Real analysis & measure theory

⚡ Key Takeaways

Real analysis rigorously develops calculus on $\mathbb{R}$ from the completeness axiom, while measure theory generalizes "length, area, volume" into a coherent framework for integration. Key result: the Lebesgue integral integrates a vastly larger class of functions than the Riemann integral and behaves beautifully under limits (Monotone/Dominated Convergence). Applies in: probability, PDEs, harmonic analysis, functional analysis, mathematical physics. Who/when: Cauchy, Weierstrass, Riemann (19th c.); Borel, Lebesgue, Carathéodory (1900–1920).

💡 The Core Insight

Riemann integration partitions the domain (the $x$ -axis); Lebesgue integration partitions the range (the $y$ -axis), grouping points by the value of $f$ . To make this work we need to measure the size of the level sets $\{x : f(x) > t\}$ . This single shift — measuring sets rather than chopping intervals — buys us robust convergence theorems, completeness of $L^p$ spaces, and a unified treatment of length, probability, and abstract integration.

DefinitionCompleteness of

\mathbb{R}

(Least Upper Bound Axiom)

A set $S \subseteq \mathbb{R}$ is bounded above if there is $M$ with $s \le M$ for all $s \in S$ . The supremum $\sup S$ is the least such bound. The defining axiom of $\mathbb{R}$ :

1

\quad S \neq \varnothing,\ S \text{ bounded above} \implies \exists\,\sup S \in \mathbb{R}.

Derivation hint: This is an axiom characterizing $\mathbb{R}$ uniquely as a complete ordered field; $\mathbb{Q}$ fails it (e.g. $\{q : q^2 < 2\}$ has no rational sup).

Definition

\sigma

-algebra and Measure

A $\sigma$ -algebra $\mathcal{M}$ on a set $X$ is a collection of subsets with $X \in \mathcal{M}$ , closed under complement and countable union:

2

\quad A \in \mathcal{M} \Rightarrow A^c \in \mathcal{M}, \qquad A_1, A_2, \ldots \in \mathcal{M} \Rightarrow \bigcup_{n=1}^\infty A_n \in \mathcal{M}.

Derivation hint: De Morgan gives closure under countable intersection; $\varnothing = X^c$ . A measure $\mu : \mathcal{M} \to [0,\infty]$ satisfies $\mu(\varnothing) = 0$ and countable additivity: for pairwise disjoint $A_n$ ,

3

\quad \mu\!\left(\bigcup_{n=1}^\infty A_n\right) = \sum_{n=1}^\infty \mu(A_n).

Derivation hint: Countable (not merely finite) additivity is the crucial strengthening that makes limit interchange possible.

DefinitionLebesgue Outer Measure and Measurability (Carathéodory)

For $E \subseteq \mathbb{R}$ , the outer measure is

4

\quad m^*(E) = \inf\left\{ \sum_{k=1}^\infty (b_k - a_k) : E \subseteq \bigcup_{k=1}^\infty (a_k, b_k) \right\}.

Derivation hint: Cover $E$ by countably many intervals; take the infimum of total length over all covers. A set $E$ is Lebesgue measurable (Carathéodory criterion) if it splits every test set additively:

5

\quad m^*(A) = m^*(A \cap E) + m^*(A \cap E^c) \quad \text{for all } A \subseteq \mathbb{R}.

Derivation hint: One inequality ( $\le$ ) is automatic from subadditivity; measurability demands the reverse $\ge$ .

DefinitionThe Lebesgue Integral

For a measurable simple function $\varphi = \sum_{i=1}^n c_i \mathbf{1}_{E_i}$ with $c_i \ge 0$ , define $\int \varphi\, d\mu = \sum_i c_i \mu(E_i)$ . For a general measurable $f \ge 0$ :

6

\quad \int_X f\, d\mu = \sup\left\{ \int_X \varphi\, d\mu : 0 \le \varphi \le f,\ \varphi \text{ simple} \right\}.

Derivation hint: Approximate $f$ from below by simple functions; this is the range-partition idea made precise. A general $f$ is integrable ( $f \in L^1$ ) iff $\int |f|\,d\mu < \infty$ , and then $\int f = \int f^+ - \int f^-$ where $f^\pm = \max(\pm f, 0)$ .

TheoremMonotone & Dominated Convergence

Let $(f_n)$ be measurable.

(MCT) If $0 \le f_1 \le f_2 \le \cdots$ and $f_n \to f$ pointwise, then

7

\quad \lim_{n\to\infty} \int f_n\, d\mu = \int f\, d\mu = \int \lim_{n\to\infty} f_n \, d\mu.

(DCT) If $f_n \to f$ pointwise and $|f_n| \le g$ for some integrable $g$ , then $f \in L^1$ and

8

\quad \lim_{n\to\infty}\int |f_n - f|\, d\mu = 0, \qquad \lim_{n\to\infty} \int f_n\, d\mu = \int f\, d\mu.

Derivation hint: MCT follows from definition [6] via simple-function approximation; DCT follows by applying Fatou's Lemma to $g + f_n$ and $g - f_n$ .

ProofMonotone Convergence Theorem

1. Setup. Let $0 \le f_1 \le f_2 \le \cdots$ with $f_n \uparrow f$ pointwise. Each $f_n$ is measurable, so $f = \sup_n f_n$ is measurable. 2. Easy inequality. Since $f_n \le f$ and integration is monotone, $\int f_n \le \int f$ . The sequence $\int f_n$ is increasing and bounded above by $\int f$ , so $L := \lim_n \int f_n \le \int f$ exists in $[0,\infty]$ . 3. Reverse inequality. Fix a simple $\varphi$ with $0 \le \varphi \le f$ and a constant $c \in (0,1)$ . Define $E_n = \{x : f_n(x) \ge c\,\varphi(x)\}$ . These sets increase to $X$ because $f_n \uparrow f \ge \varphi > c\varphi$ wherever $\varphi > 0$ . 4. Push the bound through. Then $\int f_n \ge \int_{E_n} f_n \ge c \int_{E_n} \varphi$ . By continuity of measure from below (a consequence of [3]), $\int_{E_n}\varphi \to \int_X \varphi$ as $n \to \infty$ . Hence $L \ge c \int \varphi$ . 5. Take limits. Let $c \uparrow 1$ : $L \ge \int \varphi$ . Taking the supremum over all such $\varphi$ and using [6], $L \ge \int f$ . 6. Conclude. Combining steps 2 and 5, $L = \int f$ , which is [7]. $\blacksquare$

∎

ProofFatou ⟹ Dominated Convergence

1. Fatou's Lemma (from MCT): for measurable $g_n \ge 0$ , $\int \liminf g_n \le \liminf \int g_n$ . Apply MCT to $h_k = \inf_{n\ge k} g_n \uparrow \liminf g_n$ . 2. Suppose $|f_n| \le g \in L^1$ and $f_n \to f$ . Then $g + f_n \ge 0$ and $g - f_n \ge 0$ . 3. Fatou on $g + f_n$ : $\int g + \int f \le \liminf(\int g + \int f_n) = \int g + \liminf \int f_n$ , so $\int f \le \liminf \int f_n$ . 4. Fatou on $g - f_n$ : similarly $-\int f \le \liminf(-\int f_n) = -\limsup \int f_n$ , so $\limsup \int f_n \le \int f$ . 5. Combining: $\limsup \int f_n \le \int f \le \liminf \int f_n$ , forcing $\lim \int f_n = \int f$ , which is [8]. $\blacksquare$

∎

mathematician · Henri Lebesgue | 1875–1941

In his 1902 thesis Intégrale, longueur, aire, Lebesgue introduced his measure and integral, reorganizing integration around the range of a function. His famous metaphor: a Riemann shopkeeper counts coins in the order picked from the till; a Lebesgue shopkeeper first sorts coins by denomination, then counts. This sorting — grouping by value — is exactly the range-partition. Lebesgue also proved his differentiation theorem and characterized Riemann integrability (a bounded function is Riemann integrable iff its discontinuities form a measure-zero set).

🧠 Intuition

Picture the area under a non-negative graph. Riemann slices it into thin vertical rectangles over a fine grid of the $x$ -axis. This fails badly for wild functions: the Dirichlet function $\mathbf{1}_{\mathbb{Q}}$ jumps everywhere, so every vertical sliver straddles values $0$ and $1$ and the slicing never converges.

Lebesgue slices horizontally. To compute area he asks: "for each height $t$ , how much $x$ lies above that height?" — i.e. he measures the level set $\{f > t\}$ . For $\mathbf{1}_{\mathbb{Q}}$ on $[0,1]$ , the set $\{f = 1\} = \mathbb{Q}\cap[0,1]$ has measure $0$ , so the integral is cleanly $0$ . The whole theory is built to make "how much $x$ " — the measure of these sets — meaningful, even for sets far stranger than intervals.

Figure

Riemann (vertical) vs. Lebesgue (horizontal) slicing of the area under a curve

ExampleLebesgue integral of the Dirichlet function

Given: $f = \mathbf{1}_{\mathbb{Q}}$ on $[0,1]$ , i.e. $f(x)=1$ if $x$ rational, $0$ otherwise.

Step 1. $\mathbb{Q}\cap[0,1]$ is countable: $\{q_1,q_2,\ldots\}$ . Cover $q_k$ by an interval of length $\varepsilon/2^k$ . Total length $\le \sum_k \varepsilon/2^k = \varepsilon$ .

Step 2. Since $\varepsilon$ is arbitrary, $m^*(\mathbb{Q}\cap[0,1]) = 0$ , so $m(\mathbb{Q}\cap[0,1]) = 0$ .

Step 3. $f$ is a simple function: $f = 1\cdot \mathbf{1}_{\mathbb{Q}\cap[0,1]} + 0 \cdot \mathbf{1}_{\text{irr}}$ . By definition $\int f\,dm = 1\cdot m(\mathbb{Q}\cap[0,1]) = 1\cdot 0$ .

Answer:

\boxed{\int_{[0,1]} \mathbf{1}_{\mathbb{Q}}\, dm = 0}

(By contrast the Riemann integral does not exist: upper sums $=1$ , lower sums $=0$ .)

ExampleInterchanging limit and integral via DCT

Given: $f_n(x) = \dfrac{n x}{1 + n^2 x^2}$ on $[0,1]$ . Compute $\lim_{n\to\infty}\int_0^1 f_n\,dx$ .

Step 1 (pointwise limit). For fixed $x>0$ , $f_n(x)\sim \frac{nx}{n^2x^2} = \frac{1}{nx} \to 0$ ; and $f_n(0)=0$ . So $f_n \to 0$ pointwise.

Step 2 (dominating function). By AM–GM, $1 + n^2x^2 \ge 2nx$ , so $|f_n(x)| = \frac{nx}{1+n^2x^2} \le \frac{nx}{2nx} = \tfrac12$ . Take $g \equiv \tfrac12$ , which is integrable on $[0,1]$ ( $\int_0^1 \tfrac12 = \tfrac12 <\infty$ ).

Step 3 (apply DCT). Hypotheses of [8] hold, so

\lim_{n\to\infty}\int_0^1 f_n\,dx = \int_0^1 0\,dx = 0.

Check (direct): $\int_0^1 \frac{nx}{1+n^2x^2}dx = \frac{1}{2n}\ln(1+n^2) \to 0$ . ✓

Answer:

\boxed{\lim_{n\to\infty}\int_0^1 \frac{nx}{1+n^2x^2}\,dx = 0}

⚠️ Common mistake

"Pointwise convergence lets you swap limit and integral." False without a dominating function or monotonicity. The classic failure: $f_n = n\,\mathbf{1}_{(0,1/n)}$ on $[0,1]$ converges pointwise to $0$ , yet $\int f_n = 1$ for every $n$ . So $\lim\int f_n = 1 \neq 0 = \int \lim f_n$ . There is no integrable dominator (the "spike" escapes upward), so DCT does not apply — and indeed the conclusion fails.

"Measure zero means finite/small." A set of measure zero can be uncountable (the Cantor set) and even dense (no — Cantor is nowhere dense; but $\mathbb{Q}$ is dense and measure zero). "Measure zero" is about coverable length, not cardinality.

"Every set is measurable." Under the Axiom of Choice, non-measurable sets exist (Vitali). Measure theory deliberately restricts to a $\sigma$ -algebra.

CounterexampleA non-measurable set: the Vitali set

Define $x \sim y$ iff $x - y \in \mathbb{Q}$ . This partitions $[0,1]$ into equivalence classes. By the Axiom of Choice, pick one representative from each class, forming $V \subseteq [0,1]$ . Let $\{q_k\}$ enumerate the rationals in $[-1,1]$ and set $V_k = V + q_k$ . The $V_k$ are pairwise disjoint and satisfy $[0,1] \subseteq \bigcup_k V_k \subseteq [-1,2]$ . If $V$ were measurable with measure $m(V)=c$ , translation invariance gives $m(V_k)=c$ for all $k$ , so by countable additivity [3]:

1 \le \sum_{k=1}^\infty c \le 3.

But $\sum_k c$ is either $0$ (if $c=0$ ) or $\infty$ (if $c>0$ ) — never in $[1,3]$ . Contradiction. Hence $V$ is not Lebesgue measurable.

connection

Measure theory is the foundation of Probability Theory: a probability space is a measure space with $\mu(X)=1$ , random variables are measurable functions, and expectation is the Lebesgue integral. The completeness of $L^p$ spaces feeds directly into Functional Analysis (Banach and Hilbert spaces), where $L^2$ becomes the model Hilbert space underlying Fourier Analysis. The Riesz–Markov theorem ties measures to continuous linear functionals on $C(X)$ , linking measure theory to Topology. Differentiation theory connects to Sobolev Spaces and the study of Partial Differential Equations, where weak derivatives are defined through integration. Finally, the Cantor set and measure-zero phenomena bridge into Fractal Geometry and Hausdorff dimension.

🔬 Where this appears

⏳ History

- 1821 — Cauchy publishes Cours d'analyse, introducing rigorous limits and a definition of continuity, beginning the "arithmetization of analysis."

- 1854 — Riemann defines the Riemann integral in his habilitation, via upper/lower sums on domain partitions.

- 1872 — Weierstrass constructs a continuous nowhere-differentiable function, demolishing naïve intuition and demanding

\varepsilon

–

\delta

rigor.

- 1872 — Dedekind & Cantor independently construct

\mathbb{R}

(Dedekind cuts; Cauchy sequences), cementing completeness [1].

- 1883 — Cantor introduces the Cantor set, an uncountable set of measure zero.

- 1898 — Borel develops a theory of measure on the Borel

\sigma

-algebra, with countable additivity.

- 1902 — Lebesgue completes his thesis defining outer measure [4], measurable functions, and the Lebesgue integral [6]; proves the Lebesgue differentiation theorem.

- 1906–1910 — Fatou, Vitali, Egorov prove Fatou's Lemma, exhibit the non-measurable Vitali set, and establish Egorov's theorem.

- 1918 — Carathéodory gives the abstract outer-measure criterion [5], generalizing measure to arbitrary spaces; 1933 — Kolmogorov axiomatizes probability on this foundation.

open-problem

Measure theory rests on logical foundations that are not fully forced. Whether every subset of $\mathbb{R}$ can be measurable depends on set-theoretic axioms: the existence of the non-measurable Vitali set uses the Axiom of Choice. Solovay (1970) showed that, consistent with ZF + Dependent Choice (and assuming an inaccessible cardinal), one can have a model where every set of reals is Lebesgue measurable — so non-measurability is "optional," bought with full Choice. Open and deep questions persist in geometric measure theory and related areas: the precise structure of measure-theoretic boundaries (rectifiability), the Kakeya conjecture (do Besicovitch sets containing a unit segment in every direction necessarily have full Hausdorff/Minkowski dimension?), and sharp constants in maximal inequalities. At the foundational frontier, the interplay between measure, large cardinals, and determinacy (e.g. "all projective sets are measurable" follows from determinacy axioms) remains an active research program connecting analysis to logic.

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also