Finding eigenspaces: Visual walkthrough — Hinglish deep dive

4.5.30 · D2 · Maths › Linear Algebra (Full) › Finding eigenspaces

Hum poore walkthrough mein ek hi machine use karenge: $A = (2112) .$ Yahan yeh assume nahi kiya gaya ki tum jaante ho determinant ka "matlab" kya hai — hum usse visually build karenge jahan zaroori ho.

Step 1 — Matrix ek arrow ke saath kya karta hai

KYA HAI. Ek arrow (ek vector) bas ek point-with-direction hai, origin se khicha hua. Ek matrix $A$ ek machine hai: isko ek arrow $v$ do, yeh ek naya arrow $A v$ wapas deta hai.

Yahan $v$ ka matlab hai "input arrow" aur $A v$ ka matlab hai "machine kya return karti hai."

KYUN. Pehle hum special arrows dhundhein, humein typical cheez dekhni chahiye jo hoti hai: zyaadatar arrows kisi nayi direction mein point karte hue nikalte hain. Yahi turning woh "dushman" hai jisse hum bachna chahte hain.

PICTURE. Do ordinary input arrows (black) aur $A$ unhe kahan bhejta hai (red arrows clearly ek nayi direction mein rotate ho gayi hain).

Step 2 — Woh special arrow jo mudne se mana kar deta hai

KYA HAI. Kuch input directions bahar aati hain bilkul parallel jaisi andar gayi thi. Hamare $A$ ke liye, arrow $(1 1)$ aisa hi hai: $A (11) = (2 \cdot 1 + 1 \cdot 1 1 \cdot 1 + 2 \cdot 1) = (33) = 3 (11) .$

Equation ko left se right padho: machine $A$ ne $(1 1)$ khaya aur $(3 3)$ wapas diya, jo bas wohi arrow $3$ se scale kiya hua hai. Yeh scale factor $3$ hi eigenvalue hai, jise $λ$ (Greek "lambda", "stretch amount ka naam") likhte hain.

KYUN. Yahi poora phenomenon hai. Agar hum in no-turn arrows ko algebraically characterise kar sakein, toh hum kisi bhi machine ke liye inhe dhundh sakte hain.

PICTURE. Red arrow $(1 1)$ aur uski image $(3 3)$ ek hi dashed line par hain — pure stretch, zero rotation.

Step 3 — "Same direction" ko equation mein likhna

KYA HAI. "Output input ki line par rehta hai" ka matlab hai output $=$ koi number times input: $A v = λ v .$

$A v$ — machine ka output arrow.
$λ$ — woh single number jisse arrow scale hua.
$v$ — input arrow, jo nonzero hona chahiye (zero arrow ki koi direction hi nahi preserve karne ke liye, toh woh kuch nahi batata).

KYUN. Humne ek geometric wish ("rotate mat karo") ko ek algebraic equation mein badal diya jise hum manipulate kar sakte hain. Yahi linear algebra ka poora "move" hai: picture ko symbols mein badlo, solve karo, phir picture wapas padho.

PICTURE. Wohi arrow jis par equation annotate ki gayi hai — input black mein, scaled output red mein, aur " $= λ \times$ " unke beech mein.

Step 4 — Dono sides ko ek machine par collapse karo

KYA HAI. Sab kuch ek side par le aao: $A v - λ v = 0 .$ Hum $(A - λ) v$ nahi likh sakte — $A$ numbers ka grid (ek matrix) hai, $λ$ ek single number hai; inhe subtract karna waise hi hai jaise ek apple ko ek orchard se subtract karo. Theek karne ka tarika: $λ$ ko $λ I$ se replace karo, jahan $I = (1001)$ identity machine hai (woh jo har arrow unchanged wapas karta hai, toh $I v = v$ ). Ab dono terms machines hain jo $v$ par act kar rahi hain: $(A - λ I) v = 0 .$

$A - λ I$ — ek nayi single machine: $A$ lo aur sirf diagonal entries mein se $λ$ subtract karo.
$0$ — zero arrow (origin par ek point), number zero nahi.

Hamare $A$ ke liye: $A - λ I = (2 - λ 1 1 2 - λ) .$

KYUN. Ek machine $(A - λ I)$ times ek arrow zero ke barabar — yeh ek aisi shape hai jise hum solve karna jaante hain — ek homogeneous system.

PICTURE. Nayi machine ka build-up: $A$ ka grid, minus $λ$ diagonal par slide kiya hua (red), jisse $A - λ I$ milta hai.

Step 5 — Machine ko ek direction flatten karna kyun zaroori hai

KYA HAI. Humein $(A - λ I) v = 0$ chahiye kisi nonzero $v$ ke liye. Iska matlab hai: machine $M := A - λ I$ ek nonzero arrow ko leta hai aur use origin par crush kar deta hai. Ek machine jo poori ek direction ko ek point par squeeze kar de, use singular (not invertible) kehte hain — usne information kho di hai, toh tum usse undo nahi kar sakte.

KYUN. Agar $M$ singular nahi hota, toh woh reversible hota: sirf ek arrow jise woh $0$ par bhejta woh $0$ khud hota — koi eigenvector nahi. Toh hum require karte hain ki $M$ singular ho. Yeh koi trick nahi hai; yeh woh exact condition hai jab ek no-turning arrow exist karta hai.

PICTURE. Left: ek invertible machine unit square ko ek tilted parallelogram mein spread karta hai jiska real area hai — kuch crush nahi hota. Right: ek singular machine square ko red line par flat collapse kar deta hai — area zero, ek direction annihilated.

Step 6 — Determinant hai "kitna flat?" ka meter

KYA HAI. Determinant $det (M)$ woh area measure karta hai jo machine $M$ unit square se produce karti hai (signed, par sign ko abhi ignore karo). " $M$ ek direction flat crush karta hai" wahi baat hai jo "woh area $0$ hai": $det (A - λ I) = 0.$ Haare $2 \times 2$ machine ke liye: $det (2 - λ 1 1 2 - λ) = (2 - λ) (2 - λ) - (1) (1) = (2 - λ)^{2} - 1.$

$(2 - λ) (2 - λ)$ — diagonal entries ka product.
$- (1) (1)$ — minus off-diagonal entries ka product.

Yahi characteristic equation hai — ek aisi equation jiska unknown $λ$ hai.

KYUN. Humne "woh stretch amounts dhundho jo machine ko collapse karte hain" ko ek simple single-variable $λ$ equation mein convert kar diya. Isse solve karo aur eigenvalues nikal aayenge.

Zero par set karo: $(2 - λ)^{2} = 1 \Rightarrow 2 - λ = \pm 1 \Rightarrow λ = 1 or λ = 3.$

PICTURE. Ek dial: jaise $λ$ slide karta hai, area (determinant) upar-neeche hota hai; woh exactly $λ = 1$ aur $λ = 3$ par zero touch karta hai — red zero-crossings.

Step 7 — Crushed system solve karo eigenspace nikalne ke liye

KYA HAI. Har eigenvalue ke liye, poochho ki kaun se arrows crush hote hain. $λ = 3$ lo: $A - 3 I = (2 - 3 1 1 2 - 3) = (- 1 1 1 - 1) .$ Row-reduce karke $(10 - 1 0)$ milta hai, jo $v_{1} - v_{2} = 0$ padhta hai, yaani $v_{1} = v_{2}$ . Free variable $v_{2} = t$ rakho: $v = t (11) .$ $(1 1)$ ki har scaling $t$ kaam karti hai — woh poori line eigenspace hai: $E_{3} = Null (A - 3 I) = span {(11)} .$ $λ = 1$ ke liye same karo toh $v_{1} = - v_{2}$ milta hai, toh $E_{1} = span {(- 1 1)}$ .

KYUN. $A - λ I$ ka null space hi eigenspace hai — woh sare arrows collect karta hai jinhein crushed machine $0$ par bhejti hai, jo exactly us $λ$ ke no-turn arrows hain.

PICTURE. Origin se do red lines: $E_{3}$ ( $(1 1)$ direction) aur $E_{1}$ ( $(- 1 1)$ direction) — machine ke do non-turning axes.

Step 8 — Degenerate cases jinse tum kabhi surprised mat hona

KYA HAI. Recipe kabhi nahi tootti, par answer ki shape vary karti hai. Teen cheezein ho sakti hain jab ek eigenvalue repeat karta hai ( $2 \times 2$ ke liye algebraic multiplicity $2$ ):

Machine	$A - λ I$ kya crush karta hai…	Eigenspace $E_{λ}$	$dim$
$(5005)$ , $λ = 5$	kuch bacha hi nahi rakhne ko — yeh zero machine hai, poore $R^{2}$ ko $0$ mein crush karta hai	poora plane $R^{2}$	$2$
$(5015)$ , $λ = 5$	sirf ek single line	ek line, $span {(0 1)}$	$1$
$λ = 0$ case (koi bhi singular $A$ )	$A$ khud already singular hai	$E_{0} = Null (A)$	$\geq 1$

Middle case ek trap hai: eigenvalue do baar repeat hota hai phir bhi eigenspace sirf ek line hai. Repeat count (algebraic multiplicity) eigenspace dimension (geometric multiplicity) ka sirf ek upper bound hai. Dekho Diagonalization ki gap kyun matter karta hai.

KYUN. Ek reader jisne sirf clean $λ = 1, 3$ case dekha hoga woh galti se sochega "repeat count = eigenspace dimension." Teeno cases dikhana har escape route band karta hai.

PICTURE. Teen panels side by side: (left) poora plane red shade hua — full eigenspace; (middle) ek single red line — deficient eigenspace; (right) origin se ek red line — $λ = 0$ / singular case.

Ek-picture summary

Ek frame sab kuch ek saath bandhta hai: input arrows (black) machine $A$ se takraate hain; almost sab rotate karte hain; do red eigen-lines par woh sirf stretch hote hain ( $λ = 3$ aur $λ = 1$ se). Stretch factors woh $λ$ hain jahan machine flat ho jaati hai, yaani jahan $det (A - λ I) = 0$ ; har aisi $λ$ ki crushed direction eigenspace hai.

Recall Feynman retelling — poora walkthrough simple alfazon mein

Matrix ko ek aisi hawa samjho jo har arrow ko ek nayi jagah uda le jaati hai. Zyaadatar arrows ek naye angle par swing karte hain. Par kuch directions aisi hain jaise seedhi wind mein khada flagpole — woh bas bade ya chhhote hote hain, kabhi swing nahi karte. Un directions ko dhundhne ke liye, main ek candidate "growth amount" $λ$ ko machine ki diagonal se subtract karta hoon, ek nayi machine $A - λ I$ banata hoon. Agar sahi $λ$ ke liye yeh nayi machine kisi arrow ko flatten karke kuch nahi kardi, toh woh arrow ek no-swing direction tha. "Ek direction ko kuch nahi karna" wahi baat hai jo "zero area banana," aur area ka meter determinant hai — toh main woh $λ$ dhuundhta hoon jo $det (A - λ I) = 0$ banate hain. Har aisi $λ$ ke liye main har arrow collect karta hoon jo flatten hota hai; woh collection ek poori line (ya plane) hoti hai jise eigenspace kehte hain. Kabhi kabhi ek $λ$ do baar aata hai par sirf ek line flatten karta hai — tab eigenspace repeat count se chhoti hoti hai, aur yahi last picture ki poori warning hai.

Active recall

Ek eigenvector ki direction kyun fixed rehti hai?

Kyunki

A v = λ v

output ko input ki same line par land karta hai — sirf length/sign badalta hai.

Picture mein

det (A - λ I) = 0

kaisa dikhta hai?

Machine

A - λ I

unit square ko zero area par collapse kar deta hai — woh ek poori direction ko origin par flatten kar deta hai.

A - λ I

mein identity

I

kyun insert karte hain?

Taaki hum ek matrix ko matrix se subtract karein;

A - λ

meaningless hai kyunki

A

ek grid hai aur

λ

ek single number.

A = (2112)

ke liye, eigen-lines ke naam batao.

E_{3} = span {(1 1)}

aur

E_{1} = span {(- 1 1)}

Kya ek doubly-repeated eigenvalue sirf 1D eigenspace de sakta hai?

Haan — jaise

(5015)

mein

dim E_{5} = 1

hai; geometric

\leq

algebraic.

Connections

Finding eigenspaces — woh parent recipe jise yeh page step by step draw karta hai.
Determinants — "kitna flat" ka meter jo Step 6 ko drive karta hai.
Null space and solving homogeneous systems — Step 7 ka engine.
Characteristic polynomial — woh equation jo determinant produce karta hai.
Rank-Nullity theorem — $dim E_{λ} = n - rank (A - λ I)$ deta hai.
Diagonalization — Step 8 ka deficient case kyun matter karta hai.
Symmetric matrices and spectral theorem — jab repeats hamesha space fill karte hain.