Independence of random variables — formal definition: Visual walkthrough

4.9.11 · D2 · Maths › Probability Theory & Statistics › Independence of random variables — formal definition

Shuru karne se pehle: do words jo hum baar baar use karenge.

Neeche ki saari cheezein ek flat picture pe hain: horizontal axis wo saari values hain jo $X$ le sakta hai, vertical axis wo saari values hain jo $Y$ le sakta hai. Ek dot $(x, y)$ = " $X$ ka output $x$ aur $Y$ ka output $y$ aaya".

Step 1 — Do wheels jo ek doosre ko ignore karte hain

KYA. Do spinning wheels ka picture socho. Wheel- $X$ bottom axis pe kisi bhi value pe ruk sakta hai; wheel- $Y$ side ki kisi bhi value pe.

KYUN. Hume "ek doosre se baat nahi karte" ka mental model chahiye, us se pehle ki hum ise maths mein likhen. Independence exactly yahi hai: wheel- $X$ ko rukta dekh ke tumhe wheel- $Y$ ke baare mein kuch naya nahi pata chalta.

PICTURE. Figure mein, shaded strip event hai " $X$ , $A$ mein pada" (ek vertical band, kyunki yeh sirf horizontal axis ko constrain karta hai). Doosri strip hai " $Y$ , $B$ mein pada" (ek horizontal band). Woh rectangle jahan dono overlap karte hain woh hai " $X \in A$ aur $Y \in B$ ".

Figure — Independence of random variables — formal definition

Poora derivation ek sawaal hai: us overlap rectangle ki probability kitni badi hai?

Step 2 — Independence equation, term by term

KYA. Hum claim karte hain: independent variables ke liye, overlap rectangle ki probability equals hai width-band probability times height-band probability.

$overlap rectangle P (X \in A, Y \in B) = vertical band P (X \in A) \times horizontal band P (Y \in B)$

Step 1 ki picture pe har symbol ko padhna:

$P (X \in A, Y \in B)$ — shaded overlap (dono conditions ek saath hold karti hain).
$P (X \in A)$ — sirf vertical band ki probability, $Y$ ko ignore karte hue.
$P (Y \in B)$ — sirf horizontal band ki probability, $X$ ko ignore karte hue.

Multiply KYUN karte hain? Yeh seedha Independence of events se aata hai: do events tab independent hote hain jab unka combined chance product hota hai. Ek random variable bas events " $X \in A$ " ki ek factory hai, isliye hum har band $A$ aur har band $B$ ke liye product rule demand karte hain.

PICTURE. Agla figure dikhata hai ki "sabhi $A, B$ ke liye" overkill kyun nahi hai: ek lucky rectangle ka product rule follow karna baaki sabko force nahi karta. Independence ek promise hai har rectangle ke baare mein ek saath.

Step 3 — Bands ko "sab kuch $x$ tak" tak shrink karo

KYA. Saare sets $A, B$ test karna infinite kaam hai. Toh sabse simple possible bands choose karo: $A = (- \infty, x]$ (" $X$ zyada se zyada $x$ hai") aur $B = (- \infty, y]$ (" $Y$ zyada se zyada $y$ hai").

YEH bands KYUN. Yeh nested hain — jaise jaise tum $x$ ko right slide karte ho, band sirf badhta hai. Nested "staircase" bands ki yeh akeli family baad mein har region rebuild karne ke liye kaafi hai (Step 6 batata hai kyun). Hum ek infinite test ko ek checkable test se trade kar lete hain.

Inhe Step 2 ki equation mein plug karo:

$lower-left corner box P (X \leq x, Y \leq y) = left of x P (X \leq x) \times below y P (Y \leq y)$

PICTURE. Ab overlap poora lower-left quarter hai jo corner point $(x, y)$ pe anchored hai: $x$ ke left mein sab kuch AND $y$ ke neeche.

Step 4 — Corner box ko naam do: the CDF

KYA. Us lower-left box ki probability ka ek naam hai.

Step 3 ko in names se rewrite karne par flagship result milta hai:

YEH SUNDAR KYUN HAI. Messy 2-D box do 1-D piles mein collapse ho jaata hai jinhein tum alag alag compute karke multiply kar sakte ho.

PICTURE. Figure corner box ko left-strip fraction aur bottom-strip fraction ke product ke roop mein paint karta hai — area = width $\times$ height, probability se bana hua.

Step 5 — Pile-up boxes se densities tak

KYA. Ek CDF probability pile karta hai. Ek point pe probability dekhne ke liye, hum poochte hain ki pile kitni tezi se badhta hai — woh ek derivative hai.

Derivative KYUN, aur KYUN do derivative. $\frac{\partial}{\partial x} F$ measure karta hai "kitni probability $x$ pe ek patli vertical sliver mein hai". Ise dobara $y$ mein karna, $\frac{\partial ^{2}}{\partial x \partial y} F$ , $(x, y)$ pe ek tiny tile mein probability isolate karta hai. Woh tile-density joint PDF hai.

CDF factorisation ke dono sides differentiate karo. Ek function of $x$ aur ek function of $y$ ka product cleanly split hota hai ( $x$ -derivative $y$ -factor ko ignore karta hai aur vice-versa):

$\frac{\partial ^{2}}{\partial x \partial y} [F_{X} (x) F_{Y} (y)] = f_{X} (x) F_{X}^{'} (x) f_{Y} (y) F_{Y}^{'} (y)$

PICTURE. Left: ek smooth surface jiska $(x, y)$ pe tile (slice through $x$ ) $\times$ (slice through $y$ ) ke equal hai. Right: spikes ki grid jiska height (row weight) $\times$ (column weight) hai.

Step 6 — Sirf corner boxes check karna KAAFI KYUN hai

KYA. Humne sirf nested $(- \infty, x]$ bands check kiye. Kya product rule sach mein har weird region $A, B$ ke liye hold karta hai?

KYUN haan. Koi bhi band corner boxes ko add aur subtract karke bana hai. " $a < X \leq b$ " $=$ (box up to $b$ ) $-$ (box up to $a$ ). Har Borel set aisi pieces stack karke milta hai. Kyunki product rule in additions aur subtractions ke baad bhi bachta hai, corner boxes pe hold karna use har jagah force karta hai.

PICTURE. Ek rectangle strip ko (bada corner box) $-$ (left box) $-$ (bottom box) $+$ (corner jo double-count hua use wapas add karo) ke roop mein draw kiya gaya hai — chaar boxes ka classic inclusion–exclusion. Agar har box factorise hota hai, toh strip bhi hoga.

Recall Sufficient kyun hai, sirf necessary nahi

Corner boxes saare Borel sets generate karte hain ::: isliye unpar product rule har region tak propagate hota hai, CDF test ko full independence ke equivalent bana deta hai.

Step 7 — Trap: separable formula, coupled support

KYA. Ek formula $g (x) h (y)$ jaisa lag sakta hai phir bhi variables gossip karte rehte hain — agar support ki shape unhe link karti ho.

KYUN fail hota hai. $f_{X, Y} = 2$ lo triangle $0 \leq x \leq y \leq 1$ pe. Number $2$ utna hi separable hai jitna ho sakta hai ( $2 = 1 \cdot 2$ ). Lekin region ek triangle hai, rectangle nahi: $X = 0.9$ seekhne se $Y \geq 0.9$ force ho jaata hai. Yeh information transfer hai — Step 2 ki definition band $B = (- \infty, 0.5]$ ke liye tab break ho jaati hai jab $X > 0.5$ .

Marginals compute karke multiply karke check karo: $f_{X} (x) = \int_{x}^{1} 2 d y = 2 (1 - x), f_{Y} (y) = \int_{0}^{y} 2 d x = 2 y,$ $f_{X} (x) f_{Y} (y) = 4 y (1 - x) \neq = 2 = f_{X, Y} .$

PICTURE. Triangle support (magenta) ke saath $x = 0.9$ pe ek dashed vertical line: allowed $y$ -values $0.9$ se shuru hote hain, $0$ se nahi. Rectangle test fail hota hai kyunki koi bhi product-shaped box is triangle ko exactly contain nahi karta.

Step 8 — Clean case (taaki tum success bhi dekho)

KYA. $f_{X, Y} (x, y) = e^{- x - y}$ on $x, y > 0$ .

KYUN kaam karta hai. $e^{- x - y} = e^{- x} \cdot e^{- y}$ : separable, aur support $(0, \infty) \times (0, \infty)$ ek genuine rectangle hai (ek infinite quarter-plane, phir bhi ek product). Dono boxes ticked.

$f_{X, Y} = f_{X} (x) e^{- x} \cdot f_{Y} (y) e^{- y} \Rightarrow X, Y \sim Exp (1) independently.$

PICTURE. Quarter-plane support (ek real rectangle infinity tak) ke saath contour tiles jo visibly ek $x$ -profile times ek $y$ -profile mein factor hoti hain.

Yahan independence ka matlab hai $E [X Y] = E [X] E [Y]$ , isliye covariance $0$ — lekin Covariance and correlation se yaad raho ki zero covariance akela tumhe independence wapas nahi dilaata.

Ek-picture summary

Is page ki saari cheezein logic ka ek arrow hai: event product rule → sabhi bands → nested corner boxes → CDF → derivative → PDF/PMF, rectangular-support gate se guard kiya hua.

Recall Feynman retelling (seedhe words mein)

Do fair-ground wheels independent hain agar ek ko land hote dekhne se doosre ke baare mein kuch pata nahi chalta. Ek flat map pe — wheel- $X$ neeche bottom mein, wheel- $Y$ side mein upar — "dono in ranges mein land karte hain" ek overlap rectangle hai. Independence kehta hai: overlap ki chance = left band ki chance × bottom band ki chance, aur yeh har rectangle ke liye hold karna chahiye, koi ek lucky wala nahi. Infinitely many rectangles check karna impossible lagta hai, toh hum sirf sabse simple check karte hain: "ek corner point ke down-and-left mein sab kuch". Woh pile CDF hai, aur rule ban jaata hai $F_{X, Y} = F_{X} F_{Y}$ . Ek single point mein double derivative se zoom karo aur corner boxes tiny tiles ban jaate hain: $f_{X, Y} = f_{X} f_{Y}$ . Dots ke grids ke liye yeh $p_{X, Y} = p_{X} p_{Y}$ hai. Do traps: ek formula ( $x$ mein kuch)( $y$ mein kuch) jaisa dikh sakta hai phir bhi dependent ho sakta hai agar uska support triangle ho rectangle ki jagah — tab $X$ jaanno toh $Y$ kahan reh sakta hai woh pin ho jaata hai. Aur zero covariance independence se weaker hai. Poora slogan: joint = product, everywhere, on a rectangle.

Connections

Independence of Random Variables — Formal Definition
Joint distribution and marginals
Independence of events
Conditional distributions and conditional independence
Covariance and correlation
Jointly Gaussian random variables
Sums of independent random variables — convolution