Lebesgue Integral Construction and the Monotone Convergence Theorem

Intuition Beginner

The Riemann integral asks the question backwards. It chops up the $x$-axis (the domain) into thin vertical strips, multiplies the strip width by the value of the function at a sample point, and adds the rectangles to recover the area under the curve. This works beautifully for continuous functions, where the value changes only a little across each thin strip. It runs into trouble when the function is wild: the indicator of the rationals on $[0,1]$ takes the value $1$ on a dense set and $0$ on another dense set, so every thin vertical strip contains points of both kinds and the strip-rectangle picture cannot decide what height to choose.

Lebesgue's idea was to chop up the $y$-axis (the range) instead. Slice the height between $0$ and the supremum of the function into horizontal bands, and for each band compute the measure of the set of $x$-values whose function value falls in that band. Multiply the height of the band by the measure of the corresponding domain set, and add. For the indicator of the rationals, the height $1$ is achieved on the rationals (measure zero), the height $0$ is achieved on the irrationals (measure one), so the Lebesgue integral is $1\cdot 0+0\cdot 1=0$, a clean answer where Riemann's procedure is undefined.

The construction proceeds in three steps. First, define the integral of a simple function — a finite sum of constants times indicators of measurable sets — as the obvious finite sum of values times measures of level sets. Second, define the integral of a non-negative measurable function as the supremum of integrals of simple functions sitting below it. Third, define the integral of a general measurable function by splitting it into its positive and negative parts and subtracting the two non-negative integrals.

The monotone convergence theorem is the engine that makes the second step computable. If a non-negative function is approached pointwise from below by an increasing staircase of simple functions, the theorem says that the integrals of the staircase steps climb toward the integral of the function. The supremum of integrals equals the integral of the supremum. This is a stronger and cleaner statement than anything the Riemann setting allows, and it makes the Lebesgue integral interchangeable with limits in a way the Riemann integral never could be.

A useful slogan: the Lebesgue integral is the Riemann integral made limit-stable. Whenever a sequence of integrable functions converges pointwise upward to a limit, the integrals converge to the integral of the limit. No fussing with uniform convergence, no Darboux upper-and-lower sums, no patching at jump discontinuities. The price is the upfront cost of building the measure-theoretic framework; the payoff is a theory of integration where limit-passage is a one-line citation rather than a technical struggle.

The one-sentence takeaway: chop the range instead of the domain, build the integral by approximating from below with simple functions, and the monotone convergence theorem makes limits and integrals commute.

Visual Beginner

Picture a non-negative function as a smooth hump over the $x$-axis. The Riemann picture slices the base of the hump into thin vertical strips; the Lebesgue picture slices the height of the hump into thin horizontal bands. For each horizontal band, the set of $x$-values whose function value lands in that band is some collection of intervals (or a more complicated measurable set). The measure of that collection times the band height contributes one term to the Lebesgue integral.

The monotone convergence picture is a sequence of staircases. The first staircase sits well below the smooth curve; the second is finer and climbs higher; the third is finer still and climbs higher again. Each staircase is a simple function whose integral is a finite sum of band heights times band-set measures. As the staircases climb pointwise toward the curve, their integrals climb toward the area under the curve.

Worked example Beginner

We compute the Lebesgue integral of the Dirichlet indicator function ${\chi }_{\mathbb{Q}}$ on the interval $[0,1]$. This function takes the value $1$ at every rational point and $0$ at every irrational point.

Step 1. The function is already simple. It has only two distinct values: $1$ on the rationals in $[0,1]$, and $0$ on the irrationals in $[0,1]$.

Step 2. The integral of a simple function is the sum of value times measure of level set. The set of rationals in $[0,1]$ has Lebesgue measure $0$ (it is countable, so the sum of widths of intervals covering it can be made as small as one likes). The set of irrationals in $[0,1]$ has Lebesgue measure $1$ (the complement of a set of measure $0$ in $[0,1]$).

Step 3. Add the two terms: the integral equals $1\cdot 0+0\cdot 1=0$.

Step 4. Sanity check: the Riemann integral of ${\chi }_{\mathbb{Q}}$ on $[0,1]$ is undefined because the Riemann upper sum is $1$ (any sampling strip contains rationals) and the lower sum is $0$ (any sampling strip contains irrationals). The Lebesgue procedure gives a definite value.

What this tells us: the Lebesgue integral assigns a clean numerical value to a function that the Riemann integral cannot handle, by switching from sampling-by-domain to weighing-by-range.

Check your understanding Beginner

Formal definition Intermediate+

Let $(X,\mathcal{M},\mu )$ be a measure space.

Definition (integral of a simple function). A non-negative simple function $\phi :X\to [0,\infty )$ is a measurable function with finite range. Its canonical representation is $\phi ={\sum }_{i=1}^n{c}_i{\chi }_{A_i}$ where $c_1,\dots ,c_n$ are the distinct values of $\phi$ and $A_i={\phi }^{-1}(\{c_i\})$ are the pairwise disjoint measurable level sets. The Lebesgue integral of $\phi$ over $X$ with respect to $\mu$ is $${\int }_X\phi d\mu =\sum _{i=1}^n{c}_i\mu (A_i),$$ with the convention $0\cdot \infty =0$ so that values of $\phi$ achieved only on null sets contribute zero even when $\mu (A_i)=\infty$ is paired with $c_i=0$ ^{[Folland §2.2]}.

The definition does not depend on the choice of representation: if $\phi ={\sum }_j{b}_j{\chi }_{B_j}$ is any other representation with measurable $B_j$ (not necessarily disjoint or with distinct $b_j$), refining the partition to a common refinement and re-summing produces the same value. This is the simple-function representation lemma.

Definition (integral of a non-negative measurable function). Let $f:X\to [0,\infty ]$ be measurable. The Lebesgue integral of $f$ over $X$ with respect to $\mu$ is $${\int }_{X}fd\mu =\sup \left\{{\int }_X\phi d\mu :\phi \text{ simple},0\le \phi \le f\right\}.$$ The value lies in $[0,\infty ]$. We say $f$ is integrable when ${\int }_{X}fd\mu <\infty$.

Definition (integral of a general measurable function). Let $f:X\to \bar{\mathbb{R}}$ be measurable. Write $f=f^{+}-f^{-}$ where $f^{+}=\max (f,0)$ and $f^{-}=\max (-f,0)$ are the non-negative parts. If at least one of ${\int }_X{f}^{+}d\mu$, ${\int }_X{f}^{-}d\mu$ is finite, set $${\int }_{X}fd\mu ={\int }_X{f}^{+}d\mu -{\int }_X{f}^{-}d\mu .$$ If both are finite, $f$ is called integrable and we write $f\in L^1(X,\mu )$. If both are infinite, the integral is undefined.

Definition (integral over a measurable subset). For $E\in \mathcal{M}$ and $f$ measurable, ${\int }_{E}fd\mu ={\int }_{X}f{\chi }_{E}d\mu$.

Linearity, monotonicity, and basic estimates. For non-negative measurable $f,g$ and $\alpha ,\beta \ge 0$: $${\int }_X(\alpha f+\beta g)d\mu =\alpha {\int }_{X}fd\mu +\beta {\int }_{X}gd\mu ,f\le g\text{ a.e.}⟹{\int }_{X}fd\mu \le {\int }_{X}gd\mu .$$ For general $f,g\in L^1$: $${\int }_X(\alpha f+\beta g)d\mu =\alpha {\int }_{X}fd\mu +\beta {\int }_{X}gd\mu ,\left|{\int }_{X}fd\mu \right|\le {\int }_X|f|d\mu .$$ Linearity for non-negative simple functions is direct from the definition; linearity for non-negative measurable functions follows once the monotone convergence theorem below is in hand.

Counterexamples to common slips Intermediate+

• The integral of a non-negative function can be $+\infty$. The constant function $f\equiv 1$ on $\mathbb{R}$ has ${\int }_{\mathbb{R}}fdm=+\infty$, so $f$ is measurable and non-negative but not integrable. Integrability is a finiteness condition added on top of measurability.

• Linearity for general (signed) $f$ requires integrability. If $\int f^{+}=\int f^{-}=+\infty$, the integral $\int f$ is undefined, and writing $\int (f+g)=\int f+\int g$ becomes meaningless. The hypothesis $f,g\in L^1$ is irreducible for the signed-linearity identity.

• Pointwise upper bound need not give a finite integral. The functions $f_n(x)=n{\chi }_{[0,1/n]}$ on $[0,1]$ converge pointwise to $0$ (everywhere except $x=0$) but $\int f_n=1$ for every $n$. The pointwise limit has integral $0$, but the integrals of the approximating sequence do not converge to the integral of the limit; the monotone convergence theorem requires monotonicity, not just pointwise convergence.

• Riemann-integrable does not imply Lebesgue-integrable on infinite-measure domains. The function $\sin (x)/x$ on $[0,\infty )$ is improper-Riemann-integrable (with value $\pi /2$) but not Lebesgue-integrable: ${\int }_0^{\infty }|\sin (x)/x|dx=+\infty$. The Lebesgue integral demands absolute integrability.

Key theorem with proof Intermediate+

Theorem (monotone convergence; Beppo Levi 1906). Let $(X,\mathcal{M},\mu )$ be a measure space and let $\{f_n{\}}_{n\ge 1}$ be a sequence of measurable functions $X\to [0,\infty ]$ satisfying $0\le f_1\le f_2\le \cdots$ pointwise almost everywhere, with pointwise limit $f(x)={\lim }_n{f}_n(x)={\sup }_n{f}_n(x)$ almost everywhere. Then $f$ is measurable and $${\int }_{X}fd\mu =\underset{n\to \infty }{\lim }{\int }_X{f}_{n}d\mu =\underset{n}{\sup }{\int }_X{f}_{n}d\mu .$$

Proof. By redefining the $f_n$ and $f$ on a null set, we may assume the monotonicity and convergence hold everywhere on $X$ without changing any integral. Measurability of $f={\sup }_n{f}_n$ is the countable-supremum closure of measurable functions.

Step 1 (the easy inequality). Since $f_n\le f$ for every $n$, monotonicity of the integral gives ${\int }_X{f}_{n}d\mu \le {\int }_{X}fd\mu$. Taking $n\to \infty$, ${\sup }_n{\int }_X{f}_{n}d\mu \le {\int }_{X}fd\mu$. The sequence $\{{\int }_X{f}_{n}d\mu \}$ is monotone non-decreasing (since $f_n\le f_{n+1}$ gives $\int f_n\le \int f_{n+1}$), so its limit exists in $[0,\infty ]$ and equals its supremum. Denote $L={\lim }_n{\int }_X{f}_{n}d\mu ={\sup }_n{\int }_X{f}_{n}d\mu \le {\int }_{X}fd\mu$.

Step 2 (the reverse inequality). It suffices to show that for every simple function $\phi$ with $0\le \phi \le f$, ${\int }_X\phi d\mu \le L$. Once this is proven, taking the supremum over such $\phi$ gives ${\int }_{X}fd\mu \le L$, completing the proof.

Fix a simple $\phi ={\sum }_{i=1}^k{c}_i{\chi }_{A_i}$ with $0\le \phi \le f$ and pairwise disjoint measurable $A_i$. Fix $\alpha \in (0,1)$. Define the cumulative measurable set $$E_n=\{x\in X:f_n(x)\ge \alpha \phi (x)\}.$$ The sets $E_n$ are measurable (each is the level set of the measurable function $f_n-\alpha \phi$) and form an increasing sequence: $E_n\subseteq E_{n+1}$ because $f_n\le f_{n+1}$.

Claim. ${\bigcup }_n{E}_n=X$.

Proof of claim. Fix $x\in X$. If $\phi (x)=0$, then $f_n(x)\ge 0=\alpha \phi (x)$ for every $n$, so $x\in E_n$ for every $n$. If $\phi (x)>0$, then $\phi (x)\le f(x)={\lim }_n{f}_n(x)$, and since $\alpha <1$, $\alpha \phi (x)<\phi (x)\le f(x)$. By definition of pointwise limit, $f_n(x)\to f(x)$, so eventually $f_n(x)\ge \alpha \phi (x)$, hence $x\in E_n$ for sufficiently large $n$. End of claim.

Now estimate ${\int }_X{f}_{n}d\mu$ from below by restricting attention to $E_n$: $${\int }_X{f}_{n}d\mu \ge {\int }_{E_n}{f}_{n}d\mu \ge \alpha {\int }_{E_n}\phi d\mu =\alpha \sum _{i=1}^k{c}_i\mu (A_i\cap E_n).$$ The first inequality drops the part of the integral outside $E_n$ (where $f_n\ge 0$); the second uses $f_n\ge \alpha \phi$ on $E_n$; the third is the integral of a simple function on $E_n$.

For each $i$, the sets $\{A_i\cap E_n{\}}_{n\ge 1}$ form an increasing sequence whose union is $A_i\cap {\bigcup }_n{E}_n=A_i\cap X=A_i$. Measure-theoretic continuity of increasing sequences (which holds without finite-measure hypothesis) gives $\mu (A_i\cap E_n)\to \mu (A_i)$ as $n\to \infty$. So $$\underset{n}{\lim }{\int }_X{f}_{n}d\mu \ge \alpha \sum _{i=1}^k{c}_i\mu (A_i)=\alpha {\int }_X\phi d\mu .$$ That is, $L\ge \alpha {\int }_X\phi d\mu$. Since $\alpha \in (0,1)$ was arbitrary, take $\alpha \to 1^{-}$ to conclude $L\ge {\int }_X\phi d\mu$.

Taking the supremum over all simple $\phi$ with $0\le \phi \le f$, $L\ge {\int }_{X}fd\mu$.

Combining the two steps, $L={\int }_{X}fd\mu$, as required.

$\square$

Bridge. The monotone convergence theorem builds toward 02.07.05 Fatou's lemma and the dominated convergence theorem, where the same Egorov-style cumulative-set decomposition extends from monotone to general convergent sequences via the $\mathrm{liminf}$-and-$\mathrm{limsup}$ sandwich. The central insight is that the supremum of integrals equals the integral of the supremum for non-negative increasing sequences, and this is exactly the property that makes the Lebesgue integral interchangeable with limits in a way the Riemann integral cannot match. The foundational reason the proof works is the measure-theoretic continuity of increasing sequences $E_n\uparrow X$, which holds in any measure space and does not require finite measure. Putting these together with the simple-function approximation theorem from 02.07.03, we identify the integral of a non-negative measurable function with the limit of integrals along the dyadic simple-function staircase. The bridge is between the descriptive theory (measurability as a sigma-algebra condition) and the operational theory (integrals computed as limits along constructive approximations), and the pattern generalises immediately to Banach-valued integration via the Bochner integral ^{[Bochner 1933]} and to the Itô integral for stochastic processes.

Exercises Intermediate+

Lean formalization Intermediate+

lean_status: full — Mathlib provides the full apparatus. MeasureTheory.lintegral encodes the lower Lebesgue integral of an ENNReal-valued measurable function as the supremum over simple-function approximations from below. MeasureTheory.lintegral_iSup packages the monotone convergence theorem: for a monotone sequence of measurable functions with pointwise supremum, the integral of the supremum equals the supremum of integrals. MeasureTheory.lintegral_liminf_le gives Fatou's lemma, and MeasureTheory.tendsto_integral_of_dominated_convergence gives the Bochner-valued dominated convergence theorem. The companion module records the unit's predicate names; the deeper structural theorems sit on the broader Mathlib measure-theory shelf accessed by import.

import Mathlib.MeasureTheory.Integral.Lebesgue
import Mathlib.MeasureTheory.Integral.Bochner

variable (X : Type*) [MeasurableSpace X]
variable (μ : MeasureTheory.Measure X)

abbrev CodexLintegral (f : X → ENNReal) : ENNReal := MeasureTheory.lintegral μ f

abbrev CodexIntegrable {β : Type*} [NormedAddCommGroup β] (f : X → β) : Prop :=
  MeasureTheory.Integrable f μ

Advanced results Master

The advanced theory of Lebesgue integration splits across four strands: the derived limit theorems (Fatou, dominated convergence, Vitali), the relation to Riemann integration (Lebesgue criterion, $\sin x/x$ counterexample, Henstock-Kurzweil refinement), the abstract foundations (Daniell, Bochner, Itô), and the absolute-continuity / density-with-respect-to-measure framework (the bridge to Radon-Nikodym).

Theorem 1 (Fatou's lemma; Fatou 1906). Let $(X,\mathcal{M},\mu )$ be a measure space and let $\{f_n{\}}_{n\ge 1}$ be a sequence of measurable functions $X\to [0,\infty ]$. Then $${\int }_X\underset{n\to \infty }{\mathrm{liminf}}{f}_{n}d\mu \le \underset{n\to \infty }{\mathrm{liminf}}{\int }_X{f}_{n}d\mu .$$ The inequality can be strict: with $f_n=n{\chi }_{[0,1/n]}$ on $[0,1]$, the left side is $0$ and the right side is $1$ ^{[Fatou 1906]}.

Proof. Define $g_n(x)={\inf }_{k\ge n}{f}_k(x)$. Each $g_n$ is measurable as the infimum of a countable family. The sequence is monotone non-decreasing in $n$ (taking infima over shrinking tails of an arbitrary sequence yields an increasing sequence), and its pointwise limit is ${\mathrm{liminf}}_n{f}_n$ by definition. By monotone convergence applied to $\{g_n\}$: $${\int }_X\underset{n}{\mathrm{liminf}}{f}_{n}d\mu =\underset{n}{\lim }{\int }_X{g}_{n}d\mu .$$ For each $n$, $g_n\le f_n$ pointwise, so ${\int }_X{g}_{n}d\mu \le {\int }_X{f}_{n}d\mu$. Taking $\mathrm{liminf}$: $$\underset{n}{\lim }{\int }_X{g}_{n}d\mu \le \underset{n}{\mathrm{liminf}}{\int }_X{f}_{n}d\mu .$$ Chaining the two displays gives the result. $\square$

Theorem 2 (dominated convergence; Lebesgue 1908). Let $\{f_n\}$ be measurable functions $X\to \bar{\mathbb{R}}$ with $f_n\to f$ almost everywhere, and suppose there exists $g\in L^1(X,\mu )$ with $|f_n|\le g$ almost everywhere for every $n$. Then $f\in L^1$ and $$\underset{n}{\lim }{\int }_X{f}_{n}d\mu ={\int }_{X}fd\mu ,\underset{n}{\lim }{\int }_X|f_n-f|d\mu =0.$$

Proof. Apply Fatou's lemma twice. To the non-negative sequence $g-f_n$ (with pointwise limit $g-f$): $${\int }_X(g-f)d\mu \le \underset{n}{\mathrm{liminf}}{\int }_X(g-f_n)d\mu ={\int }_{X}gd\mu -\underset{n}{\mathrm{limsup}}{\int }_X{f}_{n}d\mu .$$ Since $\int g<\infty$, subtraction gives ${\mathrm{limsup}}_n\int f_n\le \int f$. Apply Fatou to $g+f_n$ (with pointwise limit $g+f$): $${\int }_X(g+f)d\mu \le \underset{n}{\mathrm{liminf}}{\int }_X(g+f_n)d\mu ={\int }_{X}gd\mu +\underset{n}{\mathrm{liminf}}{\int }_X{f}_{n}d\mu ,$$ giving $\int f\le {\mathrm{liminf}}_n\int f_n$. Combining the two, $\int f={\lim }_n\int f_n$.

The $L^1$ convergence $\int |f_n-f|\to 0$ follows by applying the same argument to $|f_n-f|$, which is dominated by $2g$. $\square$

Theorem 3 (Vitali convergence; Vitali 1907). Let $\mu (X)<\infty$ and let $\{f_n\}$ be a sequence of measurable functions with $f_n\to f$ in measure. Then $f\in L^1$ and $\int f_n\to \int f$ if and only if the sequence is **uniformly integrable**, meaning that for every $\epsilon >0$ there exists $\delta >0$ such that for every $n$ and every measurable $E$ with $\mu (E)<\delta$, ${\int }_E|f_n|d\mu <\epsilon$, and additionally for every $\epsilon >0$ there exists a measurable $E_{\epsilon }$ with $\mu (E_{\epsilon })<\infty$ such that ${\int }_{X\setminus E_{\epsilon }}|f_n|d\mu <\epsilon$ for every $n$ ^{[Vitali 1907]}.

The Vitali theorem is sharper than DCT in the finite-measure case: it replaces the dominating-function hypothesis with the uniform-integrability condition, which is a more flexible characterisation of sequences whose integrals converge with their pointwise limits. Vitali's theorem is the standard tool in martingale theory and in the modern treatment of weak compactness in $L^1$ via the Dunford-Pettis theorem.

Theorem 4 (absolute continuity of the integral). If $f\in L^1(X,\mu )$, then for every $\epsilon >0$ there exists $\delta >0$ such that for every measurable $E$ with $\mu (E)<\delta$, $$\left|{\int }_{E}fd\mu \right|\le {\int }_E|f|d\mu <\epsilon .$$

Proof. Split $|f|=|f|{\chi }_{\{|f|\le N\}}+|f|{\chi }_{\{|f|>N\}}$. The first piece is bounded by $N$. The second piece has $\int |f|{\chi }_{\{|f|>N\}}d\mu \to 0$ as $N\to \infty$ by dominated convergence (the functions $|f|{\chi }_{\{|f|>N\}}$ converge to $0$ a.e. and are dominated by $|f|$). Pick $N$ with $\int |f|{\chi }_{\{|f|>N\}}<\epsilon /2$. Then for $\mu (E)<\epsilon /(2N)$: $${\int }_E|f|d\mu \le {\int }_E|f|{\chi }_{\{|f|\le N\}}d\mu +{\int }_E|f|{\chi }_{\{|f|>N\}}d\mu \le N\mu (E)+\epsilon /2<\epsilon .\square$$

Theorem 5 (Lebesgue criterion for Riemann integrability; Lebesgue 1902). A bounded function $f:[a,b]\to \mathbb{R}$ is Riemann-integrable if and only if $f$ is continuous almost everywhere with respect to Lebesgue measure on $[a,b]$. In that case, the Riemann and Lebesgue integrals agree: $${\int }_a^{b}f(x)dx={\int }_{[a,b]}fdm.$$

The forward direction uses the Darboux upper and lower sums and the structure of the discontinuity set as a $G_{\delta }$ in $[a,b]$. The reverse direction uses that a bounded a.e.-continuous function is the pointwise a.e. limit of its Riemann step-function approximations, which converge uniformly outside the discontinuity set. The theorem is the precise statement of why "Riemann is Lebesgue when both apply" but Lebesgue strictly extends Riemann (e.g., ${\chi }_{\mathbb{Q}}$ is discontinuous everywhere on $[0,1]$, so not Riemann-integrable, but is Lebesgue-integrable with integral $0$).

Theorem 6 (improper Riemann is not absolute Lebesgue; the $\sin x/x$ counterexample). The function $f(x)=\sin (x)/x$ on $[0,\infty )$ (extended by continuity to $f(0)=1$) is improper-Riemann-integrable with value $\pi /2$, but $|f|$ is not Lebesgue-integrable on $[0,\infty )$: ${\int }_0^{\infty }|\sin (x)/x|dm=+\infty$.

Proof of divergence. On each interval $[(k-1)\pi ,k\pi ]$, $|\sin x|\ge |\sin x_k|$ where $x_k$ is the midpoint, so $|\sin x|\ge \sin (\pi /2)=1$ for $x$ within $\pi /4$ of $x_k$. The integral of $|\sin x|/x$ on $[(k-1)\pi ,k\pi ]$ exceeds a constant times $1/(k\pi )$, and the harmonic-like sum ${\sum }_{k}1/k$ diverges.

This counterexample marks the boundary between two distinct integration theories: improper Riemann is a limit of bounded-domain integrals and can be conditionally convergent (sensitive to the order of summation, like a conditionally-convergent series), while Lebesgue is absolutely convergent by construction. The Henstock-Kurzweil "gauge" integral ^{[Henstock 1961]} resolves this by replacing the Riemann tag-and-mesh definition with a tag-with-pointwise-gauge definition, recovering all improper-Riemann integrable functions while preserving the absolute-convergence Lebesgue framework as a sub-theory.

Theorem 7 (Daniell integral; Daniell 1918). Daniell 1918 ^{[Daniell 1918]} gave an alternative functional-analytic foundation for the Lebesgue integral. Start with a vector lattice $\mathcal{L}$ of bounded real-valued functions on a set $X$ closed under pointwise minimum and maximum (the standard example: continuous functions of compact support on a locally compact Hausdorff space). Specify a positive linear functional $I:\mathcal{L}\to \mathbb{R}$ satisfying the Daniell continuity property: $f_n\downarrow 0$ pointwise on $X$ implies $I(f_n)\to 0$. The Daniell construction extends $I$ to a strictly larger class $L^1$ of "Daniell integrable" functions in such a way that the extension preserves the lattice structure and satisfies monotone convergence.

The equivalence between the Daniell extension and the measure-theoretic Lebesgue integral is the Daniell-Stone representation theorem: every Daniell integral arises from a uniquely-determined Borel measure $\mu$ on $X$ via $I(f)={\int }_{X}fd\mu$. The Daniell approach is the conceptual backbone of the Riesz-Markov-Kakutani theorem (every positive linear functional on $C_c(X)$ is integration against a Radon measure), and underlies the modern treatment of integration on locally compact groups (Bourbaki Intégration).

Theorem 8 (Bochner integral; Bochner 1933). Let $(X,\mathcal{M},\mu )$ be a measure space and let $E$ be a Banach space. A strongly measurable $f:X\to E$ (Pettis 02.07.03 sense) is Bochner integrable when ${\int }_X\Vert f{\Vert }_{E}d\mu <\infty$. The Bochner integral ${\int }_{X}fd\mu \in E$ is defined as the limit of integrals of Banach-valued simple functions ${\phi }_n={\sum }_i{v}_i{\chi }_{A_i}$ with $v_i\in E$ and $\Vert {\phi }_n-f{\Vert }_E\to 0$ in $L^1$ ^{[Bochner 1933]}.

The Bochner integral satisfies the same linearity, monotonicity (where applicable), and limit theorems as the scalar Lebesgue integral: monotone convergence for non-negative scalar-valued functions of $\Vert f_n{\Vert }_E$, Fatou, and dominated convergence all carry over. The Bochner integral is the standard integration framework for Banach-valued random variables in modern probability theory and for operator-valued functions in spectral theory and harmonic analysis on non-commutative groups. The Pettis integral provides a weaker theory when strong measurability fails (typically because the target Banach space is non-separable).

Theorem 9 (Itô integral; Itô 1944). Let $(\Omega ,\mathcal{F},P)$ be a probability space with a filtration $\{{\mathcal{F}}_t{\}}_{t\ge 0}$ and a Brownian motion $\{B_t\}$ adapted to ${\mathcal{F}}_t$. For a progressively measurable square-integrable process $\{f_t\}$, the Itô stochastic integral ${\int }_0^T{f}_{t}d{B}_t$ is defined as the $L^2(\Omega ,P)$-limit of Riemann-Itô sums ${\sum }_i{f}_{t_i}(B_{t_{i+1}}-B_{t_i})$ along progressively refined partitions, with sample points taken at the left endpoint of each partition cell ^{[Itô 1944]}.

The Itô integral built on the Bochner framework but with a load-bearing twist: the integrator $B_t$ is not of bounded variation almost surely, so the integral cannot be defined pathwise via Riemann-Stieltjes; the $L^2$-isometry property $\mathbb{E}[({\int }_0^T{f}_{t}d{B}_t{)}^2]={\int }_0^T\mathbb{E}[f_t^2]dt$ rescues the construction. The Itô integral is the foundation of stochastic differential equations and modern mathematical finance (Black-Scholes 1973, Merton 1973).

Theorem 10 (Henstock-Kurzweil "gauge" integral; Henstock 1961, Kurzweil 1957). A function $f:[a,b]\to \mathbb{R}$ is HK-integrable with integral $I$ when for every $\epsilon >0$ there is a positive "gauge" function $\delta :[a,b]\to (0,\infty )$ such that for every $\delta$-fine tagged partition $\{(t_i,[x_{i-1},x_i])\}$ (meaning $[x_{i-1},x_i]\subseteq (t_i-\delta (t_i),t_i+\delta (t_i))$), the Riemann sum satisfies $|{\sum }_{i}f(t_i)(x_i-x_{i-1})-I|<\epsilon$ ^{[Henstock 1961]}.

The HK integral strictly contains both the Lebesgue and the improper-Riemann integrals: every Lebesgue-integrable function on $[a,b]$ is HK-integrable with the same integral, every improper-Riemann integrable function is HK-integrable, and there exist HK-integrable functions (such as $f(x)=\sin (1/x^2)/x^2$ near zero) that are neither Lebesgue nor improper-Riemann integrable. The HK approach gives the cleanest version of the fundamental theorem of calculus: every derivative is HK-integrable, and ${\int }_a^b{F}^{\prime }(x)dx=F(b)-F(a)$ for every differentiable $F$, a recovery that the Lebesgue framework requires absolute continuity to achieve.

Synthesis. The Lebesgue construction is the foundational reason that twentieth-century analysis is limit-stable. The central insight is that the right primitive for an integration theory is "value times measure of level set" rather than "value times width of domain strip"; this is exactly the switch from Riemann-style sampling to Lebesgue-style weighing that the monotone convergence theorem certifies as compatible with arbitrary pointwise increasing limits. Putting these together with the three-step construction (simple, non-negative measurable, signed integrable), every classical limit-passage question reduces to one of three patterns — monotone (MCT), trapped (DCT via Fatou), or uniformly integrable (Vitali) — and the right answer is identified by which pattern the problem exhibits.

The structural pattern generalises through three escalations. First, scalar-valued integration extends to Banach-valued integration via Bochner 1933: the same three-step construction works with simple functions taking values in a Banach space, with the essential-separability hypothesis from Pettis 1938 02.07.03 guaranteeing the strong-measurability needed for the simple-function approximation. Second, deterministic integration extends to stochastic integration via Itô 1944: the integrator is replaced by a Brownian motion, the pathwise approach fails because Brownian paths are not of bounded variation, but the $L^2$-isometry property rescues the construction and identifies the stochastic integral with the Bochner integral on the path space. Third, the abstract Daniell framework 1918 shows that the entire theory can be developed functorially from a positive linear functional on a vector lattice, with the measure-theoretic Lebesgue integral and the Riesz-Markov-Kakutani representation theorem as two faces of the same coin. The bridge is between the descriptive theory (sigma-algebras, measurable sets) and the functional theory (positive linear functionals, Banach lattices), and putting them together identifies the integration theory with the right linearly-ordered functional calculus on the right function space, a pattern that recurs in non-commutative integration on operator algebras (Segal 1953, Nelson 1974) and in the modern $\infty$-categorical theory of measure spaces (Pavlov 2022).

Full proof set Master

Proposition 1 (monotone convergence for simple functions). Let $\{{\phi }_n\}$ be a non-decreasing sequence of non-negative simple functions on $(X,\mathcal{M},\mu )$ with pointwise supremum $\phi$ also simple. Then ${\int }_X\phi d\mu ={\lim }_n{\int }_X{\phi }_{n}d\mu$.

Proof. Each ${\phi }_n$ has finite range, so write ${\phi }_n={\sum }_i{c}_{n,i}{\chi }_{A_{n,i}}$ in canonical form with disjoint $A_{n,i}$ partitioning $X$. The pointwise supremum $\phi$ also has finite range (a uniform bound on the number of distinct values is provided by the convergence to a finite-range limit). Refine all partitions to a common refinement $\{B_j\}$ with each ${\phi }_n$ constant on each $B_j$; write ${\phi }_n={\sum }_j{d}_{n,j}{\chi }_{B_j}$. On each $B_j$, the sequence $\{d_{n,j}{\}}_n$ is non-decreasing with supremum $d_j$ (the value of $\phi$ on $B_j$), so $d_{n,j}\uparrow d_j$ as $n\to \infty$. Then: $${\int }_X{\phi }_{n}d\mu =\sum _j{d}_{n,j}\mu (B_j),{\int }_X\phi d\mu =\sum _j{d}_j\mu (B_j).$$ The finite sum (in $j$) of non-decreasing sequences gives ${\lim }_n{\sum }_j{d}_{n,j}\mu (B_j)={\sum }_j{d}_j\mu (B_j)$. $\square$

Proposition 2 (linearity of integral for non-negative measurable functions). For non-negative measurable $f,g:X\to [0,\infty ]$ and $\alpha ,\beta \ge 0$, ${\int }_X(\alpha f+\beta g)d\mu =\alpha {\int }_{X}fd\mu +\beta {\int }_{X}gd\mu$.

Proof. By the simple-function approximation theorem 02.07.03, choose monotone non-decreasing sequences ${\phi }_n\uparrow f$ and ${\psi }_n\uparrow g$ of non-negative simple functions. Then $\alpha {\phi }_n+\beta {\psi }_n\uparrow \alpha f+\beta g$ is a non-decreasing sequence of non-negative simple functions with pointwise limit $\alpha f+\beta g$. By the monotone convergence theorem applied to $\{\alpha {\phi }_n+\beta {\psi }_n\}$: $${\int }_X(\alpha f+\beta g)d\mu =\underset{n}{\lim }{\int }_X(\alpha {\phi }_n+\beta {\psi }_n)d\mu .$$ By Exercise 3 (linearity for simple functions), $${\int }_X(\alpha {\phi }_n+\beta {\psi }_n)d\mu =\alpha {\int }_X{\phi }_{n}d\mu +\beta {\int }_X{\psi }_{n}d\mu .$$ By MCT again applied separately to $\{{\phi }_n\}$ and $\{{\psi }_n\}$, the right side has limit $\alpha {\int }_{X}fd\mu +\beta {\int }_{X}gd\mu$.

$\square$

Proposition 3 (Fatou and DCT from MCT). Restated and proven as Theorems 1 and 2 above. The load-bearing chain of inferences: MCT (Beppo Levi 1906) is the primitive limit theorem; Fatou's lemma (1906) is the $\mathrm{liminf}$ version obtained by applying MCT to the increasing infimum-tail sequence $g_n={\inf }_{k\ge n}{f}_k$; DCT (Lebesgue 1908) is obtained by applying Fatou to $g-f_n$ and $g+f_n$ separately to bracket $\lim \int f_n$ from above and below.

Proof. See Theorems 1 and 2.

$\square$

Proposition 4 (measure-theoretic continuity of increasing sequences without finite measure). Let $\{E_n\}$ be an increasing sequence in $\mathcal{M}$ with union $E$. Then $\mu (E)={\lim }_n\mu (E_n)$.

Proof. By countable additivity, $\mu (E)=\mu (E_1)+{\sum }_{n=1}^{\infty }\mu (E_{n+1}\setminus E_n)$. The partial sum equals $\mu (E_N)$, so the series limit equals ${\lim }_N\mu (E_N)$. Combining, $\mu (E)={\lim }_N\mu (E_N)$. The proof does not use the finite-measure hypothesis (in contrast to decreasing-sequence continuity, which does). This proposition is the load-bearing measure-theoretic fact used inside the MCT proof, applied to the sets $A_i\cap E_n$.

$\square$

Proposition 5 (Riemann-Lebesgue compatibility on bounded intervals). Let $f:[a,b]\to \mathbb{R}$ be Riemann-integrable. Then $f$ is Lebesgue-integrable on $[a,b]$ with respect to Lebesgue measure, and the two integrals agree.

Proof. Riemann-integrability of $f$ implies $f$ is continuous a.e. on $[a,b]$ (Lebesgue criterion, Theorem 5). A continuous-a.e. bounded function on a compact interval is measurable (the discontinuity set is a $G_{\delta }$ of measure zero, and modification on a null set preserves measurability). Riemann's upper and lower sums $U(f,P_n)$, $L(f,P_n)$ along a sequence of partitions $P_n$ with mesh $\to 0$ converge to the common Riemann integral. By the definition of the Lebesgue integral as a supremum over simple-function approximations from below, ${\int }_{[a,b]}fdm\ge L(f,P_n)$ for every $n$ (each lower Riemann sum is the Lebesgue integral of a simple function below $f$). Symmetrically, ${\int }_{[a,b]}fdm\le U(f,P_n)$ for every $n$. Taking $n\to \infty$, the Riemann integral and the Lebesgue integral are sandwiched together and must agree. $\square$

Proposition 6 (absolute continuity of the integral implies Radon-Nikodym density). For $f\in L^1(X,\mu )$ with $f\ge 0$, the set function $\nu (E)={\int }_{E}fd\mu$ is a measure on $\mathcal{M}$ absolutely continuous with respect to $\mu$, and the Radon-Nikodym derivative $d\nu /d\mu$ equals $f$ almost everywhere.

Proof. That $\nu$ is a measure is Exercise 7 (using MCT for countable additivity). Absolute continuity $\mu (E)=0⟹\nu (E)=0$ is direct: if $\mu (E)=0$, then ${\int }_{E}fd\mu =0$ by the convention $0\cdot \infty =0$ on null sets. The full Radon-Nikodym theorem 02.07.08 gives the converse: every $\mu$-absolutely continuous $\sigma$-finite measure $\nu$ arises this way for a unique (modulo a.e. equivalence) non-negative measurable $f$. $\square$

Connections Master

• Measurable functions, simple functions, Egorov's theorem, and Lusin's theorem 02.07.03. Supplies the framework on which the Lebesgue integral is built: the pre-image-of-Borel-sets characterisation of measurable functions, the simple-function approximation theorem providing the dyadic staircase climbing toward any non-negative measurable function, and the closure properties (pointwise limits, suprema, infima) preserving measurability through every step of the integral construction.

• Lebesgue outer measure and the Carathéodory construction 02.07.02. Defines the underlying Lebesgue measure $m$ on ${\mathbb{R}}^n$ whose existence and translation invariance make the Lebesgue integral the canonical extension of the geometric notion of area. The Carathéodory completeness property of the Lebesgue sigma-algebra ensures that integrals are invariant under modification on null sets, the structural fact behind almost-everywhere equivalence in $L^1$.

• Fatou's lemma, Lebesgue dominated convergence, and Vitali convergence 02.07.05. The direct downstream consequences of the monotone convergence theorem. Fatou's lemma is the $\mathrm{liminf}$ version obtained from MCT via infimum-tail telescoping; the dominated convergence theorem combines two Fatou applications to bracket the limit. The Vitali convergence theorem refines DCT for finite-measure spaces by replacing the dominating-function hypothesis with uniform integrability.

• $L^p$ spaces, Hölder, Minkowski, and Riesz-Fischer completeness 02.07.06. Built on the Lebesgue integral constructed here. The $L^p$ norm $\Vert f{\Vert }_p=(\int |f{|}^{p}d\mu {)}^{1/p}$ depends on the Lebesgue integral for its definition; the Hölder and Minkowski inequalities give the metric structure; and the Riesz-Fischer completeness theorem (a direct consequence of MCT applied to Cauchy sequences of simple functions) makes $L^p$ a Banach space.

• Riemann integral and the Lebesgue criterion 02.04.02 pending. The classical Riemann integral on bounded intervals. The Lebesgue criterion (Theorem 5) identifies Riemann-integrability with continuity almost everywhere; the Lebesgue integral strictly extends the Riemann integral, agreeing where the Riemann integral is defined and providing values for functions like ${\chi }_{\mathbb{Q}}$ where Riemann's procedure fails. The $\sin (x)/x$ counterexample (Theorem 6) marks the boundary where the absolutely-convergent Lebesgue framework diverges from the conditionally-convergent improper-Riemann framework.

Historical & philosophical context Master

Lebesgue's 1902 PhD thesis Intégrale, longueur, aire in the Annali di Matematica Pura ed Applicata ^{[Lebesgue 1902]} introduced the integration framework that now bears his name. Lebesgue's original construction proceeded through the inner-outer measure of the graph: a bounded function on a measurable set is Lebesgue-integrable when the inner and outer approximations of its hypograph (by closed and open sets) have the same Lebesgue measure. The modern three-step construction (simple, non-negative measurable, signed integrable) was crystallised in Lebesgue's 1908 Leçons sur l'intégration second edition ^{[Lebesgue 1908]}, the canonical textbook exposition.

Beppo Levi's 1906 Rendiconti del R. Istituto Lombardo note ^{[BeppoLevi 1906]} established the monotone convergence theorem. Levi's original statement was for series of non-negative measurable functions: ${\sum }_n\int f_n=\int {\sum }_n{f}_n$ for non-negative $f_n$. The supremum-of-monotone-sequence version is the equivalent restatement obtained by setting the supremum equal to a single function. The theorem is sometimes called the Beppo Levi theorem in the Italian tradition and the monotone convergence theorem in the modern English tradition; the two names refer to the same result.

Fatou's 1906 Acta Mathematica paper ^{[Fatou 1906]} on trigonometric and Taylor series included as a lemma the inequality $\int \mathrm{liminf}{f}_n\le \mathrm{liminf}\int f_n$ that is now the standard Fatou's lemma. The lemma was a tool in Fatou's main result on the boundary behaviour of bounded analytic functions in the unit disc (the Fatou theorem on radial limits), and only later was extracted as a general measure-theoretic result.

Vitali's 1907 Rendiconti del Circolo Matematico di Palermo paper ^{[Vitali 1907]} gave a sharper version of dominated convergence for finite-measure spaces. Vitali's uniform-integrability condition replaces the dominating function with a quantitative absolute-continuity bound on the integrals over small-measure sets, allowing convergence in cases where no integrable upper envelope exists. The Vitali convergence theorem is the standard tool in martingale theory and in the modern characterisation of weakly compact subsets of $L^1$ via the Dunford-Pettis theorem.

Lebesgue's own 1908 monograph contains the dominated convergence theorem and the absolute-continuity of the integral property, completing the foundational set of limit theorems. The descriptive characterisation of Riemann-integrability as continuity almost everywhere (Lebesgue 1902, Theorem 5) was the headline result of Lebesgue's thesis and gave the first precise account of why some functions (continuous, monotone) admit a classical integral while others (Dirichlet's ${\chi }_{\mathbb{Q}}$) require the new theory.

Daniell's 1918 Annals of Mathematics paper ^{[Daniell 1918]} gave an alternative foundation. Daniell started with a vector lattice of bounded functions and a positive linear functional satisfying the Daniell continuity property, and extended the functional to a strictly larger class by a procedure parallel to the Carathéodory extension of an outer measure. The Daniell-Stone representation theorem identifies every Daniell integral with integration against a Borel measure; this is the conceptual backbone of the Riesz-Markov-Kakutani theorem (Riesz 1909 Comptes Rendus 149, Markov 1938, Kakutani 1941 Ann. Math. 42) characterising positive linear functionals on $C_c(X)$ as integrals against Radon measures on locally compact Hausdorff spaces.

Bochner's 1933 Fundamenta Mathematicae paper ^{[Bochner 1933]} extended the Lebesgue integral to Banach-valued functions. The Bochner construction follows the same three-step pattern as the scalar Lebesgue integral, with simple functions taking values in a Banach space, the strong-measurability hypothesis from Pettis 1938 02.07.03 guaranteeing the simple-function approximation, and the integrability condition $\int \Vert f{\Vert }_{E}d\mu <\infty$ replacing the scalar finiteness condition. The Bochner integral underlies modern probability theory of Banach-valued random variables (Ledoux-Talagrand 1991 Probability in Banach Spaces), spectral theory of operator-valued functions (Dunford-Schwartz 1958-71 Linear Operators), and harmonic analysis on non-commutative groups.

Itô's 1944 Proceedings of the Imperial Academy Tokyo paper ^{[Ito 1944]} extended the Bochner framework to stochastic integration. The Itô integral ${\int }_0^T{f}_{t}d{B}_t$ against a Brownian motion $B_t$ cannot be defined pathwise because Brownian paths are not of bounded variation almost surely; instead, Itô used the $L^2$-isometry $\mathbb{E}[(\int fdB{)}^2]=\int \mathbb{E}[f^2]dt$ to define the integral as an $L^2$-limit of left-endpoint Riemann-Itô sums. The Itô integral is the foundation of stochastic differential equations and modern mathematical finance (Black-Scholes 1973 J. Polit. Econ. 81, Merton 1973 Bell J. Econ. 4).

Henstock 1955-61 and Kurzweil 1957 ^{[Henstock 1961]} introduced the gauge integral, an alternative refinement of the Riemann definition that strictly contains both the Lebesgue and improper-Riemann integrals. The HK construction uses a positive gauge function $\delta :[a,b]\to (0,\infty )$ instead of a uniform mesh, and the resulting integral integrates every derivative — a property the Lebesgue framework requires absolute continuity to achieve. The HK integral is the cleanest version of the fundamental theorem of calculus for general differentiable functions and has growing applications in non-absolutely-convergent integration theory (Bartle 2001, Gordon 1994).

The structural story of integration theory is a hundred-year escalation: scalar Lebesgue 1902 → countable-additivity machinery (Beppo Levi, Fatou, Vitali, Lebesgue) 1906-1908 → Banach-valued Bochner 1933 → stochastic Itô 1944 → Daniell-functional 1918 → operator-algebraic non-commutative (Segal 1953 Ann. Math. 57, Nelson 1974) → modern $\infty$-categorical formulations (Pavlov 2022). Each step extends the previous to a more general setting while preserving the load-bearing limit theorems (MCT, Fatou, DCT), and the result is a body of theorems that organises every limit-passage question in analysis under one of the three canonical patterns (monotone, dominated, uniformly integrable).

Bibliography Master

@article{Lebesgue1902,
  author  = {Lebesgue, Henri},
  title   = {Int\'egrale, longueur, aire},
  journal = {Annali di Matematica Pura ed Applicata},
  volume  = {7},
  year    = {1902},
  pages   = {231--359}
}

@article{BeppoLevi1906,
  author  = {Levi, Beppo},
  title   = {Sopra l'integrazione delle serie},
  journal = {Rendiconti del Reale Istituto Lombardo di Scienze e Lettere},
  volume  = {15},
  series  = {5},
  year    = {1906},
  pages   = {775--780}
}

@article{Fatou1906,
  author  = {Fatou, Pierre},
  title   = {S\'eries trigonom\'etriques et s\'eries de Taylor},
  journal = {Acta Mathematica},
  volume  = {30},
  year    = {1906},
  pages   = {335--400}
}

@article{Vitali1907,
  author  = {Vitali, Giuseppe},
  title   = {Sull'integrazione per serie},
  journal = {Rendiconti del Circolo Matematico di Palermo},
  volume  = {23},
  year    = {1907},
  pages   = {137--155}
}

@book{Lebesgue1908,
  author    = {Lebesgue, Henri},
  title     = {Le\c{c}ons sur l'int\'egration et la recherche des fonctions primitives},
  edition   = {2},
  publisher = {Gauthier-Villars},
  address   = {Paris},
  year      = {1908}
}

@article{Daniell1918,
  author  = {Daniell, Percy J.},
  title   = {A general form of integral},
  journal = {Annals of Mathematics},
  volume  = {19},
  series  = {2},
  year    = {1918},
  pages   = {279--294}
}

@article{Bochner1933,
  author  = {Bochner, Salomon},
  title   = {Integration von Funktionen, deren Werte die Elemente eines Vektorraumes sind},
  journal = {Fundamenta Mathematicae},
  volume  = {20},
  year    = {1933},
  pages   = {262--276}
}

@article{Ito1944,
  author  = {It\^o, Kiyosi},
  title   = {Stochastic integral},
  journal = {Proceedings of the Imperial Academy Tokyo},
  volume  = {20},
  year    = {1944},
  pages   = {519--524}
}

@article{Henstock1961,
  author  = {Henstock, Ralph},
  title   = {Definitions of {R}iemann type of the variational integrals},
  journal = {Proceedings of the London Mathematical Society},
  volume  = {11},
  series  = {3},
  year    = {1961},
  pages   = {402--418}
}

@book{Halmos1950,
  author    = {Halmos, Paul R.},
  title     = {Measure Theory},
  publisher = {Van Nostrand},
  year      = {1950},
  note      = {Reprinted as Springer GTM 18, 1974}
}

@book{Folland1999,
  author    = {Folland, Gerald B.},
  title     = {Real Analysis: Modern Techniques and Their Applications},
  edition   = {2},
  publisher = {Wiley},
  year      = {1999}
}

@book{Rudin1987,
  author    = {Rudin, Walter},
  title     = {Real and Complex Analysis},
  edition   = {3},
  publisher = {McGraw-Hill},
  year      = {1987}
}

@book{Bogachev2007,
  author    = {Bogachev, Vladimir I.},
  title     = {Measure Theory, Volume 1},
  publisher = {Springer},
  year      = {2007}
}

@book{RoydenFitzpatrick2010,
  author    = {Royden, H. L. and Fitzpatrick, P. M.},
  title     = {Real Analysis},
  edition   = {4},
  publisher = {Pearson},
  year      = {2010}
}

@book{Tao2016,
  author    = {Tao, Terence},
  title     = {Analysis II},
  edition   = {3},
  publisher = {Hindustan Book Agency},
  year      = {2016}
}