Egorov's Theorem: Almost Uniform Convergence

The precise statement

Let (X, ℳ, μ) be a measure space with μ(X) < ∞, and let f, f₁, f₂, … : X → ℝ (or ℂ, or any separable metric space) be measurable functions with fₙ → f pointwise μ-almost everywhere. Then for every ε > 0 there exists a measurable set E ∈ ℳ such that:

• μ(X \ E) < ε, and
• fₙ → f uniformly on E: sup_x∈E |fₙ(x) − f(x)| → 0 as n → ∞.

The phrase used for this conclusion is almost uniform convergence: uniform convergence after excising a set of arbitrarily small measure. Note the order of quantifiers — ε is chosen first, then the exceptional set E depends on ε. You cannot in general take ε → 0 and get one common good set; the leftover set shrinks in measure but need not become null. The finiteness of μ is essential and cannot simply be dropped, only weakened to the assumption that the fₙ are supported (up to the relevant convergence) on a set of finite measure.

The picture: trading domain for control

Pointwise convergence is a local, unquantified promise: for each point x separately, the values fₙ(x) eventually get close to f(x) — but the rate can vary wildly from point to point, and there may be no single N that works everywhere. Uniform convergence is the global promise: one N handles all x at once.

Egorov says the gap between these is measure-theoretically thin. Imagine the domain colored by how slow convergence is at each point. The points where convergence is arbitrarily slow can only occupy a small total measure. So peel those off — a set you can make as small as you like in measure — and on the bulk of the domain the rate is uniformly controlled.

• What you give up: a set of measure < ε (which can be spread out, not an interval).
• What you gain: a single N(δ) valid for all remaining points.

It is the measure-theoretic analogue of the intuition that "bad behavior concentrates on a small set."

The mechanism of the proof

Discard the null set where convergence fails, so assume fₙ → f everywhere. The engine is a family of tail sets. For integers n, k ≥ 1 define

E_n,k = { x : |f_m(x) − f(x)| < 1/k for all m ≥ n }.

For each fixed k, the sets E_n,k increase with n, and because fₙ → f at every point, ⋃_n E_n,k = X. Now the crucial step where finiteness enters: by continuity from below, μ(E_n,k) → μ(X), so μ(X \ E_n,k) → 0. Since μ(X) < ∞ these complements have finite, vanishing measure, so choose n(k) with

μ(X \ E_n(k),k) < ε / 2ᵏ.

Set E = ⋂_k E_n(k),k. By countable subadditivity μ(X \ E) ≤ ∑_k ε/2ᵏ = ε. And on E: given δ > 0 pick k with 1/k < δ; then for all m ≥ n(k) and all x ∈ E ⊂ E_n(k),k we have |fₘ(x) − f(x)| < 1/k < δ. That single N = n(k) works uniformly on E. ∎

A worked example

Take X = [0, 1] with Lebesgue measure and fₙ(x) = xⁿ. Pointwise, fₙ(x) → 0 for x ∈ [0, 1) and fₙ(1) = 1, so fₙ → f a.e. where f = 0 (the single point x = 1 is null). Convergence is not uniform on [0, 1): sup_[0,1) xⁿ = 1 for every n, so the sup never goes to 0. The obstruction is entirely near x = 1, where xⁿ decays arbitrarily slowly.

Egorov's remedy: fix ε > 0 and delete the interval (1 − ε, 1]. On E = [0, 1 − ε] we have

sup_x∈E xⁿ = (1 − ε)ⁿ → 0,

so fₙ → 0 uniformly on E, and μ([0,1] \ E) = ε. As ε → 0 the deleted interval shrinks but never vanishes — you cannot get uniform convergence on all of [0, 1), only after removing a sliver near the trouble spot. This is Egorov's conclusion made completely explicit: a small excised set converts non-uniform pointwise limit into a uniform one.

Why finiteness matters: the counterexample

Drop μ(X) < ∞ and the theorem fails outright. Take X = ℝ with Lebesgue measure and the marching bumps fₙ = 𝟙_{[n, n+1]}, the indicator of [n, n+1]. For every fixed x, once n > x we have fₙ(x) = 0, so fₙ → 0 pointwise everywhere — the hypothesis holds perfectly.

But there is no set E with μ(ℝ \ E) < ∞ (let alone < ε) on which convergence is uniform. Any such E must meet [n, n+1] for infinitely many n (otherwise its complement contains infinitely many disjoint unit intervals and has infinite measure), and on each such n the value fₙ = 1 somewhere in E, so the sup over E stays ≥ 1. Uniform convergence is impossible off any small set. The mass simply escapes to infinity; there is no finite budget to confine the slow-converging region.

• Fix: Egorov survives on σ-finite spaces only in weakened, localized forms — you can apply it on each finite-measure piece, not globally.
• Related: this is the same escape-to-infinity that defeats the dominated convergence theorem without an integrable dominating function.

Significance and connections

Egorov's theorem is a workhorse for converting weak convergence hypotheses into usable uniform estimates. Its most famous consequence is Lusin's theorem: a measurable function on [a,b] agrees with a continuous function off a set of arbitrarily small measure. One standard proof approximates the measurable function by continuous ones pointwise, then uses Egorov to make the approximation uniform off a small set — where uniform limits of continuous functions are continuous.

• Convergence in measure: almost uniform convergence implies convergence in measure, so Egorov cleanly bridges pointwise a.e. and in-measure convergence on finite spaces.
• Integration theory: it supplies the uniform control behind quick proofs of the bounded convergence theorem and various interchange-of-limit results.
• Probability: on a probability space (μ(X) = 1 is automatically finite) it says almost-sure convergence is "almost uniform," a fact used in ergodic theory and the analysis of stochastic limits.

It is the precise measure of how far pointwise convergence is from uniform convergence — the answer is: only an ε of the domain.

Modes of convergence for measurable functions on a finite measure space, and how Egorov relates them

Mode of convergence	Definition (informal)	Egorov's role
Pointwise a.e.	fₙ(x) → f(x) for μ-a.e. x	The hypothesis Egorov starts from
Uniform on X	sup over X of |fₙ − f| → 0	The strong conclusion Egorov almost delivers
Almost uniform	∀ε ∃E with μ(Eᶜ) < ε and uniform convergence on E	Exactly Egorov's conclusion
In measure	μ(|fₙ − f| ≥ δ) → 0 for all δ > 0	Implied by almost uniform (hence by Egorov)
Uniform off a null set	uniform on X \ N with μ(N) = 0	STRONGER than Egorov — generally false (see counterexample)