Rank of a matrix — definition, row rank = column rank theorem: Visual walkthrough

4.5.12 · D2 · Maths › Linear Algebra (Full) › Rank of a matrix — definition, row rank = column rank theore

Step 0 — Woh words jo hum use karenge (use karne se pehle build kiye)

Kisi bhi formula mein koi symbol aane se pehle, chaliye usse ek picture se anchor karte hain.

Ek vector sirf origin se ek arrow hai, ya equally numbers ki ek list jaise $(1, 2, 3)$ . Numbers batate hain ki har axis ke along kitna chalna hai.
Kuch vectors ka ek linear combination matlab hai: har ek ko ek number se scale karo, phir unhe add kar do. Kuch arrows ko stretch karte hue aur unhe tip-to-tail rakhte hue imagine karo.
Vectors ke ek set ka span woh har point hai jo tum unke linear combinations se reach kar sakte ho. Ek nonzero arrow ek line span karta hai; do alag directions mein point karne wale arrows ek plane span karte hain.
Vectors independent hain agar unme se koi bhi doosron se reachable na ho — har ek genuinely nayi direction add karta hai. Agar ek arrow baaki ke span ke andar lie karta hai, toh woh dependent hain.
Ek space ka basis independent arrows ka ek chhotese-chhotha set hai jiska span poora space hai. Basis mein arrows ki sankhya dimension hai.

Yeh alag-alag worlds mein rehte hain ( $R^{n}$ vs $R^{m}$ ), possibly alag sizes ke. Hamara goal: dikhana ki yeh do dimensions ek hi integer hain. Yeh genuinely surprising hai, isliye iska picture-proof deserve karta hai.

Figure — Rank of a matrix — definition, row rank = column rank theorem

In spaces ke baare mein aur: Row Space and Column Space. Independence aur bases pe: Linear Independence and Basis.

Step 1 — Ek concrete matrix dekho aur haath se dono taraf count karo

KYA. Kisi bhi theorem se pehle, chaliye physically ek chhoti matrix ke liye independent rows aur independent columns count karte hain, taaki claim abstract na rahe.

$A = 123246134$

KYUN. Ek mana hua example das faith-ke-proofs se zyada kaam ka hai. Hum ek baar equality ko witness karna chahte hain, phir usse explain karna chahte hain.

PICTURE. Neeche ki figure mein, teen columns arrows ke roop mein draw kiye gaye hain. Columns $1$ aur $3$ genuinely alag directions mein point karte hain; column $2 = 2 \times$ column $1$ (yeh column 1 ki line ke upar lie karta hai, bas lamba). Toh sirf 2 independent column-directions hain.

$(2 1) 3$ — ise $a_{1}$ kaho, pehli direction.
$(4 2) 6 = 2 a_{1}$ — koi nayi direction nahi (dependent).
$(3 1) 4$ — ise $a_{3}$ kaho, $a_{1}$ line se alag point karta hai.

Column rank $= 2$ . Hum Step 2 mein confirm karenge ki row rank $= 2$ hai.

Step 2 — Elimination bina guessing ke count reveal karta hai

KYA. $A$ ko row operations use karke reduce karo — ek row ko (khud minus doosri row ka ek multiple) se replace karte hue — jab tak har nonzero row uske upar wali row se aur right mein shuru na ho (echelon form). Phir nonzero rows count karo.

123246134 R_{2} - 2 R_{1} R_{3} - 3 R_{1} 100200111 R_{3} - R_{2} \underline{1} 00 200 1 \underline{1} 0

Arrows mein term by term:

$R_{2} - 2 R_{1}$ matlab naya row 2 $=$ purana row 2 $- 2 \times$ row 1. $2 \times$ exactly isliye choose kiya gaya hai taaki leading $1$ ke neeche ka leading $2$ zero ho jaye.
$R_{3} - 3 R_{1}$ usi leading $1$ ke neeche ke $3$ ko zero karta hai.
$R_{3} - R_{2}$ duplicate $(0, 0, 1)$ ko hatata hai taaki ek chhupa hua dependence ek sachcha zero row ban jaye.

Underlined entries $\underline{1}$ pivots hain — har surviving row ki pehli nonzero entry.

KYUN row operations yahan legal hain. Har naya row purani rows ka ek linear combination hai, aur purani rows nayi rows se recover ki ja sakti hain. Toh rows ka span kabhi nahi badlta — row space, aur isliye uski dimension (row rank), untouched rehti hai. (Poori kahani: Gaussian Elimination & Echelon Form.)

PICTURE. Figure mein row space ek shaded plane ke roop mein dikhaya gaya hai. Har operation ek row ko us plane ke andar slide karta hai; plane khud tilt nahi hoti. Jab teesra row $0$ mein collapse ho jaata hai, toh hum jaante hain ki plane sirf 2-dimensional thi: row rank $= 2$ . Yeh Step 1 ke column count se match karta hai — lekin kyun match karna zaroori hai woh Steps 3–7 mein hai.

Recall Apni pakad check karo

Original matrix mein nonzero rows count karna galat kyun hai? ::: Woh nonzero ho sakte hain phir bhi dependent hon (jaise $(1, 2)$ aur $(2, 4)$ ); sirf echelon form sachchi independence expose karta hai. Row operations ne kaunsi quantity preserve ki? ::: Row space, isliye row rank.

Step 3 — Column rank ko ek naam do aur ek column basis pakdo

KYA. Ab se, general matrix. Maano

$r = column rank of A .$

Yahan $r$ bas "kitni independent column-directions exist hain" hai. $r$ actual columns choose karo jo independent hain aur baaki sab ko span karte hain — inhe $c_{1}, \dots, c_{r}$ kaho. Yeh column space ka ek basis hai.

KYUN. Ek basis ek minimal instruction set hai: parent note ka LEGO idea. $A$ ka har column inhi $r$ arrows se bana ja sakta hai aur inse kam mein nahi.

PICTURE. Column space ko ek $r$ -dimensional flat ke roop mein draw kiya gaya hai ( $r = 2$ ke liye, $R^{m}$ ke andar ek plane). Basis arrows $c_{1}, \dots, c_{r}$ highlight kiye gaye hain; baaki har column us flat ke andar ek dot hai, highlighted arrows ke mix ke roop mein reachable hai.

In basis columns ko side by side ek skinny matrix mein stack karo:

$C = [c_{1} c_{2} \dots c_{r}], C is m \times r .$

Har $c_{k}$ mein $m$ entries hain (yeh $R^{m}$ mein rehta hai) → $m$ rows.
Inki sankhya $r$ hai → $r$ columns. Isliye $C$ , $m \times r$ hai.

Step 4 — $A$ ka har column $C$ times ek short recipe hai

KYA. Kyunki $c_{1}, \dots, c_{r}$ column space span karte hain, $A$ ka column $j$ (ise $A_{j}$ kaho) unka koi combination hai:

$A_{j} = x_{1 j} c_{1} + x_{2 j} c_{2} + \dots + x_{r j} c_{r} .$

Har symbol padh ke:

$A_{j}$ — original matrix ka $j$ -wa column (ek arrow $R^{m}$ mein).
$x_{k j}$ — basis arrow $c_{k}$ ka amount jo $A_{j}$ build karne ke liye chahiye. Ek akela number.
Poora right side — " $r$ ingredients ko in proportions mein mix karo". Yeh exactly matrix multiplication columns combine karne ke roop mein pada hai.

KYUN. Hum " $A_{j}$ span mein rehta hai" (ek picture) ko " $A_{j} = C \cdot (recipe)$ " (ek equation) mein badal rahe hain, taaki algebra ise aage le ja sake.

PICTURE. Ek chosen column ke liye, figure mein target arrow $A_{j}$ ko $c_{1}$ aur $c_{2}$ ke scaled copies se tip-to-tail reconstruct kiya gaya hai. Chhote numbers $x_{1 j}, x_{2 j}$ scaling label karte hain.

Har column ki recipe ko ek matrix $R$ mein collect karo, jiska $j$ -wa column $(x_{1 j}, \dots, x_{r j})$ hai:

$A = C R, C : m \times r, R : r \times n .$

$R$ mein $r$ rows hain (ek per basis ingredient) aur $n$ columns hain (ek recipe per original column) → $r \times n$ .
Shared inner dimension $r$ column rank hai. Yeh yaad rakho — yahi poora trick hai.

Step 5 — Wahi factorization secretly rows ko bhi control karti hai

KYA. Multiplication $A = C R$ ko columns ki bajaye rows se padha ja sakta hai. $A$ ka Row $i$ , $C$ ke row $i$ ko $R$ mein multiply karne ke barabar hai:

$(row i of A) = (row i of C) R .$

Symbol by symbol:

$(row i of C)$ — sirf $r$ numbers ki ek chhoti list.
Ise $R$ se multiply karna $R$ ke $r$ rows ko unhi $r$ numbers ko weights ke roop mein use karke mix karta hai.
Toh result — $A$ ka row $i$ — $R$ ke $r$ rows ka ek linear combination hai.

KYUN yeh mushkil direction khatam karta hai. $A$ ka har row unhi $r$ rows of $R$ se bana hai. Toh $A$ ke saare rows sirf $r$ vectors ke span ke andar rehte hain. $r$ vectors ka span ki dimension at most $r$ hoti hai:

$row rank (A) \leq r = column rank (A) .$

PICTURE. Left panel: $A$ ke columns $C$ ke $r$ columns se nikalte hain. Right panel: $A$ ke rows $R$ ke $r$ rows se nikalte hain. Ek hi chhota bottleneck $r$ dono worlds ko feed karta hai — woh shared waist hi reason hai ki dono ranks ek dusre se chain hain.

Step 6 — Doosri inequality paane ke liye matrix flip karo

KYA. Humne har matrix ke liye row rank $\leq$ column rank prove kiya. Ab wahi result transpose $A^{T}$ pe apply karo (rows columns ban jaate hain, columns rows ban jaate hain).

$row rank (A^{T}) \leq column rank (A^{T}) .$

Har side ko wapas translate karo:

$A^{T}$ ke rows hi $A$ ke columns hain ⇒ $row rank (A^{T}) = column rank (A)$ .
$A^{T}$ ke columns hi $A$ ke rows hain ⇒ $column rank (A^{T}) = row rank (A)$ .

Substitute karke:

$column rank (A) \leq row rank (A) .$

KYUN. Step 5 ne ek taraf " $\leq$ " diya; transpose karne se doosri taraf " $\leq$ " milta hai. Do opposite inequalities dono tabhi hold kar sakti hain jab do numbers equal hon.

$row rank (A) = column rank (A) = rank (A)$

PICTURE. Figure mein $A$ aur $A^{T}$ ko diagonal ke across mirror images ke roop mein dikhaya gaya hai, jisme do inequality arrows beech mein milte hain aur ranks ko pin karte hain.

Step 7 — Degenerate aur edge cases (reader ko kabhi surprise nahi hona chahiye)

KYA. Har woh corner check karo jahan argument dol sakta hai.

Case $r = 0$ (zero matrix). Koi independent columns nahi. Toh column space sirf single point ${0}$ hai, dimension $0$ . Row space bhi ${0}$ hai. Dono ranks $= 0$ ; factorization $A = C R$ degenerate ho jaati hai lekin equality $0 = 0$ hold karti hai. ✔

Case full row rank ( $r = m$ ). Saare rows independent; $C$ , $m \times m$ hai aur invertible hai, $R$ mein $m$ rows hain. Column rank bhi $= m \leq n$ . Neeche di gayi bound ke saath consistent.

Case full column rank ( $r = n$ ). Saare columns independent; kuch bhi collapse nahi hota. Yahan $rank = n \leq m$ .

Case square aur $r = n = m$ (full rank square). Tab map koi bhi direction nahi khoता — yeh invertible hai.

Universal bound. Sab cases mein,

$0 \leq rank (A) \leq min (m, n) .$

Kyun? Columns $n$ arrows hain jo $R^{m}$ mein rehte hain: tumhare paas arrows se zyada independent arrows nahi ho sakte ( $\leq n$ ), na hi us room ki dimension se zyada jisme woh rehte hain ( $\leq m$ ). Toh rank $min (m, n)$ ke neeche squeeze hoti hai.

PICTURE. $0$ se $min (m, n)$ tak ek number line jisme ek slider mark karta hai ki ek matrix ki rank kahan ho sakti hai, aur teen named landmarks (zero matrix, full column rank, full row rank) uspe pin kiye gaye hain.

Ek-picture summary

Upar sab kuch ek single diagram mein compress hota hai: matrix $A$ width $r$ ki ek waist se squeeze hoti hui. Columns waist ke left face se nikalte hain; rows right face se nikalte hain; waist ki width hi rank hai, dono ke liye shared.

Recall Poore walkthrough ki Feynman retelling

Ek matrix ko ek doorway ke roop mein imagine karo. Ek side pe tum count karo ki kitne sachche alag columns squeeze hote hain — us number ko $r$ kaho. Ab trick: tum poori matrix ko " $r$ core columns, phir ek recipe sheet jo batati hai unhe kaise mix karein" ke roop mein rebuild kar sakte ho — wahi hai $A = C R$ . Lekin us recipe sheet ko side se dekho: usmein bhi exactly $r$ rows hain, aur matrix ki har actual row bas unhi $r$ recipe-rows ka ek blend hai. Toh rows ko bhi $r$ se zyada independent directions ki zaroorat nahi pad sakti. Rows $\leq$ columns. Phir matrix ko uski diagonal pe flip karo aur identical argument chalao columns $\leq$ rows paane ke liye. Dono "less-than-or-equal" statements ek saath tabhi true ho sakti hain jab dono counts ek hi number hon — rank. Rows aur columns ne width $r$ ki waist pe haath milaye, aur wahi handshake theorem hai.

Connections

Row Space and Column Space — woh do spaces jinhein humne measure kiya.
Gaussian Elimination & Echelon Form — kyun row ops row rank preserve karte hain (Step 2).
Linear Independence and Basis — $r$ basis columns ka matlab (Step 3).
Matrix Multiplication as Linear Combination — $A = C R$ ke peeche ki reading (Steps 4–5).
Rank–Nullity Theorem — bound $rank \leq n$ aage kahan le jaati hai.
Invertible Matrix Theorem — full-rank square case (Step 7).

Visual walkthrough — Rank of a matrix — definition, row rank = column rank theorem