Invertible matrix theorem — 12+ equivalent conditions: Visual walkthrough

4.5.25 · D2 · Maths › Linear Algebra (Full) › Invertible matrix theorem — 12+ equivalent conditions

Step 1 — Matrix ek arrow ko kya karta hai

KYA. Ek square matrix $A$ ek machine hai. Tum isme ek arrow $x$ daalo (plane mein ek point, likha jaata hai $x = (x_{1}, x_{2})$ ), aur woh ek naya arrow wapas deta hai, jise hum $A x$ likhte hain. "Square" ka matlab hai ki yeh plane-arrows ko plane-arrows mein le jaata hai (inputs aur outputs ki saman sankhya) — yeh word baad mein matter karega.

KYUN. Theorem mein sab kuch secretly is machine ke baare mein ek sawaal hai: kya tum ise hamesha undo kar sakte ho? kya yeh kabhi do arrows ko ek mein crush karta hai? kya yeh har arrow tak pahunch sakta hai? Toh kisi bhi algebra se pehle, hume machine ko action mein picture karna hoga.

PICTURE. Neeche, faint grid input plane hai. Bright grid woh hai jahan $A$ us plane ko bhejtaa hai. Amber arrow ek input $x$ hai; cyan arrow uska output $A x$ hai.

Figure — Invertible matrix theorem — 12+ equivalent conditions

Step 2 — $A x$ bas columns ko mix karne ki ek recipe hai

KYA. $A$ ko uske columns mein split karo — woh arrows jo tumhe pure directions $(1, 0)$ aur $(0, 1)$ dene par milte hain. Unhe $a_{1}$ aur $a_{2}$ kaho. Phir machine ek rule follow karti hai:

$A x = kitna a_{1} x_{1} pehla column a_{1} + kitna a_{2} x_{2} doosra column a_{2} .$

KYUN. Yeh pure linear algebra mein sabse important re-reading hai. Iska matlab hai ki output columns ka ek weighted blend hai aur kuch nahi. Isliye $A$ ke baare mein sawaal $a_{1}, a_{2}$ columns ke baare mein sawaal ban jaate hain: kya woh independent directions mein point karte hain? kya unke blends har jagah pahunchte hain?

PICTURE. Output arrow (cyan) $a_{1}$ ke saath $x_{1}$ steps aur phir $a_{2}$ ke saath $x_{2}$ steps chalke banta hai — instructions ka ek chhota parallelogram.

Span picture se link: saare possible $A x$ ka set exactly columns ka span hai.

Step 3 — Achhi machine vs. collapsing machine

KYA. Do cases, aur sirf do.

Achhi (invertible): dono columns $a_{1}, a_{2}$ genuinely alag directions mein point karte hain. Bright grid ek poori, un-squashed plane hai.
Collapsing (singular): $a_{2}$ usi line par hai jis par $a_{1}$ hai (maano $a_{2} = 3 a_{1}$ ). Har blend $x_{1} a_{1} + x_{2} a_{2}$ usi ek line par land karta hai. Poori plane ek line par flatten ho jaati hai.

KYUN. Yeh flattening theorem ki har failing condition ki root cause hai. Jab tum plane ko ek line par crush hote dekhte ho, saare 17 conditions ek saath fail ho jaate hain — yahi walkthrough unpack karegi.

PICTURE. Left: independent columns, area preserved. Right: parallel columns, shaded square ek segment mein collapse ho jaata hai (zero area).

Step 4 — Null space: kaun se arrows zero mein crush ho jaate hain?

KYA. Machine se pucho: kaun se inputs $x$ zero arrow $0$ ke roop mein nikalta hain? Woh collection null space hai, $Nul A = {x : A x = 0}$ .

Achhi machine: sirf $x = 0$ hi $0$ par map hota hai. Null space ek single point hai.
Collapsing machine: inputs ki poori ek line $0$ par map hoti hai (woh saare arrows jo flattening origin par squash karta hai).

KYUN. "Kya hum machine ko undo kar sakte hain?" wahin same hai jaisa "kya machine kabhi do alag inputs ko ek saath crush karti hai?" Agar do inputs $u \neq = v$ same output dete hain, toh $A u = A v$ , isliye $A (u - v) = 0$ jahan $u - v \neq = 0$ — ek nonzero crushed arrow. Isliye fat null space exactly wahi reason hai kyun tum machine ko undo nahi kar sakte. Yeh condition (4) aur condition (6) ka milna hai, one-to-one.

PICTURE. Collapsing machine ke liye, amber line null space hai: iske har arrow par landing origin dot par hoti hai.

Dekho injective: one-to-one $⟺$ null space sirf ${0}$ hai.

Step 5 — Independence aur "crushing" ek hi fact hain

KYA. Steps 2–4 ko chain karo. $A x = 0$ ka matlab hai $x_{1} a_{1} + x_{2} a_{2} = 0$ — columns ka ek blend jo zero ke barabar hai. Step 3 ki definition se compare karo: independent columns ka matlab hai aise blend mein sirf all-off wala hota hai.

$null space ek point hai (Step 4) A x = 0 only at x = 0 ⟺ columns alag ways mein point karte hain (Step 3) a_{1}, \dots, a_{n} independent$

KYUN. Humne do facts prove nahi kiye — humne notice kiya ki woh do tarike se likha hua ek hi sentence hai. Yeh condition (4) $⟺$ (5) hai. Yaad karne ke liye koi formula nahi; dono boxes ki equality Step 2 ki column-blend reading se forced hai.

PICTURE. Usi scene ke do views: left par, null space ek point/line ke roop mein; right par, columns independent/parallel arrows ke roop mein. Ek amber bracket unhe baandhta hai.

Step 6 — Row reduction: "independent" ko " $A \sim I$ " mein badalna

KYA. Row reduction pivots dhundhta hai — matrix ko tidy karne ke baad har row mein leading nonzero entry. Independent columns of a square matrix ka matlab hai koi column free nahi hai, isliye har column ek pivot earn karta hai. Ek $n \times n$ matrix ke liye, $n$ pivots har row aur har column ko bharte hain, aur tidied form (RREF) sirf identity $I_{n}$ ho sakti hai.

$har E_{i} = ek reversible row move E_{k} \dots E_{2} E_{1} A = I .$

KYUN. Har row move reversible hai (tum hamesha swap back, unscale, ya un-add kar sakte ho), isliye har $E_{i}$ khud invertible hai. $B = E_{k} \dots E_{1}$ set karne par $B A = I$ milta hai — haath se banaya hua ek honest inverse. Yeh arrow $2 \Rightarrow 1$ hai: row-equivalent-to- $I$ inverse manufacture karta hai.

PICTURE. Diagonal par descending pivots ki ek staircase; bright cells pivots hain, ek per row aur column.

Step 7 — Determinant: machine jo area rakhta hai

KYA. Determinant $det A$ woh signed area hai (2-D mein) ya volume (higher-D mein) jo shape ka milta hai jab machine ko unit square feed karo. Achhi machine us square ko kisi nonzero area ke parallelogram mein stretch/rotate karti hai. Collapsing machine use ek line par flatten karti hai — area $= 0$ .

$det A \neq = 0 ⟺ unit square positive area rakhta hai ⟺ koi collapse nahi .$

KYUN. Isliye $det$ chhote matrices ke liye sabse sasta test hai: yeh directly ek number se flattening measure karta hai. Row moves area ko sirf nonzero factors se scale karte hain aur $det I = 1$ , isliye $A \sim I ⟺ det A \neq = 0$ . Note karo ki theorem sirf is baat ki parwah karta hai ki area zero hai ya nahi, kabhi kitna chhota (ek $1 0^{- 9}$ ka $det$ abhi bhi fully invertible hai).

PICTURE. Left parallelogram: shaded, nonzero area, $det \neq = 0$ . Right: parallelogram ek amber segment mein collapse, area $0$ .

Step 8 — Eigenvalue $0$ : ek arrow jo machine apni line par rakhta hai lekin zero tak shrink karta hai

KYA. Ek eigenvalue $λ$ (dekho Eigenvalues and Eigenvectors) ek number hai jiske liye koi nonzero arrow $v$ obey karta hai $A v = λ v$ — machine $v$ ko usi ki line par rakhti hai aur sirf $λ$ se scale karti hai. Agar $λ = 0$ , toh $A v = 0$ ek nonzero $v$ ke liye: ek crushed arrow, exactly Step 4 ka fat null space.

$det A = λ_{1} λ_{2} \dots λ_{n} .$

KYUN. Ek single zero factor poora product kill kar deta hai, isliye $0$ eigenvalue hona $⟺ det A = 0 ⟺$ collapse. Yeh condition (14) usi picture mein neatly slide hona hai: "0 eigenvalue nahi hai" ek aur mask hai "no crush" ke liye.

PICTURE. Collapsing machine apne zero-eigenvector $v$ (amber) ke saath seedha origin tak shrink hoti hui.

Step 9 — Degenerate warning: non-square machines deal tod deti hain

KYA. Upar sab kuch quietly square use karta raha. Ek non-square machine — maano $3$ inputs, $2$ outputs, ya $2$ inputs, $4$ outputs — one-to-one ya onto ho sakti hai lekin kabhi dono nahi, isliye theorem simply apply nahi hota.

Tall ( $4 \times 2$ ): sirf $2$ columns $4$ -D space mein; unke blends zyada se zyada ek $2$ -D plane bharte hain, isliye woh har output tak nahi pahunch sakte. Onto nahi, hence not invertible even if columns independent hain.
Wide ( $2 \times 4$ ): $4$ columns $2$ -D space mein zaroor dependent honge (bahut saare arrows bahut kam room mein), isliye hamesha ek nonzero crushed input hoga. One-to-one nahi.

KYUN. IMT ka magic — injective aur surjective ka coincide karna — " $n$ pivots saari rows aur saare columns bharte hain" se aata hai, jiske liye equal counts chahiye. Square todo, equivalence todi.

PICTURE. Left: tall map, columns sirf 3-D space ke andar ek plane bharte hain. Right: wide map, plane mein chaar arrows overlap karne par majboor.

Ek-picture summary

Har mask same coin hai. Achhi machine area rakhti hai, information rakhti hai, har jagah pahunchti hai; collapsing machine flatten karti hai, ek line ko zero tak crush karti hai, aur zyatdar space miss karti hai. Ek flattening event saare failing conditions ko ek saath trip karta hai.

Recall Feynman: poora walkthrough seedhe shabdon mein

$A$ ko ek aisi machine socho jo arrows ko reshuffle karti hai. Uske columns woh hain jahan woh pure directions bhejtii hai, aur har output bas unhi columns ka blend hai tumhare dials $x_{1}, x_{2}, \dots$ ke saath. Ek achhi machine ke columns genuinely alag ways mein point karte hain: jo zero arrow deta hai woh sirf woh blend hai jab har dial off ho. Iska matlab hai kuch crush nahi hota, tum hamesha input rebuild kar sakte ho, unit square real area rakhta hai (isliye determinant zero nahi), koi arrow zero tak scale nahi hota (isliye $0$ eigenvalue nahi hai), aur blends space mein har arrow tak pahunch sakte hain. Ek buri machine mein ek repeated direction hoti hai: plane ek line par flatten ho jaata hai, inputs ki poori ek line origin par crush ho jaati hai, area zero ho jaata hai, $0$ eigenvalue ban jaata hai, aur space ke bade regions unreachable ho jaate hain — har failing test ek saath fire hota hai. Aur yeh sab sirf tab kaam karta hai jab machine mein utne hi inputs ho jaane outputs hain: ek non-square machine ek side se reversible ho sakti hai lekin kabhi dono se nahi, isliye theorem quietly side ho jaata hai.

Recall Quick self-check

Kyun ek nonzero null-space vector immediately invertibility kill kar deta hai? ::: Iska matlab hai ki do alag inputs ek output share karte hain ( $A (u - v) = 0$ ), isliye information lost ho jaati hai aur machine undo nahi ho sakti. Kyun ek invertible square matrix ke liye RREF $I$ ke barabar hona chahiye? ::: Independent columns har column mein ek pivot dete hain; $n \times n$ grid mein $n$ pivots har row aur column bharte hain, sirf identity chhod kar. $det A = 0$ ka geometrically kya matlab hai? ::: Machine unit square/cube ko zero area/volume par flatten karti hai — space ek lower dimension mein collapse ho jaati hai. IMT ek $4 \times 2$ matrix par kyun apply nahi hota? ::: Yeh square nahi hai; uske columns independent ho sakte hain lekin kabhi $R^{4}$ span nahi kar sakte, isliye injective aur surjective ab coincide nahi karte.