The Lebesgue Integral Introduction

One-line statement: The Lebesgue integral measures the "area under a curve" by partitioning the range (y-axis values) rather than the domain (x-axis), enabling integration of far more functions than the Riemann integral. Key result in words: A function is Lebesgue integrable if it can be approximated from below by simple (step-like) functions whose measured "horizontal slices" sum to a finite value, and the integral is the supremum of these sums. Where it applies: Probability theory, Fourier analysis, functional analysis, partial differential equations, and any setting requiring robust convergence theorems for sequences of functions. Who/when: Henri Lebesgue, doctoral thesis Intégrale, longueur, aire (1902), Paris.

The Lebesgue integral generalizes the Riemann integral by reorganizing how we accumulate the values of a function. Instead of slicing the domain into thin vertical strips (the Riemann approach), Lebesgue slices the range into horizontal bands and asks: "How large is the set of inputs where the function lies in this band?" Answering that question requires Measure Theory — a rigorous notion of the "size" of subsets of the real line. The Lebesgue integral matters because it possesses powerful convergence theorems (Monotone Convergence, Dominated Convergence) that the Riemann integral lacks, making limits and integrals interchangeable under mild hypotheses. A surprising fact: the Dirichlet function (1 on rationals, 0 on irrationals), which is not Riemann integrable, is trivially Lebesgue integrable with integral 0.

The Core Insight

Riemann partitions the domain; Lebesgue partitions the range. To integrate a non-negative function

f

, fix a height-level

y

and measure the set

{x : f (x) > y}

. Stacking these measured sets across all heights — like accounting for a sum of money by sorting coins by denomination rather than by the order received — yields a quantity stable under limits. The price of admission is that we must first know how to measure arbitrary sets, which is the content of Lebesgue Measure.

Formal Statement

Measurable Function

Let $(X, M, μ)$ be a measure space: $X$ a set, $M$ a $σ$ -algebra of "measurable" subsets, and $μ : M \to [0, \infty]$ a measure. A function $f : X \to [- \infty, \infty]$ is measurable if for every $α \in R$ ,

{x \in X : f (x) > α} \in M .

⟳The condition for all

α

is equivalent to preimages of all Borel sets being measurable; use that the half-lines

(α, \infty]

generate the Borel

σ

-algebra on

[- \infty, \infty]

Here $M$ is closed under complements and countable unions, and $μ (\emptyset) = 0$ with countable additivity. On $R$ , $μ$ is Lebesgue Measure and $M$ the Lebesgue-measurable sets.

Simple Function and Its Integral

A simple function is a measurable function taking finitely many values $a_{1}, \dots, a_{n} \geq 0$ , written

s (x) = i = 1 \sum n a_{i} 1_{A_{i}} (x), A_{i} = {x : s (x) = a_{i}},

where $1_{A_{i}}$ is the indicator function of the measurable set $A_{i}$ . Its Lebesgue integral over $E \in M$ is

\int_{E} s d μ = i = 1 \sum n a_{i} μ (A_{i} \cap E) .

⟳This is the only definition consistent with linearity and with

\int_{E} 1_{A} d μ = μ (A \cap E)

; verify it does not depend on the chosen representation of

s

by refining two partitions to a common one.

The Lebesgue Integral of a Non-negative Function

For a measurable $f : X \to [0, \infty]$ , define

\int_{E} f d μ = sup {\int_{E} s d μ : 0 \leq s \leq f, s simple} .

⟳The supremum is over all simple functions dominated by

f

; monotonicity of

μ

guarantees the set of values is bounded below by

0

and the sup is well-defined in

[0, \infty]

For general measurable $f$ , split $f = f^{+} - f^{-}$ where $f^{+} = max (f, 0)$ and $f^{-} = max (- f, 0)$ . Then

\int_{E} f d μ = \int_{E} f^{+} d μ - \int_{E} f^{-} d μ,

provided not both terms are $+ \infty$ . The function $f$ is Lebesgue integrable on $E$ iff

\int_{E} ∣ f ∣ d μ = \int_{E} f^{+} d μ + \int_{E} f^{-} d μ < \infty.

⟳Since

∣ f ∣ = f^{+} + f^{-}

and both

f^{\pm} \leq ∣ f ∣

, finiteness of

\int ∣ f ∣

forces finiteness of each piece; the space of such functions is denoted

L^{1} (μ)

Monotone Convergence Theorem (MCT)

If ${f_{n}}$ is a sequence of measurable functions with $0 \leq f_{1} \leq f_{2} \leq \dots$ and $f_{n} \to f$ pointwise, then

n \to \infty lim \int_{E} f_{n} d μ = \int_{E} f d μ = \int_{E} n \to \infty lim f_{n} d μ .

⟳The limit and supremum can be exchanged; prove

\geq

trivially from monotonicity, and

\leq

by approximating

f

from below with a simple

s

scaled by

c < 1

and the sets

E_{n} = {f_{n} \geq cs}

increasing to

E

Dominated Convergence Theorem (DCT)

Let $f_{n} \to f$ pointwise a.e., and suppose there exists an integrable $g$ with $∣ f_{n} ∣ \leq g$ for all $n$ . Then $f$ is integrable and

n \to \infty lim \int_{E} f_{n} d μ = \int_{E} f d μ, n \to \infty lim \int_{E} ∣ f_{n} - f ∣ d μ = 0.

⟳Apply Fatou's Lemma to the non-negative sequences

g + f_{n}

and

g - f_{n}

; the dominating

g

controls both directions of the inequality, sandwiching the limit.

Why It Holds / Derivation

Construction of the Integral via the Simple Function Approximation Theorem

We justify that the supremum in [4] is genuinely computable by exhibiting an increasing sequence of simple functions converging to any non-negative measurable $f$ .

Step 1 — Dyadic slicing of the range. For each

n \in N

, partition

[0, 2^{n})

into bands of height

2^{- n}

and cap at

2^{n}

. Define

s_{n} (x) = ⎩ ⎨ ⎧ \frac{k - 1}{2 ^{n}} n if \frac{k - 1}{2 ^{n}} \leq f (x) < \frac{k}{2 ^{n}}, k = 1, \dots, n 2^{n}, if f (x) \geq n .

⟳Each set

{(k - 1) / 2^{n} \leq f < k / 2^{n}}

is measurable as the difference of two superlevel sets from [1], so

s_{n}

is a legitimate simple function.

Step 2 — Monotonicity. Refining

n \to n + 1

halves each band, so

s_{n} \leq s_{n + 1}

everywhere. The approximation always undershoots:

s_{n} \leq f

. Step 3 — Pointwise convergence. If

f (x) < \infty

, then for

n > f (x)

we have

0 \leq f (x) - s_{n} (x) < 2^{- n} \to 0

. If

f (x) = \infty

, then

s_{n} (x) = n \to \infty = f (x)

. Step 4 — Integral via MCT. By the Monotone Convergence Theorem [7],

\int_{E} f d μ = n \to \infty lim \int_{E} s_{n} d μ = n \to \infty lim k \sum \frac{k - 1}{2 ^{n}} μ ({\frac{k - 1}{2 ^{n}} \leq f < \frac{k}{2 ^{n}}} \cap E) .

⟳Each

\int_{E} s_{n} d μ

is a finite sum computed by [3]; the limit exists in

[0, \infty]

because the sequence of integrals is increasing.

This proves the supremum in [4] is attained as a limit, and gives a concrete recipe: the Lebesgue integral is the limit of "height-band measure" sums. $■$

Henri Léon Lebesgue | 1875–1941

Lebesgue, a student at the École Normale Supérieure, revolutionized analysis with his 1902 thesis under the indirect influence of Émile Borel's measure theory and Camille Jordan's content. His insight — "I have to pay a sum; I empty my pockets and put coins and bills on the table by value, then sum" — captures the range-partition philosophy. He resolved long-standing pathologies (term-by-term integration of series, the fundamental theorem of calculus for absolutely continuous functions) and gave precise meaning to "almost everywhere." His name now attaches to measure, integral, dominated convergence, the Lebesgue differentiation theorem, and $L^{p}$ spaces.

Building Intuition

Picture earning income over a year and wanting your total. The Riemann accountant goes day by day (slicing the time-domain), adding each day's receipts. The Lebesgue accountant instead asks: "On how many days did I earn between \ $0 an d$ 10? Between \ $10 an d$ 20?" — then multiplies each income-band by the number of days it occurred. Both reach the same total when both methods work. But if your daily income is wildly erratic (jumping between a fixed set of values in no orderly pattern), the day-by-day Riemann sums refuse to settle, while the Lebesgue band-by-band sorting still produces a clean answer. The Lebesgue method is robust precisely because measure doesn't care about the arrangement of points, only about how much total "stuff" lands in each band.

Riemann vertical strips (domain partition) versus Lebesgue horizontal bands (range partition)

<svg viewBox="0 0 760 340" xmlns="http://www.w3.org/2000/svg">
<rect width="760" height="340" fill="#0a0e17"/>
<!-- Left: Riemann -->
<text x="120" y="28" fill="#e2e8f0" font-size="16" text-anchor="middle">Riemann: slice domain (x)</text>
<line x1="40" y1="280" x2="340" y2="280" stroke="#94a3b8" stroke-width="1.5"/>
<line x1="40" y1="280" x2="40" y2="60" stroke="#94a3b8" stroke-width="1.5"/>
<text x="345" y="285" fill="#e2e8f0" font-size="12">x</text>
<!-- vertical strips -->
<g fill="#94a3b8" fill-opacity="0.18" stroke="#94a3b8" stroke-width="0.6">
<rect x="60" y="230" width="28" height="50"/>
<rect x="88" y="200" width="28" height="80"/>
<rect x="116" y="160" width="28" height="120"/>
<rect x="144" y="140" width="28" height="140"/>
<rect x="172" y="155" width="28" height="125"/>
<rect x="200" y="185" width="28" height="95"/>
<rect x="228" y="215" width="28" height="65"/>
</g>
<path d="M60 235 Q150 110 256 220" fill="none" stroke="#d4a853" stroke-width="2.5"/>
<text x="200" y="120" fill="#d4a853" font-size="13">f(x)</text>
<!-- Right: Lebesgue -->
<text x="560" y="28" fill="#e2e8f0" font-size="16" text-anchor="middle">Lebesgue: slice range (y)</text>
<line x1="430" y1="280" x2="730" y2="280" stroke="#94a3b8" stroke-width="1.5"/>
<line x1="430" y1="280" x2="430" y2="60" stroke="#94a3b8" stroke-width="1.5"/>
<text x="424" y="70" fill="#e2e8f0" font-size="12">y</text>
<!-- horizontal bands -->
<g stroke="#94a3b8" stroke-width="0.6" stroke-dasharray="3 3">
<line x1="430" y1="230" x2="730" y2="230"/>
<line x1="430" y1="190" x2="730" y2="190"/>
<line x1="430" y1="150" x2="730" y2="150"/>
<line x1="430" y1="110" x2="730" y2="110"/>
</g>
<!-- shaded band showing {x : f>y} measure -->
<rect x="490" y="150" width="160" height="40" fill="#d4a853" fill-opacity="0.25" stroke="#d4a853" stroke-width="0.6"/>
<path d="M450 235 Q540 110 646 220" fill="none" stroke="#d4a853" stroke-width="2.5"/>
<text x="560" y="172" fill="#e2e8f0" font-size="11" text-anchor="middle">μ{f &gt; y}·Δy</text>
<text x="600" y="120" fill="#d4a853" font-size="13">f(x)</text>
</svg>

Worked Examples

The Dirichlet Function — Lebesgue Integrable, Not Riemann

Given:

f : [0, 1] \to R

defined by

f (x) = 1

x \in Q

f (x) = 0

x \in / Q

. Use Lebesgue measure

μ = m

. Step 1 — Riemann fails. Every subinterval contains both rationals and irrationals, so the upper Darboux sum is

1

and the lower is

0

for every partition. They never agree, so

f

is not Riemann integrable. Step 2 — Identify the level sets.

f

is already simple:

f = 1 \cdot 1_{Q \cap [0, 1]} + 0 \cdot 1_{(Q^{c}) \cap [0, 1]}

. Step 3 — Measure the sets. The rationals are countable, hence

m (Q \cap [0, 1]) = 0

. Therefore

m (irrationals \cap [0, 1]) = 1

. Step 4 — Apply [3].

\int_{[0, 1]} f d m = 1 \cdot m (Q \cap [0, 1]) + 0 \cdot m (irrationals) = 1 \cdot 0 + 0 \cdot 1.

\int_{[0, 1]} f d m = 0 (dimensionless)

Interchanging Limit and Integral via MCT

Given:

f_{n} (x) = \frac{n x}{1 + n ^{2} x ^{2}} \cdot 1_{[0, 1]} (x)

? — instead take the cleaner monotone case

f_{n} (x) = x^{1/ n}

[0, 1]

, increasing to

f (x) = 1

for

x \in (0, 1]

. Compute

lim_{n} \int_{0}^{1} x^{1/ n} d m

. Step 1 — Verify hypotheses. For fixed

x \in (0, 1)

x^{1/ n}

increases toward

x^{0} = 1

n \to \infty

(since

1/ n ↓ 0

and

0 < x < 1

). Each

f_{n} \geq 0

and measurable (continuous). MCT [7] applies. Step 2 — Compute each integral (Lebesgue agrees with Riemann for continuous functions):

\int_{0}^{1} x^{1/ n} d m = [\frac{x ^{1/ n + 1}}{1/ n + 1}]_{0}^{1} = \frac{1}{1 + 1/ n} = \frac{n}{n + 1} .

Step 3 — Take the limit.

n \to \infty lim \int_{0}^{1} f_{n} d m = n \to \infty lim \frac{n}{n + 1} = 1.

Step 4 — Confirm via MCT.

\int_{0}^{1} f d m = \int_{0}^{1} 1 d m = m ((0, 1]) = 1

. The two match exactly, as guaranteed.

n \to \infty lim \int_{0}^{1} x^{1/ n} d m = 1 (dimensionless)

Common Misconceptions

"Every Riemann integrable function is Lebesgue integrable, and vice versa." Only the first half is true on a bounded interval: if

f

is Riemann integrable on

[a, b]

, it is Lebesgue integrable with the same value. The converse fails — the Dirichlet function above is Lebesgue but not Riemann integrable. For improper integrals the relationship inverts in subtle ways:

\int_{0}^{\infty} \frac{s i n x}{x} d x

converges as an improper Riemann integral (to

π /2

) but is not Lebesgue integrable, because

\int_{0}^{\infty} \frac{s i n x}{x} d x = \infty

and the Lebesgue integral requires absolute integrability per [6].

Why Boundedness of $|f_n|$ Alone Is Insufficient — the Moving Bump

Let $f_{n} (x) = n \cdot 1_{(0, 1/ n)} (x)$ on $[0, 1]$ . Then $f_{n} (x) \to 0$ pointwise (for any fixed $x > 0$ , eventually $1/ n < x$ , and $f_{n} (0) = 0$ always). So $\int_{0}^{1} lim_{n} f_{n} d m = \int_{0}^{1} 0 d m = 0$ . But

\int_{0}^{1} f_{n} d m = n \cdot m ((0, \frac{1}{n})) = n \cdot \frac{1}{n} = 1 for every n .

⟳Use [3] with the single nonzero value

a_{1} = n

on the set

A_{1} = (0, 1/ n)

of measure

1/ n

Thus $lim_{n} \int f_{n} = 1 \neq = 0 = \int lim_{n} f_{n}$ . The interchange fails because there is no integrable dominating function $g$ : any $g \geq sup_{n} f_{n}$ would need $g (x) \geq n$ near $0$ , forcing $\int g = \infty$ . This is exactly why DCT [8] requires a dominating envelope — MCT does not apply either since ${f_{n}}$ is not monotone.

Connections

The Lebesgue integral sits at the crossroads of several theories. It is built atop Lebesgue Measure and the abstract Measure Theory of $σ$ -algebras. Its convergence theorems power Lp Spaces, the complete normed spaces $L^{p} (μ)$ central to Functional Analysis. In probability, $\int_{X} X d P$ is the expected value — Probability Theory is literally measure theory with $μ (X) = 1$ . The integral also clarifies the Fundamental Theorem of Calculus via absolutely continuous functions and the Lebesgue differentiation theorem, and it provides the natural setting for Fourier Analysis where Parseval's identity demands $L^{2}$ completeness that the Riemann integral cannot supply.

Real-World Applications

Probability and statistics. Expectation $E [X] = \int_{Ω} X d P$ is a Lebesgue integral; the Strong Law of Large Numbers and martingale convergence rely on DCT and MCT. Used in finance for option pricing (risk-neutral expectations).
Signal processing & Fourier analysis. The space $L^{2} (R)$ of finite-energy signals is complete only with the Lebesgue integral, enabling the Plancherel theorem behind MP3/JPEG transform coding.
Quantum mechanics. Wavefunctions live in $L^{2}$ ; the inner product $⟨ ψ, ϕ ⟩ = \int \overline{ψ} ϕ d m$ and normalization $\int ∣ ψ ∣^{2} d m = 1$ are Lebesgue integrals.
Partial differential equations. Sobolev spaces $W^{k, p}$ , the foundation of finite-element methods (used in aircraft and bridge engineering), are defined via Lebesgue integrability of weak derivatives.
Machine learning. Expected loss $E_{(x, y) \sim D} [ℓ (f (x), y)]$ is a Lebesgue integral over the data distribution; convergence of empirical risk uses dominated-convergence arguments.
Ergodic theory & dynamical systems. Birkhoff's ergodic theorem equates time averages with space averages via Lebesgue integration against an invariant measure — applied to statistical mechanics.
Image reconstruction (tomography). The Radon transform and its inversion are formulated in $L^{p}$ spaces, underpinning CT and MRI scan algorithms.

Historical Development

1854 — Bernhard Riemann formalizes the Riemann integral in his Habilitationsschrift, the first rigorous definition of the integral.
1881–1884 — Vito Volterra constructs a bounded derivative that is not Riemann integrable, exposing limits of the Riemann theory.
1898 — Émile Borel develops the theory of measure for "Borel sets," assigning length consistently to a large class of subsets.
1901 — Lebesgue announces his integral in a Comptes Rendus note, "Sur une généralisation de l'intégrale définie."
1902 — Lebesgue's doctoral thesis Intégrale, longueur, aire establishes Lebesgue measure and the integral, including the dominated and monotone convergence ideas.
1906 — Pierre Fatou proves Fatou's Lemma in his study of analytic functions and trigonometric series.
1907 — Frigyes Riesz and Ernst Fischer prove the completeness of $L^{2}$ (Riesz–Fischer theorem), cementing the integral's role in Fourier theory.
1910 — Lebesgue's Sur l'intégration des fonctions discontinues extends the theory to higher dimensions and differentiation of the integral.
1918–1933 — Maurice Fréchet, Constantin Carathéodory (1918, outer-measure construction), and Andrey Kolmogorov (1933, Grundbegriffe) abstract the theory into general measure spaces and axiomatic probability.

Subtleties & Limits

Non-measurable sets and the limits of "measuring everything." Using the Axiom of Choice, Giuseppe Vitali (1905) constructed a subset of

[0, 1]

that is not Lebesgue measurable — so [1] genuinely restricts which functions we can integrate. The Banach–Tarski paradox (1924) sharpens this: in

R^{3}

a ball can be decomposed into finitely many (necessarily non-measurable) pieces and reassembled into two balls, showing finitely-additive volume cannot extend to all sets in a rotation-invariant way. Robert Solovay (1970) proved that, consistent with ZF (without full Choice), one can have a model where every set of reals is Lebesgue measurable — so the existence of non-measurable sets is intimately tied to the strength of the axioms one accepts. Open meta-questions remain about which extensions of measure (e.g., to non-

σ

-finite or non-locally-compact settings) preserve the convergence theorems, and how Lebesgue integration interacts with constructive/computable analysis where the supremum in [4] may not be effectively attainable.