Determinants — cofactor expansion along any row - column: Visual walkthrough

4.5.21 · D2 · Maths › Linear Algebra (Full) › Determinants — cofactor expansion along any row - column

Hum yeh assume nahi karenge ki tum jaante ho determinant "kya hota hai". Hum numbers ki ek grid dekhkar ek seedha sawal poochhte hain.

Step 1 — Yeh number jo hum dhoondh rahe hain, hai kya?

KYA. Ek square block of numbers lo — jitni rows utni hi columns. Isse hum matrix kehte hain. Ek $2 \times 2$ matrix mein 2 rows aur 2 columns hote hain:

A = (a c b d) .

Yahan $a, b, c, d$ sirf numbers hain, aur rows left-to-right jaati hain, columns top-to-bottom.

KYUN. Ek matrix ek aise machine ki tarah hai jo flat space ko stretch aur shear karti hai: yeh chhote unit square (jiske corners $(0, 0), (1, 0), (0, 1), (1, 1)$ par hain) ko ek tedhe parallelogram mein bhej deti hai. Determinant ek single number hai jo us parallelogram ka signed area measure karta hai — machine kitna area badhati ya ghatati hai. Agar parallelogram ek line mein squash ho jaye, uska area $0$ hai, aur $det A = 0$ .

PICTURE. Blue unit square orange parallelogram ban jaata hai. Uska area hi determinant hai.

Figure — Determinants — cofactor expansion along any row - column

Jo bhi bada hoga usse is $2 \times 2$ base case pe wapas laya jayega. Yahi poora plan hai.

Step 2 — Sacchi definition: har row se ek aur har column se ek pick karo

KYA. $2 \times 2$ se aage jaane ke liye hume determinant ki seedhi, scratch-se definition chahiye — Leibniz formula. Yeh kehta hai: har possible tarika banao jisme har row se ek entry chuno aisa ki koi do chosen entries ek column share na karein, chosen entries ko multiply karo, ek sign lagao, aur aaise saare products ko jodo.

KYUN. Parallelogram-area idea (Step 1) meaning hai, lekin kisi bhi size ke liye compute karne ke liye hume raw numbers mein formula chahiye. "Ek per row, ek per column" exactly wahi algebra hai jo area-under-a-linear-map ka hai — koi entry do baar use nahi hoti, koi line do baar count nahi hoti.

PICTURE. Ek $3 \times 3$ grid ke liye $3! = 6$ legal choices hain. Har ek teen cells ka set hai, ek per row, ek per column — jaise chessboard par teen non-attacking rooks. Teen patterns par $+$ aata hai, teen par $-$ .

Step 1 se check karo: $2 \times 2$ ke liye exactly do permutations hain. "Row 1→col 1, row 2→col 2" deta hai $+ a d$ . "Row 1→col 2, row 2→col 1" ek swap hai, toh $sgn = - 1$ , deta hai $- b c$ . Total $a d - b c$ . ✓ Base case hi Leibniz hai.

Step 3 — Minus sign sach mein kahan se aata hai?

KYA. Humne abhi claim kiya ki anti-diagonal choice ko $-$ milta hai. Chalte hain dekhte hain kyun, taaki aage ka checkerboard obvious lage.

KYUN. Sign parity (even/odd swap count) of the permutation hai. Ek swap = ek $- 1$ . Do swaps $+ 1$ par cancel ho jaate hain. Yahi ek mechanism hai jo har determinant mein har sign produce karta hai — koi exception nahi.

PICTURE. Dono rooks ko main diagonal par shuru karo (identity, sign $+$ ). Unhe anti-diagonal par slide karo: yeh do columns ka ek swap hai. Ek swap → parity flip hoti hai → sign $-$ ho jaata hai.

Step 4 — Sum ko sort karo "row 1 kahan jaata hai?" ke hisaab se

KYA. Ab hum ek $n \times n$ matrix ke bade Leibniz sum ko buckets mein kaatte hain, is hisaab se ki row-1 entry kis column mein baith ti hai. Us column ko $j = σ (1)$ kehte hain.

det A = j = 1 \sum n a_{1 j} σ \in S_{n} σ (1) = j \sum sgn (σ) k = 2 \prod n a_{k, σ (k)} .

Symbols padhte hain:

$\sum_{j = 1}^{n}$ :: outer sum — row 1 ke har possible landing column $j$ ke liye ek bucket.
$a_{1 j}$ :: row-1 entry, aage pull out ki gayi (woh us bucket ke har term mein common hai).
inner $\sum_{σ (1) = j}$ :: saare tarike jisse rows $2, \dots, n$ ab bhi legally fill ho sakti hain, jab row 1 column $j$ par lock ho.
$\prod_{k = 2}^{n} a_{k, σ (k)}$ :: un remaining rows ki entries ka product.

KYUN. Koi term create ya destroy nahi hoti — humne sirf $n!$ products ko $n$ labelled folders mein daal diya. Har original term exactly ek folder mein jaati hai (joh uske $j$ se match karti hai).

PICTURE. Row 1 ka rook kahin toh jayega. Uske column ka har choice ek folder paint karta hai; baaki rooks board ke remaining hissay ko us column aur row 1 ko avoid karte hue fill karte hain.

Step 5 — Har folder ek chhota determinant hai: minor saamne aata hai

KYA. Row-1 rook ko column $j$ mein freeze karo. Ab row 1 aur column $j$ delete karo. Bache hue rooks ek $(n - 1) \times (n - 1)$ board par baithe hain, us chhote board ki ek per row aur ek per column fill karte hue — jo exactly us chhote matrix ka Leibniz sum hai. Woh chhota determinant minor $M_{1 j}$ hai.

KYUN. Jab row 1 aur column $j$ chale jaate hain, inner sum bade matrix ke baare mein rehna band kar deta hai aur ek genuine chhota determinant ban jaata hai — woh recursion jo hume chahiye tha. Aise hi ek $n \times n$ problem $(n - 1) \times (n - 1)$ problems mein shrink hoti hai.

PICTURE. Row 1 aur column $j$ ko grey karo; neeche-daayein survive karta $2 \times 2$ block hi minor hai.

Lekin sign mein ek leftover twist hai, aur wahi twist checkerboard hai. Step 6.

Step 6 — Sign ko untwist karne se checkerboard $(- 1)^{1 + j}$ milta hai

KYA. Ek folder ke andar, full permutation $σ$ ka sign sirf chhote board par leftover permutation ka sign nahi hota. Chhote board ko naturally padhne ke liye (uske columns order mein $1, 2, \dots$ ), pehle column $j$ ko raste se slide karna padta hai, use apne left waale $j - 1$ columns se hopaate hue. Har hop ek swap hai → ek factor of $- 1$ . Row 1 ki position ke saath milke yeh $(- 1)^{1 + j}$ ban jaata hai.

ek folder σ (1) = j \sum sgn (σ) k \geq 2 \prod a_{k, σ (k)} = (- 1)^{1 + j} M_{1 j} = C_{1 j} .

$(- 1)^{1 + j}$ :: entry $(1, j)$ ko top-left corner mein laane ke liye swap count.
$M_{1 j}$ :: Step 5 se chhota determinant.
$C_{1 j}$ :: inke product — cofactor, yaani signed minor.

KYUN. Ek column ko $k$ doosre columns ke past slide karna $k$ adjacent swaps hain, aur har swap sign flip karta hai (Step 3). Row 1 pehle se top par hai, jo exponent mein " $1$ " contribute karta hai. Yahi checkerboard $+, -, +, \dots$ ki poori origin hai.

PICTURE. Rook $(1, j)$ mein ek-ek column-swap karte hue left chalta hai, har hop $\times (- 1)$ stamp karta hai, jab tak corner nahi pahunch jaata. Hops count karo = $j - 1$ ; row 1 jodo = exponent $1 + j$ .

Step 4 mein folder result wapas substitute karke:

Step 7 — KISI BHI row aur KISI BHI column se kyun kaam chalta hai

KYA. Humne sirf row 1 ke saath expand kiya. Row 3, ya column 2, same number guarantee kyun karta hai?

KYUN (do moves).

Koi bhi row. Leibniz sum saari rows ko symmetrically treat karta hai — row 1 ko special kuch nahi tha, sirf hamara choice tha ki pehle uske hisaab se bucket karein. Iske bajaye row $i$ se bucket karo aur identical argument deta hai $det A = \sum_{j} a_{ij} C_{ij}$ . Row $i$ ko top par slide karne ke extra swaps $i - 1$ hops add karte hain, exactly isliye exponent $1 + j$ nahi, $i + j$ hota hai.
Koi bhi column. Transpose karna (rows↔columns flip karna) Leibniz sum ko untouched chhod deta hai, isliye $det A = det A^{T}$ . $A$ ka column expansion $A^{T}$ ka row expansion hai — jo pehle se prove ho chuka hai. Isliye har column bhi kaam karta hai.

PICTURE. Same board, teen alag tarike se expand kiya (row 1, row 3, column 2), saari ek identical single number par funnel karti hain.

Step 8 — Degenerate aur edge cases (kuch chhuta nahi)

KYA & KYUN. Derivation tab poori hoti hai jab weird inputs bhi sahi behave karein.

Zero entry $a_{ij} = 0$ . Uska poora term $a_{ij} C_{ij} = 0$ drop ho jaata hai. Yahi 80/20 rule hai: sabse zyada zeros wali line ke saath expand karo taaki compute karne se pehle terms kill ho jaayein.
Zero row (ya zero column). Us line par har $a_{ij}$ zero hai, toh uske saath expand karne se $det A = 0$ milta hai. Geometrically parallelogram flat squash ho jaata hai → area $0$ → matrix singular hai.
$1 \times 1$ matrix. $det (a) = a$ . Recursion yahan bottom out hoti hai; $2 \times 2$ formula asal mein do $1 \times 1$ minors hai.
Do equal rows. Unhe swap karna kuch nahi badalta, phir bhi swap sign flip karna chahiye (Step 3): $det A = - det A \Rightarrow det A = 0$ . Collapse picture se consistent hai.

PICTURE. Left: ek zero column parallelogram ko ek segment mein flatten kar deta hai (area 0). Right: do equal rows force karte hain $det = - det = 0$ .

Ek-picture summary

Is page ki har cheez ek flow mein aati hai: bada determinant → row 1 se sort karo → freeze karo → delete karo → sign-hop → chhote determinants → recurse.

Recall Feynman retelling (ek 12-saal ke bachche ko bolo)

Determinant ek number hai jo batata hai ki ek grid ek shape ko kitna stretch karta hai. Ise find karne ke liye, grid par "rooks" rakho taaki koi do ek row ya column share na karein — har legal placement numbers ka ek product deti hai, aur har ek ko $+$ ya $-$ milta hai is hisaab se ki tumne columns ko tidy karne ke liye kitni baar swap kiya (even swaps $= +$ , odd $= -$ ). Un saare signed products ko add karo: wahi determinant hai. Clever shortcut: fix karo ki top row ka rook kahan jaata hai, phir apne haath se us row aur column ko dhak do — jo bachta hai woh ek chhota grid hai jise tum usi tarah solve karte ho. Ek column dhakna jo, maano, 2 steps baayein se hai, 2 swaps leta hai, isi se checkerboard $+, -, +$ aata hai. Aur kyunki rooks ko farak nahi padta ki tumne kis row ya column se sort karna shuru kiya, tum kisi bhi line ke saath sweep kar sakte ho — sabse zyada zeros wali choose karo, kam kaam karo, har baar same number pao.

Connections

Leibniz formula for determinants — woh ground truth jise humne expand kiya (Steps 2–3).
Properties of determinants (multilinearity, alternation) — swap-flips-sign fact (Step 3) hi alternation hai.
Determinant via row reduction — woh fast cousin jo pehle zeros banata hai.
Adjugate matrix and inverse — har cofactor ko ek matrix mein package karta hai.
Cramer's rule — $A x = b$ solve karta hai inhi cofactors se.
Invertibility and singular matrices — Step 8 ka $det A = 0$ collapse.