02.07.03 · analysis / measure-theory

Measurable Functions, Simple Functions, Egorov's Theorem, and Lusin's Theorem

shipped3 tiersLean: full

Anchor (Master): Halmos, Measure Theory §III; Bogachev, Measure Theory Vol. 1 §2.1-2.3

Intuition Beginner

A measurable function is the right kind of function to integrate. Continuous functions are smooth enough to integrate by the older Riemann method, but plenty of useful functions are not continuous: the indicator of the rational numbers, the limit of a wiggling sequence, the pointwise sum of infinitely many step jumps. Measure theory needs a wider class. Measurable functions are exactly that class, and the definition is built from the sigma-algebra framework of the previous two units.

The rule is short. A function $f$ between two measurable spaces is measurable when the pre-image of every measurable set in the target lands in the sigma-algebra of the domain. Pre-images are the test, not images. The phrase "pre-image of every open set is open" is the topological version. Swapping "open" for "measurable" on both sides gives the measure-theoretic version. Continuity asks for openness; measurability asks for membership in the rule book.

The reason this is the right class becomes clear once you start adding, multiplying, and taking limits. The sum of two measurable functions is measurable. The product is measurable. The pointwise supremum of a countable family is measurable. The pointwise limit, if it exists everywhere, is measurable. Continuity has none of these closure properties for free: the pointwise limit of a sequence of continuous functions can be wildly discontinuous. Measurability handles all of these by routine sigma-algebra closure arguments.

Simple functions are the easiest measurable functions. A simple function takes only finitely many distinct values, and each value is achieved on a measurable set. You can write any simple function as a finite sum: each term is a constant times the indicator of a measurable set. Indicators are the smallest building blocks; simple functions are finite combinations of them; and every non-negative measurable function is a pointwise increasing limit of simple functions. That last fact is the engine of the Lebesgue integral.

Two famous results sharpen the picture. Egorov's theorem says that if a sequence of measurable functions converges almost everywhere on a set of finite measure, then off a small exceptional set the convergence is uniform. Lusin's theorem says that any measurable function on a region of the line is continuous on the complement of a small set. In short: measurable functions are almost continuous, and almost-everywhere convergence is almost uniform. The word "almost" hides a measure-zero or arbitrarily-small-measure gap, but the rest of the picture is very clean.

A useful slogan from Littlewood organises this picture into three rules of thumb. Every measurable set is almost an interval, every measurable function is almost continuous, every almost-everywhere convergent sequence is almost uniformly convergent. The word "almost" in each rule covers an exceptional set of arbitrarily small measure that we are willing to throw away. The three rules together explain why measure-theoretic analysis works: the difficult objects (measurable sets, measurable functions, pointwise limits) reduce, modulo a small error, to the easy objects (intervals, continuous functions, uniform limits) whose handling is classical. Whenever a measure-theoretic argument feels tractable, one of Littlewood's principles is usually doing the work.

The one-sentence takeaway: measurable functions are the closure of continuous functions under all the operations integration cares about, and the simple-function ladder is the staircase by which any of them can be approximated.

Visual Beginner

The mental picture is a staircase climbing under a curve. Start with a non-negative function on the real line, drawn as a smooth hump. Slice the height axis into a few horizontal levels, say $0$ to $1$ then $1$ to $2$ then $2$ to $3$ . On the domain, the set of points where the function lies between $0$ and $1$ is some collection of intervals; the set where the function lies between $1$ and $2$ is another collection; and so on. Place a constant value (the lower edge of each height-band) over each of these domain pieces. The resulting step function sits underneath the curve and approximates it from below.

Now refine the slicing. Cut the height axis into bands of width $1/2$ , then $1/4$ , then $1/ 2^{n}$ . The staircase gets finer; each step is a thinner sliver; the approximation climbs closer to the curve. The bottom panel shows the Egorov picture: a sequence of measurable functions whose graphs wiggle but settle pointwise to a limit; the wiggling is fast on a small set (shaded) and uniformly small elsewhere.

Worked example Beginner

We approximate the function $f (x) = x$ on $[0, 1]$ by simple functions using dyadic slicing.

Step 1. Cut the height into $4$ equal bands of width $1/4$ : the bands $[0, 1/4)$ , $[1/4, 1/2)$ , $[1/2, 3/4)$ , $[3/4, 1)$ on the height axis, plus the top value $1$ .

Step 2. For each band $[k /4, (k + 1) /4)$ with $k = 0, 1, 2, 3$ , the set of $x$ in $[0, 1]$ with $f (x) = x$ falling into this band is the interval $[k /4, (k + 1) /4)$ .

Step 3. Define the simple function $φ_{2} (x) = k /4$ for $x \in [k /4, (k + 1) /4)$ , with $φ_{2} (1) = 1$ . This staircase has four steps of height $0$ , $1/4$ , $1/2$ , $3/4$ .

Step 4. Refine to width $1/8$ . The new simple function $φ_{3}$ has eight steps with heights $0$ , $1/8$ , $2/8$ , $3/8$ , $4/8$ , $5/8$ , $6/8$ , $7/8$ over intervals of length $1/8$ . It approximates $f (x) = x$ more closely.

Step 5. At slicing width $1/ 2^{n}$ , the simple function $φ_{n} (x) = ⌊ 2^{n} x ⌋ / 2^{n}$ satisfies $0 \leq f (x) - φ_{n} (x) < 1/ 2^{n}$ for every $x \in [0, 1]$ . As $n$ grows, the difference shrinks uniformly to zero.

What this tells us: the function $f (x) = x$ is the increasing pointwise limit of explicit simple functions, and the approximation is uniform on a bounded interval.

Check your understanding Beginner

Formal definition Intermediate+

Let $(X, M)$ and $(Y, N)$ be measurable spaces. A map $f : X \to Y$ is $(M, N)$ -measurable when $f^{- 1} (B) \in M$ for every $B \in N$ ^{[Folland §2.1]}. When $Y = R$ or $\overset{ˉ}{R} = [- \infty, + \infty]$ with the Borel sigma-algebra, $f$ is called Borel-measurable, and when the domain sigma-algebra is the Lebesgue sigma-algebra $f$ is called Lebesgue-measurable.

For real-valued functions the definition reduces to a one-parameter condition. The rays $(a, \infty)$ for $a \in R$ generate $B (R)$ , so $f : (X, M) \to (R, B (R))$ is measurable if and only if ${x \in X : f (x) > a} \in M$ for every $a \in R$ . Equivalent formulations use ${f \geq a}$ , ${f < a}$ , or ${f \leq a}$ ; each generating family produces the same sigma-algebra.

Closure under algebraic operations. If $f, g : X \to R$ are measurable and $c \in R$ , then $f + g$ , $c f$ , $f g$ , $∣ f ∣$ , $max (f, g)$ , and $min (f, g)$ are measurable. The argument for $f + g$ uses the identity ${x : f (x) + g (x) > a} = q \in Q ⋃ {x : f (x) > q} \cap {x : g (x) > a - q},$ a countable union of measurable rectangles.

Closure under countable limits. If ${f_{n}}_{n \in N}$ is a sequence of measurable functions $X \to \overset{ˉ}{R}$ , then $sup_{n} f_{n}$ , $in f_{n} f_{n}$ , $lim sup_{n} f_{n}$ , and $lim inf_{n} f_{n}$ are measurable, and the limit $lim_{n} f_{n}$ is measurable wherever it exists. The supremum identity is ${x : n sup f_{n} (x) > a} = n ⋃ {x : f_{n} (x) > a} .$ The other identities follow by complementation, the formula $lim sup_{n} f_{n} = in f_{k} sup_{n \geq k} f_{n}$ , and its $lim inf$ counterpart.

Almost-everywhere equivalence. Given a measure $μ$ on $M$ , two functions $f, g : X \to \overset{ˉ}{R}$ are equal $μ$ -almost everywhere (written $f = g$ a.e.) when $μ ({x : f (x) \neq = g (x)}) = 0$ . Almost-everywhere equality is an equivalence relation, and measurability is preserved within an equivalence class when the underlying measure is complete (every subset of a null set is measurable).

Simple functions. A simple function $φ : X \to R$ is a measurable function taking only finitely many distinct values. Equivalently, $φ = \sum_{i = 1}^{n} c_{i} χ_{A_{i}}$ where the $A_{i}$ are pairwise disjoint measurable sets and $χ_{A}$ denotes the characteristic function of $A$ (the function equal to $1$ on $A$ and $0$ off $A$ ). The canonical representation uses the distinct values $c_{1}, \dots, c_{n}$ of $φ$ and the level sets $A_{i} = φ^{- 1} ({c_{i}})$ .

Counterexamples to common slips Intermediate+

Pointwise limits of continuous functions are measurable but need not be continuous. The function $f (x) = χ_{Q} (x)$ is the pointwise limit of $f_{n} (x) = cos^{2 n} (n! π x)$ , but $f$ is discontinuous at every real point.
Image of a measurable set need not be measurable. The standard counterexample uses a continuous map $R \to R$ sending a measurable set to a non-Lebesgue-measurable set; the converse direction (pre-image of measurable) is the load-bearing one.
Pointwise convergence does not imply uniform convergence. The sequence $f_{n} (x) = x^{n}$ on $[0, 1]$ converges pointwise to $χ_{{1}}$ but not uniformly. Egorov's theorem is the partial rescue.
Egorov fails on infinite-measure spaces. The sequence $f_{n} = χ_{[n, n + 1]}$ on $R$ converges to $0$ pointwise but no exceptional set of arbitrarily small measure makes the convergence uniform.

Key theorem with proof Intermediate+

Theorem (simple-function approximation). Let $(X, M)$ be a measurable space and $f : X \to [0, \infty]$ a non-negative measurable function. There exists a sequence of simple functions ${φ_{n}}_{n \geq 1}$ with $0 \leq φ_{1} \leq φ_{2} \leq \dots \leq f$ and $φ_{n} (x) \to f (x)$ for every $x \in X$ . If $f$ is bounded, the convergence is uniform on $X$ .

Proof. Fix $n \geq 1$ . Slice the half-line $[0, n]$ into $n \cdot 2^{n}$ dyadic bands of width $2^{- n}$ , and bundle everything above $n$ into a single top band. Concretely, define $φ_{n} (x) = {k \cdot 2^{- n} n if k \cdot 2^{- n} \leq f (x) < (k + 1) \cdot 2^{- n}, 0 \leq k < n \cdot 2^{n}, if f (x) \geq n .$

Each level set ${x : k \cdot 2^{- n} \leq f (x) < (k + 1) \cdot 2^{- n}} = f^{- 1} ([k \cdot 2^{- n}, (k + 1) \cdot 2^{- n}))$ is measurable, since $f$ is measurable and the half-open interval is Borel. The cap set ${x : f (x) \geq n} = f^{- 1} ([n, \infty])$ is likewise measurable. So $φ_{n}$ is a finite sum of measurable indicator functions times non-negative constants, hence simple and non-negative.

Monotonicity. Compare $φ_{n}$ and $φ_{n + 1}$ . On the cap region ${f \geq n + 1}$ , $φ_{n + 1} = n + 1 > n = φ_{n}$ . On the band ${n \leq f < n + 1}$ , $φ_{n} = n$ and $φ_{n + 1}$ takes values in $[n, n + 1)$ at the level $n + j \cdot 2^{- (n + 1)}$ for the relevant $j$ , so $φ_{n + 1} \geq n = φ_{n}$ . On the lower bands ${f < n}$ , halving the band width can only refine the floor: a value $f (x) \in [k \cdot 2^{- n}, (k + 1) \cdot 2^{- n})$ lies in $[2 k \cdot 2^{- (n + 1)}, (2 k + 1) \cdot 2^{- (n + 1)})$ or $[(2 k + 1) \cdot 2^{- (n + 1)}, (2 k + 2) \cdot 2^{- (n + 1)})$ , so $φ_{n + 1} (x) \in {2 k \cdot 2^{- (n + 1)}, (2 k + 1) \cdot 2^{- (n + 1)}}$ , and both options exceed or equal $φ_{n} (x) = k \cdot 2^{- n} = 2 k \cdot 2^{- (n + 1)}$ .

Pointwise convergence. Fix $x \in X$ .

Case 1: $f (x) = + \infty$ . Then $f (x) \geq n$ for every $n$ , so $φ_{n} (x) = n \to \infty = f (x)$ .

Case 2: $f (x) < \infty$ . Choose $N$ large enough that $f (x) < N$ . For every $n \geq N$ , the value $f (x)$ falls in some band $[k \cdot 2^{- n}, (k + 1) \cdot 2^{- n})$ with $k < n \cdot 2^{n}$ , and $φ_{n} (x) = k \cdot 2^{- n}$ satisfies $0 \leq f (x) - φ_{n} (x) < 2^{- n}$ . As $n \to \infty$ , the gap shrinks to $0$ , so $φ_{n} (x) \to f (x)$ .

Uniformity for bounded $f$ . If $sup_{X} f \leq M < \infty$ , then for every $n \geq M$ and every $x$ , $f (x) < n$ falls into some band of width $2^{- n}$ , so $sup_{X} ∣ f - φ_{n} ∣ < 2^{- n}$ . The convergence is uniform on $X$ .

$□$

Bridge. The dyadic simple-function approximation builds toward 02.07.04 the Lebesgue integral, where the integral of a non-negative measurable function is defined as the supremum of integrals of simple functions below it, and the existence of the approximation here is exactly what guarantees the supremum is attained as a limit rather than as a transfinite construction. The foundational reason the dyadic recipe works is that the rationals are dense in the reals, so refining the level partition by halving the bandwidth produces successively finer approximations from below. This is exactly the same constructive pattern as the Carathéodory completion from 02.07.02, where approximation from outside built the Lebesgue sigma-algebra. The bridge is between the descriptive theory (measurability as a sigma-algebra condition) and the constructive theory (every measurable function is a limit of computable building blocks), and putting these together with the algebraic-closure properties identifies the measurable functions with the smallest function class containing the simple functions and closed under monotone limits.

Exercises Intermediate+

Exercise 3 (medium, symbolic).

Let $(X, M, μ)$ be a measure space and let $f_{n}, f : X \to R$ be measurable with $f_{n} \to f$ almost everywhere. Show that $f$ is measurable provided $μ$ is complete.

Hint

Define $g (x) = lim sup_{n} f_{n} (x)$ . Show $g$ is measurable everywhere. Show $g = f$ off a null set $N$ . Use completeness to make $f$ measurable on $N$ .

Answer

The function $g (x) = lim sup_{n} f_{n} (x)$ is measurable everywhere, since $lim sup$ of measurable functions is measurable.

Let $N = {x : f_{n} (x) \neq \to f (x)}$ . By hypothesis $μ (N) = 0$ . Off $N$ , $f_{n} (x) \to f (x)$ and also $g (x) = lim_{n} f_{n} (x) = f (x)$ .

On $X ∖ N$ , $f = g$ , which is measurable.

On $N$ , $f$ may differ from $g$ . For any Borel $B \subseteq R$ , $f^{- 1} (B) = (f^{- 1} (B) ∖ N) \cup (f^{- 1} (B) \cap N)$ . The first piece equals $g^{- 1} (B) ∖ N$ , which is measurable. The second piece is a subset of the null set $N$ , so by completeness it is measurable. Hence $f^{- 1} (B) \in M$ , and $f$ is measurable.

Exercise 5 (medium, symbolic).

State and prove the negative-part / positive-part decomposition. Given a measurable $f : X \to R$ , define $f^{+} = max (f, 0)$ and $f^{-} = max (- f, 0)$ . Show $f = f^{+} - f^{-}$ , $∣ f ∣ = f^{+} + f^{-}$ , and both $f^{+}, f^{-}$ are measurable.

Hint

The decomposition is pointwise. Measurability follows from the closure properties of measurable functions under $max$ and negation.

Answer

Pointwise, if $f (x) \geq 0$ : $f^{+} (x) = f (x)$ , $f^{-} (x) = 0$ , so $f^{+} - f^{-} = f (x)$ and $f^{+} + f^{-} = f (x) = ∣ f (x) ∣$ . If $f (x) < 0$ : $f^{+} (x) = 0$ , $f^{-} (x) = - f (x) > 0$ , so $f^{+} - f^{-} = 0 - (- f (x)) = f (x)$ and $f^{+} + f^{-} = - f (x) = ∣ f (x) ∣$ . Both identities hold.

Measurability: $- f$ is measurable since ${x : - f (x) > a} = {x : f (x) < - a} \in M$ . The function $0$ is measurable (constant). The maximum of two measurable functions is measurable. Hence $f^{+} = max (f, 0)$ and $f^{-} = max (- f, 0)$ are measurable.

Exercise 7 (hard, symbolic).

Construct a sequence of measurable functions on $R$ that converges to $0$ pointwise everywhere but where no exceptional set of Lebesgue measure less than $1$ makes the convergence uniform. Conclude that Egorov's theorem requires finite measure.

Hint

Use marching characteristic functions on disjoint intervals shifting to infinity, as the standard counterexample. Verify that off any set of small measure, the supremum norm of the tail stays bounded away from zero.

Answer

Define $f_{n} (x) = χ_{[n, n + 1]} (x)$ . At each fixed $x \in R$ , $f_{n} (x) = 1$ only for the single index $n \in [⌊ x ⌋ - 1, ⌊ x ⌋ + 1]$ , so $f_{n} (x) = 0$ for $n > x$ , and $f_{n} (x) \to 0$ pointwise.

Suppose $E \subseteq R$ has Lebesgue measure less than $1$ and we claim $f_{n} \to 0$ uniformly on $R ∖ E$ . Then for every $ε \in (0, 1)$ there is $N$ with $∣ f_{n} (x) ∣ < ε$ for all $n \geq N$ and all $x \in R ∖ E$ .

Take $ε = 1/2$ . For $n \geq N$ and $x \in [n, n + 1]$ , $f_{n} (x) = 1 > ε$ , so $[n, n + 1] \subseteq E$ would be forced. But $E$ has measure less than $1$ , so $E$ cannot contain $[n, n + 1]$ (whose Lebesgue measure is $1$ ). Pick any $x \in [n, n + 1] ∖ E$ ; then $f_{n} (x) = 1 \geq ε$ , violating uniform convergence on $R ∖ E$ .

This contradicts the uniform-convergence assumption. So no exceptional set of measure less than $1$ produces uniform convergence, and Egorov's hypothesis $μ (X) < \infty$ cannot be dropped.

Exercise 8 (medium, symbolic).

Show that if $f : (X, M) \to (R, B (R))$ is measurable, then so is ${x : f (x) \in B}$ for every Borel $B \subseteq R$ . Use this to construct a measurable function whose graph ${(x, f (x)) : x \in X}$ in $X \times R$ is measurable with respect to the product sigma-algebra.

Hint

The first part is the definition of measurability unpacked. For the second part, the graph is the level set of the measurable function $(x, y) \mapsto y - f (x)$ at the value $0$ .

Answer

By definition ${x : f (x) \in B} = f^{- 1} (B)$ , which is in $M$ since $f$ is measurable.

For the graph: equip $X \times R$ with the product sigma-algebra $M \otimes B (R)$ . The projections $π_{X} : X \times R \to X$ and $π_{R} : X \times R \to R$ are measurable. The function $h (x, y) = y - f (π_{X} (x, y)) = π_{R} (x, y) - f (π_{X} (x, y))$ is a difference of measurable functions, hence measurable.

The graph of $f$ is $h^{- 1} ({0})$ , the pre-image of a Borel singleton, which is measurable in $M \otimes B (R)$ .

Exercise 9 (hard, symbolic).

Prove Egorov's theorem in full. Let $(X, M, μ)$ be a measure space with $μ (X) < \infty$ , and let $f_{n}, f : X \to R$ be measurable with $f_{n} \to f$ almost everywhere. Show that for every $ε > 0$ there exists a measurable set $E_{ε} \subseteq X$ with $μ (E_{ε}) < ε$ on whose complement $f_{n} \to f$ uniformly.

Hint

Define the "bad" sets $E_{n, k} = {x : ∣ f_{n} (x) - f (x) ∣ \geq 1/ k}$ , and use measure-theoretic continuity (subsets shrinking to a null set have measure going to zero on a finite-measure space).

Answer

We may assume $f_{n} \to f$ everywhere on $X$ by replacing $X$ with $X$ minus a null set; this changes $μ (X)$ by zero.

For $k, m \geq 1$ , define $E_{m, k} = n \geq m ⋃ {x : ∣ f_{n} (x) - f (x) ∣ \geq 1/ k} .$

Each $E_{m, k}$ is measurable (countable union of measurable level sets). As $m \to \infty$ , the sets $E_{m, k}$ are decreasing: $E_{m + 1, k} \subseteq E_{m, k}$ .

For each $x \in X$ , $f_{n} (x) \to f (x)$ gives an $N (x, k)$ such that $∣ f_{n} (x) - f (x) ∣ < 1/ k$ for all $n \geq N (x, k)$ ; that is, $x \in / E_{m, k}$ for $m \geq N (x, k)$ . So $⋂_{m} E_{m, k} = \emptyset$ .

Since $μ (X) < \infty$ and $E_{m, k} ↓ \emptyset$ , measure-theoretic continuity gives $μ (E_{m, k}) \to 0$ as $m \to \infty$ .

Fix $ε > 0$ . For each $k \geq 1$ , pick $m_{k}$ such that $μ (E_{m_{k}, k}) < ε / 2^{k}$ . Set $E_{ε} = ⋃_{k} E_{m_{k}, k}$ . Then $μ (E_{ε}) \leq \sum_{k} ε / 2^{k} = ε$ .

On $X ∖ E_{ε}$ , for every $k \geq 1$ and every $n \geq m_{k}$ , $x \in / E_{m_{k}, k}$ means $∣ f_{n} (x) - f (x) ∣ < 1/ k$ . So for every $k \geq 1$ , $sup_{x \in X ∖ E_{ε}} ∣ f_{n} (x) - f (x) ∣ < 1/ k$ for $n \geq m_{k}$ . This is the definition of uniform convergence on $X ∖ E_{ε}$ .

Lean formalization Intermediate+

lean_status: full — Mathlib provides the full apparatus. Measurable encodes the pre-image-of-measurable-set definition, MeasureTheory.SimpleFunc encodes simple functions with their canonical disjoint-level-set representation, MeasureTheory.SimpleFunc.approx constructs the dyadic approximation sequence used in the Key theorem proof, and MeasureTheory.Egorov packages the Egorov uniform-convergence rescue under finite measure. The companion module records the unit's predicate names; the deeper structural theorems (Lusin's theorem in full generality on Polish spaces, the Pettis measurability characterisation) sit on the broader Mathlib measure-theory shelf accessed by import.

import Mathlib.MeasureTheory.MeasurableSpace.Basic
import Mathlib.MeasureTheory.Function.SimpleFunc
import Mathlib.MeasureTheory.Function.Egorov

variable (X : Type*) [MeasurableSpace X]
variable (Y : Type*) [MeasurableSpace Y]

abbrev CodexMeasurable (f : X → Y) : Prop := Measurable f

abbrev CodexSimpleFunc : Type _ := MeasureTheory.SimpleFunc X ENNReal

Advanced results Master

The advanced theory of measurable functions splits across four strands: descriptive (Lusin's theorem and almost-continuity), structural (Littlewood's three principles), modal (convergence in measure versus a.e. versus uniform), and the vector-valued generalisation (Pettis measurability).

Theorem 1 (Lusin, 1912). Let $f : R^{n} \to R$ be Lebesgue-measurable and let $A \subseteq R^{n}$ be Lebesgue-measurable with $m (A) < \infty$ . For every $ε > 0$ there exists a closed set $K \subseteq A$ with $m (A ∖ K) < ε$ such that the restriction $f ∣_{K}$ is continuous on $K$ ^{[Lusin 1912]}.

The proof has three stages. First, reduce to the case where $f$ is a simple function on $A$ . For a simple function $φ = \sum_{i} c_{i} χ_{A_{i}}$ with $A_{i}$ disjoint measurable, choose closed sets $K_{i} \subseteq A_{i}$ with $m (A_{i} ∖ K_{i}) < ε / (n \cdot 2^{i})$ using inner regularity of Lebesgue measure (every measurable set is the union of an $F_{σ}$ set and a null set, and closed subsets approximate from inside). Set $K = ⋃_{i} K_{i}$ ; the $K_{i}$ are pairwise disjoint and at positive distance (by closedness in a compact-or-finite-measure ambient), so $φ ∣_{K}$ is locally constant on each $K_{i}$ , hence continuous on $K$ . Second, approximate a general measurable $f$ by simple functions $φ_{k}$ converging to $f$ in measure (via the simple-function approximation theorem and Egorov). Third, apply Tietze's extension theorem ^{[Carathéodory 1918]} to extend $f ∣_{K}$ to a continuous function $\tilde{f} : R^{n} \to R$ .

Theorem 2 (Egorov, 1911). Let $(X, M, μ)$ be a measure space with $μ (X) < \infty$ , and let $f_{n}, f : X \to R$ be measurable with $f_{n} \to f$ almost everywhere. For every $ε > 0$ there exists $E_{ε} \in M$ with $μ (E_{ε}) < ε$ and $f_{n} \to f$ uniformly on $X ∖ E_{ε}$ ^{[Egorov 1911]}. Full proof: Exercise 8.

The finite-measure hypothesis is irreducible: Exercise 7 gives the standard $χ_{[n, n + 1]}$ counterexample on $R$ . The theorem also has a sigma-finite refinement: on a sigma-finite measure space, every finite-measure subset admits an Egorov rescue.

Theorem 3 (Littlewood's three principles). Littlewood 1944 ^{[Littlewood 1944]} organised the qualitative content of measure theory into three slogans. (i) Every Lebesgue-measurable set of finite measure in $R^{n}$ is approximately a finite union of bounded open intervals (or rectangles): for every $ε > 0$ there is a finite union $U_{ε}$ of bounded open intervals with $m (A △ U_{ε}) < ε$ . (ii) Every Lebesgue-measurable function on a finite-measure set is approximately continuous: this is exactly Lusin's theorem. (iii) Every almost-everywhere convergent sequence of measurable functions on a finite-measure set is approximately uniformly convergent: this is Egorov.

Each principle has the same structural form: a property defined by countable operations on a generating family extends, up to arbitrarily small measure, to the property of the generating family itself. The principles organise the working measure-theorist's reflexes and identify the descriptive boundary at which approximation arguments work.

Theorem 4 (modes of convergence). On a measure space $(X, M, μ)$ with $μ (X) < \infty$ , the following implications hold between modes of convergence of measurable functions $f_{n} \to f$ . Uniform convergence implies $L^{\infty}$ convergence implies $L^{p}$ convergence (for any $1 \leq p < \infty$ ) implies convergence in measure. Almost-everywhere convergence implies convergence in measure but not the reverse; convergence in measure implies a subsequence converges almost everywhere (Riesz subsequence theorem). $L^{p}$ convergence implies convergence in measure but not the reverse ^{[Folland §2.4]}. Under domination (Lebesgue dominated convergence, deferred to 02.07.05), a.e. convergence upgrades to $L^{p}$ convergence.

The role of finite measure shows up at the top of this diagram: in infinite measure, $L^{p}$ convergence for different $p$ no longer chains (the inclusion $L^{p} \subseteq L^{q}$ fails), and convergence in measure decouples from $L^{p}$ convergence.

Concrete counterexamples populate the diagram. The "typewriter" sequence $f_{n} = χ_{[k / 2^{j}, (k + 1) / 2^{j}]}$ enumerated by $n = 2^{j} + k$ on $[0, 1]$ converges to $0$ in $L^{p}$ for every $p \in [1, \infty)$ but at no point of $[0, 1]$ converges pointwise: every point is visited infinitely often by the marching window. This shows $L^{p}$ does not imply almost everywhere. The Riesz subsequence theorem rescues a subsequence: choose $n_{j} = 2^{j}$ , and then $f_{n_{j}} = χ_{[0, 1/ 2^{j}]}$ converges to $0$ pointwise on $(0, 1]$ .

Conversely, the sequence $g_{n} = n χ_{[0, 1/ n]}$ on $[0, 1]$ converges to $0$ pointwise (everywhere except at $x = 0$ ), hence almost everywhere, and converges to $0$ in measure (the bad set ${g_{n} \geq ε} = [0, 1/ n]$ for $ε < n$ has shrinking measure), but $∥ g_{n} ∥_{1} = 1$ for every $n$ , so the sequence does not converge in $L^{1}$ . This shows a.e. convergence does not imply $L^{p}$ convergence without domination.

Theorem 5 (Pettis measurability, 1938). Let $(X, M)$ be a measurable space, let $E$ be a Banach space 02.11.04, and let $f : X \to E$ be a function. The following are equivalent: (a) $f$ is strongly measurable, meaning the pointwise limit of a sequence of $E$ -valued simple functions $φ_{n}$ with finite range; (b) $f$ is weakly measurable (every $φ \circ f$ for $φ \in E^{*}$ is measurable) AND $f$ has essentially separable range (lies in a separable subspace except on a null set) ^{[Pettis 1938]}.

The theorem extends the simple-function approximation theorem to Banach-valued functions, with the essential-separability hypothesis controlling the cardinality difficulty: an unrestricted weakly measurable Banach-valued function can fail to be a pointwise limit of simple functions when its range is too large. Pettis's result is the gateway to the Bochner integral, the Banach-valued analogue of the Lebesgue integral, and underlies modern probability theory of Banach-valued random variables.

Proof sketch. The forward implication (a) $\Rightarrow$ (b) is direct: strongly measurable $f$ is a pointwise limit of simple functions $φ_{n}$ , each with finite range in $E$ , so the closed linear span of $⋃_{n} range (φ_{n})$ is a separable subspace containing the range of $f$ almost everywhere, and every $φ \circ f = lim_{n} φ \circ φ_{n}$ is a pointwise limit of measurable scalar functions, hence measurable.

For (b) $\Rightarrow$ (a), reduce to a separable target $E_{0} \subseteq E$ . Let ${e_{n}}_{n \geq 1}$ be a countable dense subset of $E_{0}$ . Define $φ_{n} (x) = e_{k (n, x)}$ where $k (n, x)$ is the smallest index $k \leq n$ minimising $∥ f (x) - e_{k} ∥$ (with measurable selection: each level set ${k (n, \cdot) = k}$ is a Borel set in $M$ because the norm distance $∥ f (x) - e_{k} ∥$ is measurable by weak measurability and Pettis's lemma identifying the norm with $sup_{∥ φ ∥_{E^{*}} \leq 1} ∣ φ (f (x)) ∣$ over a countable dense subset of the unit ball). Each $φ_{n}$ has finite range; $φ_{n} (x) \to f (x)$ for every $x$ since the $e_{k}$ are dense.

The essential-separability hypothesis is not removable. If $E$ is non-separable (for instance $E = ℓ^{\infty} ([0, 1])$ ) and $f$ is the canonical evaluation $x \mapsto δ_{x} \in ℓ^{\infty} ([0, 1])^{*} =$ (Banach dual), then $f$ is weakly measurable in many natural settings but its range ${δ_{x} : x \in [0, 1]}$ is not separable, and no sequence of simple functions converges to $f$ pointwise. The Pettis integral provides a weaker integration theory valid in this non-separable setting.

Theorem 6 (Vitali-Carathéodory). Let $f : R^{n} \to R$ be Lebesgue-integrable. For every $ε > 0$ there exist functions $u, v : R^{n} \to R$ with $u$ upper semicontinuous and bounded above, $v$ lower semicontinuous and bounded below, $u \leq f \leq v$ everywhere, and $\int (v - u) d m < ε$ ^{[Carathéodory 1918]}.

The proof builds $v$ by approximating the positive part $f^{+}$ from above and $u$ by approximating $f^{-}$ from above, using the regularity of Lebesgue measure (every measurable set is approximable from outside by open sets and from inside by closed sets). The theorem sharpens Lusin's "almost continuous" to "enveloped between two semicontinuous functions" and is the standard tool for converting integrability statements into compactness statements.

A concrete construction: write $f = f^{+} - f^{-}$ and assume $f \geq 0$ by symmetry. By the simple-function approximation theorem, find a simple $φ = \sum_{i} c_{i} χ_{A_{i}}$ with $\int (f - φ) < ε /3$ and $c_{i} > 0$ . For each measurable $A_{i}$ , pick an open $U_{i} \supseteq A_{i}$ with $m (U_{i} ∖ A_{i}) < ε / (3 n c_{i} \cdot 2^{i})$ (outer regularity) and a closed $K_{i} \subseteq A_{i}$ with $m (A_{i} ∖ K_{i}) < ε / (3 n c_{i} \cdot 2^{i})$ (inner regularity). Set $v = \sum_{i} c_{i} χ_{U_{i}}$ , lower semicontinuous as a sum of indicators of open sets times positive constants; set $u = \sum_{i} c_{i} χ_{K_{i}}$ , upper semicontinuous as a sum of indicators of closed sets. Then $u \leq φ \leq v$ pointwise on $R^{n}$ off a null set, $\int (v - φ) < ε /3$ , $\int (φ - u) < ε /3$ , and $\int (f - φ) < ε /3$ , so $\int (v - u) < ε$ . The semicontinuity is the key qualitative gain: semicontinuous functions are exactly the pointwise infima/suprema of continuous functions, and this is the structural fact that bridges measurable functions to the Stone-Čech and Riesz-Markov-Kakutani frameworks.

Theorem 7 (existence of non-Borel Lebesgue-measurable functions). There exists a function $f : R \to R$ that is Lebesgue-measurable but not Borel-measurable.

Proof outline. The Lebesgue sigma-algebra $L$ strictly contains the Borel sigma-algebra $B$ (every subset of the Cantor set is Lebesgue-measurable; most subsets of the Cantor set are not Borel by cardinality). Pick $A \in L ∖ B$ , for instance the image under the Cantor function of a non-Borel subset of $[0, 1]$ . The characteristic function $χ_{A} : R \to R$ is Lebesgue-measurable (since $χ_{A}^{- 1} ({1}) = A \in L$ ) but $χ_{A}^{- 1} ({1}) = A \in / B$ , so $χ_{A}$ is not Borel-measurable.

Theorem 8 (Luzin's $N$ -property and absolute continuity). A measurable function $f : [a, b] \to R$ satisfies Luzin's $N$ -property when $f$ maps Lebesgue null sets to Lebesgue null sets. A theorem of Banach asserts that a continuous function of bounded variation satisfies the $N$ -property if and only if it is absolutely continuous ^{[Bogachev §5.8]}. The $N$ -property is the failure mode of the Cantor function: continuous, monotone, of bounded variation, but maps the Cantor set (a null set) onto $[0, 1]$ (a positive-measure set), hence not absolutely continuous and not satisfying the $N$ -property.

Theorem 9 (convergence in measure — Riesz subsequence theorem). Let $(X, M, μ)$ be a measure space and $f_{n}, f : X \to R$ measurable. The sequence ${f_{n}}$ converges to $f$ in measure when $μ ({x : ∣ f_{n} (x) - f (x) ∣ \geq ε}) \to 0$ as $n \to \infty$ for every $ε > 0$ . If $f_{n} \to f$ in measure, then a subsequence $f_{n_{k}} \to f$ almost everywhere ^{[Riesz 1909]}. Conversely, on a finite-measure space, $f_{n} \to f$ almost everywhere implies $f_{n} \to f$ in measure.

Proof outline. Choose indices $n_{k}$ recursively so that $μ ({x : ∣ f_{n_{k}} (x) - f (x) ∣ \geq 1/ 2^{k}}) < 1/ 2^{k}$ . Define $A_{k} = {x : ∣ f_{n_{k}} (x) - f (x) ∣ \geq 1/ 2^{k}}$ . By the Borel-Cantelli lemma, $μ (lim sup_{k} A_{k}) = 0$ . Off $lim sup_{k} A_{k}$ , $x$ lies in finitely many $A_{k}$ , so eventually $∣ f_{n_{k}} (x) - f (x) ∣ < 1/ 2^{k}$ , giving $f_{n_{k}} (x) \to f (x)$ .

The Riesz subsequence theorem identifies almost-everywhere convergence as the topology-of-pointwise-convergence companion to convergence in measure, and is the key lemma in upgrading $L^{p}$ -convergence statements to almost-everywhere statements along subsequences.

Theorem 10 (composition of measurable functions: the asymmetry). A composition $g \circ f$ of a Borel-measurable $g : R \to R$ with a measurable $f : (X, M) \to (R, B (R))$ is measurable. The composition need not be measurable when $g$ is only Lebesgue-measurable: the pre-image $g^{- 1} (B)$ of a Borel set $B$ may fail to be Borel, and then $f^{- 1} (g^{- 1} (B))$ may fail to be in $M$ unless $M$ already contains every Lebesgue set ^{[Royden-Fitzpatrick §3.5]}.

The standard counterexample uses the Cantor function $c : [0, 1] \to [0, 1]$ , which sends a non-Lebesgue-measurable subset of the Cantor set to a non-Lebesgue-measurable subset of $[0, 1]$ . Composing with an appropriate Lebesgue-measurable indicator produces a non-measurable function. This asymmetry motivates the standard practice of working with Borel-measurable representatives when composing functions, and underlies the regularity-of-measure approach to the Lebesgue density theorem.

Synthesis. The structural picture is that measurability is a sigma-algebra membership condition with three layers of refinement. The foundational reason measurable functions form the right class for integration is that the sigma-algebra structure is exactly the closure property under countable operations needed to make Lebesgue's dominated and monotone convergence theorems hold, and this is exactly why Riemann integration breaks where Lebesgue integration succeeds. The simple-function approximation builds the staircase by which any non-negative measurable function climbs to a limit, identifying the cone of non-negative measurable functions with the closure of the cone of non-negative simple functions under pointwise increasing limits.

Putting these together with Littlewood's three principles, the central insight is that measurable functions are not new objects but old objects (continuous functions, simple functions, indicator functions of intervals) reached by countable operations and modulo small-measure exceptions. The bridge is between the descriptive theory (sigma-algebras, Borel hierarchy) and the operational theory (Lusin, Egorov, Vitali-Carathéodory); each operational result identifies a continuous-function approximation with a measurable function up to a small error in measure. This pattern generalises to Banach-valued functions via Pettis 1938: strong measurability is exactly the closure of finite-range simple functions under pointwise limits, with the essential-separability hypothesis taming the cardinality difficulty. The Bochner integral and modern Banach-valued probability theory build on this scaffold, appearing again in 02.11.04 as the integration theory of Banach-valued random variables, and putting the pieces together identifies the abstract measurability framework with both the classical Lebesgue integral and its non-commutative and infinite-dimensional descendants.

Full proof set Master

Proposition 1 (algebraic closure of measurable functions). Let $(X, M)$ be a measurable space and $f, g : X \to R$ measurable. Then $f + g$ , $c f$ , $f g$ , $∣ f ∣$ , $max (f, g)$ , $min (f, g)$ are all measurable.

Proof. For $f + g$ : by density of $Q$ in $R$ , ${x : f (x) + g (x) > a} = q \in Q ⋃ ({x : f (x) > q} \cap {x : g (x) > a - q}) .$ Each set in the union is measurable (intersection of two measurable level sets); the union is countable, so the result is measurable.

For $c f$ : if $c > 0$ , ${c f > a} = {f > a / c} \in M$ ; if $c < 0$ , ${c f > a} = {f < a / c} \in M$ ; if $c = 0$ , $c f \equiv 0$ has measurable level sets by definition (each is either $\emptyset$ or $X$ ).

For $f g$ : use the polarisation identity $f g = \frac{1}{4} ((f + g)^{2} - (f - g)^{2})$ . So it suffices to show $h^{2}$ is measurable for measurable $h$ . For $a \geq 0$ , ${h^{2} > a} = {h > a} \cup {h < - a}$ ; for $a < 0$ , ${h^{2} > a} = X$ . Both are measurable.

For $∣ f ∣$ : ${∣ f ∣ > a} = {f > a} \cup {f < - a}$ for $a \geq 0$ ; for $a < 0$ , ${∣ f ∣ > a} = X$ .

For $max (f, g)$ and $min (f, g)$ : ${max (f, g) > a} = {f > a} \cup {g > a}$ ; ${min (f, g) > a} = {f > a} \cap {g > a}$ .

$□$

Proposition 2 (countable-supremum closure). Let ${f_{n}}_{n \in N}$ be a sequence of measurable functions $X \to \overset{ˉ}{R}$ . Then $sup_{n} f_{n}$ , $in f_{n} f_{n}$ , $lim sup_{n} f_{n}$ , $lim inf_{n} f_{n}$ are measurable. The pointwise limit $lim_{n} f_{n}$ , where it exists, is measurable.

Proof. For $sup_{n} f_{n}$ : ${sup_{n} f_{n} > a} = ⋃_{n} {f_{n} > a}$ , a countable union of measurable sets.

For $in f_{n} f_{n}$ : ${in f_{n} f_{n} < a} = ⋃_{n} {f_{n} < a}$ , similarly.

For $lim sup$ and $lim inf$ : $lim sup_{n} f_{n} = in f_{k} sup_{n \geq k} f_{n}$ and $lim inf_{n} f_{n} = sup_{k} in f_{n \geq k} f_{n}$ , both compositions of the above.

For $lim_{n} f_{n}$ on the set $E = {x : lim_{n} f_{n} (x) exists}$ : $E = {lim sup_{n} f_{n} = lim inf_{n} f_{n}}$ is the set where two measurable functions agree, hence measurable. On $E$ , $lim_{n} f_{n} = lim sup_{n} f_{n} = lim inf_{n} f_{n}$ as measurable functions.

$□$

Proposition 3 (Egorov from finite measure). Restated and proven in Exercise 8 above. For self-containment, the load-bearing step is the measure-theoretic continuity: $E_{m, k} ↓ \emptyset$ as $m \to \infty$ , and $μ (X) < \infty$ forces $μ (E_{m, k}) \to 0$ .

Proof. See Exercise 8 for the complete argument.

$□$

Proposition 4 (Lusin reduction to simple functions). Lusin's theorem (Theorem 1 above) reduces, modulo Egorov and Tietze, to the special case of simple functions. For a simple function $φ = \sum_{i = 1}^{n} c_{i} χ_{A_{i}}$ with disjoint $A_{i}$ , the closed set $K = ⨆_{i} K_{i}$ with $K_{i} \subseteq A_{i}$ closed and $m (A_{i} ∖ K_{i}) < ε / n$ satisfies $m (A ∖ K) < ε$ , and $φ ∣_{K}$ is continuous because the $K_{i}$ are at positive distance from each other in $R^{n}$ (closed subsets of disjoint open neighbourhoods of disjoint compact sets, by inner regularity).

Proof. By inner regularity of Lebesgue measure (every measurable set in $R^{n}$ contains a closed subset of approximately the same measure), pick $K_{i} \subseteq A_{i}$ closed with $m (A_{i} ∖ K_{i}) < ε / n$ . Then $K = ⋃ K_{i}$ is closed (finite union of closed sets) and $m (A ∖ K) = \sum_{i} m (A_{i} ∖ K_{i}) < n \cdot ε / n = ε$ . On $K$ , $φ$ takes the constant value $c_{i}$ on each $K_{i}$ . The $K_{i}$ are pairwise disjoint closed subsets of $R^{n}$ ; by Hausdorffness, the function $φ ∣_{K}$ is continuous (each fibre is open in the relative topology of $K$ ). Tietze's theorem extends $φ ∣_{K}$ to a continuous function on all of $R^{n}$ .

$□$

Proposition 5 (Lusin general case from the simple-function case). The reduction from a general measurable $f$ to a simple-function instance uses Egorov plus the simple-function approximation theorem. We sketch the load-bearing steps.

Proof. By inner regularity, restrict to a set $A^{'} \subseteq A$ with $m (A ∖ A^{'}) < ε /3$ on which $f$ is bounded (the sets ${∣ f ∣ \leq N}$ exhaust $A$ as $N \to \infty$ , so pick $N$ with $m (A ∖ {∣ f ∣ \leq N}) < ε /3$ and set $A^{'} = A \cap {∣ f ∣ \leq N}$ ). On $A^{'}$ , the simple-function approximation theorem gives simple functions $φ_{k} \to f$ uniformly (since $f$ is bounded on $A^{'}$ ).

Apply Proposition 4 to each $φ_{k}$ : pick a closed set $K_{k} \subseteq A^{'}$ with $m (A^{'} ∖ K_{k}) < ε / (3 \cdot 2^{k})$ on which $φ_{k}$ is continuous. Set $K = ⋂_{k} K_{k}$ ; then $K$ is closed and $m (A^{'} ∖ K) \leq \sum_{k} m (A^{'} ∖ K_{k}) < ε /3$ , so $m (A ∖ K) < 2 ε /3$ .

On $K$ , every $φ_{k}$ is continuous, and $φ_{k} \to f$ uniformly. The uniform limit of continuous functions is continuous, so $f ∣_{K}$ is continuous.

Refine $K$ if needed (by inner regularity again) so that $m (A ∖ K) < ε$ . Tietze's theorem extends $f ∣_{K}$ to a continuous function on all of $R^{n}$ .

$□$

Proposition 6 (measure-theoretic continuity of decreasing sequences). Let $μ$ be a measure on $(X, M)$ , and let ${E_{n}}$ be a decreasing sequence in $M$ with $μ (E_{1}) < \infty$ . Then $μ (⋂_{n} E_{n}) = lim_{n} μ (E_{n})$ .

Proof. Set $E = ⋂_{n} E_{n}$ . By countable additivity applied to the disjoint decomposition $E_{1} = E \cup ⨆_{n} (E_{n} ∖ E_{n + 1})$ , $μ (E_{1}) = μ (E) + n = 1 \sum \infty μ (E_{n} ∖ E_{n + 1}) .$ The series telescopes: $\sum_{n = 1}^{N - 1} μ (E_{n} ∖ E_{n + 1}) = μ (E_{1}) - μ (E_{N})$ , using finite additivity and $μ (E_{N}) < μ (E_{1}) < \infty$ . Taking $N \to \infty$ , $μ (E_{1}) = μ (E) + μ (E_{1}) - N lim μ (E_{N}) .$ Subtracting the finite quantity $μ (E_{1})$ from both sides yields $μ (E) = lim_{N} μ (E_{N})$ . The finite-measure hypothesis $μ (E_{1}) < \infty$ is essential: without it, the subtraction step is undefined, and on infinite-measure spaces the conclusion can fail (take $E_{n} = [n, \infty)$ on $R$ with Lebesgue measure; then $⋂_{n} E_{n} = \emptyset$ but $μ (E_{n}) = \infty$ for every $n$ ).

This proposition is the load-bearing measure-theoretic fact behind Egorov: the sets $E_{m, k}$ in the Egorov proof form a decreasing sequence with intersection a null set, and finite measure forces $μ (E_{m, k}) \to 0$ .

$□$

Connections Master

σ-algebra, measurable space, and the Borel σ-algebra 02.07.01. Supplies the underlying measurable-space structure on which every measurability statement in this unit rests. The Borel sigma-algebra on $R$ is the target sigma-algebra for real-valued measurable functions, and the ray-generated characterisation reduces the measurability test to the one-parameter condition ${f > a} \in M$ used throughout the proofs here.
Lebesgue outer measure and the Carathéodory construction 02.07.02. Defines what "measure zero" means and equips the Lebesgue sigma-algebra with the completeness property needed for the almost-everywhere equivalence relation. Lusin's inner-regularity step (closed sets approximate measurable sets from inside) is exactly the Carathéodory-completion property in concrete form on $R^{n}$ .
Lebesgue integral 02.07.04. Built on the simple-function approximation theorem proven here. The Lebesgue integral of a non-negative measurable function $f$ is defined as $\int f = sup {\int φ : 0 \leq φ \leq f, φ simple}$ , and the existence of the approximating sequence here is the load-bearing fact that makes the supremum a limit attained along a definite sequence rather than a transfinite construction.
Continuous map 02.01.02. Every continuous function between topological spaces with their Borel sigma-algebras is Borel-measurable, but the converse fails substantively: pointwise limits, indicator functions of Borel sets, and the derivative of a differentiable function are all Borel-measurable but in general discontinuous. Lusin's theorem in this unit closes the gap up to measure: every measurable function is the restriction of a continuous function to a set of arbitrarily small complement.
Banach spaces 02.11.04. Pettis 1938 extends the scalar-valued measurable-function framework here to Banach-valued functions, with the essential-separability hypothesis controlling the cardinality difficulty. The Bochner integral built on Pettis measurability is the Banach-valued analogue of the Lebesgue integral, and underlies modern probability theory of random variables taking values in function spaces and operator-algebra states.

Historical & philosophical context Master

Lebesgue's 1904 Leçons sur l'intégration ^{[Lebesgue 1904]} introduced the measurable-function framework as the natural domain of the integral he had defined in his 1902 thesis. Lebesgue's notion of measurability was inner-outer approximation: a bounded function on a measurable set is measurable when the inner and outer approximations of its graph (by open and closed sets) have the same measure. The modern pre-image-of-Borel-sets characterisation is equivalent and was crystallised in subsequent expositions.

Egorov's 1911 note in the Comptes Rendus ^{[Egorov 1911]} proved what Lebesgue had used implicitly: almost-everywhere convergence on a finite-measure set is almost uniform. Severini independently published the same result a few months earlier in the Atti dell'Accademia Gioenia, and the result is sometimes called the Severini-Egorov theorem in the Italian and Russian traditions. The proof uses measure-theoretic continuity of decreasing sequences, which is the load-bearing structural fact of countably additive measures and an immediate consequence of countable additivity plus the finite-measure hypothesis.

Lusin's 1912 note in the Comptes Rendus ^{[Lusin 1912]} established the descriptive companion: a measurable function on $R^{n}$ is continuous on the complement of a set of arbitrarily small measure. Lusin's proof in the original note used Egorov plus the inner-regularity of Lebesgue measure; the Tietze-extension step (extending the continuous restriction to a continuous function on all of $R^{n}$ ) is sometimes packaged into the theorem statement, sometimes presented as a separate corollary. The theorem extends to any Polish space carrying a Radon measure, a generalisation due to Bourbaki and developed in the modern setting by Schwartz 1973 and Bogachev 2007.

Carathéodory's 1918 Vorlesungen über reelle Funktionen ^{[Carathéodory 1918]} systematised the framework: measurable functions, almost-everywhere equivalence, simple-function approximation, and the Vitali-Carathéodory enveloping theorem. The abstract measurable-space approach (general $(X, M)$ , not just $R^{n}$ ) was Carathéodory's, and it allowed the same framework to apply to probability theory (Kolmogorov 1933) and integration on locally compact groups (Haar 1933, Weil 1940).

Pettis's 1938 Trans. AMS paper ^{[Pettis 1938]} extended measurability to Banach-valued functions. Pettis identified the essential-separability hypothesis as the load-bearing distinction between weak and strong measurability and showed that under this hypothesis the simple-function approximation extends to Banach-valued functions, opening the Bochner integral and Banach-valued probability theory. The non-separable case admits the Pettis integral, weaker than Bochner, used in spectral theory and operator algebras.

Littlewood's 1944 Lectures on the Theory of Functions ^{[Littlewood 1944]} codified the three principles as the qualitative organising slogans of measure theory. Littlewood's principles capture the structural pattern that every approximation argument in real analysis follows: a property of measurable objects extends, up to arbitrarily small measure, to the property of the simpler generating objects (intervals, continuous functions, uniform-convergence sequences) they are built from. The principles remain a teaching device and a working heuristic, and modern monographs (Royden-Fitzpatrick 2010, Bogachev 2007) continue to organise their expositions around them.

The descriptive set theory of measurable functions deepens through the analytic and projective hierarchies. Souslin's 1917 work on analytic sets (continuous images of Borel sets, equivalently projections of Borel subsets of $R^{2}$ ) showed that the class of Borel-measurable sets is not closed under continuous images, and the strict containment $B (R) ⊊ Σ_{1}^{1}$ opened the field of descriptive set theory. Analytic and Borel functions admit a uniform measurability theory (Lusin separation, Lusin uniformisation) that extends the Lusin theorem in this unit; the modern reference is Kechris 1995 Classical Descriptive Set Theory.

The Bochner integral built on Pettis's 1938 measurability theorem became the standard integration framework for Banach-valued random variables in modern probability theory (Ledoux-Talagrand 1991 Probability in Banach Spaces), for operator-valued functions in spectral theory (Dunford-Schwartz 1958-71 Linear Operators, three volumes), and for the integration theory of vector-valued $L^{p}$ spaces (Hytönen-van Neerven-Veraar-Weis 2016-17, two volumes). The non-separable Pettis integral remains the standard tool when the range Banach space is too large to admit the strong-measurability hypothesis.

Bibliography Master

@article{Egorov1911,
  author  = {Egorov, Dmitri F.},
  title   = {Sur les suites des fonctions mesurables},
  journal = {Comptes Rendus de l'Acad\'emie des Sciences de Paris},
  volume  = {152},
  year    = {1911},
  pages   = {244--246}
}

@article{Lusin1912,
  author  = {Lusin, Nikolai N.},
  title   = {Sur les propri\'et\'es des fonctions mesurables},
  journal = {Comptes Rendus de l'Acad\'emie des Sciences de Paris},
  volume  = {154},
  year    = {1912},
  pages   = {1688--1690}
}

@book{Lebesgue1904,
  author    = {Lebesgue, Henri},
  title     = {Le\c{c}ons sur l'int\'egration et la recherche des fonctions primitives},
  publisher = {Gauthier-Villars},
  address   = {Paris},
  year      = {1904}
}

@book{Caratheodory1918,
  author    = {Carath\'eodory, Constantin},
  title     = {Vorlesungen \"uber reelle Funktionen},
  publisher = {Teubner},
  address   = {Leipzig},
  year      = {1918}
}

@article{Pettis1938,
  author  = {Pettis, B. J.},
  title   = {On integration in vector spaces},
  journal = {Transactions of the American Mathematical Society},
  volume  = {44},
  year    = {1938},
  pages   = {277--304}
}

@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{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{Littlewood1944,
  author    = {Littlewood, J. E.},
  title     = {Lectures on the Theory of Functions},
  publisher = {Oxford University Press},
  year      = {1944}
}

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

Prerequisites

02.07.01
02.07.02
02.01.02

Tier anchors

beginner: Tao, Analysis II §7; pictures of simple-function ladders approximating a curve
intermediate: Folland, Real Analysis 2e §2.1; Royden, Real Analysis 4e §3.3-3.6
master: Halmos, Measure Theory §III; Bogachev, Measure Theory Vol. 1 §2.1-2.3

References

Folland — Real Analysis: Modern Techniques and Their Applications, 2e · §2.1, Measurable functions
Royden, Fitzpatrick — Real Analysis, 4e · §3.3-3.6, Lebesgue measurable functions
Halmos — Measure Theory · §III, Measurable functions
Bogachev — Measure Theory, Vol. 1 · §2.1-2.3, Measurable functions and convergence modes
Tao — Analysis II · §7, Lebesgue measure and measurable functions
Egorov — Sur les suites des fonctions mesurables · C. R. Acad. Sci. Paris 152 (1911), 244-246
Lusin — Sur les propriétés des fonctions mesurables · C. R. Acad. Sci. Paris 154 (1912), 1688-1690
Lebesgue — Leçons sur l'intégration et la recherche des fonctions primitives · Gauthier-Villars 1904
Pettis — On integration in vector spaces · Trans. Amer. Math. Soc. 44 (1938), 277-304
Vitali — Una proprietà delle funzioni misurabili · Rendiconti del R. Istituto Lombardo 38 (1905), 599-603
Carathéodory — Vorlesungen über reelle Funktionen · Teubner, Leipzig 1918
Littlewood — Lectures on the Theory of Functions · Oxford 1944, §IV (the three principles)

Lean module

Codex.Analysis.MeasureTheory.MeasurableFunctionsEgorovLusin

Estimated time

beginner: 18m
intermediate: 50m
master: 80m