Ratio test — proof, limitations: Visual walkthrough

4.3.12 · D2 · Maths › Calculus III — Sequences & Series › Ratio test — proof, limitations

Step 1 — Ek picture mein series kya hoti hai?

WHAT. Hum har term $a_{n}$ ko ek coloured bar mein badal dete hain jiska height $∣ a_{n} ∣$ hai, phir unhe stack karte hain.

WHY. "Kya series converge karti hai?" ek visible sawaal ban jaata hai: kya stacked tower kisi finite ceiling tak pahunchti hai, ya bina bound ke badhti rehti hai? Hum $∣ a_{n} ∣$ (size, sign ignore karke) use karte hain kyunki vertical bars ki sizes hi pile up hoti hain — signs baad mein handle kiye jaate hain.

WHAT IT LOOKS LIKE. Figure mein, har bar ek term hai. Agar bars kaafi tezi se shrink karein, toh running total (dayi taraf climbing dashed line) ek ceiling par flatten ho jaati hai.

Figure — Ratio test — proof, limitations

Yahan $∣ a_{n} ∣$ = ek bar ki height, aur $S_{N}$ = ab tak total stacked height. Poora proof is baare mein hai ki bar heights ko shrink kaise control karein.

Step 2 — Woh ek number jise hum dekhte hain: shrink-ratio

WHAT. Hum har step par maapte hain ki agla bar current bar ke comparison mein kitna tall hai.

WHY division aur subtraction nahi? Kyunki geometric shrink multiplicative hoti hai: "har bar pehle wale ka $70%$ hai" ek ratio statement hai, difference nahi. $0.7$ ka ratio matlab hai "agla bar paane ke liye $0.7$ se multiply karo." Yeh bilkul wahi language hai jo geometric series bolti hai, aur geometric series woh akeeli series hain jinhe hum closed form mein sum karna jaante hain. Toh division woh tool hai jo us answer ki shape se match karta hai jise hum dhoondh rahe hain.

WHAT IT LOOKS LIKE. Figure mein, ratio agla bar ka height ÷ is bar ka height hai — do paas-paas bars ko compare karta ek bracket ke roop mein draw kiya gaya.

Agar ratio chhota hai ( $0.3$ ): agla bar ek stub hai — tez shrink.
Agar ratio $1$ ke paas hai: agla bar lagbhag utna hi tall — barely shrink ho raha hai.
Agar ratio $1$ se upar hai: agla bar taller hai — tower badh raha hai.

Step 3 — Long-run ratio $L$

WHAT. Hum shrink-ratio ko infinity tak follow karte hain aur woh number record karte hain jis par yeh settle ho jaata hai.

WHY. Pehle kuch bars kuch bhi kar sakte hain — convergence decide karta hai tail (bars jo bahut door hain). $L$ noisy start ko throw away karta hai aur tail ki sacchi shrink habit ko ek number mein capture karta hai.

WHAT IT LOOKS LIKE. Shrink-ratio ko $n$ ke against plot karo. Yeh shuru mein wiggle karta hai, phir $L$ height par horizontal line ki taraf flatten ho jaata hai. Teen lines matter karti hain: $L$ neeche $1$ se, $L = 1$ , $L$ upar $1$ se.

Poora verdict is baat par depend karta hai ki woh flat line level $1$ ke relative kahan land karti hai.

Step 4 — $L < 1$ : gap mein ek fence-rail $r$ choose karo

WHAT. Hum ek number $r$ choose karte hain jahan $L < r < 1$ aur ise $L$ ke upar ek horizontal line ke roop mein draw karte hain.

WHY ek fixed $r$ , sirf $L$ hi nahi? Kyunki "ratio approaches $L$ " allow karta hai ki yeh temporarily $L$ se thoda upar poke kare. Lekin yeh us rail ke upar kabhi poke nahi kar sakta jo strictly upar baithe. Woh strict headroom hi hai jo agla inequality permanent banata hai — humein ek fixed wall chahiye, aur $L < 1$ guarantee karta hai ki ek exist karti hai (gap mein koi bhi $r$ kaam karta hai, jaise midpoint $r = \frac{L + 1}{2}$ ).

WHAT IT LOOKS LIKE. Shrink-ratios ki settling curve rail $r$ ke neeche dip karti hai aur kisi point $n = N$ ke baad hamesha wahan rehti hai.

Step 5 — Rails ko chain karke ek geometric fence banao

WHAT. Bar $N$ se shuru karke, hum "next $<$ $r \times$ current" ko repeatedly cash in karte hain.

WHY. Ek application deta hai $∣ a_{N + 1} ∣ < r ∣ a_{N} ∣$ . Lekin hum chahte hain $k$ steps baad ke bar ka formula, kyunki poori tail sum karne ke liye $k$ mein ek pattern chahiye, sirf ek step nahi. Inequality ko $k$ baar repeat karna exactly woh pattern manufacture karta hai: ek $r^{k}$ .

WHAT IT LOOKS LIKE. Actual bars (violet) geometric fence-posts (orange) ke neeche hain jinki heights $∣ a_{N} ∣, r ∣ a_{N} ∣, r^{2} ∣ a_{N} ∣, \dots$ hain — fence geometrically droops aur bars kabhi use touch nahi karte.

Step 6 — Fence sum karo: ek finite ceiling

WHAT. Hum poori geometric fence add karte hain aur ek closed number paate hain, phir note karte hain ki humara tower chhota hai.

WHY yeh comparison? Yeh Comparison Test mechanism hai: positive bars ka ek stack, jiska har bar ek convergent stack ke matching bar se chhota hai, khud bhi converge karna chahiye. Geometric humaara "reference tower" hai kyunki woh ek hai jiski infinite height hum exactly likh sakte hain.

WHAT IT LOOKS LIKE. Dono running totals chadhte hain; orange geometric total ceiling $\frac{∣ a _{N} ∣}{1 - r}$ par flatten hota hai, aur violet real total uske neeche rehta hai — toh woh bhi flatten hota hai.

Step 7 — $L > 1$ : tower badhta hai, term test use karo

WHAT. Hum $1 < r < L$ choose karte hain aur observe karte hain ki bade $n$ ke liye $∣ a_{n + 1} ∣ > r ∣ a_{n} ∣ > ∣ a_{n} ∣$ hai.

WHY term test? Ek iron law hai: agar ek series converge karti hai, toh uske terms $0$ tak girne chahiye. (Agar bars vanish na hote, toh aap forever non-vanishing amounts add karte rehte — infinite total.) Yahan bars grow karte hain, toh $∣ a_{n} ∣ \neq \to 0$ , toh yeh necessary condition fail hoti hai, toh series nahi converge kar sakti — yeh diverge karti hai.

WHAT IT LOOKS LIKE. Bars upar chadhte hain, shrink-ratio curve level $1$ ke upar float karti hai, aur running total bina kisi ceiling ke shoot off kar jaata hai.

$L = \infty$ ka case usi story ka hissa hai, aur bhi zyada violently: ratio har rail se aage nikal jaata hai.

Step 8 — $L = 1$ : fence collapse ho jaati hai (asli limitation)

WHAT. Hum do towers dikhate hain, dono ka shrink-ratio limit exactly $1$ hai, jo alag end karte hain.

WHY dono dikhao? Yeh prove karne ke liye ki $L = 1$ "secretly diverges" ya "secretly converges" nahi hai — yeh truly ambiguous hai. Ratio test power-law terms $1/ n^{p}$ ke liye blind hai kyunki unka ratio $(\frac{n}{n + 1})^{p} \to 1$ hamesha hota hai.

WHAT IT LOOKS LIKE. Do shrink-ratio curves dono neeche se level $1$ par flatten hoti hain — phir bhi ek tower ceiling tak pahunchta hai ( $\sum 1/ n^{2}$ ) aur doosra kabhi nahi pahunchta ( $\sum 1/ n$ ).

Ek-picture summary

Sab kuch ek canvas par: possible $L$ values ki number line, woh fence jo aap har region mein bana sakte hain (ya nahi bana sakte), aur verdict.

$L < 1$ — ek rail $r$ $L$ ke upar gap mein fit hoti hai; bars $r^{k} ∣ a_{N} ∣$ ke neeche droop karte hain; converges.
$L = 1$ — koi rail kisi bhi side fit nahi hoti; inconclusive.
$L > 1$ (incl. $\infty$ ) — bars badhte hain, terms $0$ miss karte hain; diverges.

Recall Feynman: poora walkthrough plain words mein

Socho blocks stack kar rahe ho. Pehle main har block ke liye maap karta hoon ki pichle block ki height ka woh kya fraction hai — wahi shrink-ratio hai. Main us fraction ko pile mein bahut door tak follow karta hoon; jis number par yeh settle hota hai woh $L$ hai. Agar $L$ $1$ se neeche hai, toh spare room hai, toh main $L$ aur $1$ ke beech ek rail $r$ thok deta hoon; kisi point ke baad har block pichle wale se $r$ times se kam hota hai, jo blocks ko heights $r^{k}$ times ek starting block ki geometric fence ke neeche force kar deta hai. Us geometric fence ka ek known finite total hai, $\frac{∣ a _{N} ∣}{1 - r}$ , aur kyunki mere real blocks chhote hain, mere tower ki ek ceiling hai — yeh converge karta hai. Agar $L$ $1$ se upar hai, toh blocks shrink hone ki jagah grow karte hain, toh woh kabhi zero tak nahi girte, aur non-vanishing blocks ki tower infinitely tall hai — yeh diverge karta hai. Agar $L$ exactly $1$ hai, toh koi rail kisi bhi side fit nahi hoti: maine do real towers dikhaye hain, $\sum 1/ n$ aur $\sum 1/ n^{2}$ , dono $L = 1$ ke saath lekin ek infinite aur ek finite — toh ratio akele inhe honestly alag nahi kar sakta. $1$ se kam: kaam ho gaya. $1$ se zyada: gaya kaam. $1$ ke barabar: koi faayda nahi.

Connections

Geometric Series — woh reference tower jiski sum hum borrow karte hain.
Comparison Test — woh rule jo ek chhote tower ko "finite" inherit karne deta hai.
Term Test (nth-term divergence) — woh jo $L > 1$ case ko condemn karta hai.
Root Test — sibling test jo ratios ki jagah $n$ -th roots use karta hai.
Radius of Convergence — jahan yeh poori machine power series ke liye aim karti hai.
p-Series and Integral Test, Raabe's Test — $L = 1$ blind spot ke liye sharper rulers.