Visual walkthrough — Measure theory — Lebesgue measure (intro)

4.10.25 · D2 · Maths › Advanced Topics (Elite Level) › Measure theory — Lebesgue measure (intro)

---

Step 1 — Sawaal: points ke ek set ki "length" kya hoti hai?

YEH MUSHKIL KYUN HAI. Ek interval $[a,b]$ ke liye jawaab obvious hai: size hai $b-a$, left end $a$ se right end $b$ tak ki door. Lekin bikre hue points ki ek bheed ke liye koi "ends" nahi hain subtract karne ke liye. Hume ek aisa machine chahiye jo kisi bhi $A$ par kaam kare aur phir bhi intervals par $b-a$ de.

PICTURE. Green blob hamaara mystery set $A$ hai. Neeche woh ek case hai jis par hum pehle se trust karte hain: interval $[a,b]$ apni honest length $b-a$ ke saath.

---

Step 2 — Ek honest move: set ko intervals se cover karo

COVERS KYUN. Intervals ek hi cheez hain jinki length hum jaante hain ($\ell =b-a$). Toh trick hai: anjaan blob ko un cheezon se approximate karo jinhe hum JAANTE hain. Agar intervals $A$ ko poora dhaamp lein, toh intervals ki total length zaroor kam se kam utni badi hai jitni $A$ ki koi bhi honest "length" ho sakti hai. Cover se hum kabhi undershoot nahi karte — yahi ek directional guarantee hai.

PICTURE. Chalk-blue intervals $I_1,I_2,I_3,\dots$ green blob par bikhre hue. $A$ ka har point kisi interval ke neeche aata hai. Dhyan do intervals overlap ho sakte hain aur $A$ se bahar bhi nikal sakte hain — dono allowed hain.

Ek cover ki cost woh total chalk hai jo humne kharch ki: $\text{cost}={\sum }_{n=1}^{\infty }\ell (I_n)={\sum }_{n=1}^{\infty }(b_n-a_n).$

---

Step 3 — Squeeze: sabse tight cover lo (infimum)

INFIMUM KYUN, MINIMUM KYUN NAHI? Koi ek "sabse sasta" cover nahi ho sakta — aksar aap cost ko thoda aur chota kar sakte hain intervals ko thoda aur shrink karke, hamesha ke liye. Toh koi minimum nahi hota. Iski jagah hum infimum ($\inf$) lete hain: woh sabse bada number jisse koi cover neeche nahi ja sakta — woh "floor" jiske paas costs aati hain lekin shayad kabhi touch na karain. Woh floor $A$ ka honest size hai.

PICTURE. Ek hi blob ke teen covers: fat/expensive (pink), medium (yellow), tight/cheap (blue). Ek arrow dikhata hai costs neeche ek floor line ki taraf ja rahi hain — woh floor hai ${\mu }^{\ast }(A)$.

---

Step 4 — Pehla sanity check: ek akele point ki cost kuch nahi

YEH ZERO KYUN HONA CHAHIYE. Ek point ki koi width nahi hoti. Hum use kisi bhi width ke chhote interval mein qaed kar sakte hain aur phir woh width $0$ ki taraf shrink kar sakte hain.

PICTURE. Point $p$ (ek chalk dot) ek shrinking interval $(p-\epsilon ,p+\epsilon )$ ke andar; width $2\epsilon$ dot par collapse ho rahi hai.

$\{p\}\subseteq (p-\epsilon ,p+\epsilon ),\ell =(p+\epsilon )-(p-\epsilon )=2\epsilon .$

• $\epsilon$ ("epsilon") koi bhi tiny positive number hai jo hum choose karte hain.
• Yeh akela interval pehle se $p$ ko cover karta hai, toh yeh cost $2\epsilon$ ka ek legal cover hai.

Kyunki ${\mu }^{\ast }(\{p\})\le 2\epsilon$ har $\epsilon >0$ ke liye, aur ${\mu }^{\ast }\ge 0$ hamesha, floor forced hai ki ${\mu }^{\ast }(\{p\})=0$ ho. ✔

---

Step 5 — $\epsilon /2^n$ ka jaadu: $\mathbb{Q}$ ki bhi cost kuch nahi

YEH KYUN KAAM KARTA HAI. $\mathbb{Q}$ countable hai, toh hum fractions ko line up kar sakte hain: $q_1,q_2,q_3,\dots$. Ab har fraction ko uska apna tiny interval do, lekin unhe geometrically shrink karo — $n$-wein ko width $\epsilon /2^n$ milti hai. Ek shrinking geometric series ki finite total hoti hai, aur hum us total ko jitna chahein utna chota bana sakte hain.

PICTURE. Fractions line par dots ke roop mein; interval $n$ ki width $\epsilon /2^n$ hai toh widths row mein aadhi hoti jaati hain: $\epsilon /2,\epsilon /4,\epsilon /8,\dots$. Ek stacked bar dikhata hai widths sirf $\epsilon$ mein add ho rahi hain.

$\mathbb{Q}\subseteq {\bigcup }_{n=1}^{\infty }(q_n-\frac{\epsilon }{2^{n+1}},q_n+\frac{\epsilon }{2^{n+1}}),\text{width of interval }n=\frac{\epsilon }{2^n}.$

$\text{total cost}={\sum }_{n=1}^{\infty }\frac{\epsilon }{2^n}=\epsilon \underset{=1}{\underbrace{(\frac{1}{2}+\frac{1}{4}+\frac{1}{8}+\cdots )}}=\epsilon .$

• Har term $\epsilon /2^n$ woh width hai jo akeli fraction $q_n$ ko cover karti hai.
• Geometric sum $\frac{1}{2}+\frac{1}{4}+\frac{1}{8}+\cdots$ exactly $1$ ke barabar hai — isliye poora cover sirf $\epsilon$ cost karta hai.

Kyunki ${\mu }^{\ast }(\mathbb{Q})\le \epsilon$ har $\epsilon >0$ ke liye, hum paate hain ${\mu }^{\ast }(\mathbb{Q})=0$. Uncountable set ke liye jo ki measure zero bhi hai, dekho Cantor set — yeh dikhata hai ki "bada count" aur "bada measure" unrelated hain.

---

Step 6 — ${\mu }^{\ast }$ abhi kaafi kyun nahi: the leak

YEH HAMARE WISHLIST KE LIYE FATAL KYUN HAI. Hamaari wishlist mein exact additivity thi: disjoint pieces ke sizes exactly whole mein add up hone chahiye. Sirf "$\le$" ke saath, ${\mu }^{\ast }$ wishlist item 3 fail karta hai. Culprit sets pathological monsters hain jaise Vitali set (Axiom of Choice se build hua).

PICTURE. Ek set $A$ ek wall se $A\cap E$ (blue) aur $A\cap E^c$ (pink) mein katta hua. Ek nice $E$ ke liye do chalk bars exactly $A$-bar ke upar stack hoti hain. Ek bad $E$ ke liye pink+blue bars $A$-bar se zyada ho jaati hain — woh gap leak hai.

Kisi bhi $E$ ke liye, sub-additivity free direction deti hai ${\mu }^{\ast }(A)\le {\mu }^{\ast }(A\cap E)+{\mu }^{\ast }(A\cap E^c).$

• $A\cap E$ = $A$ ka woh hissa jo $E$ ke andar hai; $A\cap E^c$ = woh hissa jo bahar hai ($E^c$, $E$ ka complement hai).
• "$\le$" hamesha hold karta hai kyunki do halves ke covers $A$ ke ek cover mein jud jaate hain.

Missing "$\ge$" exactly wahi hai jo good sets ko leaky ones se alag karta hai.

---

Step 7 — Carathéodory ka filter: sirf clean-splitting sets rakhon

YEH EXACT TEST KYUN. Humne dekha ki "$\le$" automatic hai. Toh equality maangna actually extra "$\ge$" maangna hai — yaani $E$ kisi bhi possible probe $A$ ke khilaf kabhi leak nahi karta. Is test se pass hone wale sets exactly add kiye ja sakte hain, jo exactly wishlist item 3 restore karta hai.

PICTURE. Saare sets ki universe ek bade board ki tarah; iske andar, ek fenced region $\mathcal{M}$ = measurable sets (intervals, Borel sets, aur aur) glow karte hue; Vitali monster fence ke bahar baitha hai.

Parent note se rewards: measurable sets $\mathcal{M}$ ek Sigma-algebra form karte hain (complement aur countable unions ke under closed), inme har interval aur isliye har Borel set aata hai, aur $\mathcal{M}$ par hamaara size machine finally countably additive ban jaata hai: $\mu \left({⨆}_n{E}_n\right)={\sum }_n\mu (E_n)(\text{disjoint }{E}_n).$

Yahi woh payoff hai jo Lebesgue integral ko power karta hai aur un integrals ko rescue karta hai jo Riemann integration nahi kar sakta.

---

Ek-picture summary

Ek board par poora pipeline: blob → use intervals se cover karo → cost ko ek floor tak squeeze karo (${\mu }^{\ast }$) → leaky sets filter karo (Carathéodory) → exact additivity pao ($\mu$).

Recall Feynman retelling — plain words mein bolo

Main jaanna chahta hoon ki blob $A$ kitni line leti hai. Main sirf ek interval ki length par trust karta hoon, jo hai right end minus left end. Toh main blob par tiny intervals ki ek list — ek cover — bichhaata hoon aur unki lengths add karta hoon. Woh total kabhi too small nahi hota, sirf possibly too big, toh main cover ko shrink karta rehta hoon aur un saari totals ka bilkul bottom leta hoon: infimum. Woh floor hai outer measure ${\mu }^{\ast }$, aur yeh har set ke liye kaam karta hai. Main ise ek point par check karta hoon (width $2\epsilon$ se cover karo, $0$ tak shrink karo → size $0$) aur saari fractions par (fraction $n$ ko width $\epsilon /2^n$ do; halving widths sirf $\epsilon$ mein add hoti hain → size $0$, chahe fractions har jagah hon). Lekin ${\mu }^{\ast }$ mein ek flaw hai: ek monster set ko do disjoint parts mein todna parts ke sizes ko whole se zyada shoot kara sakta hai — ek leak. Toh main sirf woh sets $E$ rakhta hoon jo har probe set ko exactly split karte hain (Carathéodory). Un acche sets par, disjoint pieces ke sizes perfectly add hote hain. Woh perfect adding hi poori baat hai — yahi Lebesgue measure aur Lebesgue integral ko kaam karwaata hai.

Recall Quick self-test

• Hum intervals se kyun cover karte hain, blob ko directly measure kyun nahi karte? ::: Intervals ek hi cheez hain jinki length ($b-a$) par hum pehle se trust karte hain; covers over-estimate karte hain, kabhi under-estimate nahi.
• Infimum kyun, minimum kyun nahi? ::: Aksar aap cover ko thoda forever shrink kar sakte hain, toh koi ek cheapest cover exist nahi karta; infimum woh floor hai jiske paas woh costs aati hain.
• ${\mu }^{\ast }(\mathbb{Q})=0$ kyun hai, density ke bawajood? ::: Countability hume fraction $n$ ko width $\epsilon /2^n$ dene deti hai; geometric widths sirf $\epsilon \to 0$ mein total hoti hain. Density $\ne$ measure.
• Carathéodory ka test ${\mu }^{\ast }$ ke upar kya add karta hai? ::: "$\ge$" direction — koi measurement leak nahi — sub-additivity ko $\mathcal{M}$ par exact countable additivity mein badalta hai.