Visual walkthrough — Eigenvalues of A — system modes, stability

3.5.30 · D2 · Physics › Guidance, Navigation & Control (GNC) › Eigenvalues of A — system modes, stability

---

Step 1 — "Coupled" ka matlab asal mein kya hota hai

KYA. Hamare paas do numbers hain jo hamare system ko abhi describe karte hain. Inhe $x_1$ aur $x_2$ bol lo. Ek spacecraft ke liye socho: $x_1$ = attitude kitna off hai jahan chahte hain wahan se, aur $x_2$ = woh error kitni tezi se change ho raha hai. Dono milke ek single arrow (ek state vector) banate hain: $\mathbf{x}=(x_1,x_2)$, jo ek 2-D plane par rehta hai.

Woh rule jo humein batata hai ki arrow kaise move karta hai, woh hai: $\dot{\mathbf{x}}=A\mathbf{x}.$

• $\mathbf{x}$ :: arrow (hum kahan hain), plane par ek point.
• $\dot{\mathbf{x}}$ :: woh choti velocity arrow jo us point se attached hai — state agli instant mein kis taraf aur kitni tezi se slide karta hai.
• $A$ :: woh machine jo tumhari position padhti hai aur tumhe tumhari velocity wapas deti hai.

YEH MUSHKIL KYUN HAI. Figure dekho: zyaadatar points par velocity arrow (orange) position arrow (blue) ke saath align nahi karta. Machine $A$ ne tumhara arrow liya aur use rotate kar diya. Woh rotation hi coupling hai — $x_1$ ka future $x_2$ par depend karta hai aur vice-versa, isliye inhe ek ek karke solve nahi kar sakte.

PICTURE.

---

Step 2 — Woh khaas directions jahan rotation gayab ho jaata hai

KYA. Zyaadatar arrows $A$ ke through rotate hote hain. Lekin kuch special directions ho sakte hain jahan output arrow bilkul usi taraf point karta hai jis taraf input — $A$ use sirf stretch ya shrink karta hai, kabhi turn nahi karta. Aisa ek direction $\mathbf{v}$ satisfy karta hai: $A\mathbf{v}=\lambda \mathbf{v}.$

• $\mathbf{v}$ :: woh special direction, jise eigenvector kehte hain ("eigen" = German mein apna/proper).
• $\lambda$ :: stretch factor, eigenvalue. $\lambda =2$ arrow ko double karta hai, $\lambda =0.5$ aadha karta hai, $\lambda =-1$ flip karta hai.

YEH KYUN DHOONDH RAHE HAIN. Agar state kabhi $\mathbf{v}$ ki taraf point kare, to uski velocity $A\mathbf{v}=\lambda \mathbf{v}$ bhi $\mathbf{v}$ ki taraf point karegi. State sirf apni line ke saath slide kar sakti hai — woh kabhi nahi jaayegi. Ek 2-D tangle bas ek rail par 1-D problem ban jaata hai.

PICTURE. Green arrows eigen-directions hain: input aur output parallel hain. Gray arrows ordinary directions hain: output twist ho jaata hai.

---

Step 3 — Solution guess karna (the ansatz)

KYA. Hum ek aisa solution guess karte hain jo ek fixed shape $\mathbf{v}$ rakhe aur sirf apni length time ke saath change kare: $\mathbf{x}(t)=e^{\lambda t}\mathbf{v}.$

• $\mathbf{v}$ :: frozen direction — state kabhi is line se nahi jaata.
• $e^{\lambda t}$ :: ek single scalar "volume knob" jo arrow ko grow ya shrink karta hai jab $t$ chalta hai.
• Dono milke :: ek aisa arrow jo ek rail par rehta hai aur bas andar-bahar saans leta hai.

YEH GUESS KYUN KIYA. Humne abhi prove kiya (Step 2) ki eigen-direction ke saath motion 1-D hai. To ek solution jo poori tarah us rail par rehta hai woh sabse simple cheez hai jo possibly kaam kar sakti hai — aur exponential $e^{\lambda t}$ woh ek function hai jis ka rate of change khud uske proportional hai, bilkul wahi jo ek "fixed percentage par grow karne wali" rail demand karti hai.

Exponential aur kuch nahi kyun? Humein ek knob $f(t)$ chahiye jisme $\dot{f}=\lambda f$ ho (velocity size ke proportional — hamare ODE ka 1-D version). Woh ek hi function jo yeh satisfy karta hai woh hai $f(t)=e^{\lambda t}$. Yahi poora reason hai ki $e$ linear systems mein kyun aata hai.

PICTURE. State green rail par rehti hai; sirf uski length $e^{\lambda t}$ curve par chalti hai.

---

Step 4 — Substitute karo aur dekho vectors cancel ho jaate hain

KYA. Guess ko $\dot{\mathbf{x}}=A\mathbf{x}$ mein plug karo. Pehle guess differentiate karo: $\dot{\mathbf{x}}=\underset{\text{knob ki speed}}{\underbrace{\lambda e^{\lambda t}}}\mathbf{v}.$ Ab demand karo ki yeh physics maan kar chale: $\lambda e^{\lambda t}\mathbf{v}=A{e}^{\lambda t}\mathbf{v}.$

• Left side :: woh velocity jo hamara guess predict karta hai.
• Right side :: woh velocity jo machine $A$ asal mein deta hai.
• Inhe equal set karna :: guess ko honest hone par majboor karna.

KYUN. Ek guess tabhi solution hai jab woh equation har jagah satisfy kare. Yeh line woh demand hai jo likhi gayi hai.

PICTURE. Do velocity arrows — predicted (blue) aur machine-produced (orange) — coincide hone chahiye.

Kyunki $e^{\lambda t}$ ek positive number hai jo kabhi zero nahi hota, hum ise dono sides se cancel kar sakte hain: $A\mathbf{v}=\lambda \mathbf{v}.$ Puri time-dependence gayab ho gayi. Guess tab hi kaam karta hai jab $(\lambda ,\mathbf{v})$ ek eigenpair ho — bilkul Step 2 ke special directions. Clock $A$ mein pehle se hi baka hua tha.

---

Step 5 — Ek real (nonzero) direction ki condition

KYA. $A\mathbf{v}=\lambda \mathbf{v}$ ko rearrange karo sab kuch left la kar: $(A-\lambda I)\mathbf{v}=0.$

• $I$ :: identity matrix — "kuch mat karo" machine, isliye $\lambda I$ matlab "har direction mein $\lambda$ se scale karo."
• $A-\lambda I$ :: woh machine jo "$A$ apply karo, phir pure $\lambda$-stretch minus karo."
• $0$ :: zero arrow (origin par ek dot).

Humne $\mathbf{v}=0$ ban kar diya (origin ek direction nahi hai). To humein ek nonzero arrow chahiye jo $A-\lambda I$ origin tak crush kar de.

DETERMINANT KYUN. Ek matrix ek nonzero arrow ko zero tak crush karta hai tabhi jab woh singular ho — jab woh plane ko ek line (ya ek point) par flatten kare, saari area collapse karke. Woh number jo "kitni area bachti hai" measure karta hai woh hai determinant. Area zero ho jaana matlab: $\boxed{\det (A-\lambda I)=0.}$ Yeh characteristic equation hai. Dekho Characteristic Polynomial & Determinants ki kyun determinant exact area-scaling factor hai.

PICTURE. Left: ek healthy $A-\lambda I$ area rakhta hai, sirf origin origin par map hota hai — koi eigenvector nahi. Right: sahi $\lambda$ unit square ko ek line par collapse karta hai, isliye poori green rail origin par map hoti hai — eigenvector mil gaya.

---

Step 5b — Jab rails khatam ho jaayein: defective matrices

KYA. Sum-of-modes formula ne chup chap assume kiya tha ki hamare paas $n$ independent rails hain. Kabhi kabhi nahi hote: do roots coincide karte hain lekin sirf ek eigenvector share karte hain. Plane sirf rails se build nahi ho sakta — ek direction missing hai.

• diagonalizable :: $n$ independent eigenvectors exist karte hain → clean sum $\sum c_i{e}^{{\lambda }_{i}t}{\mathbf{v}}_i$.
• defective :: $n$ se kam eigenvectors → ek rail missing → ek extra $t{e}^{\lambda t}$ term aata hai.

STABILITY KE LIYE YEH KYUN MATTER KARTA HAI. Ek missing rail flow mein ek shear force karta hai, aur solution ek $t$ ka factor utha leta hai: $\mathbf{x}(t)=(c_1\mathbf{v}+c_2(t\mathbf{v}+\mathbf{w}))e^{\lambda t}.$ Woh extra $t$ ek otherwise-borderline mode ko growing mode mein badal sakta hai — yahi poora reason hai ki "axis par repeated eigenvalue" ko Step 7 mein dobara dekhna padta hai.

Zaroori distinction. Ek repeated eigenvalue automatically defective nahi hota. For example $A=\left(\begin{array}{cc}\lambda & 0\\ 0& \lambda \end{array}\right)$ mein $\lambda$ do baar hai lekin har direction ek eigenvector hai — yeh diagonalizable hai, koi $t$ term nahi, bilkul theek behave karta hai. Defectiveness ek missing-eigenvector ki problem hai, repeated-value ki problem nahi.

PICTURE. Left: repeated value lekin do rails (diagonalizable, pure scaling). Right: repeated value, ek rail — flow shear karta hai aur ek $t$ saath khiinchta hai.

---

Step 6 — Ek complex eigenvalue padhna: $\sigma$ aur $\omega$

KYA. Fundamental theorem of algebra ki wajah se kuch roots complex ho sakte hain. Ek ko likhte hain: $\lambda =\sigma +j\omega .$

• $\sigma$ :: real part — growth/decay rate.
• $\omega$ :: imaginary part — mode kitni tezi se spin karta hai.
• $j$ :: imaginary unit, $j^2=-1$ (control engineers $j$ use karte hain taaki $i$ current ke liye free rahe).

Ise knob mein daalo: $e^{\lambda t}=e^{\sigma t}{e}^{j\omega t}=e^{\sigma t}(\cos \omega t+j\sin \omega t).$

• $e^{\sigma t}$ :: envelope — ek pure grow/shrink curve.
• $e^{j\omega t}=\cos \omega t+j\sin \omega t$ :: ek point jo unit circle par spin karta hai — length hamesha exactly $1$.

SIRF $\sigma$ HI STABILITY DECIDE KYUN KARTA HAI. Magnitude hai: $|e^{\lambda t}|=e^{\sigma t}\cdot \underset{=1}{\underbrace{|\cos \omega t+j\sin \omega t|}}=e^{\sigma t}.$ Spinning wala part kabhi length nahi badalta — woh arrow ko hamesha ke liye ghuma sakta hai lekin use bada ya chhota nahi kar sakta. Sirf envelope $e^{\sigma t}$ magnitude control karta hai. Isliye $\sigma$ ki sign akeli bataati hai grow hoga ya khatam. Yeh Damping Ratio and Natural Frequency se link karta hai jahan $\sigma =-\zeta {\omega }_n$ hai.

PICTURE. Envelope $e^{\sigma t}$ (dashed) oscillation $\cos \omega t$ ko andar squeeze kar raha hai.

---

Step 6b — Complex modes secretly real hote hain

KYA. Hamare matrix $A$ mein real entries hain, aur physical state $\mathbf{x}(t)$ ek real arrow hai — koi imaginary spacecraft nahi. Phir bhi ek eigenvalue complex nikla, $\lambda =\sigma +j\omega$. Complex numbers ek real motion kaise describe kar sakte hain?

YEH KAISE KAM KARTA HAI. Real matrix ke liye complex eigenvalues conjugate pairs mein aate hain: agar $\lambda =\sigma +j\omega$ ek root hai, to $\bar{\lambda }=\sigma -j\omega$ bhi hai, conjugate eigenvector $\bar{\mathbf{v}}$ ke saath. Mode sum mein yeh do partners saath aate hain. Agar state real shuru hoti hai, unke coefficients bhi conjugate hote hain, $c_2={\bar{c}}_1$, aur do complex modes milke real ho jaate hain:

= 2\,e^{\sigma t}\big(\mathbf a\cos\omega t - \mathbf b\sin\omega t\big),$$ jahan $\mathbf a,\mathbf b$, $c_1\mathbf v$ ke real aur imaginary parts hain. - Imaginary halves :: exactly cancel ho jaate hain (ek number plus uska conjugate real hai, uske real part ka do guna). - Jo bachta hai :: ek real envelope $e^{\sigma t}$ times real sine/cosine — ek sachcha decaying (ya growing) oscillation. **PICTURE.** Do conjugate spirals (real axis ke across mirror images) milke ek flat real oscillation banate hain. --- ## Step 7 — Complex plane ka verdict (sab cases) **KYA.** Har eigenvalue ko complex plane par ek dot ki tarah plot karo: horizontal = $\sigma$, vertical = $\omega$. **Vertical axis finish line hai.** - $\sigma<0$ (axis ke left) :: envelope $e^{\sigma t}$ shrink karta hai → mode khatam → **stable**. - $\sigma>0$ (axis ke right) :: envelope grow karta hai → mode explode karta hai → **unstable**. - $\sigma=0$ (axis par) :: envelope flat $1$ hai → mode na khatam hota na grow karta → **marginal**, *basharte* ki us root ke paas eigenvectors ka full set ho (Step 5b). **HAR EIGENVALUE KO PASS KARNA KYUN ZAROORI HAI.** Full solution modes ka *sum* hai. Ek growing mode poore sum ko infinity tak kheench leta hai, chahe baaki kitne bhi acche se behave karo. Isliye stability = **saare** dots open left half-plane mein. [[Routh–Hurwitz Criterion]] ke algebraic shortcut aur [[Transfer Functions and Poles]] ki pole language se compare karo. **Degenerate case — axis par repeated eigenvalue.** Yahan Step 5b ka diagonalizable-vs-defective split verdict decide karta hai: - **Defective** repeat on the axis (rail missing): solution ek $t\,e^{0t}=t$ ramp carry karta hai jo bina kisi bound ke grow karta hai → **unstable**. Example $\begin{pmatrix}0&1\\0&0\end{pmatrix}$ ka double $\lambda=0$ hai lekin sirf ek eigenvector hai. - **Diagonalizable** repeat on the axis (rails ka full set): koi $t$ term nahi, solutions bounded rehte hain → **marginally stable**. Example $A=\mathbf 0$ ka double $\lambda=0$ hai lekin har direction ek eigenvector hai, isliye $\mathbf x(t)=\mathbf x(0)$ hamesha ke liye wahin baitha rehta hai. Isliye ek repeated axis eigenvalue *unstable sirf tab hai jab defective ho* — yeh mat maano ki ramp zaroor aayega. **PICTURE.** Har quadrant/case uske behaviour sketch ke saath, plus axis-repeat ke dono flavours. > [!mnemonic] LEFT hai RIGHT > **LEFT** half-plane mein dots ($\sigma<0$) system ko **RIGHT** kaam karaate hain — zero par settle. > Right side ka koi bhi dot, ya axis par baitha ek *defective* repeated dot, trouble ki nishani hai. --- ## Ek-picture summary Ek hi canvas par sab kuch: coupled field → ek special green rail jahan $A$ sirf scale karta hai → us rail par knob $e^{\lambda t}$ → uski speed $\sigma$ se decide → left/right verdict. > [!recall]- Feynman: poori walkthrough seedhe alfaaz mein > Ek hawadar maidan socho jahan ghaas ka har tinka dikhata hai ki wahan hawa kis taraf chalti hai — yahi hai > $A$ jo position ko velocity mein badalta hai, aur yeh ek chakradar gadbad hai kyunki directions ek dusre par push karti hain. > Lekin kuch magic lines hain jahan hawa seedha line ke saath hi chalti hai. Ek patta unhi lines par rakho aur > woh sirf us line ke upar ya neeche slide kar sakta hai, kabhi bhatak nahi sakta. Kitni tezi se slide hota hai > woh ek number hai, $\lambda$. Agar woh number patte ko center ki taraf kheenche, to line "safe" hai; agar > patte ko infinity tak dhakele, to line "dangerous" hai. Kabhi kabhi do magic lines ek mein collapse ho jaati hain > aur ek line missing ho jaati hai — tab patta *khud waqt ke saath sideways kheencha jaata hai* (woh extra $t$), > yahi reason hai ki ek borderline case secretly dangerous ho sakta hai. Aur jab ek line patte ko spiral karaaye, > woh spiral imaginary part $\omega$ hai; kyunki hawa real hai, spirals hamesha mirror pairs mein aate hain jo > milke ek asli real wobble banate hain. Real systems yeh saari lines mix karte hain, isliye hum har ek check karte hain: > agar *sab* andar kheenchen, to maidan shant ho jaata hai (stable); agar ek bhi bahar dhakele, sab kuch > eventually ud jaata hai (unstable). > [!recall]- Quick self-test > Step 4 mein hum $e^{\lambda t}$ cancel kyun kar sakte hain? ::: Kyunki $e^{\lambda t}$ strictly positive hai aur kabhi zero nahi hota, isliye dono sides ko isse divide karna legal hai, aur $A\mathbf v=\lambda\mathbf v$ bachta hai. > $\det(A-\lambda I)=0$ kyun (aur koi aur condition kyun nahi)? ::: Ek nonzero arrow sirf tabhi $\mathbf 0$ tak crush ho sakta hai jab matrix area flatten kare, yaani singular ho, yaani determinant zero ho. > Do eigenvalues hain $-1\pm 100j$. Stable hai ya nahi? ::: Stable — real part $-1<0$ hai; bada $\omega=100$ sirf tezi se oscillate karaata hai, stability par koi asar nahi. > Ek matrix ka ek repeated eigenvalue exactly $\lambda=0$ par hai. Verdict? ::: Depend karta hai — agar defective hai (ek eigenvector) to $t$ ramp grow karega → unstable; agar diagonalizable hai (full rails, jaise $A=\mathbf 0$) to solutions wahin rehte hain → marginally stable. > Complex eigenvalue phir bhi real motion kyun deta hai? ::: Kyunki real $A$ mein conjugate eigenpairs hote hain; do partner modes milte hain to unke imaginary parts cancel ho jaate hain, ek real $e^{\sigma t}\cos/\sin$ oscillation bachta hai. > Yeh guarantee kya karta hai ki $\det(A-\lambda I)=0$ ke roots hain hi? ::: Fundamental theorem of algebra: ek degree-$n$ polynomial ke exactly $n$ roots hote hain complex numbers par, multiplicity count karke.