Noesis
4.10.25Advanced Topics (Elite Level)

Measure theory — Lebesgue measure (intro)

2,185 words10 min readdifficulty · medium1 backlinks
🇮🇳 Yeh note Hinglish mein padho →

WHY do we even need this?

WHAT we want — a wishlist for "length" μ\mu:

  1. μ([a,b])=b−a\mu([a,b]) = b-a (agrees with intervals)
  2. μ(∅)=0\mu(\varnothing) = 0, μ≥0\mu \ge 0
  3. Countable additivity: disjoint A1,A2,⋯⇒μ ⁣(⋃An)=∑μ(An)A_1, A_2,\dots \Rightarrow \mu\!\left(\bigcup A_n\right) = \sum \mu(A_n)
  4. Translation invariance: μ(A+t)=μ(A)\mu(A + t) = \mu(A)

It turns out you cannot have all four for every subset of R\mathbb{R} (the Vitali set breaks it). The fix: only measure a rich class of "nice" sets — the measurable sets.

HOW we build it: outer measure first

Deriving the basic properties from scratch

(P1) Monotonicity: if A⊆BA \subseteq B then μ∗(A)≤μ∗(B)\mu^*(A) \le \mu^*(B). Why? Every cover of BB is also a cover of AA. So the set of cover-sums for AA contains that of BB; an inf over a bigger set is ≤\le. Done.

(P2) Countable sub-additivity: μ∗ ⁣(⋃nAn)≤∑nμ∗(An)\mu^*\!\left(\bigcup_n A_n\right) \le \sum_n \mu^*(A_n). Why & How (the ε/2n\varepsilon/2^n trick): Fix ε>0\varepsilon>0. For each AnA_n pick a cover {In,k}k\{I_{n,k}\}_k with ∑kℓ(In,k)≤μ∗(An)+ε2n.\sum_k \ell(I_{n,k}) \le \mu^*(A_n) + \frac{\varepsilon}{2^n}. Why this step? By definition of inf, a cover beating μ∗(An)\mu^*(A_n) by any margin exists; we choose margin ε/2n\varepsilon/2^n so the errors form a geometric series. The combined family {In,k}n,k\{I_{n,k}\}_{n,k} is a countable cover of ⋃nAn\bigcup_n A_n, so μ∗ ⁣(⋃nAn)≤∑n,kℓ(In,k)≤∑n(μ∗(An)+ε2n)=∑nμ∗(An)+ε.\mu^*\!\Big(\bigcup_n A_n\Big) \le \sum_{n,k}\ell(I_{n,k}) \le \sum_n\Big(\mu^*(A_n)+\tfrac{\varepsilon}{2^n}\Big) = \sum_n \mu^*(A_n) + \varepsilon. Let ε→0\varepsilon \to 0. ■\blacksquare

(P3) μ∗([a,b])=b−a\mu^*([a,b]) = b-a.

  • "≤\le": the single cover (a−ε,b+ε)(a-\varepsilon, b+\varepsilon) has length b−a+2εb-a+2\varepsilon, so μ∗≤b−a\mu^*\le b-a.
  • "≥\ge" is the subtle direction (needs compactness): any open cover of the compact [a,b][a,b] has a finite subcover, whose lengths must total ≥b−a\ge b-a (else a gap is left uncovered). So μ∗≥b−a\mu^*\ge b-a. Together: =b−a=b-a. ✔ matches wishlist item 1.
Figure — Measure theory — Lebesgue measure (intro)

HOW we pick the "nice" sets: the Carathéodory criterion

Worked examples

Common mistakes (steel-manned)

Active recall

Recall Try before peeking
  • State the wishlist of 4 properties we want for "length".
  • Define μ∗(A)\mu^*(A) from scratch.
  • Why does μ∗(Q)=0\mu^*(\mathbb{Q}) = 0? (Give the ε/2n\varepsilon/2^n argument.)
  • State Carathéodory's measurability criterion and explain why the split must add exactly.
  • Which property upgrades from sub-additive (all sets) to additive (measurable sets)?
What is Lebesgue outer measure μ∗(A)\mu^*(A)?
The infimum of ∑nℓ(In)\sum_n \ell(I_n) over all countable covers of AA by open intervals — the tightest total cover length.
Why is μ∗\mu^* only sub-additive, not additive, on all sets?
Pathological sets (e.g. Vitali set) can "leak" measure; full additivity holds only on measurable sets in the σ\sigma-algebra M\mathcal M.
State Carathéodory's measurability criterion.
EE is measurable iff μ∗(A)=μ∗(A∩E)+μ∗(A∩Ec)\mu^*(A)=\mu^*(A\cap E)+\mu^*(A\cap E^c) for every test set AA.
What is μ∗(Q)\mu^*(\mathbb{Q}) and why?
00; cover the nn-th rational by an interval of length ε/2n\varepsilon/2^n, total ≤ε→0\le\varepsilon\to0. Countable sets are null.
What is μ([a,b])\mu([a,b]) and which direction needs compactness?
b−ab-a; the lower bound "≥\ge" uses the finite subcover of the compact [a,b][a,b].
Which key property fails for Riemann but holds for Lebesgue?
Countable additivity of the size/integral — lets us integrate wild functions like 1Q\mathbf 1_{\mathbb Q}.
Is every uncountable set of positive measure?
No — the Cantor set is uncountable yet has measure 00.
What structure do the measurable sets form?
A σ\sigma-algebra (closed under complement and countable unions), containing all Borel sets.
Recall Feynman: explain to a 12-year-old

Imagine measuring how much "ribbon" a set of points needs. To measure any blob, you throw lots of tiny ribbon-pieces (intervals) on top until the whole blob is hidden, then add their lengths. You try every possible way and keep the stingiest total — that's the "size". For a normal stick from 0 to 1, the size is 1, obviously. But here's the cool trick: the fractions (1/2, 1/3, 1/4…) can be listed one-by-one, so you cover the nn-th one with a ribbon half as small as the last. Add 12+14+18+…\tfrac12+\tfrac14+\tfrac18+\dots and even all the fractions together need almost no ribbon — their size is 00! Some weird blobs are so messy you can't measure them honestly; we just agree to only measure the well-behaved ones.

Connections

  • Riemann integration — what Lebesgue measure was invented to fix
  • Lebesgue integral — built on measurable functions and this measure
  • Sigma-algebra — the structure of measurable sets
  • Borel sets — smallest σ\sigma-algebra containing open intervals; all measurable
  • Axiom of Choice & Vitali set — source of non-measurable sets
  • Cantor set — uncountable, measure zero
  • Outer regularity — measure approximated by open supersets
  • Countable vs uncountable — why Q\mathbb Q is null but [0,1][0,1] is not

Concept Map

motivates

requires

wishlist for length mu

includes

includes

includes

breaks all four

forces restriction to

built via

defined as

gives

gives

proved by

restricted to

Riemann fails on Dirichlet fn

Lebesgue idea: chop y-axis

measure sets first

4 desired properties

countable additivity

translation invariance

agrees with intervals

Vitali set

measurable sets

outer measure mu star

inf over countable open covers

monotonicity

countable sub-additivity

epsilon over 2 to n trick

Hinglish (regional understanding)

Intuition Hinglish mein samjho

Dekho, Lebesgue measure ka idea simple hai: hum kisi bhi set AA ka "length" naapna chahte hain, na sirf interval ka. Trick yeh hai — AA ko chhote-chhote open intervals se dhak do (cover karo), un sabki lengths add karo, aur sabse kam total dene wala cover choose karo. Yahi inf⁡\inf wali cheez ko outer measure μ∗(A)\mu^*(A) bolte hain. Interval [a,b][a,b] ke liye yeh exactly b−ab-a nikalta hai, jaisa hona chahiye.

Sabse mast example: rational numbers Q\mathbb{Q}. Yeh dense hain (har jagah faile hain), phir bhi inka measure 00 hai! Kaise? Q\mathbb{Q} countable hai, toh nn-th rational ko ε/2n\varepsilon/2^n length ke interval se dhako. Total =ε/2+ε/4+⋯=ε= \varepsilon/2 + \varepsilon/4 + \dots = \varepsilon, aur ε→0\varepsilon \to 0. Matlab dense hone se size nahi banta — yeh galatfehmi se bachna. Isi wajah se Lebesgue theory mein 1Q\mathbf 1_{\mathbb Q} ka integral 00 aa jaata hai, jo Riemann nahi kar paata.

Ek problem hai: har set ko honestly naap nahi sakte (Vitali set jaise weirdo sets sab tod dete hain). Iska solution Carathéodory ne diya — set EE "measurable" tabhi hai jab woh har test set AA ko do hisson mein clean split kare aur dono ka measure exactly add ho: μ∗(A)=μ∗(A∩E)+μ∗(A∩Ec)\mu^*(A)=\mu^*(A\cap E)+\mu^*(A\cap E^c). Aise ache sets ka collection σ\sigma-algebra banata hai, aur unhi par countable additivity (disjoint sets ke measures add hote hain) milti hai.

Yaad rakhne ka mantra: COVER → SQUEEZE → SPLIT. Cover karo intervals se, squeeze karo inf⁡\inf se (outer measure), aur split-test karo (measurable sets). Yeh sab Lebesgue integral ki neenv hai — pehle sets naapo, phir functions integrate karo. Bahut powerful tool hai analysis aur probability theory mein.

Go deeper — visual, from zero

D1FoundationsEvery symbol from zero · 1,735 wordsD2Visual walkthroughThe derivation in pictures · 2,256 wordsD3Worked examplesEvery scenario covered · 3,819 wordsD4ExercisesGraded, full solutions · 3,142 wordsD5Question bankConcept traps, quick reveals · 1,546 words

Test yourself — Advanced Topics (Elite Level)

L1 RecognitionL2 RecallL3 ProductionL4 ApplicationL5 Mastery

Connections

Probability space — sample space Ω, sigma-algebra F, measure P — Kolmogorov axiomsMaths · 4.9.1