the theory in the 1880s–1895, establishing the necessary and sufficient characterization and applying it to the existence of solutions of differential equations and to the calculus of variations. Arzelà's work on the term-by-term integration of series and on functions of lines (functionals) places him among the founders of functional analysis, alongside Vito Volterra. The theorem fuses their contributions: Ascoli's concept, Arzelà's completeness.

Building Intuition

🧠 Intuition

Picture each fnf_n as a wiggly graph over KK. Uniform boundedness keeps all the graphs trapped in a horizontal band — they can't run off to infinity. Equicontinuity caps how fast any of them can wiggle: no matter which function you pick, moving a little in xx changes the height by only a little, with the same "little" working for the whole family. A bounded but wildly oscillating family (frequencies →∞\to\infty) escapes compactness because no subsequence settles down. Equicontinuity forbids the runaway oscillation, so the graphs are "tame" enough that a diagonal-selected subsequence — which already agrees at a dense scaffold of points — has nowhere to misbehave in the gaps. In finite dimensions, bounded ⇒\Rightarrow relatively compact (Heine–Borel). In C(K)C(K) that fails; equicontinuity is exactly the extra ingredient that restores it.

Figure
Equicontinuity tames oscillation: bounded-but-wild sin⁡(nx)\sin(nx) escapes compactness, while an equicontinuous family is trapped in a controllable tube.
Bounded but NOT equicontinuous +1 -1 sin(nx): frequency → ∞, no uniform limit Equicontinuous family tube width → 0 all graphs share one modulus of continuity ⇒ uniformly convergent subsequence exists

Worked Examples

ExampleExtracting a Convergent Subsequence from an Integral Family

Given. For n≥1n\ge 1 define fn:[0,1]→Rf_n:[0,1]\to\mathbb{R} by

fn(x)=∫0xcos⁡(nt)1+t2 dt.f_n(x)=\int_0^x \frac{\cos(n t)}{1+t^2}\,dt.

Question. Show {fn}\{f_n\} has a uniformly convergent subsequence and estimate the equicontinuity modulus.

Step 1 — Uniform bound. Since ∣cos⁡(nt)/(1+t2)∣≤1|\cos(nt)/(1+t^2)|\le 1,

∣fn(x)∣≤∫011 dt=1⇒∥fn∥∞≤1.|f_n(x)|\le \int_0^1 1\,dt = 1 \quad\Rightarrow\quad \|f_n\|_\infty \le 1.

So the family is uniformly bounded by 11.

Step 2 — Equicontinuity via a common Lipschitz constant. For x<x′x<x',

∣fn(x′)−fn(x)∣=∣∫xx′cos⁡(nt)1+t2dt∣≤∣x′−x∣.|f_n(x')-f_n(x)| = \left|\int_x^{x'}\frac{\cos(nt)}{1+t^2}dt\right| \le |x'-x|.

Every fnf_n is 11-Lipschitz, so given ε\varepsilon choose δ=ε\boxed{\delta=\varepsilon} (independent of nn): equicontinuous.

Step 3 — Conclude. By Arzelà–Ascoli, a subsequence fnk→gf_{n_k}\to g uniformly on [0,1][0,1]. (Here in fact fn→0f_n\to 0 uniformly by Riemann–Lebesgue–type cancellation, consistent with g≡0g\equiv 0.)

Boxed answer.   {fn}\;\{f_n\} is uniformly bounded (≤1\le 1) and 11-Lipschitz-equicontinuous ⇒\Rightarrow relatively compact; a uniformly convergent subsequence exists, with modulus δ(ε)=ε\delta(\varepsilon)=\varepsilon.

ExamplePeano Existence Theorem via Arzelà–Ascoli

Given. The ODE y′=F(t,y)y'=F(t,y), y(0)=y0y(0)=y_0, with FF continuous and bounded ∣F∣≤M|F|\le M on a box R=[0,T]×[y0−b, y0+b]R=[0,T]\times[y_0-b,\,y_0+b], T≤b/MT\le b/M.

Step 1 — Euler/approximate solutions. Construct piecewise-linear approximations yny_n with mesh 1/n1/n satisfying yn(0)=y0y_n(0)=y_0 and slopes from FF. Each yny_n is MM-Lipschitz in tt.

Step 2 — Uniform bound. ∣yn(t)−y0∣≤MT≤b|y_n(t)-y_0|\le M T \le b, so ∥yn∥∞≤∣y0∣+b\|y_n\|_\infty \le |y_0|+b. Uniformly bounded.

Step 3 — Equicontinuity. From the MM-Lipschitz property, ∣yn(t)−yn(s)∣≤M∣t−s∣|y_n(t)-y_n(s)|\le M|t-s| for all nn. Choose δ=ε/M\delta=\varepsilon/M: equicontinuous, independent of nn.

Step 4 — Apply A–A. A subsequence ynk→yy_{n_k}\to y uniformly on [0,T][0,T]. Passing to the limit in the integral form

ynk(t)=y0+∫0tF(s,ynk(s)) ds+o(1)y_{n_k}(t)=y_0+\int_0^t F\big(s,y_{n_k}(s)\big)\,ds + o(1)

gives y(t)=y0+∫0tF(s,y(s)) dsy(t)=y_0+\int_0^t F(s,y(s))\,ds, i.e. yy solves the ODE.

Boxed answer.   \;A C1C^1 solution exists on [0,T][0,T] with T=min⁡(T0,b/M)T=\min(T_0, b/M) — Peano's theorem, proved purely by compactness, with no Lipschitz / uniqueness assumption on FF.

Common Misconceptions

⚠️ Common mistake

Pointwise boundedness is not uniform boundedness in general, and pointwise convergence of a subsequence does not give uniform convergence — equicontinuity is what bridges them. Two further traps: 1. Dropping compactness of the domain. On a non-compact domain (e.g. R\mathbb{R}) uniform boundedness + equicontinuity does not imply a uniformly convergent subsequence — mass can escape to infinity. One needs a locally uniform / σ\sigma-compact version with a diagonal argument over an exhaustion. 2. Equicontinuity is a property of the family, not of single functions. Every fnf_n can be uniformly continuous while {fn}\{f_n\} fails to be equicontinuous.

CounterexampleBounded + Each Continuous, but NOT Equicontinuous

Let fn(x)=sin⁡(nx)f_n(x)=\sin(nx) on [0,π][0,\pi]. Each is smooth, ∥fn∥∞=1\|f_n\|_\infty=1 (uniformly bounded). But {fn}\{f_n\} is not equicontinuous: near any point the slope ncos⁡(nx)n\cos(nx) blows up, so no single δ\delta works for all nn. Indeed {fn}\{f_n\} has no uniformly (nor even pointwise) convergent subsequence — by ∫0π(sin⁡nx−sin⁡mx)2=π\int_0^\pi(\sin nx-\sin mx)^2 = \pi for n≠mn\ne m, no subsequence is Cauchy in L2L^2, hence none in sup norm. Compactness fails precisely because equicontinuity fails.

Escape-to-infinity counterexample on a non-compact domain. gn(x)=ϕ(x−n)g_n(x)=\phi(x-n) on R\mathbb{R}, ϕ\phi a fixed bump. Uniformly bounded and equicontinuous, yet the only pointwise limit is 00 and convergence is not uniform; no subsequence converges uniformly on R\mathbb R. Compactness of the domain is essential.

Connections

connection

Arzelà–Ascoli is the prototype of all compactness-by-extraction theorems and sits in a web of related results:

  • <span class="wiki-link-missing">Heine–Borel Theorem</span> — finite-dimensional ancestor; A–A is its replacement in C(K)C(K), with equicontinuity standing in for the boundedness that no longer suffices.
  • <span class="wiki-link-missing">Stone–Weierstrass Theorem</span> — companion structural theorem on C(K)C(K): density (approximation) versus A–A's compactness (selection).
  • <span class="wiki-link-missing">Banach–Alaoglu Theorem</span> — the dual phenomenon: weak-* compactness of bounded sets; A–A can be derived as a special case in suitable settings.
  • <span class="wiki-link-missing">Rellich–Kondrachov Theorem</span> — Sobolev-space analogue: compact embedding W1,p↪LpW^{1,p}\hookrightarrow L^p on bounded domains, proved using A–A-type equicontinuity estimates via Sobolev bounds.
  • <span class="wiki-link-missing">Montel&#39;s Theorem</span> — complex-analytic refinement: a locally bounded family of holomorphic functions is normal, because Cauchy estimates supply equicontinuity automatically.
  • <span class="wiki-link-missing">Modulus of Continuity</span> — equicontinuity = a common modulus of continuity for the whole family.

Real-World Applications

🔬 Where this appears

Historical Development

⏳ History
- 1883–1884 — Giulio Ascoli introduces equicontinuity ("egual continuità") and proves that an equicontinuous, bounded family of curves admits a uniformly convergent subsequence (Le curve limiti di una varietà data di curve).
- 1889 — Cesare Arzelà extends and clarifies the convergence theory of sequences of functions; develops the notion of "convergenza in egual grado" (uniform convergence) systematically.
- 1895 — Arzelà states the full necessary-and-sufficient characterization and applies it; the combined result becomes known as the Arzelà–Ascoli theorem.
- 1890s–1900s — Vito Volterra and the Italian school use the theorem in the nascent theory of functionals ("functions of lines"), an embryo of functional analysis.
- 1906 — Maurice Fréchet casts the theorem in abstract metric-space language in his thesis, founding general topology and giving compactness its modern setting.
- 1932 — Stefan Banach's Théorie des opérations linéaires situates A–A within the structural theory of function spaces.
- 1937 — M. H. Stone and the Bourbaki program generalize C(K)C(K) theory; the compactness form for compact Hausdorff KK is standardized.
- 1940s–50s — Generalizations to uniform spaces and to families valued in compact metric spaces (Bourbaki, Kelley) reach full modern abstraction.
- 1960s– — Sobolev-space analogues (Rellich–Kondrachov) and geometric versions (Gromov compactness, 1981) extend the equicontinuity philosophy across analysis and geometry.

Subtleties & Limits

open-problem

Beyond classical compact domains. The theorem's reach depends delicately on the target and domain structure:

  • Infinite-dimensional or non-locally-compact targets. For maps into a general metric space YY, one must add that for each xx the set {f(x):f∈F}\{f(x):f\in\mathcal F\} is relatively compact in YY — "uniform boundedness" alone is insufficient when YY is infinite-dimensional. Finding sharp, easily-checkable target conditions in Banach-valued settings remains delicate.
  • Vector-valued and operator-valued analogues. Characterizing compactness of families of operators or of C(K,X)C(K,X)-valued maps in the strong/weak topologies blends A–A with weak-compactness theory (Dunford–Pettis, Eberlein–Šmulian); clean criteria are case-specific.
  • Quantitative / effective versions. Translating "a subsequence converges" into explicit rates (effective Arzelà–Ascoli) requires modulus-of-continuity bookkeeping and is an active topic in computable analysis and reverse mathematics, where the theorem is known to be non-constructive (it implies weak König's lemma).
  • Optimal regularity substitutes. In PDE, replacing equicontinuity by fractional Sobolev or Besov compactness (Aubin–Lions–Simon lemmas for time-dependent problems) continues to be refined for nonlinear evolution equations.

See Also

Heine–Borel Theorem · Stone–Weierstrass Theorem · Banach–Alaoglu Theorem · Montel's Theorem · Rellich–Kondrachov Theorem · Modulus of Continuity · Uniform Convergence · Peano Existence Theorem · Aubin–Lions Lemma