Foundations — Law of Large Numbers — weak and strong

4.9.16 · D1 · Maths › Probability Theory & Statistics › Law of Large Numbers — weak and strong

Yeh Law of Large Numbers — weak and strong ka ground-floor page hai. Parent note bahut saara notation ek saath throw karta hai: $X_i$, $\mu$, ${\sigma }^2$, $\mathbb{E}$, ${\bar{X}}_n$, $\to$ do alag flavours mein, $\mathbb{P}(\dots )$, $\epsilon$, aur summation sign $\sum$. Neeche hum har ek ko scratch se build karte hain, ek aisi order mein jahan har symbol use hone se pehle define ho jaata hai.

0. Woh picture jis par hum baar baar wapas aate hain

Kisi bhi symbol se pehle, dekho yeh poora topic kis baare mein hai: ek running average jo shuru mein wild hota hai aur phir calm ho jaata hai.

Jagged cyan line har nayi trial ke baad abhi-tak-ka-average hai. Dekho yeh left side par bahut zyada swing karta hai (kam trials) aur right side par amber line ke saath chipak jaata hai (zyada trials). Amber horizontal line woh number hai jis taraf average ja raha hai. Inhi do cheezon ko precisely naam dena is page ka poora kaam hai.

1. Ek random variable — $X$

Simple words mein. Ek coin flip karo. Outcome hai "heads" ya "tails" — words hain, numbers nahi. Arithmetic (averaging!) karne ke liye humein numbers chahiye, isliye hum ek translation agree karte hain: heads $\to 1$, tails $\to 0$. Woh translation-rule hi random variable $X$ hai.

Picture mein. Ek machine socho: outcome andar jaata hai slot mein, ek number bahar nikalta hai.

Topic ko kyun chahiye. Law of Large Numbers numbers average karta hai. Tum "heads, tails, heads" average nahi kar sakte. Tum $1,0,1$ average kar sakte ho. Toh step one hamesha yahi hai: outcomes ko numbers mein badlo ek random variable se.

2. Kaafi saare copies, aur subscript $X_i$

Simple words mein. Subscript sirf ek label hai — trial number 1, 2, 3, … Cinema mein seat numbers se zyada kuch mysterious nahi.

Picture mein. Section 1 ki identical machines ki ek row, har ek apna number ugal rahi hai, left se right label ki gayi hain.

Topic ko kyun chahiye. LLN repetition ke baare mein hai. Humein 1st, 2nd, … results ke naam chahiye taaki hum unhe add kar sakein.

Inhi copies se attached do words: i.i.d.

Topic ko kyun chahiye. "Identical" guarantee karta hai ki ek sachcha mean $\mu$ hai jis par converge karna hai. "Independent" woh cheez hai jo random wiggles ko cancel karaata hai pile up hone ki jagah — yeh literally poore law ka engine hai (tum ise variance calculation mein cross-terms ko khatam karte dekhoge).

3. Probability — $\mathbb{P}(\dots )$

Simple words mein. $\mathbb{P}(A)$ padho "$A$ ki probability". Jo bhi brackets ke andar hai woh ek event hai — outcomes ka description jo tumhe care karte hain, jaise "$X=1$" (heads) ya "$|{\bar{X}}_n-\mu |>0.1$" (average bahut door hai).

Picture mein. $0$ se $1$ tak ek bar; event ki likelihood uspar kahin ek mark hai.

Topic ko kyun chahiye. Weak Law ek statement hai ek probability ke zero shrink hone ke baare mein. Is symbol ke bina tum ise likh bhi nahi sakte.

4. Expectation / mean — $\mathbb{E}[X]$ aur $\mu$

Yeh woh target hai jis taraf poora topic point karta hai, isliye hum slowly jaate hain.

Sum kyun, aur kyun yahi tool? Hum possible values ka "centre of gravity" chahte hain. Har value ko kitni baar hota hai us hisaab se weight karna, phir add karna, exactly see-saw ka balance-point hai — zyada heavy (zyada likely) values balance ko apni taraf kheenchte hain. Woh balance point woh hai jis par averages ki ek badi pile ko baithna chahiye. Koi aur combination ka woh balancing property nahi hai, isliye expectation, (maan lo) middle value nahi, sahi tool hai.

Picture mein. Values as weights on a ruler; $\mu$ woh point hai jahan ruler balance karta hai.

Topic ko kyun chahiye. $\mu$ wahi number hai jis par sample average converge karta hai. Yeh Figure 1 mein amber line hai.

5. Summation sign — $\sum$

Simple words mein. Yeh compact "inhe sab add karo" hai. ${\sum }_{i=1}^3{X}_i=X_1+X_2+X_3$.

Picture mein. Ek hopper jo machine-outputs ki poori row nigal jaata hai aur ek total ugalata hai.

Topic ko kyun chahiye. Average hota hai total divided by count. Humein ek clean way chahiye "pehle $n$ results ka total" likhne ka.

6. Sample mean — ${\bar{X}}_n$

Ab hum show ke star ko un parts se assemble kar sakte hain jo hum already jaante hain.

Simple words mein. Pehle $n$ results add karo, jitne the utne se divide karo. Upar bar ka matlab hai "ka average", subscript $n$ matlab "ab tak $n$ trials use karke".

Picture mein. Yeh exactly woh jagged cyan line hai Figure 1 mein. Har nayi trial top mein ek term add karti hai aur $n$ ko ek bump karti hai, curve par ek naya dot deti hai.

Topic ko kyun chahiye. Poora Law of Large Numbers yeh claim hai "${\bar{X}}_n$ approaches $\mu$". Yeh symbol sab kuch ka left-hand side hai.

7. Spread — variance ${\sigma }^2$

Square kyun, aur kyun yahi tool? Hum "typical wobble size" ke liye ek number chahte hain. Agar hum raw distance $X-\mu$ average karte, positives aur negatives cancel ho jaate aur hamesha $0$ dete — useless. Squaring sign remove karta hai taaki kuch cancel na ho, aur bade misses ko zyada punish karta hai. Isliye variance, average signed distance nahi, woh tool hai jo spread measure karta hai.

Picture mein. $\mu$ ke aas paas dots ke cloud ki width: tight cloud mein chota ${\sigma }^2$, wide scattered cloud mein bada ${\sigma }^2$.

Topic ko kyun chahiye. ${\bar{X}}_n$ ke settle hone ki speed variance se govern hoti hai: parent ka engine $\mathrm{Var}({\bar{X}}_n)={\sigma }^2/n$ shrink karta hai kyunki hum ${\sigma }^2$ ko $n$ se divide karte hain. Bada spread = slow settling.

8. Tolerance $\epsilon$ aur "door hona"

Vertical bars. $|z|$ absolute value hai: yeh sign hata deta hai, $z$ ki $0$ se distance deta hai. Toh $|{\bar{X}}_n-\mu |$ bas "average aur true mean kitne door hain" hai, chahe koi bhi bada ho.

Picture mein. Amber line ke around $\epsilon$ half-width ka ek funnel. "$\epsilon$ se zyada door hona" matlab cyan dot funnel se bahar jhank raha hai.

Topic ko kyun chahiye. Convergence ka matlab hai "eventually har funnel ke andar, chahe funnel kitna bhi thin ho". $\epsilon$ woh hai jisse hum "chahe kitna bhi thin ho" kehte hain.

9. Do arrows: convergence — $\to$

Parent do alag senses mein convergence use karta hai. Dono ka matlab hai "close hota hai aur close rehta hai", lekin woh closeness alag-alag measure karte hain. Details Modes of Convergence mein hain; yahan sirf woh minimum hai jo tumhe chahiye.

• In probability (Weak Law): $\mathbb{P}(|{\bar{X}}_n-\mu |>\epsilon )\to 0$. Ek fixed bade $n$ ke liye, funnel se bahar jhankne ka chance tiny hai.
• Almost surely (Strong Law): $\mathbb{P}({\lim }_{n\to \infty }{\bar{X}}_n=\mu )=1$. Ek poori infinite path dekho; certainty ke saath woh eventually funnel mein hamesha ke liye ghus jaati hai.

Topic ko dono kyun chahiye. Weak aur Strong Laws mein sirf woh arrow alag hai jo use hota hai. In dono ko samajhna parent note ka point hai.

Prerequisite map

Kaise padhen. Left side par sab raw vocabulary hai; har arrow ka matlab hai "build karne ke liye zaroori hai". Sab do convergence statements mein funnel hote hain, jo hain hi Weak aur Strong Laws. Variance branch convergence ki speed feed karta hai, Chebyshev's Inequality proof ko power deta hai.

Yeh aage kahan jaate hain

• $\mathbb{E}$, $\mathbb{P}$, variance $\Rightarrow$ Markov's Inequality aur Chebyshev's Inequality (woh tools jo Weak Law prove karte hain).
• Do arrows $\Rightarrow$ Modes of Convergence (weak vs strong rigorous tarike se).
• $\mu$ ke around fluctuation shape $\Rightarrow$ Central Limit Theorem.
• Almost-sure machinery $\Rightarrow$ Borel–Cantelli Lemma.
• Integrals estimate karne ke liye averaging $\Rightarrow$ Monte Carlo Methods.
• Classic misuse $\Rightarrow$ Gambler's Fallacy.

Equipment checklist

Self-test: right side cover karo, answer do, phir reveal karo.