Foundations — Central Limit Theorem — statement, proof sketch, significance

4.9.15 · D1 · Maths › Probability Theory & Statistics › Central Limit Theorem — statement, proof sketch, significanc

0. "Random variable" kya hota hai?

Kisi bhi Greek letter se pehle, hume woh object samajhna hai jise baaki sab describe karte hain.

Picture yeh sochiye: ek machine hai jisme ek lever hai. Usse kheencho aur ek number bahar aata hai. Aapko pehle se nahi pata kaun sa number aayega, lekin aapko rules pata hain ki har number kitna likely hai.

• Ek die roll: $X\in \{1,2,3,4,5,6\}$, sab equally likely.
• Ek coin jisme heads ke liye $1$ aur tails ke liye $0$ flag hai: $X\in \{0,1\}$ — yeh Bernoulli building block hai.

Topic ko yeh kyun chahiye: CLT ek statement hai bahut saari aisi machines ke baare mein, isliye pehle ek ko naam dena zaroori hai.

1. Probability operator $P$, aur likelihood kaise store hoti hai

Outputs ko average karne se pehle hume yeh batana hoga ki har output kitna likely hai.

Machine ke do flavours hain, aur dono apni likelihood alag tarike se store karte hain:

Picture: ek PMF spikes (bars) ki ek row hoti hai jinki heights ka sum $1$ hota hai; ek PDF ek smooth curve hoti hai jiska total shaded area $1$ hota hai.

Topic ko yeh kyun chahiye: baad ke har formula mein outputs ko unki likelihood se weight kiya jaata hai, isliye hume "likelihood" ka matlab fix karna hai — aur note karna hai ki yeh dono discrete aur continuous machines ke liye kaam karta hai.

2. Mean $\mu$ aur symbol $E[\cdot ]$

Picture: mean probability picture ka balance point hai — woh jagah jahan bars (ya curve ke neeche ka area) ek pencil tip par balance karein.

Tool $E[\cdot ]$ kyun, na ki sirf "meri data ka average"? Data ka average ek aisa number hai jo har baar sample karne par badal jaata hai. $E[X]$ woh sachchi fixed target hai — machine ki ek property, kisi ek sample ki nahi. CLT poori tarah is baare mein hai ki noisy data-average is fixed $\mu$ ke around kaise nachhta hai.

3. Variance ${\sigma }^2$ aur standard deviation $\sigma$

Mean batata hai machine kahan centre karti hai. Yeh nahi batata ki outputs kitne wild hain.

Picture: $\sigma$ probability picture ka typical "arm span" hai — ek single output aksar balance point se kitna dur jaata hai.

Topic ko yeh kyun chahiye: CLT require karta hai ki ${\sigma }^2$ finite aur positive ho. Agar spread infinite ho (jaise Cauchy Distribution mein), toh poora theorem collapse ho jaata hai — mean kabhi settle nahi hota.

4. "i.i.d." — independent and identically distributed

Yeh chaar-letter tag classical CLT ka load-bearing assumption hai. Pehle counter $n$ fix karo:

Picture: identical, alag-alag wired machines ki ek row, har ek ko ek baar pull kiya — koi bhi do machines wire se connected nahi.

Topic ko yeh kyun chahiye: independence wahi cheez hai jo proof mein expectations ke product ko factor karne deti hai (dekho Characteristic Functions), aur variances ko add hone deti hai. "Identically distributed" thoda relax karo toh aapko Lindeberg–Feller Condition chahiye.

5. Sample mean ${\bar{X}}_n$ aur kyun uski spread $\sigma /\sqrt{n}$ hai

Ab woh crucial fact jo parent note assert karta hai, Section 3 ke do variance rules se build hota hai:

Picture: jaise aap zyada machines use karte ho, average ka histogram tight hokar $\mu$ ke around aa jaata hai — uska arm span $1/\sqrt{n}$ ki tarah shrink hota hai.

$\sqrt{n}$ itna important kyun hai: kyunki spread shrink hoti hai (yeh Law of Large Numbers hai) lekin sirf slowly, hum sirf $\mu$ subtract nahi kar sakte — hume ek zero-par-shrink hoti cheez milegi. Hume exactly is standard error se rescale karna hoga taaki ek stable, non-trivial shape mile. Yahi woh Standard Error hai jo topic mein har jagah aata hai.

6. Standardizing: symbol $Z_n$

Words mein: mera sample average sach se kitne standard errors dur hai?

Topic ko yeh kyun chahiye: CLT jo limit prove karta hai woh ek fixed shape hai, isliye hume $\mu$ aur $\sigma$ (jo har problem mein alag hote hain) strip karne hote hain. $Z_n$ woh stripped, universal version hai.

7. Target shape: $N(0,1)$, $\Phi$, aur "converges in distribution"

Picture: $\varphi$ woh pahadi hai; $\Phi$ shaded area ka running total hai jaise aap left se right sweep karte ho — yeh $0$ se upar chhada ke $1$ tak jaata hai.

Topic ko yeh kyun chahiye: yeh arrow CLT ke conclusion ka exact matlab hai. Jo bhi downstream hai — Confidence Intervals, Hypothesis Testing — z and t tests — woh $\Phi$ ki values se padhe jaate hain.

Prerequisite map

Equipment checklist

Khud ko tick do sirf tab agar aap yeh out loud answer kar sako: