Optimization — constrained, unconstrained, real-world problems: Visual walkthrough

4.1.31 · D2 · Maths › Calculus I — Limits & Derivatives › Optimization — constrained, unconstrained, real-world proble

Yeh poore optimization recipe ka visual engine hai. Key ideas Critical Points & Fermat's Theorem, Second Derivative & Concavity, aur Taylor Series Expansion se connect hote hain.

Step 1 — "Slope" ka matlab kya hai (ek pahaadi ki tilt)

KYA. Ek curved path draw karein — socho ek roller-coaster track jo graph ke roop mein bana ho. Horizontal axis hai aap kitna aage gaye (isko $x$ bolte hain). Vertical axis hai aap kitna upar hain (isko $f (x)$ bolte hain, padha jaata hai " $f$ of $x$ " — position $x$ par height).

KYUN. Isse pehle ki hum kahein "slope zero hai", humein agree karna hoga ki slope hai kya. Slope = track ek jagah par kitna steeply tilt karta hai. Flat road ka slope $0$ hota hai; diwar ka slope bahut bada hota hai.

PICTURE. Figure mein, curve par ek point chunein aur ek straight ruler wahan rakho jo curve ko sirf wahan touch kare — woh ruler wahan ka tangent line hai. Uski tilt hi us point par slope hai. Hum tilt ko rise over run ke roop mein measure karte hain:

$slope = \frac{rise}{run} = \frac{change in height}{change in position} .$

$rise$ — aap kitna upar gaye (lal vertical segment).
$run$ — aap kitna aage gaye (neela horizontal segment).
unka ratio — steepness. Thode aage jaane par zyada upar = steep.

Figure — Optimization — constrained, unconstrained, real-world problems

Step 2 — Ek interior peak par slope kyun vanish ho jaata hai

KYA. Ek hilltop ke upar left-to-right chalein jo strictly inside interval mein hai (edge par nahi) aur tangent ruler ko dekho.

KYUN. Top se thoda pehle aap abhi bhi chadhte hain, toh tangent upar tilt karta hai: $f^{'} > 0$ . Top ke thoda baad aap utarte hain, toh woh neeche tilt karta hai: $f^{'} < 0$ . Kyunki humne assume kiya hai ki $f^{'}$ continuous hai (standing assumption dekhein), isme intermediate-value property hai: yeh seedha positive value se negative par nahi ja sakta — iske beech mein $0$ se guzarna zaroori hai. Woh crossing point hi peak hai.

PICTURE. Figure mein teen rulers hain: green (upar, peak ke baayein), yellow (bilkul flat, peak par), red (neeche, peak ke daayein).

$chadhna f^{'} > 0 ⟶ summit f^{'} = 0 ⟶ utarna f^{'} < 0$

left term — positive slope, upar ja raha hai.
middle term — zero tilt ka exact instant, flat ruler.
right term — negative slope, neeche ja raha hai.

Step 3 — Trap: flat ka matlab HAMESHA top ya bottom nahi hota

KYA. $f (x) = x^{3}$ ko $x = 0$ par dekho.

KYUN. Hum blindly nahi keh sakte "flat = extremum". $x^{3}$ ka slope hai $f^{'} (x) = 3 x^{2}$ , jo $x = 0$ par $0$ hai — sach mein flat. Lekin $f^{'} = 3 x^{2} \geq 0$ dono taraf hai, toh curve poore time chadh raha hota hai; sirf ruk jaata hai thodi der ke liye.

PICTURE. Origin par tangent flat hai (yellow), phir bhi dono taraf green arrows ek hi direction mein point karte hain (upar-left neeche hai, upar-right upar hai). Koi summit nahi, koi dip nahi.

Yahan $x^{3}$ genuinely ek inflection point hota hai — lekin sirf isliye nahi ki $f^{''} (0) = 0$ hai. Jo cheez ise inflection banati hai woh yeh hai ki concavity sign change karti hai: $f^{''} (x) = 6 x$ $x < 0$ ke liye negative (cap) aur $x > 0$ ke liye positive (cup) hai. $f^{''}$ ka sign-change dono taraf hi real definition hai; $f^{''} (c) = 0$ akela sirf ek hint hai. Toh $f^{'} (c) = 0$ sirf $c$ ko ek candidate banata hai; hume phir bhi classify karna hoga.

Step 4 — Curvature kaisi dikhti hai (jo missing information hai)

KYA. Flat tangent batata hai ki curve ruk raha hai; yeh nahi batata ki curve kis taraf bend karta hai. Hume ek doosra, independent number chahiye: slope khud kitna change ho raha hai.

KYUN. Tilt ko right mein jaate dekho. Valley ke paas tilt negative se positive hoti hai — slope badh raha hai. Peak ke paas tilt positive se negative hoti hai — slope ghad raha hai. "Slope kitni tezi se change ho raha hai" exactly derivative of the derivative hai.

PICTURE. Figure mein do bowls hain. Green cup (paani rok sakta hai) mein slope left-to-right badhta hai. Red cap (paani gira dega) mein slope girta hai. Is bending ko concavity kehte hain.

Step 5 — Taylor ke ruler se classification test build karna

KYA. Hum critical point $c$ mein zoom karte hain aur aas-paas ki height ko ek simple polynomial se approximate karte hain. Yeh Taylor Series Expansion hai — yeh ek messy curve ko ek point ke paas best-fitting parabola se replace karta hai.

KYUN yeh tool aur koi nahi? Hum tiny step $h$ ke liye change $f (c + h) - f (c)$ ka sign jaanna chahte hain. Tangent line (first order) yahan kisi kaam ki nahi hai — critical point par line flat hoti hai, kuch nahi batati. Parabola (second order) sabse simple shape hai jo curvature information carry karta hai, aur curvature exactly wahi cheez hai jo cup ko cap se alag karti hai. Toh hum $h^{2}$ tak ke terms rakhte hain:

$f (c + h) = f (c) + f^{'} (c) h + \frac{1}{2} f^{''} (c) h^{2} + \dots$

$f (c + h)$ — $c$ se small step $h$ door ki height.
$f (c)$ — $c$ par height (hamara reference level).
$f^{'} (c) h$ — tangent-line correction. Critical point par yeh $0$ hai kyunki $f^{'} (c) = 0$ .
$\frac{1}{2} f^{''} (c) h^{2}$ — parabola correction, pehla surviving term.
$\dots$ — aur bhi chhote terms, tiny $h$ ke liye negligible.

PICTURE. Figure mein true curve (blue) aur uska fitted parabola (yellow) $c$ par ek doosre se chipke hue hain; $c$ ke paas woh indistinguishable hain, toh parabola ki shape hi curve ki shape hai.

Step 6 — Sign padhna: test khud nikalta hai

KYA. Step 5 ke formula mein $f^{'} (c) = 0$ set karein aur dekho kya bachta hai:

$f (c + h) - f (c) \approx \frac{1}{2} f^{''} (c) h^{2} .$

KYUN. Yahan khoobsoorat baat hai. Chahe aap left ( $h < 0$ ) ya right ( $h > 0$ ) step karein, square karne se $h^{2} > 0$ ho jaata hai. Aur $\frac{1}{2} > 0$ hamesha. Toh height change ka poora sign sirf $f^{''} (c)$ se decide hota hai — ek number dono taraf ka verdict deta hai.

$h^{2}$ — kisi bhi nonzero step ke liye hamesha positive, left ya right.
$\frac{1}{2}$ — positive constant, kuch nahi badlata.
$f^{''} (c)$ — akela factor jo $+$ ya $-$ ho sakta hai; yahi verdict hai.

PICTURE. Agar $f^{''} (c) > 0$ : dono taraf $c$ se upar hain → valley (local min). Agar $f^{''} (c) < 0$ : dono taraf $c$ se neeche hain → peak (local max). Figure mein parabola upar khulta hai (green, min) aur neeche (red, max), do step arrows $\pm h$ ke saath jo higher / lower land karte hain.

Step 7 — Degenerate cases: $f^{''} (c) = 0$ , aur poore flat plateaus

KYA. Do degeneracies Steps 5–6 se bachke nikal jaati hain. (a) $f^{'} (c) = 0$ aur $f^{''} (c) = 0$ dono: ruling term zero hai aur parabola flat hai, toh koi verdict nahi milta. (b) ek plateau: $f^{'} \equiv 0$ poore interval par, toh critical points isolated nahi hain — flat stretch ka har point critical hai.

KYUN. Case (a) mein hum flat parabola par aa gaye; case (b) mein test karne ke liye koi single "point" nahi hai. Dono mein, reliable move hai seedha har taraf $f^{'}$ ka sign dekhna.

PICTURE — first-derivative sign test. $c$ ke across ek test point chalao aur $f^{'}$ ka sign record karo:

$f^{'}$ jaata hai $+ \to -$ : slope pehle badhta hai phir girta hai → peak (local max).
$f^{'}$ jaata hai $- \to +$ : slope pehle girta hai phir badhta hai → valley (local min).
$f^{'}$ same sign rakhta hai (jaise $x^{3}$ ke liye $+ \to +$ ): koi extremum nahi — ek flat inflection (concavity flip hoti hai, lekin height ek hi direction mein chalti rehti hai).

Do named degeneracies, resolved:

$f (x) = x^{3}$ : $f^{'} = 3 x^{2} \geq 0$ dono taraf → same sign → inflection, extremum nahi. (Concavity flipping se confirm: $f^{''} = 6 x$ sign change karta hai.)
$f (x) = x^{4}$ : $f^{'} = 4 x^{3}$ , $x < 0$ ke liye $-$ aur $x > 0$ ke liye $+$ → sign jaata hai $- \to +$ → genuine minimum, chahe $f^{''} (0) = 0$ aur concavity flip nahi karti. Yeh figure ka fourth panel hai.

Plateaus. Agar $f^{'} \equiv 0$ poore interval mein, $f$ wahan constant hai. Har plateau point simultaneously ek (weak) local max aur min hota hai. Yeh decide karne ke liye ki plateau summit hai ya shelf, sign test iske dono ends par lagao: left edge ke thoda baayein aur right edge ke thoda daayein $f^{'}$ ka sign check karo — same logic, point ki jagah block par applied.

Ek-picture summary

Ek graph par har idea. Ek single wavy curve carry karti hai: ek valley (green, $f^{'} = 0, f^{''} > 0$ ), ek peak (red, $f^{'} = 0, f^{''} < 0$ ), aur ek flat inflection (yellow, $f^{'} = 0$ , concavity flips). Har ek ke neeche tiny tangent ruler flat drawn hai, aur local best-fit parabola verdict dikhata hai. Margin note yaad dilaata hai ki endpoints aur plateaus ke liye alag sign/boundary check chahiye. Yeh parent recipe ka poora decision procedure ek frame mein hai: dhundho jahan ruler flat ho jaaye, phir dekho bowl kis taraf face karta hai — aur edges kabhi mat bhoolo.

Recall Feynman retelling — plain words mein poora walkthrough

Ek roller-coaster socho. Slope bas yahi hai ki track aapke paon ke neeche kitna tilted feel karta hai — uphill positive, downhill negative, flat zero. Track ke bilkul beech mein pahadi ki top par pahunchne ke liye aapko upar jaana band karna hoga pehle neeche jaana shuru karne se, aur kyunki tilt smoothly change hoti hai (kabhi teleport nahi karta), ek instant ke liye track bilkul dead flat hota hai: yahi hai "derivative zero hai." Lekin dhyan raho — agar top track ke bilkul end par hai, toh aap tab bhi uphill tilt kar rahe ho jab track khatam ho jaaye, toh ends ko apne aap check karna hoga. Aur flatness akele ek dhoka hai: track flat ho ke chadh sakta rehta hai (woh hai $x^{3}$ , ek shelf, summit nahi). Toh aapko ek doosra clue chahiye: track kis taraf curve karta hai — cup matlab aap dip mein ho, dome matlab top par. Math trick yeh hai ki apne spot ke paas track ko ek simple parabola maan lo (Taylor ka idea): ek baar flat part cancel ho jaata hai, sirf $\frac{1}{2} f^{''} h^{2}$ bachta hai, aur $h^{2}$ hamesha positive hai, toh curvature ka sign akele cup ya dome batata hai. Agar curvature bhi zero hai, andaza lagana bandh karo aur bas dekho ki slope across chalte waqt sign flip karta hai ya nahi: plus-then-plus inflection hai ( $x^{3}$ ), minus-then-plus valley hai ( $x^{4}$ ). Aur agar track poore stretch par flat ho — ek plateau — toh bas uske dono ends par tilt check karo.

Recall Quick self-check

Smooth interior extremum par $f^{'} (c) = 0$ kyun hona zaroori hai? ::: Kyunki $f^{'}$ continuous hai, woh ek sign se doosre par slide karti hai aur zero se guzarna zaroori hai (intermediate-value property). Kahan extremum mein $f^{'} \neq = 0$ ho sakta hai? ::: Interval ke endpoint / boundary par — zero-slope test sirf interior points ke liye hai. Ek hi kaun sa number cup vs cap decide karta hai jab $f^{'} (c) = 0$ ? ::: $f^{''} (c)$ ka sign, kyunki $\frac{1}{2} h^{2} > 0$ hamesha hota hai. Kya $f^{''} (c) = 0$ akele inflection ka matlab hai? ::: Nahi — inflection ke liye $f^{''}$ ko $c$ ke across actually sign change karna hoga; $x^{4}$ mein $f^{''} (0) = 0$ hai lekin genuine minimum hai. Jab dono $f^{'} (c) = 0$ aur $f^{''} (c) = 0$ hon to kya karte hain? ::: Dono taraf first-derivative sign test use karo. Poore plateau ( $f^{'} \equiv 0$ on an interval) ko kaise classify karte hain? ::: $f^{'}$ ka sign uske dono ends ke thoda bahar check karo. $0$ par $x^{3}$ kya hai? ::: Ek flat inflection — critical hai, concavity flip hoti hai, lekin max ya min nahi hai.