Visual walkthrough — Observability matrix — rank test

3.5.33 · D2 · Physics › Guidance, Navigation & Control (GNC) › Observability matrix — rank test

Yeh page parent result ka flagship picture-tour hai. Agar yahan koi word unfamiliar lage, toh use yahan hi build kiya jaata hai — pehli line se.

---

Step 1 — Woh box jo andar se nahi dekh sakte

KYA HAI. Humare paas ek machine hai jiske andar ko numbers ki ek list describe karti hai jise hum state kehte hain, likha jaata hai $x$. Agar $n$ hidden numbers hain, toh hum kehte hain ki $x$ ${\mathbb{R}}^n$ mein rehta hai — ise ek $n$-dimensional room mein kahin point karne wale arrow (ek vector) ki tarah socho. Hum $x$ ko directly nahi dekh sakte. Hume sirf ek sensor par ek reading milti hai, numbers ki ek chhoti list $y$.

KYU. "Kya hum start figure out kar sakte hain?" ka saara maamla is gap par tika hai: kaafi saare hidden $x$, bahut kam visible $y$. Hume poochna hai ki kya thodi si readings bohot saare unknowns ko pin down kar sakti hain.

PICTURE. Box mein hidden gears hain ($x$ ke components); ek baahri needle $y$ dikhati hai.

Yahan $p$ hai kitne sensor numbers hum read karte hain; often $p=1$ hota hai (ek single dial).

---

Step 2 — Jo bhi input karta hai usse hatao

KYA HAI. Poora system hai $\dot{x}=Ax+Bu,y=Cx+Du$. $\dot{x}$ par dot ka matlab hai "samay mein $x$ ke change ki rate". $u$ hai input — knobs jo hum khud ghumaate hain. Hum $u=0$ set karte hain.

KYU. Kyunki $u$ humne choose kiya, $y$ par iska poora effect pehle se humein pata hai — hum use perfectly subtract kar sakte hain. Jo bacha woh machine ka free response hai: woh apne aap kahan se start hua wahan se kaise drift karta hai. Observability sirf uss free part ke baare mein hai, isliye $B,D,u$ simply question se gayab ho jaate hain. Sirf $A$ aur $C$ reh jaate hain.

PICTURE. Jana-pehchana input-contribution ek sticker ki tarah chheel diya jaata hai, clean free-response curve chodke.

$$y(t)=\underset{\text{dial wiring}}{\underbrace{C}}\underset{\text{how }x\text{ drifts in time}}{\underbrace{e^{At}}}\underset{\text{the prize}}{\underbrace{x(0)}}$$

---

Step 3 — Ek instant kaafi nahi hota

KYA HAI. Single instant $t=0$ par reading hai $y(0)=Cx(0)$.

KYU AKELA FAIL HOTA HAI. $C$ mein sirf $p$ rows hain (maan lo $1$). Ek equation $y(0)=Cx(0)$ sirf ek direction mein $x(0)$ ka shadow fix karta hai. Saare arrows $x(0)$ jo ek jaisa shadow daalte hain identical reading dete hain — hum unhe alag nahi kar sakte. Hume aur equations chahiye.

PICTURE. Candidate arrows ki ek poori line saari ek jaisi $y(0)$ project karti hain — dial unhe distinguish nahi kar sakta.

$$y(0)=\underset{\text{1 row}}{\underbrace{C}}x(0)\Rightarrow \text{only 1 direction of the }n\text{ pinned down.}$$

---

Step 4 — Dekho needle kitni tezi se hilaati hai (differentiate karo)

KYA HAI. $y$ ki time-derivatives $t=0$ par lo. Derivative $\dot{y}$ ka matlab hai "reading kitni tezi se change hoti hai". Kyunki $\frac{d}{dt}{e}^{At}=A{e}^{At}$, har derivative ek aur factor $A$ neeche kheench laata hai:

$$\begin{array}{rlrl}y(0)& =Cx(0)& & \text{(the reading)}\\ \dot{y}(0)& =CAx(0)& & \text{(its speed)}\\ \ddot{y}(0)& =C{A}^{2}x(0)& & \text{(its acceleration)}\\ & ⋮& & \\ y^{(k)}(0)& =C{A}^{k}x(0)& & (k\text{-th derivative})\end{array}$$

KYU. Har derivative same unknown $x(0)$ mein ek bilkul naya equation hai, lekin ek nayi coefficient row $C{A}^k$ ke saath. Har genuinely nayi row $x(0)$ ki ek nayi direction ki taraf point karti hai jise hum ab dekh sakte hain. Position tumhe $C$ deta hai; velocity tumhe $CA$ deta hai; aur aage bhi — needle ki motion hidden gears bahar nikaaal deti hai.

PICTURE. Har derivative order ek nayi arrow direction ko capture hote hue light up karta hai.

$y^{(k)}(0)=C{A}^{k}x(0)$ mein term-by-term:

• $y^{(k)}(0)$ — reading ka $k$-th derivative, ek number jo hum measure kar sakte hain.
• $C$ — dial wiring (fixed).
• $A^k$ — linkage $k$ baar apply hua; yahi hai jo hume fresh directions mein rotate karta hai.
• $x(0)$ — abhi bhi wahi ek prize jo solve karna hai.

---

Step 5 — Kyun hum $A^{n-1}$ par ruk jaate hain (Cayley–Hamilton)

KYA HAI. Hum forever differentiate karte nahi rehte. Hum $(n-1)$-th derivative par ruk jaate hain, rows $C,CA,\dots ,C{A}^{n-1}$ use karte hue.

KYU. Cayley–Hamilton theorem kehta hai ki har $n\times n$ matrix apni khud ki characteristic equation satisfy karta hai, jise rearrange karne par:

$$A^n=c_{n-1}{A}^{n-1}+\cdots +c_{1}A+c_{0}I.$$

Toh $A^n$ sirf lower powers ka recipe hai. $C$ se multiply karo: row $C{A}^n$ un rows ka mix hai jo pehle se hamare paas hain. Yeh zero new directions contribute karta hai. $n-1$ ke baad har derivative redundant hai.

PICTURE. Rows ki seedhi barhti jaati hai, phir ek ceiling se takra jaati hai — $C{A}^{n-1}$ ke baad ke naaye rungs purane waalon par flat ho jaate hain.

---

Step 6 — Prize ke liye solve karo: rank test

KYA HAI. Measured derivatives ko ek tall equation mein stack karo:

$$\underset{\text{known column}}{\underbrace{\left[\begin{array}{c}y(0)\\ \dot{y}(0)\\ ⋮\\ y^{(n-1)}(0)\end{array}\right]}}=\underset{\mathcal{O}}{\underbrace{\left[\begin{array}{c}C\\ CA\\ ⋮\\ C{A}^{n-1}\end{array}\right]}}x(0).$$

$x(0)$ ko uniquely solve karna possible hai iff $\mathcal{O}$ $n$-dimensional room ki har direction tak pahunchta hai — yaani uske columns independent hain, yaani uska rank $n$ ke barabar hai.

KYU. Rank $=n$ ka matlab hai $\mathcal{O}$ kisi bhi nonzero arrow ko zero tak flatten nahi karta: uska null space sirf $\{0\}$ hai. Do alag starts kabhi identical readings nahi de sakte. $n$ se kam hone ka matlab koi arrow invisibly nikal jaata hai.

PICTURE. Full rank = map $\mathcal{O}$ injective hai; har distinct $x(0)$ ek distinct reading-stack par land karta hai.

$$\boxed{(A,C)\text{observable}⟺\mathrm{rank}(\mathcal{O})=n}$$

Yahi woh yes/no gate hai jis par Kalman Filter ko chahiye uske state estimate converge karne se pehle, aur PBH test isi check ka eigenvalue-flavoured version deta hai.

---

Step 7 — Degenerate case: ek gear jo dial ko kabhi nahi hilata

KYA HAI. Maan lo $\mathrm{rank}(\mathcal{O})=r<n$. Toh ek nonzero arrow $x_0$ hai jiske liye $\mathcal{O}{x}_0=0$ hai.

KYU FATAL HAI. Agar $\mathcal{O}{x}_0=0$ toh har row use khatam kar deti hai: $C{x}_0=0,CA{x}_0=0,\dots$. Aage feed karo aur $y(t)=C{e}^{At}{x}_0\equiv 0$ sabhi time ke liye. Yeh arrow ek unobservable state hai — ek internal motion jo needle ko kabhi nahi hilaata.

PICTURE. Do decoupled identical modes ($A=2I$, single dial): difference-direction null space mein slide karta hai aur reading se gayab ho jaata hai.

---

Ek-picture summary

Poora tour ek canvas par: input chheel diya → free response → differentiate karke directions harvest karo → ladder $C,CA,\dots$ Cayley–Hamilton se $n-1$ par cap → independent rungs count karo. Rank $=n$ ⇒ har gear dekha gaya (observable); rank $<n$ ⇒ ek hidden gear (ek null-space arrow jo $y\equiv 0$ read karta hai).

Recall Feynman: poora walkthrough simple words mein

Tumhare paas ek sealed box hai spinning gears ke saath aur bahar sirf ek dial hai. Pehle tum kuch bhi jo tumne box ke saath kiya (knobs) use ignore karo — woh effect tumhe pata hai, subtract kar do. Ab dial ko apne aap drift karte dekho. Ek baar padhne se thoda pata chalta hai: gears ka ek shadow. Lekin dekho kitni tezi se hilaata hai, aur kitni tezi se woh change hota hai — har nayi "speed of the speed" ek fresh clue hai jo un gears ki taraf point karti hai jo pehle nahi dikh rahe the. Tum clues collect karte rehte ho, lekin ek mathematical guarantee hai (Cayley–Hamilton) ki $n-1$ speeds ke baad genuinely nayi koi clue nahi bachi — baad wala sab repeat ho jaata hai. Saari woh clue-rows ko ek tall table $\mathcal{O}$ mein stack karo. Truly independent rows count karo. Agar woh count gears ki tadaad $n$ ke barabar hai, dial secretly har gear reveal karta hai — box observable hai, aur tum iske exact starting spin par wapas ja sakte ho. Agar count kam pad jaaye, gears ka koi combination perfectly cancel ho jaata hai aur dial ko kabhi nahi hilata — woh hidden motion hamesha ke liye invisible hai, aur chahe kitna bhi dekho woh recover nahi hoga.

---

Connections

• Observability matrix — rank test (parent result)
• Cayley–Hamilton theorem (kyun ladder $A^{n-1}$ par rukti hai)
• Controllability matrix — rank test (mirror-image derivation)
• Rank and null space of a matrix ("rank $=n$" aur "kernel $=\{0\}$" ka matlab)
• Kalman Filter (converge karne ke liye observability chahiye)
• Kalman decomposition (Step 7 ke unobservable arrows ko isolate karta hai)
• State-space representation ($\dot{x}=Ax+Bu$ ki language)
• PBH test (rank test ka eigenvalue-based twin)