Visual walkthrough — Eigenvalues and eigenvectors — characteristic polynomial

4.5.29 · D2 · Maths › Linear Algebra (Full) › Eigenvalues and eigenvectors — characteristic polynomial

Step 1 — Ek matrix ek arrow ke saath kya karta hai

KYA HAI. Sabse basic cheez se shuru karte hain: ek arrow (ek vector) jo origin se nikalta hai. Matrix $A$ ek machine hai: isko arrow $\mathbf{x}$ do, aur yeh ek naya arrow nikaalti hai, jise hum $A\mathbf{x}$ kehte hain.

KYUN. Pehle humen dekhna hoga ki ek ordinary arrow ke saath kya hota hai — tabhi hum special arrows dhundh sakte hain. Zyaadatar arrows naye direction mein nikal ke aate hain — machine ne unhe ghuma diya.

TASVEER. Figure mein, cyan arrow aapka input $\mathbf{x}$ hai. Amber arrow output $A\mathbf{x}$ hai. Dhyan do ki amber arrow alag direction mein point kar raha hai — machine ne use rotate bhi kiya aur lamba bhi kiya.

$A\mathbf{x}$ mein term by term: letter $A$ machine hai, arrow $\mathbf{x}$ uske daayein taraf andar daali jaane wali cheez hai, aur dono ko saath likhne ka matlab hai "$A$ ko $\mathbf{x}$ pe apply karo".

Step 2 — Woh jadui directions jahan kuch ghumta nahi

KYA HAI. Saare input arrows mein se, kuch special ones aisi same line pe nikal ke aate hain jis pe gaye the. Machine ne unhe ghuma nahi — sirf lamba ya chhota kiya (ya ulta flip kiya).

KYUN. Inhe naam dena zaroori hai, kyunki yeh matrix ka "grain" — woh directions jinhein woh simply treat karta hai — dikhate hain. Aise arrow ko eigenvector kehte hain, aur stretch factor ko eigenvalue $\lambda$.

TASVEER. Cyan arrow $\mathbf{x}$ aur amber arrow $A\mathbf{x}$ ab same dashed white line pe hain. Output sirf input hai jo $\lambda$ se scale hua hai (yahan $\lambda =2$, yani do guna lamba, same direction).

Equals sign hi poori demand hai: "machine ka output input ki line pe wapas aana chahiye."

Step 3 — Zero arrow kyun forbidden hai

KYA HAI. Sochte hain: agar hum $\mathbf{x}=0$, yani bina length waala arrow, allow kar den to kya hoga?

KYUN. Daalo ise: $A0=0$ aur $\lambda 0=0$ har number $\lambda$ ke liye. Toh equation $A0=\lambda 0$ jo bhi $\lambda$ ho, sach hai. Iska matlab $\lambda$ meaningless ho jaata hai — har number "qualify" kar leta. Isliye zero arrow ko bahar nikaalo.

TASVEER. Zero arrow origin pe sirf ek dot hai (amber drawn). Koi direction nahi, koi line nahi jis pe lie kare — $\lambda$ ke describe karne ke liye kuch hai hi nahi.

Step 4 — Sab kuch ek taraf lo

KYA HAI. $A\mathbf{x}=\lambda \mathbf{x}$ lo aur daayein side subtract karo: $A\mathbf{x}-\lambda \mathbf{x}=0.$

KYUN. Hum arrow $\mathbf{x}$ ko factor out karna chahte hain, jaise $5x-3x$ se $x$ factor out karte ho. Factor karne ke liye, saare $\mathbf{x}$ terms ek saath ek side pe hone chahiye, zero ke barabar.

TASVEER. Geometrically: agar $A\mathbf{x}$ aur $\lambda \mathbf{x}$ same arrow hain, toh $\lambda \mathbf{x}$ ko tip-to-tail ulta $A\mathbf{x}$ ke against rakhne par tum exactly origin pe aate ho — woh white closing loop $0$ tak.

• $A\mathbf{x}-\lambda \mathbf{x}$ ::: output arrow minus scaled input arrow.
• $=0$ ::: yeh perfectly cancel ho jaate hain — difference ki koi length nahi.

Step 5 — Identity trick: $\lambda$ ko matrix mein badalna

KYA HAI. Hum $(A-\lambda )\mathbf{x}$ likhna chahenge, lekin nahi likh sakte: $A$ numbers ka ek grid (matrix) hai aur $\lambda$ ek akela number. Ek number ko grid se subtract karna undefined hai — yeh alag worlds mein rehte hain. Repair: $\lambda$ ko $\lambda I$ se replace karo, jahan $I$ identity matrix hai (woh machine jo kuch nahi badaiti).

KYUN. Kyunki $I\mathbf{x}=\mathbf{x}$, $\lambda I$ se multiply karna wohi kaam karta hai jo $\lambda$ se multiply karna: $\lambda I\mathbf{x}=\lambda \mathbf{x}$. Lekin ab $\lambda I$ ek genuine grid hai, toh $A-\lambda I$ ek legal matrix subtraction hai. Determinants baad mein dekhna; abhi bas ek real matrix chahiye.

TASVEER. Identity $I$ ko draw kiya gaya hai as $\mathbf{x}$ khud pe land karta hua (arrow unchanged). Ise $\lambda$ se scale karne par amber arrow $\lambda \mathbf{x}$ milta hai — $\lambda I\mathbf{x}$ se identical.

$A\mathbf{x}-\lambda I\mathbf{x}=0⟹(A-\lambda I)\mathbf{x}=0.$

• $I$ ::: identity matrix — diagonal pe $1$s, baaki $0$s; kisi bhi arrow ko untouched chhodta hai.
• $\lambda I$ ::: ek matrix jo har arrow ko $\lambda$ se stretch karta hai.
• $A-\lambda I$ ::: ek honest matrix, $A$ ki har diagonal entry se $\lambda$ subtract karke banta hai.
• $\mathbf{x}$ ko factor out karna ab legal hai, kyunki dono terms (matrix)$\times \mathbf{x}$ hain.

Step 6 — Ek matrix kab ek arrow ko zero par crush karta hai?

KYA HAI. $M=A-\lambda I$ rename karo. Humen ek nonzero arrow $\mathbf{x}$ chahiye jiske liye $M\mathbf{x}=0$ ho. Iska matlab $M$ ko ek real arrow ko origin pe collapse karna hoga.

KYUN. Agar $M$ reversible (invertible) hota, toh hum ise undo kar sakte: dono sides ko $M^{-1}$ se multiply karo aur $\mathbf{x}=M^{-1}0=0$ milega — forbidden zero arrow pe wapas forced! Toh $M$ reversible ka opposite hona chahiye: ise kisi direction ko nothing mein squash karna hoga. Aise $M$ ka nonzero null space hota hai. Aisa collapsing matrix singular kehlata hai.

TASVEER. Left panel: ek invertible $M$ chhote square ko nonzero-area parallelogram mein map karta hai — kuch collapse nahi hota, sirf $0$ hi $0$ pe map hota hai. Right panel: ek singular $M$ poore square ko ek line pe flatten karta hai — ek poora direction (cyan) origin pe crush ho jaata hai. Wahi crushed direction hamara eigenvector hai.

Step 7 — Singular ka matlab zero determinant

KYA HAI. Hum kaise test karein ki $M=A-\lambda I$ space collapse karta hai, bina tasveer banaye? Determinant use karo. Determinant $\det (M)$ machine ka area (ya volume) scaling measure karta hai.

YEH TOOL KYUN. Humen ek aisa number chahiye jo "space flat ho gaya" chilla ke bolе. Determinant exactly woh number hai: agar $M$ square ko line pe squash karta hai, toh enclosed area $0$ ho jaata hai, toh $\det (M)=0$. Aur koi cheez itni clearly collapse detect nahi karti — isliye determinant, naki trace ya koi simpler cheez, yahan sahi instrument hai.

TASVEER. Unit square (area $1$) parallelogram mein map hota hai jiska area $|\det M|$ hai. Jab $\lambda$ eigenvalue ki taraf slide karta hai, dekho parallelogram line mein thin hota hua — area $\to 0$.

$\det (A-\lambda I)=0.$

• $\det (\cdot )$ ::: ek matrix ka area/volume-scaling number.
• $=0$ ::: scaling collapse ho gayi — space flat ho gayi — matrix singular hai.

Step 8 — Equation ko ek curve ki tarah padhna

KYA HAI. Kyunki ek $n\times n$ matrix ka determinant uski entries ke products se banta hai, $p(\lambda )=\det (A-\lambda I)$ $\lambda$ mein degree $n$ ka ek polynomial nikalta hai: the characteristic polynomial.

KYUN. Polynomial ek curve hai jise hum graph kar sakte hain, aur uske horizontal axis ke crossings exactly woh $\lambda$ hain jo $\det =0$ banate hain — eigenvalues. Toh ek abstract collapse condition ban jaata hai "woh jagah dhundho jahan curve zero hit kare".

TASVEER. $A=\left(\begin{array}{cc}2& 1\\ 1& 2\end{array}\right)$ ke liye $p(\lambda )={\lambda }^2-4\lambda +3$ milta hai. Amber parabola axis ko $\lambda =1$ aur $\lambda =3$ par cross karta hai — do eigenvalues, cyan dots se mark kiye.

• $p(\lambda )$ ::: characteristic polynomial, curve ki height.
• roots of $p$ ::: woh $\lambda$-values jahan curve $0$ ko touch kare = eigenvalues.

Roots ka sum $1+3=4=\mathrm{tr}(A)$ hai aur product $1\cdot 3=3=\det (A)$ — trace aur determinant se free sanity check. (Yeh bhi dekho Cayley–Hamilton Theorem: matrix apna khud ka $p$ satisfy karta hai.)

Step 9 — Degenerate cases (kabhi chhode nahi jaate)

KYA HAI. Teen special situations jo derivation mein survive honi chahiye.

Case A — repeated root. $A=\left(\begin{array}{cc}3& 1\\ 0& 3\end{array}\right)$ se $p(\lambda )=(3-\lambda {)}^2$ milta hai: curve $\lambda =3$ par axis ko touch karta hai bina cross kiye (double root). Algebraic multiplicity $2$, lekin sirf ek eigenvector direction.

Case B — no real roots. Rotation $A=\left(\begin{array}{cc}0& -1\\ 1& 0\end{array}\right)$ se $p(\lambda )={\lambda }^2+1$ milta hai: parabola axis ko upar se float karta hai, kabhi touch nahi karta. Koi real eigenvalue nahi — roots $\pm i$ hain, complex world mein. Ek real matrix jo har arrow ko ghuma deta hai uska koi real eigen-direction nahi hota, exactly jaisa tasveer warn karti hai.

Case C — identity itself. $A=I$ se $A\mathbf{x}=\mathbf{x}$ milta hai har arrow ke liye: $\lambda =1$ saare directions mein kaam karta hai. $p(\lambda )=(1-\lambda {)}^2$, aur har nonzero arrow eigenvector hai.

TASVEER. Shared axes pe teen parabolas: do baar cross karna (distinct real), ek baar touch karna (repeated), bilkul clear float karna (complex).

Ek-tasveer summary

Har arrow ek step dikhata hai: eigen-equation $\to$ ek taraf lo $\to$ $I$ daalo $\to$ collapse demand karo $\to$ determinant zero $\to$ curve ki roots padho.

Flashcards