Measure Theory

Lebesgue Integral

Partition the codomain, not the domain — handles characteristic function of ℚ that Riemann can't

The Lebesgue integral, introduced by Henri Lebesgue in his 1902 PhD thesis, generalizes the Riemann integral by partitioning the function's range (codomain) rather than the domain. For a non-negative measurable function f: ℝ → ℝ, ∫ f dμ = sup over simple functions s ≤ f of Σ s · μ(level sets). Handles functions Riemann cannot — most famously, the indicator function of ℚ ∩ [0,1] is Lebesgue integrable (= 0, since ℚ has measure zero) but not Riemann integrable. Dominated Convergence Theorem (DCT) provides clean swap of limit and integral when |f_n| ≤ g, ∫ g < ∞ — making Lebesgue indispensable in analysis, probability (E[X] is a Lebesgue integral on (Ω, F, P)), Fourier analysis (L²), and PDE.

  • AuthorLebesgue 1902
  • Methodpartition codomain
  • GeneralityRiemann ⊂ Lebesgue
  • Indicator of ℚLebesgue OK
  • DCTclean limit/integral swap
  • L^p spacescompletion w.r.t. ‖·‖_p

Watch the 60-second explainer

A condensed visual walkthrough — narrated, captioned, under a minute.

Why the Lebesgue integral matters

  • Probability theory. Kolmogorov's 1933 axioms make probability into measure theory; expectation E[X], variance, conditional expectation are all Lebesgue integrals on (Ω, F, P).
  • Fourier analysis. L²(ℝ) is the natural home of Fourier transforms; Plancherel's theorem (‖f‖₂ = ‖f̂‖₂) is a statement about Lebesgue integrals.
  • Sobolev spaces and PDE. H^k(Ω) consists of L²-functions whose distributional derivatives are also L²; weak solutions to PDE live here.
  • Ergodic theory. Time averages along trajectories equal space averages with respect to an invariant measure — a statement that requires Lebesgue integration to even formulate.
  • Convergence theorems. Monotone convergence (MCT), Fatou's lemma, dominated convergence (DCT) — three swap-limit-and-integral results that are far more powerful than uniform convergence.
  • Functional analysis. The Banach spaces L^p, the Hilbert space L², dual pairings, weak topology — all built on Lebesgue integration.
  • Quantum mechanics. Wave functions live in L²(ℝ³); inner products ⟨ψ, φ⟩ = ∫ ψ̄ φ dx are Lebesgue integrals; rigorous QM cannot be done with Riemann.

Common misconceptions

  • Lebesgue extends Riemann always. No — improper Riemann integrals can disagree. ∫_0^∞ sin(x)/x dx converges as improper Riemann (= π/2) but ∫|sin(x)/x| diverges, so it's not Lebesgue-integrable.
  • You don't need measure theory. Yes you do — σ-algebras and measures are prerequisite; that's why measure theory is taught before Lebesgue integration.
  • Lebesgue is computational. In practice it's symbolic — you almost never compute Lebesgue integrals via the definition; you prove existence and use the same formulas as Riemann when both apply.
  • Every function is Lebesgue-integrable. No — non-measurable functions exist (Vitali sets, requires AC), and even measurable functions need ∫|f| < ∞ to be Lebesgue-integrable.
  • Lebesgue is the most general integral. No — there are extensions (Henstock-Kurzweil, Itô stochastic integrals, Daniell integral) that handle cases Lebesgue doesn't.
  • L^p spaces consist of functions. Strictly, of equivalence classes — two functions equal almost everywhere are identified. Pointwise values aren't well-defined in L^p.

Frequently asked questions

Why partition the codomain instead of domain?

Riemann partitions the x-axis into vertical strips and asks how f varies on each strip; Lebesgue partitions the y-axis (the values of f) into horizontal strips and asks for the measure of x where f lands in each strip. Lebesgue's reformulation works for any function that's measurable, no matter how wildly it oscillates. A useful analogy: counting coins by sweeping along a counter (Riemann) versus stacking same-denomination coins together first then summing (Lebesgue). Both give the right total for nice cases — but only the second works for unsorted, jumbled distributions.

What is a measurable function?

Given a measurable space (X, A) (a set X with a σ-algebra A of subsets), a function f : X → ℝ is measurable if f⁻¹(B) ∈ A for every Borel set B ⊂ ℝ. Equivalently, all level sets {x : f(x) > c} are measurable. Continuous functions are measurable. Pointwise limits of measurable functions are measurable. Non-measurable functions exist but require the Axiom of Choice (Vitali sets) — so any function you can write down explicitly is measurable. Measurability is the bare minimum for the Lebesgue integral to make sense.

When is a Riemann-integrable function Lebesgue-integrable?

Every Riemann-integrable function on a bounded interval is Lebesgue-integrable, and the values agree (Lebesgue characterization: f is Riemann-integrable on [a, b] iff f is bounded and discontinuous only on a set of measure zero). For improper Riemann integrals on unbounded domains the picture splits: ∫_0^∞ sin(x)/x dx = π/2 as an improper Riemann integral but is NOT Lebesgue-integrable (∫|sin(x)/x| dx diverges). Lebesgue's framework is essentially absolute integrability — it cares about ∫|f|, not delicate cancellations.

What is the dominated convergence theorem?

DCT: if measurable f_n → f pointwise (almost everywhere) and there exists an integrable g with |f_n| ≤ g for all n, then ∫ f_n → ∫ f. The dominator g controls oscillation uniformly; pointwise convergence then suffices to swap limit and integral. Compare to uniform convergence (Riemann's swap criterion): DCT is much weaker — it only requires pointwise convergence and a fixed dominator. DCT, combined with monotone convergence and Fatou's lemma, are the three convergence theorems that make Lebesgue indispensable. Without DCT, even basic Fourier-series convergence proofs become tedious.

What are L^p spaces?

L^p(X, μ) for 1 ≤ p ≤ ∞ is the space of measurable functions with ‖f‖_p = (∫|f|^p dμ)^(1/p) finite (sup essential for p = ∞). Functions equal almost everywhere are identified. L^p is a Banach space (complete normed); L² is a Hilbert space (with inner product ∫ f·ḡ dμ). Completion under ‖·‖_p of continuous functions yields L^p; without Lebesgue, this completion can't be described by formulas — the limits don't exist as Riemann-integrable functions. Foundational to functional analysis, signal processing, quantum mechanics, and PDE.

Why is Lebesgue essential for probability theory?

A probability space is a measure space (Ω, F, P) with P(Ω) = 1. Random variables are measurable functions X : Ω → ℝ. Expectation E[X] is exactly the Lebesgue integral ∫ X dP. Without measure theory, you can't define probability for events more complex than countable unions of intervals, and you can't take limits of expectations through DCT or Fatou. Modern probability — martingales, ergodic theory, stochastic calculus, central limit theorem proofs — is built on Kolmogorov's 1933 axiomatization, which IS Lebesgue measure theory applied to (Ω, F, P).