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

WHAT is rank?

WHY define it three ways? Each face is useful in a different situation:

Row space tells you about equations / constraints.
Column space tells you what outputs $Ax$ are reachable.
Pivots give a computational way to find rank.

HOW to compute rank (worked first, theory after)

Worked example Compute rank by elimination

$A=\begin{pmatrix}1&2&1\\2&4&3\\3&6&4\end{pmatrix}$ Step 1: $R_2 \to R_2 - 2R_1$ , $R_3 \to R_3 - 3R_1$ . Why? Row operations don't change the row space (each new row is a combination of old rows), so they preserve rank. $\to\begin{pmatrix}1&2&1\\0&0&1\\0&0&1\end{pmatrix}$ Step 2: $R_3 \to R_3 - R_2$ . Why? Kill the duplicate to expose independence. $\to\begin{pmatrix}1&2&1\\0&0&1\\0&0&0\end{pmatrix}$ Step 3: Count pivots: positions $(1,1)$ and $(2,3)$ . So $\operatorname{rank}(A)=2$ . Why? Two nonzero rows that are independent ⇒ row space is 2-dimensional.

WHY does row rank = column rank? (Derivation from scratch)

Let $r$ = column rank of $A$ ( $m\times n$ ). We prove row rank $\le r$ , then by symmetry get equality.

Step 1 — pick a column basis. Let $\{c_1,\dots,c_r\}$ be a basis of the column space. Form $C=[c_1\;\cdots\;c_r]$ , an $m\times r$ matrix.

Why? Every column of $A$ is a combination of these $r$ vectors — that's what "basis of column space" means.

Step 2 — factor $A=CR$ . Each column $A_j$ equals $C$ times a coefficient vector. Stack those coefficient vectors as columns of an $r\times n$ matrix $R$ . Then $A = C\,R,\qquad C:m\times r,\; R:r\times n.$

Why this is the key trick? We've written $A$ as a product where the middle dimension is exactly the column rank $r$ . This factorization is forced to "leak" information about the rows too.

Step 3 — read off the rows. Row $i$ of $A$ is row $i$ of $C$ times $R$ : $(\text{row } i \text{ of } A) = (\text{row } i \text{ of } C)\,R.$ So every row of $A$ is a linear combination of the $r$ rows of $R$ .

Why does that finish it? The rows of $A$ live in the span of just $r$ vectors (the rows of $R$ ). Therefore $\text{row rank}(A) \le r = \text{column rank}(A).$

Step 4 — symmetry. Apply the same argument to $A^{T}$ . Rows of $A^{T}$ are columns of $A$ and vice versa, so: $\text{column rank}(A) = \text{row rank}(A^T) \le \text{column rank}(A^T) = \text{row rank}(A).$ Combining both inequalities: $\boxed{\;\text{row rank}(A) = \text{column rank}(A) = \operatorname{rank}(A)\;}$

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

Forecast-then-Verify

Common mistakes (Steel-manned)

Key bounds (derive, don't memorize)

Recall Feynman: explain to a 12-year-old

Imagine a matrix is a list of toy LEGO instructions, each row tells you how to build one thing. Some instructions are just copies or mixes of others — they don't teach you anything new. The rank is how many genuinely new instructions there are. The magic? Whether you count "new instructions" by looking at the rows OR by looking at the columns, you get the same number every single time. It's like counting people in a room by counting heads or counting pairs of shoes — you trust they match, and here they always do.

Flashcards

What are the three equivalent definitions of rank?

dim(row space) = dim(column space) = number of pivots in echelon form.

State the central theorem of rank.

Row rank = column rank for every matrix; this common value is the rank.

In the proof, what factorization is used?

A=CR

where

C

m\times r

(column basis),

R

r\times n

; inner dimension

r

= rank.

Why does

A=CR

prove row rank

\le

column rank?

Row

i

A

= (row

i

C

)·

R

, so all rows are combinations of the

r

rows of

R

Why is rank invariant under row operations?

Row ops replace rows by linear combinations of rows, never changing the row space (hence its dimension).

Bound on rank of an

m\times n

matrix?

0\le \operatorname{rank}(A)\le \min(m,n)

\operatorname{rank}(AB)

bound?

\le \min(\operatorname{rank}A,\operatorname{rank}B)

Where do you get a basis of the column space from?

The ORIGINAL columns in the pivot positions (not the reduced ones).

Common false claim about rank and addition?

\operatorname{rank}(A+B)=\operatorname{rank}A+\operatorname{rank}B

is FALSE; only

\le

holds.

Rank–nullity statement?

\operatorname{rank}(A)+\dim\ker(A)=n

(number of columns).

Connections

Row Space and Column Space
Gaussian Elimination & Echelon Form
Rank–Nullity Theorem
Linear Independence and Basis
Matrix Multiplication as Linear Combination
Invertible Matrix Theorem (full rank ⇔ invertible for square $A$ )

Concept Map

Hinglish (regional understanding)

Intuition Hinglish mein samjho

Dekho, rank ka matlab simple hai: matrix ek transformation machine hai, aur rank batata hai ki output mein kitni genuinely independent directions bachti hain. Agar do rows ya columns ek doosre ke multiple hain, toh woh koi nayi information nahi dete — unhe count nahi karte. Rank nikaalne ka aasaan tareeka: matrix ko row echelon form mein le jao aur pivots gino. Jitne pivots, utna rank.

Ab sabse khoobsurat baat — row rank = column rank hamesha. Yeh coincidence nahi, ek deep theorem hai. Proof ka dil hai factorization $A = CR$ , jahan $C$ ke columns column space ka basis hain (size $m\times r$ ) aur $R$ ki rows independent (size $r\times n$ ). Jaise hi humne yeh likha, har row of $A$ automatically $R$ ki $r$ rows ka combination ban jaati hai — matlab row rank bhi $r$ se zyada nahi ho sakti. Phir same cheez $A^T$ pe lagao, aur dono inequalities milke equality de deti hain.

Common galti: log original matrix ki non-zero rows gin lete hain. Galat! Non-zero rows bhi dependent ho sakti hain — echelon form mein laao tabhi sahi count milega. Doosri galti: column space ka basis reduced matrix se uthaana — nahi, pivot positions dekho aur original matrix ke wahi columns lo.

Yeh topic kyun important hai? Rank batata hai system of equations ka solution unique hai, infinite hai ya nahi (rank-nullity), aur square matrix invertible hai ki nahi (full rank ⇔ invertible). Poore Linear Algebra ka backbone hai bhai.

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

WHAT is rank?

HOW to compute rank (worked first, theory after)

WHY does row rank = column rank? (Derivation from scratch)

Forecast-then-Verify

Common mistakes (Steel-manned)

Key bounds (derive, don't memorize)

Flashcards

Connections

Concept Map

Hinglish (regional understanding)

Go deeper — visual, from zero

Test yourself — Linear Algebra (Full)

Connections