Constraints using Lagrange multipliers: Visual walkthrough

2.1.10 · D2 · Physics › Analytical Mechanics › Constraints using Lagrange multipliers

Shuru karne se pehle, ek honest promise: neeche sab kuch geometry hai. Poori derivation arrows aur un flat sheets ki kahaani hai jinpe woh slide kar sakte hain. Yeh image apne dimag mein rakho.

Step 1 — Ek configuration ek made-up space mein ek single point hai

KYA: hum "ek chalti system" ko "coordinate space mein ek chalte dot" se replace karte hain.

KYU: kyunki jab sab kuch ek space mein dot ho jaata hai, toh constraints shapes ban jaate hain aur forces arrows ban jaate hain. Isse physics, geometry mein badal jaati hai — jise hum draw kar sakte hain.

PICTURE: figure do axes $q_{1}, q_{2}$ draw karta hai (hum $n = 2$ use karte hain taaki paper pe fit ho) aur ek orange dot — poori machine, abhi is waqt.

Step 2 — Ek constraint ek curve hai jis par dot rehna chahiye

KYA: humne equation $f = 0$ ko plane ke through ek green curve ke roop mein draw kiya.

KYU: is method ka poora point yahi hai ki constraint ko visible rakha jaaye, use clever coordinates ke andar chhupaane ke bajaye. Toh hume literally woh surface dekhni chahiye jis par dot trapped hai.

PICTURE: green curve = ${f = 0}$ . Orange dot us par baitha hai. Curve se bahar jaana forbidden territory hai (shaded gray). Dimag mein rakho: radius $R$ ka hoop, toh $f = r - R = 0$ — curve khud circle hai.

Step 3 — Gradient: surface ke seedha bahar jaata ek arrow

KYA: dot par, humne red arrow $\nabla f = (\frac{\partial f}{\partial q _{1}}, \dots, \frac{\partial f}{\partial q _{n}})$ draw kiya.

PERPENDICULAR KYU? Surface ke saath ek tiny step $δ q$ lo — $f$ nahi badlega, toh us direction mein iska slope zero hai. Nonzero slope wali ek hi direction hai woh hai surface ke across. Gradient slopes collect karta hai, toh woh us across-direction ke saath align ho jaata hai. Yahi poora reason hai ki normal force sirf ek hi direction mein point kar sakti hai.

PICTURE: green curve, red gradient arrow perpendicularly bahar shoot karta hua, aur ek faint blue arrow curve ke saath flat leta hua (allowed direction — Step 4).

Recall "

$\partial$ " kyun, " $d$ " kyun nahi? Kyunki $f$ ek saath kai $q$ 's par depend karta hai. $\partial f / \partial q_{k}$ ka matlab hai "sirf $q_{k}$ ko wiggle karo, baaki ko freeze karo." Yeh single-variable wiggle exactly wahi hai jo " $q_{k}$ direction mein slope" chahiye. ::: partial derivative = ek axis ke saath slope jab baaki held still hain.

Step 4 — Allowed wiggles: virtual displacements jo surface ke tangent hain

KYA: humne $δ q$ ko green curve par flat lete hue ek chhote blue arrow ke roop mein draw kiya.

TIME FREEZE MOVE KYU? Real motion surface ko idhar-udhar khich sakti hai; hum test karna chahte hain "abhi is waqt available moves mein se, kaun si legal hain?" $t$ ko freeze karna geometry ko clock se alag karta hai.

PICTURE: blue $δ q$ curve ke tangent; red $\nabla f$ perpendicular. Woh $9 0^{\circ}$ par milte hain — woh right angle agle step ka key fact hai.

Step 5 — d'Alembert hume ek doosra perpendicular arrow deta hai

KYA: hum poore bracket ko $E_{k}$ naam dete hain — equations of motion ka "leftover". Unhe ek arrow $E = (E_{1}, \dots, E_{n})$ mein stack karo.

YEH ZERO KYU HOTA HAI? Ek free system ke liye har $δ q_{k}$ independent hota, toh har $E_{k}$ khud hi vanish karna padta — woh ordinary Euler–Lagrange equation hai. Lekin ek constraint ke neeche $δ q_{k}$ chained hote hain (Step 4), toh hum sirf yeh conclude kar sakte hain ki poora sum zero hai.

YEH KAISA DIKHTA HAI: sum ka zero hona $E \cdot δ q = 0$ hai — arrow $E$ bhi blue allowed direction ke perpendicular hai.

Step 6 — Same line ke ⟂ do arrows parallel hone chahiye — aur $λ$ paida hota hai

KYA: hum $E = λ \nabla f$ set karte hain, component by component.

YEH FORCED KYU HAI (CHOOSE NAHI KIYA): parallelism yahan ek theorem hai, assumption nahi. Ek hi freedom bachi hai woh hai length ratio, aur woh single unknown number Lagrange multiplier hai.

PICTURE: red $\nabla f$ aur thoda lamba red $E$ same line par draw kiye gaye; ek bracket unke length ratio $λ$ ko label karta hai.

Step 7 — Counting: $n$ equations, $n + 1$ unknowns — constraint rent pay karta hai

KYA: hum green curve ki apni equation wapas pile mein add karte hain.

KYU: har unknown ko ek equation chahiye. Extra unknown $λ$ jo humne banaya woh exactly us extra equation $f = 0$ se balance hota hai jo hum sab along carry kar rahe the. Balanced ledger.

PICTURE: ek chhota tally — $n$ EL rows blue mein, $1$ constraint row green mein, $n$ coordinate unknowns + $1$ multiplier unknown — arrows unhe match up karte hue.

Step 8 — Edge & degenerate cases (reader ko kabhi stranded mat chhodna)

Case B — gradient vanish ho jaata hai ( $\nabla f = 0$ ). Tab "off-surface" direction undefined hai — constraint us point par degenerate hai (jaise $f = r^{2}$ at $r = 0$ ). Method ke paas kuch point karne ko nahi, toh $λ \nabla f = 0$ aur wahaan koi force recover nahi ki ja sakti. Fix: $f$ ko rewrite karo taaki uska gradient nonzero ho ( $f = r - R$ use karo, $f = (r - R)^{2}$ nahi).

Case C — tumne already coordinate eliminate kar liya. Agar tumne constraint ko independent coordinates mein embed kar liya, toh $\partial f / \partial q = 0$ (koi redundant $q$ nahi bachi differentiate karne ke liye), $λ$ zero se multiply ho jaata hai aur meaningless hai. Fix: redundant coordinate rakho jab bhi constraint force chahiye.

Case D — velocity-only (non-holonomic) constraints. Agar constraint $\sum_{k} a_{k} \overset{q}{˙}_{k} = 0$ hai aur kisi bhi $f (q) = 0$ mein integrate nahi ho sakti, toh koi surface nahi aur koi gradient nahi. Tum velocity form use karte ho $E_{k} = \sum_{a} λ_{a} a_{ak},$ jahan coefficients $a_{ak}$ woh role play karte hain jo gradient ne play kiya. Dekho Holonomic vs Non-holonomic Constraints.

Recall "gradient = λ · gradient" pehle kahaan dekha hai?

Yeh constrained optimization ka exact skeleton hai: constrained extremum par, $\nabla (objective) = λ \nabla (constraint)$ . Woh equality constraints ke saath KKT Conditions hai. Mechanics aur optimization geometry share karte hain: leftover constraint ke normal ke parallel hai. ::: same parallel-gradients picture, alag words.

Ek-picture summary

Upar sab kuch yeh ek image hai: ek green surface, us ke tangent ek blue allowed-direction, aur do red arrows ( $E$ physics se, $\nabla f$ geometry se) jo same perpendicular line par lie karne ke liye forced hain — unka length ratio $λ$ hai, aur $λ \nabla f$ woh push hai jo surface deliver karta hai.

Recall Feynman: 12-saal ke bacche ke liye poora walkthrough

Ek curved rail mein marble socho. Rail ek green line hai jise marble kabhi nahi chhodh sakta. Abhi marble sirf rail ke saath slide kar sakta hai — use blue direction kaho. Do alag arrows seedhe rail ke across point karte hain: ek hai "physics leftover" (jo marble ki motion demand karti hai), doosra rail ka apna "mujhse door kaun si taraf hai" arrow. Flat 2D mein rail ke seedhe across sirf ek direction hai, toh dono arrows use par lie karne chahiye — unke sirf alag lengths hain. Woh length ratio ek chhota honest number $λ$ hai. Aur yahan payoff hai: rail marble ko sirf sideways push kar sakti hai, seedhe apne aap ke across — same red direction — toh $λ$ literally rail kitna hard push karta hai woh hai. $λ$ dekho: jab woh zero hota hai, rail pushing band kar deti hai aur marble ud jaata hai.

Visual walkthrough — Constraints using Lagrange multipliers