Visual walkthrough — Complex eigenvalues — rotation-scaling interpretation

4.5.32 · D2 · Maths › Linear Algebra (Full) › Complex eigenvalues — rotation-scaling interpretation

---

Step 1 — "Eigenvalue" actually poochh kya raha hai, aur kyun ek spinner ke paas koi nahi hota

KYA HAI. Ek matrix $A$ ek machine hai: isko ek arrow $v$ do (ek "vector" — origin se ek arrow), yeh ek naya arrow $Av$ nikalta hai. Ek eigenvector ek special input arrow hai jiska output usi line pe point karta hai — sirf stretch ya flip hota hai. Stretch factor hi eigenvalue $\lambda$ hai:

$Av=\lambda v.$

• $A$ :: the machine (numbers ki ek table jo arrows pe act karti hai).
• $v$ :: input arrow, zero arrow nahi ho sakta.
• $\lambda$ :: output kitni baar lamba hai (aur same/opposite direction mein).

KYUN CARE KARTE HAIN. Agar aisi koi line exist karti hai, toh machine us pe "boring" hai — bas arrows ko andar bahar slide karti hai. Lekin kuch machines ke paas aisi koi line hi nahi hoti: ek machine jo spin karti hai har arrow ko. Spinning ke baad, koi bhi arrow wahan point nahi karta jahan se shuru hua tha, isliye koi real $\lambda$ nahi hota.

PICTURE. Left pe, ek stretch machine: red arrow apni direction maintain karta hai, sirf bada hota hai — woh direction ek eigenvector hai. Right pe, ek spinner: har arrow tilt ho jaata hai, toh koi arrow eigenvector ke roop mein survive nahi karta.

---

Step 2 — Numbers kahan se aate hain: the characteristic equation

KYA HAI. $\lambda$ find karne ke liye, $Av=\lambda v$ ko $(A-\lambda I)v=0$ mein rewrite karo, jahan $I$ identity hai (the "do-nothing" machine). Ek nonzero arrow $v$ sirf tabhi zero ho jaata hai jab machine $A-\lambda I$ degenerate ho — uska determinant $0$ ho:

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

$A=\left(\begin{array}{cc}a& b\\ c& d\end{array}\right)$ ke liye yeh multiply hokar banta hai

$\underset{\text{squared term}}{\underbrace{{\lambda }^2}}-\underset{\text{trace }\mathrm{tr}A}{\underbrace{(a+d)}}\lambda +\underset{\det A}{\underbrace{(ad-bc)}}=0.$

• $\mathrm{tr}A=a+d$ :: the trace, diagonal ka sum — dekho Trace and Determinant.
• $\det A=ad-bc$ :: the determinant, machine ka area-scaling.

YEH SHAPE KYUN. Ek $2\times 2$ matrix $\lambda$ mein ek quadratic deta hai, toh zyada se zyada do eigenvalues. Quadratic formula unhe deta hai:

$\lambda =\frac{\mathrm{tr}A\pm \sqrt{(\mathrm{tr}A{)}^2-4\det A}}{2}.$

PICTURE. Parabola $y={\lambda }^2-(\mathrm{tr}A)\lambda +\det A$. Jahan yeh horizontal axis ko cross karta hai woh real eigenvalues hain. Jab parabola axis ke upar completely float karta hai toh woh kabhi cross nahi karta — koi real roots nahi — aur wahi hai spinner.

---

Step 3 — Ek complex $\lambda$ ko pehle point ke roop mein, phir spin-zoom ke roop mein padhna

KYA HAI. Jab $\Delta <0$, toh $\sqrt{\Delta }=\sqrt{-|\Delta |}=i\sqrt{|\Delta |}$, jahan $i$ woh invented number hai jiske liye $i^2=-1$. Toh do eigenvalues hain

$\lambda =\alpha \pm i\beta ,\alpha =\frac{\mathrm{tr}A}{2},\beta =\frac{\sqrt{|\Delta |}}{2}.$

• $\alpha$ :: real part (point $\lambda$ ka horizontal coordinate).
• $\beta$ :: imaginary part (vertical coordinate).

PLOT KYUN KARTE HAIN. Ek complex number ek plane mein ek point hai. Plane mein har point ko kitna bahar hai aur kis angle pe bhi describe kiya ja sakta hai — uska polar form (dekho Complex Numbers - Polar Form):

$r=|\lambda |=\sqrt{{\alpha }^2+{\beta }^2}(\text{distance}),\theta =\arg \lambda =atan2(\beta ,\alpha )(\text{angle}).$

Hum $\sqrt{{\alpha }^2+{\beta }^2}$ use karte hain kyunki woh origin se $(\alpha ,\beta )$ tak ke arrow ki Pythagorean length hai — legs $\alpha ,\beta$ wala right triangle. Hum $atan2$ use karte hain (plain $\arctan$ nahi) kyunki $\arctan (\beta /\alpha )$ quadrant II aur IV mein distinguish nahi kar sakta — woh har $180°$ pe repeat hota hai. $atan2$ dono $\alpha$ aur $\beta$ ke signs dekhta hai angle ko correct quadrant mein place karne ke liye.

PICTURE. Point $\lambda =\alpha +i\beta$ apne right triangle ke saath: horizontal leg $\alpha$, vertical leg $\beta$, hypotenuse $r$, opening angle $\theta$.

---

Step 4 — Ek complex equation ko do real equations mein split karna

KYA HAI. Hamare paas do eigenvalues hain, $\alpha +i\beta$ aur $\alpha -i\beta$. Hum deliberately neeche wala grab karte hain, $\lambda =\alpha -i\beta$ (minus sign wali branch). Uska eigenvector bhi complex hai; use $v=x+iy$ likho jahan $x$ aur $y$ ordinary real arrows hain. $Av=\lambda v$ mein plug karo:

$A(x+iy)=(\alpha -i\beta )(x+iy).$

Right side expand karo $i^2=-1$ use karke:

$Ax+iAy=(\alpha x+\beta y)+i(\alpha y-\beta x).$

Kyunki $A$, $x$, $y$ sab real hain, $Ax$ aur $Ay$ real hain. Do complex cheezein exactly tabhi equal hoti hain jab unke real parts match karein aur unke imaginary parts match karein. Woh ek complex equation do real equations ban jaati hai:

$\boxed{Ax=\alpha x+\beta y},\boxed{Ay=-\beta x+\alpha y}.$

• real part :: $Ax=\alpha x+\beta y$ — machine arrow $x$ ko kaise move karti hai.
• imaginary part :: $Ay=-\beta x+\alpha y$ — machine arrow $y$ ko kaise move karti hai.

YEH POORA TRICK KYUN HAI. Complex algebra do real geometric facts ban gayi do real arrows ke baare mein. Yeh bridge hai "$i$" se "pictures" tak.

PICTURE. Do real arrows $x$ (mint) aur $y$ (coral), aur machine $A$ har ek ko kahan bhejti hai — $\alpha ,\beta$ weights ke saath $x$ aur $y$ ka blend.

---

Step 5 — Do facts ko ek matrix equation mein package karna

KYA HAI. $x$ aur $y$ ko side by side columns ke roop mein ek matrix $P=\left(\begin{array}{cc}x& y\end{array}\right)$ mein rakho. Toh "har column pe $A$ apply karo" bas $AP$ hai, jiske columns $Ax$ aur $Ay$ hain. Hamare do boxed facts kehte hain woh columns hain $\alpha x+\beta y$ aur $-\beta x+\alpha y$. Lekin woh exactly wohi hain jo tumhe milta hai $P$ ko right mein ek chhoti $2\times 2$ table se multiply karne pe:

$AP=P\left(\begin{array}{cc}\alpha & -\beta \\ \beta & \alpha \end{array}\right).$

• chhoti table ka top-left column $(\alpha ,\beta )$ :: $\alpha x+\beta y=Ax$ banata hai. ✓
• right column $(-\beta ,\alpha )$ :: $-\beta x+\alpha y=Ay$ banata hai. ✓

Dono sides ko right mein $P^{-1}$ se multiply karo (jo humne abhi prove kiya ki exist karta hai):

$A=PC{P}^{-1},C=\left(\begin{array}{cc}\alpha & -\beta \\ \beta & \alpha \end{array}\right).$

KYUN. $A=PC{P}^{-1}$ matlab: $x,y$ frame mein change karo ($P^{-1}$), $C$ karo, wapas change karo ($P$). Yeh exactly Diagonalization hai — lekin diagonal ki jagah, hum block $C$ pe land karte hain. Saari interesting geometry $C$ mein rehti hai.

PICTURE. The sandwich: input frame → $P^{-1}$ $x,y$ coordinates mein untwist karta hai → $C$ cleanly act karta hai → $P$ wapas retwist karta hai.

---

Step 6 — Kyun $C$ hai hi rotate-then-scale

KYA HAI. $C=\left(\begin{array}{cc}\alpha & -\beta \\ \beta & \alpha \end{array}\right)$ dekho. Yaad karo Step 3 ke triangle se $\alpha =r\cos \theta$ aur $\beta =r\sin \theta$. Substitute karo aur common $r$ bahar nikalo:

$C=\left(\begin{array}{cc}r\cos \theta & -r\sin \theta \\ r\sin \theta & r\cos \theta \end{array}\right)=r\underset{\text{pure rotation by }\theta }{\underbrace{\left(\begin{array}{cc}\cos \theta & -\sin \theta \\ \sin \theta & \cos \theta \end{array}\right)}}.$

• brace mein matrix :: standard rotation matrix — yeh har arrow ko $\theta$ se turn karta hai bina length change kiye (dekho Rotation Matrices).
• bahar factor $r$ :: uniform scaling — har arrow ki length $r$ se multiply hoti hai.

Toh $C=$ $\theta$ se rotate karo, phir $r$ se scale karo. Aur $r=|\lambda |$, $\theta =\arg \lambda$ — eigenvalue ka modulus aur argument.

YEH FACTOR KYUN KARNA CHAHIYE. Uss special pattern ki koi bhi table (equal diagonal $\alpha$, opposite off-diagonals $\mp \beta$) guaranteed hai $r$ times ek rotation, kyunki $(\alpha ,\beta )$ ko hamesha polar form mein likha ja sakta hai. Case-by-case check karne ki zarurat nahi — algebra force karta hai.

PICTURE. Ek grid $\theta$ se rotate hota hua aur $r$ se ek hi baar blow up hota hua: unit circle ek bada circle ban jaata hai $\theta$ se turned.

---

Step 7 — Har case: $r<1$, $r=1$, $r>1$, aur degenerate corners

KYA HAI. Eigenvalue plane mein kahin pe rehta hai; origin se uski distance $r$ decide karti hai repeated application $A,A^2,A^3,\dots$ ki shape (ek discrete Linear Dynamical Systems — uske Phase Portraits):

| $\lambda$ kahan baitha hai | $r=|\lambda |$ | Har step kya karta hai | Orbit shape | |---|---|---|---| | Unit circle ke andar | $r<1$ | spin + shrink | spiral inward | | Unit circle pe | $r=1$ | sirf spin | closed ellipse | | Unit circle ke bahar | $r>1$ | spin + grow | spiral outward |

Degenerate corners:

• $\beta =0$ :: bilkul bhi complex nahi — do eigenvalues ek hi real $\lambda =\alpha$ mein collapse ho jaate hain, aur $C$ collapse ho jaata hai $\lambda I=\left(\begin{array}{cc}\alpha & 0\\ 0& \alpha \end{array}\right)$ mein: invariant lines ke along pure scaling by $\alpha$, $x$ aur $y$ ka koi coupling nahi. Agar $\alpha >0$ toh har arrow bas in-place lengthen/shorten hota hai; agar $\alpha <0$ toh har eigen-line origin ke through flip ho jaati hai (us line pe ek $180°$ turn) aur saath mein scale bhi hoti hai — ek "spin-like" corner, lekin yeh $0$ ke through reflection hai, plane ka genuine rotation nahi.
• $\alpha =0,\beta \ne 0$ :: $\theta =\pm 90°$, pure quarter-turn $r$ se scaled (Worked Example 1 mein $r=1$, $\theta =90°$ tha).
• $\lambda =0$ :: $r=0$, machine arrows ko origin pe collapse kar deti hai.

$\alpha$ ki jagah iteration ke liye $r$ KYUN. $n$ steps ke baad, lengths $r^n$ se multiply hoti hain. Agar instead continuously flow karo ($\dot{v}=Av$), growth real part $\alpha =\mathrm{Re}\lambda$ se set hoti hai — alag sawaal hai, inhe confuse mat karo.

PICTURE. Teen orbits side by side: inward spiral ($r<1$), ellipse ($r=1$), outward spiral ($r>1$), sab same seed arrow se start hote hue.

---

Ek-picture summary

Complex eigenvalue $\lambda =\alpha -i\beta$, ek point ke roop mein plot hoke, hi woh instruction hai. Uski length $r$ batata hai har step kitna scale karta hai; uska angle $\theta$ batata hai har step kitna spin karta hai. Matrix $P$ (columns $x,y$ eigenvector se) bas woh frame hai jismein yeh clean spin-zoom hota hai; sandwiching deta hai $A=PC{P}^{-1}$.

Recall Feynman retelling — poora walkthrough simple words mein

Humne ek matrix se poocha, "kya koi arrow hai jise tum sirf stretch karte ho?" Ek spinner ne kaha "koi real wala nahi." Toh humne phir bhi stretch factor ke liye solve kiya aur ek complex number mila — real line se hatke ek point. Humne woh point plot kiya. Center se uski distance ne bataya spinner har turn pe kitna bada ya chhota hota hai; uske angle ne bataya woh kitna turn karta hai. Prove karne ke liye, humne complex eigenvector ko do real arrows $x$ aur $y$ ke roop mein likha, aur ek complex equation do honest real equations mein split ho gayi machine ke $x$ aur $y$ ko push karne ke baare mein. Woh do arrows kabhi same direction mein point nahi kar sakte (warna eigenvector ek real arrow hota, jo ek real spinner ke paas nahi ho sakta), toh hum safely unhe $P$ mein stack kar sakte hain aur $P^{-1}$ se undo kar sakte hain — sab kuch $A=PC{P}^{-1}$ mein turn karte hue, jahan $C$ kuch nahi hai sirf "angle se rotate karo, distance se scale karo." Phir humne teen fates dekhe — spiral in, loop forever, spiral out — aur woh flat corners jahan spin khatam ho jaata hai. Poora secret yahi hai: ek complex eigenvalue ek disguised spin-and-zoom hai.