Visual walkthrough — Continuous random variables — PDF, CDF, percentiles

4.9.7 · D2 · Maths › Probability Theory & Statistics › Continuous random variables — PDF, CDF, percentiles

---

Step 1 — "Density curve" hoti kya hai

KYA. Hum ek curve $y=f(x)$ draw karte hain. Yahan $x$ wo value hai jo humari random cheez le sakti hai (maan lo, ek randomly chosen fraction jo $0$ ke paas se zyada $1$ ke paas land karne ki dono guna zyada chances rakhti hai), aur height $f(x)$ batati hai ki probability wahan kitni thickly pile up hai.

KYU ek curve aur probabilities ki list nahi. Ek continuous cheez (ek exact length, ek exact time) ki infinitely many possible values hoti hain. Agar har ek ko positive probability milti, toh total $1$ se zyada ho jaata. Toh koi single value probability carry nahi kar sakti — balki hum probability ko bread pe jam ki tarah spread karte hain, aur har jagah jam ki thickness describe karte hain. Woh thickness hi density hai, likha jaata hai $f(x)$.

PICTURE. Curve $f(x)=2x$ left pe $0$ se right pe $2$ tak rise karti hai. Notice karo ki height $2$ tak pahunchti hai — 1 se bhi badi — aur yeh bilkul theek hai, kyunki height density hai, probability nahi.

---

Step 2 — AREA probability kyun hai (mass-density argument, drawn)

KYA. Curve ka $x$ aur $x+\Delta x$ ke beech ek thin vertical strip lo. Uska area approximately height $\times$ width $=f(x)\Delta x$ hai. Hum claim karte hain ki woh chota area hi us strip mein land karne ki probability hai.

KYU. Parent note ka rod analogy yaad karo: chunk ka mass $=$ density $\times$ length. Same algebra yahan. Density $f(x)$ ki units hain "probability per unit $x$"; multiply karo width $\Delta x$ se (units of $x$) aur units cancel hokar pure probability bachti hai. Toh:

$$\underset{\text{chance of landing in strip}}{\underbrace{P(x\le X\le x+\Delta x)}}\approx \underset{\text{thickness}}{\underbrace{f(x)}}\cdot \underset{\text{strip width}}{\underbrace{\Delta x}}.$$

KYU integral, ordinary multiplication nahi. Ek thin strip ek approximation hai kyunki $f$ strip ke across change hoti hai. Interval $[a,b]$ exactly paane ke liye, ise infinitely many infinitely-thin strips mein slice karo aur unke areas add karo — woh infinitely-thin-slices ka addition exactly wahi hai jo integral ${\int }_a^{b}f(x)dx$ ka matlab hai. Isliye hum calculus ki taraf jaate hain, ordinary multiplication ki nahi.

$$P(a\le X\le b)={\int }_a^{b}f(x)dx(\text{a se b tak saari strip areas ka sum}).$$

PICTURE. Ek orange strip highlighted; uske neeche, strip ko baarik aur baarik kaata jaata hai jab tak smooth shaded area na ban jaaye.

---

Step 3 — Total area 1 equal hona chahiye (aur "zero at a point" kaisa dikhta hai)

KYA. Kuch na kuch toh hoga hi, toh saari probability mila ke $1$ hoti hai. Area language mein:

$${\int }_{-\infty }^{\infty }f(x)dx=1.$$

Apna example check karo: ${\int }_0^{1}2xdx=[x^2{]}_0^1=1-0=1$. ✓ Term $[x^2{]}_0^1$ antiderivative hai jo right edge pe evaluate hua minus left edge.

EK single point ki probability zero kyun. "Point" ek strip hai jiska width $\Delta x=0$ hai. Area $=f(c)\cdot 0=0$. Toh $P(X=c)={\int }_c^{c}f=0$. Yahi drawn reason hai ki continuous variables ke liye $<$ vs $\le$ ka distinction kabhi matter nahi karta.

PICTURE. $f(x)=2x$ ke neeche ka poora region $[0,1]$ pe shaded (ek triangle of area $\frac{1}{2}\cdot 1\cdot 2=1$ — tum calculus skip karke triangle-area formula bhi use kar sakte ho!), aur ek single red vertical line "zero width" ke saath jo koi area carry nahi karti.

---

Step 4 — Area sweep karke CDF banana

KYA. Ab fixed intervals ke baare mein poochna band karo aur ek running question poochho: "$x$ ke left mein kitni probability hai?" Us running total ko $F(x)$ kaho:

$$F(x)=P(X\le x)={\int }_{-\infty }^{x}f(t)dt.$$

• $F(x)$ ::: the area accumulated from the far left up to $x$ — the Cumulative Distribution Function.
• $t$ ::: a dummy sliding variable that runs from $-\infty$ up to the endpoint $x$. We rename the inside variable so it isn't confused with the moving right edge $x$.

KYU. Ek running total har interval question ko subtraction mein convert kar deta hai (Step 6) aur, crucially, wahi quantity hai jo hum percentiles find karne ke liye invert karte hain (Step 7). Yeh master accumulator hai.

Hamare example ke liye, $0\le x\le 1$ pe:

$$F(x)={\int }_0^{x}2tdt=[t^2{]}_0^x=x^2.$$

Lower limit $0$ hai (na ki $-\infty$) kyunki $f=0$ hai $0$ ke left mein — wahan koi area accumulate nahi hoti.

PICTURE. Left panel: $2x$ curve with position $x$ pe ek teal moving edge aur uske peeche area shaded. Right panel: us shaded area ki value ek new curve ki tarah plot ki — $0$ se $1$ tak climb karti hai, aur woh climbing curve hi $F(x)=x^2$ hai.

---

Step 5 — $F$ sirf kyun rise karta hai, $0$ pe start hota hai, aur $1$ pe finish hota hai

KYA. Har CDF ke baare mein teen shape-facts, Step 4 ki picture se seedhe padhe:

1. $F(-\infty )=0$ — abhi kuch accumulate nahi hua.
2. $F(+\infty )=1$ — sab kuch accumulate ho gaya (Step 3 ka total).
3. $F$ non-decreasing hai — jaise teal edge right slide karta hai woh sirf zyada area sweep kar sakta hai (kyunki $f\ge 0$ hai, area kabhi subtract nahi hoti).

KYU isme jumps bhi nahi hote (continuous RV ke liye). $F$ mein ek jump matlab kisi chunk ki probability zero width ke saath appear ho — woh exactly $P(X=c)>0$ hai, jo Step 3 ne forbid kiya tha. Toh $F$ ka graph ek unbroken climb hai.

PICTURE. $F(x)$ curve left-to-right trace ki: $0$ pe flat, ek smooth rise, $1$ pe flat — arrows ke saath "sirf upar jaata hai."

---

Step 6 — Ulta karna: $F$ ki slope $f$ wapas deti hai, aur difference intervals deta hai

KYA (differentiate). $F$ ko $f$ pile up karke banaya gaya tha. Toh jis rate se $F$ kisi point pe climb karta hai woh exactly hai ki wahan kitni density add ho rahi hai — woh slope hai. Fundamental Theorem of Calculus se:

$$f(x)=\frac{d}{dx}F(x)=F^{\prime }(x).$$

Check karo: $\frac{d}{dx}{x}^2=2x=f(x)$. ✓ $F$ jitna steeper hoga, $f$ utna taller hoga.

KYA (subtract). Fixed interval ke liye, $a$ aur $b$ ke beech area = ($b$ tak area) minus ($a$ tak area):

$$P(a\le X\le b)={\int }_a^{b}f=F(b)-F(a).$$

Example: $P(0.2\le X\le 0.5)=F(0.5)-F(0.2)=0.5^2-0.2^2=0.25-0.04=\mathrm{0.21.}$

PICTURE. Top: kisi $x$ pe $F$ ki slope (tangent line), caption ke saath "slope yahan $=$ $f$ ki height wahan." Bottom: do shaded areas $F(b)$ aur $F(a)$ jinka difference middle strip hai.

---

Step 7 — $F$ pe sideways jaake percentiles padhna

KYA. Ek percentile CDF question ko ulta kar deta hai. CDF poochta hai "diye gaye position $x$ ke liye, uske peeche kitna area hai?" Percentile poochta hai "diye gaye amount of area $p$ ke liye, position $x_p$ kahan hai?" Formally:

$$F(x_p)=p⟺x_p=F^{-1}(p),$$

jahan $F^{-1}$ inverse CDF hai (quantile function).

• $p$ ::: a target fraction of probability, e.g. $0.5$ for the median.
• $x_p$ ::: the value on the $x$-axis where that fraction has accumulated.

KYU yahan always solvable hai. Kyunki $f>0$ hai $[0,1]$ across, $F$ strictly rise karta hai (kabhi flat nahi), toh har height $p$ exactly ek $x_p$ pe hit hoti hai — ek unique answer.

Hamare example ka Median: solve karo $F(m)=m^2=0.5$, toh $m=\sqrt{0.5}\approx 0.707$. Notice karo median midpoint $0.5$ se aage baithti hai — kyunki density right lean karti hai, zyada "jam" right pe hai, toh halfway split right shift ho jaati hai.

PICTURE. $F$ curve: $y$-axis pe UP jaao $p=0.5$ tak, RIGHT travel karo jab tak curve hit na ho, phir DOWN jaao $x_p=0.707$ padhne ke liye. Percentiles ek horizontal-then-vertical read hain; CDF ek vertical-then-horizontal read hai.

---

Step 8 — Degenerate cases jo tumhe phir bhi handle karne honge

KYA. Real problems tumhe flat spots, edges, aur support ke bahar ke cases dete hain. Inhe explicitly handle karo:

• Support ke bahar. Hamare example ke liye $F(x)=0$ for $x<0$ aur $F(x)=1$ for $x>1$. Agar koi "$150\%$ percentile" maange, toh aisa koi $p$ exist nahi karta — percentiles sirf $0<p<1$ ke liye exist karte hain.
• Flat density (uniform). $[0,1]$ pe Uniform distribution ke liye, $f=1$ (constant) aur $F(x)=x$ (ek straight diagonal). Percentiles trivially padhte hain: $x_p=p$. Yeh "no lean" case hai — median exactly $0.5$ pe baithta hai.
• Endpoints koi probability carry nahi karte. $P(X\le 0)=F(0)=0$ aur $P(X\ge 1)=1-F(1)=0$. Edge ko include ya exclude karna kuch change nahi karta (Step 3).
• Ek gap jahan $f=0$. Agar density kisi stretch pe zero hai, $F$ wahan flat hai — woh briefly accumulate karna band kar deta hai. Tab $F$ us gap pe strictly increasing nahi hai, aur exactly flat spot pe land karne wala percentile unique nahi hai. (Haara $2x$ example yeh avoid karta hai kyunki support throughout $f>0$ hai.)

PICTURE. Do mini-panels side by side: left = uniform (flat $f$, diagonal $F$); right = ek density with ek zero-gap jo $F$ mein flat plateau produce karta hai.

---

Ek-picture summary

Sab kuch ek canvas pe: density $f=2x$ (orange) ek shaded strip ke saath = probability; CDF $F=x^2$ (teal) us area ko sweep karke build hua; aur median sideways trick se read kiya gaya (plum). Teen characters, ek story: density spreads, cumulative collects, percentile asks where.

Recall Feynman retelling — simple words mein poora walkthrough

Picture karo ki 1 kg jam bread pe spread kar rahe ho, lekin right pe zyada thick. Har jagah ki thickness PDF $f$ hai — aur thickness bahut badi ho sakti hai ("1" se zyada) jab tak total jam 1 kg ho. Kisi stretch mein land karne ka chance paane ke liye thickness nahi padhte; us strip ko scoop karke weigh karte hain — woh weight area hai, ek integral. Ab ek knife far left edge se right ki taraf chalao aur ab tak scooped saari cheez ka running weight track karo: woh running total CDF $F$ hai. Yeh $0$ se start hota hai, sirf climb karta hai (jam un-scoop nahi ho sakta), aur $1$ pe khatam hota hai jab saab sweep ho jaaye. Agar thickness wapas chahiye, dekho running total kitni fast climb kar raha hai — steep climb matlab thick jam; yeh differentiation hai. Finally, median dhundne ke liye thickness curve pe mat dhoondo — running-total curve pe height $0.5$ tak jao, curve ke across slide karo, neeche drop karo, aur woh jagah ke dono sides pe exactly aadha jam hai. Kyunki jam right lean karta tha, woh halfway spot ($0.707$) thoda right of middle baithta hai.

---

Connections

• Continuous random variables — PDF, CDF, percentiles (parent — yeh page uska central link visually derive karta hai)
• Fundamental Theorem of Calculus (Step 6: accumulated area ki slope density wapas deti hai)
• Uniform distribution (Step 8: flat-density / diagonal-CDF base case)
• Exponential distribution (support-from-0 ka ek aur density jo same tarah sweep hota hai)
• Normal distribution (iska CDF $\Phi$ exactly Step 7 ke sideways trick se padha jaata hai)
• Quantile function and inverse-transform sampling (Step 7 ka $F^{-1}$ action mein)
• Expectation and Variance of continuous RVs (next: in areas ko $x$ se weight karna)
• Discrete random variables — PMF (is integrating story ka summing cousin)

Concept Map