LECTURE - 29

ANJAN GUPTA

Throughout the lecture F stands for C or R.

Definition 1. Let A = (ai j ) be an m×n matrix with coefficients in F. Assume that αi = (ai1 , ai2 , . . . , ain )
 a1 j 
 
 a 
 2j
denotes the i-th row vector of A and β j =  ..  denotes the j-th column vector of A for i =
 . 
 
am j
1, 2, . . . , m and j = 1, 2, . . . , n.
The row space of A is defined to be the subspace of Fn generated by the row vectors α1 , α2 , . . . , αm
and is denoted by R(A). Therefore
R(A) = L(α1 , . . . , αm ).
The dimension of the row space R(A) is called the row rank of A, i.e.
Row rank of A = dim R(A).
The column space of A is defined to be the subspace of Fm generated by the column vectors
β1 , β2 , . . . , βn and is denoted by C(A). Therefore
C(A) = L(β1 , . . . , βn ).
The dimension of the column space C(A) is called the column rank of A, i.e.
Column rank of A = dim C(A).
The following lemma is left to the reader as an exercise.
Lemma 2. Let A be a matrix. Then row rank of A = column rank of AT .

1 2 0 1

Example 3. We consider the matrix A = 0 1 3 1. The row vectors of A are α1 = (1, 2, 0, 1),
2 0 1 2
 
1
 
2
 
0
α2 = (0, 1, 3, 1) and α3 = (2, 0, 1, 2). The column vectors of A are β1 = 0, β2 = 1, β3 = 3
2 0 1
 
1
 
and β4 = 1. Therefore the row space of A is R(A) = L(α1 , α2 , α3 ) and the column space of A is
2
C(A) = L(β1 , β2 , β3 , β4 ).

Date: 08/05/2020.

1
LECTURE - 29 ANJAN GUPTA

To compute the row rank of A we need to find the dimension of R(A). We consider the equality
xα1 + yα2 + zα3 = θ for some scalars x, y, z ∈ R. We have
x(1, 2, 0, 1) + y(0, 1, 3, 1) + z(2, 0, 1, 2) = (0, 0, 0, 0).
This yields the following system of linear equations :
x + 2z = 0
2x + y = 0
3y + z = 0
x + y + 2z = 0.
We apply elementary row operations on the coefficient matrix as follows :
1 0 2 1 0 2  1 0 2 
     
2 1 0 R02 =R2 −2R1 0 1 −4 R03 =R3 −3R2 0 1 −4 R01 =R1 − 132 R3 , R02 =R2 + 134 R3
0 3 1 −−R−0 −−−−−→ 
   −−−−−−−−→   −−−−−−−−−−−−−−−−−−→
=R 4 −R1 0 3 1  R0 =R −R
4 2
 0 0 13  R04 =R4 − 13
4
R3
  4 
 
 4 
 
1 1 2 0 1 0 0 0 4
1 0 0  1 0 0
   
0 1 0  R03 = 131 R3 0 1 0
0 0 13 −−−−−−→ 0 0 1
   
   
0 0 0 0 0 0
The equivalent system of equations is x = 0, y = 0, z = 0. Therefore the row vectors α1 , α2 , α3
are linearly independent. Consequently {α1 , α2 , α3 } is a basis of R(A) and the row rank of A is
dim R(A) = 3.
To compute the column rank of A, we need to find the dimension of the column space C(A). We
consider the equality aβ1 + bβ2 + cβ3 + dβ4 = θ for some scalars a, b, c, d ∈ R. Therefore we have
 
1
 
2
 
0
   
1 0
a 0 + b 1 + c 3 + d 1 = 0 .

 
 
 
 
 

2 0 1 2 0
Consequently we obtain the following system of linear equations :
a + 2b + d = 0
b + 3c + d = 0
2a + c + 2d = 0.
We apply elementary row operations on the coefficient matrix as follows :
11 
1 2 0 1 R0 =R3 −2R1 1 2 0 1 R =R −2R 1 0 −6 −1 R0 =R1 + 6 R3 1 0 0
      
3 1 1 2 1 13
13 

0 1 3 1 −−−−−−−−→ 0 1 3 1 −−−−−−−−→ 0 1 3 1  −−−−−−−−−→ 0 1 0 1 
 R0 =R +4R  13 
0 −4 1 0 3 3 2 0 0 13 4 R2 =R2 + 13 R3 0 0 13
    0 3  
2 0 1 2 4
1 0 0 11
 
R03 = 13
1 13
R3 

−−−−−−→ 0 1 0 13  .
1

0 0 1 134
 

2
LECTURE - 29 ANJAN GUPTA

The equivalent system of equations is

11
a+ d=0
13
1
b+ d =0
13
4
c+ d =0
13

If we put d = −13, we get a = 11, b = 1, c = 4. Therefore we have 11β1 +β2 +4β3 −13β4 = θ. This
implies that β4 = 13
11
β1 + 131 β2 + 13 4
β3 . It follows that the column space of A is C(A) = L(β1 , β2 , β3 ).
We need to see if the vectors β1 , β2 , β3 are linearly independent. We consider the equality pβ1 +
qβ2 + rβ3 = θ for some scalars p, q, r ∈ R. This is equivalent to the matrix equation :
   
 p 0
(β1 β2 β3 )  q  = 0 .
r 0

1 2 1

The matrix (β1 β2 β3 ) is 0 1 1 and its determinant is 1(1 × 2 − 1 × 0) − 0(2 × 2 − 1 × 0) + 2(2 ×
2 0 2
1 − 1 × 1) = 4 , 0. Therefore the matrix (β1 β2 β3 ) is invertible, so we must have p = q = r = 0.
Thus it follows that the vectors β1 , β2 , β3 are linearly independent. We deduce that {β1 , β2 , β3 } is a
basis of C(A) and the column rank of A is 3.
The rank of A is also 3 because the minor det(β1 β2 β3 ) is nonzero.
Question 4. Let A be an m × n matrix over F. Are row rank and column rank of A the same? If yes,
is the common value of row rank and column rank of A equal to rank of A?
First we try to find the answer for row reduced echelon matrices.
Lemma 5. Let A be a row reduced echelon matrix of size m × n over F. Then
Row rank of A = Column rank of A = rank A.
Proof. Let the i-th row of A be αi and the j-th column of A be β j for i = 1, . . . , m and j = 1, . . . , n.
Then the row space of A is R(A) = L(α1 , . . . , αm ) and the column space of A is C(A) = L(β1 , . . . , βn ).
We assume that the first r row vectors α1 , . . . , αr are nonzero row vectors and the rest are zero
row vectors. We also assume that the leading 1 of the i-th row appears in the ki -th column for
i = 1, . . . , r. Then the rank of A is r by Proposition 5, Lecture 15.
Now R(A) = L(α1 , . . . , αr ) since αr+1 = . . . = αm = θ. We consider the equality
c1 α1 + . . . + cr αr = θ.
The ki -th coordinate of αi is 1 but that of any other row vector is zero. Comparing k1 , k2 , . . . , kr
coordinates of both sides of the above equality we obtain c1 = . . . = cr = 0. Therefore the vectors
α1 , . . . , αr are linearly independent. It follows that {α1 , . . . , αr } is a basis of R(A) and the row rank
of A is r.
We observe that L(βk1 , βk2 , . . . , βkr ) ⊂ C(A) ⊂ Fr , where Fr is the subspace of Fm consisting of
column vectors whose last m − r coordinates are zero. Now βk j is a unit column vector whose j-th

3
LECTURE - 29 ANJAN GUPTA

entry is 1 and the rest are zero. This implies that {βk j : j = 1, . . . , r} is a basis of Fr . Therefore we
obtain Fr = L(βk1 , βk2 , . . . , βkr ) ⊂ C(A) ⊂ Fr . It follows that C(A) = Fr and consequently the column
rank of A is r.
Hence we prove that Row rank of A = Column rank of A = rank A = r. 
The next results are focussed on proving that row rank, column rank and rank remain invariant
under elementary row operations.
Lemma 6. Let A be an l × m matrix over F and B be an m × n matrix over the same F. Then
R(AB) ⊂ R(B) ⊂ Fn .
Proof. Let the row vectors of B be β1 , β2 , . . . , βm . Then the row space of B is R(B) = L(β1 , . . . , βm ).
Clearly R(B) ⊂ Fn . Let A = (ai j )l×m . If the row vectors of AB are α1 , . . . , αl , then they are given by
αi = ai1 β1 + ai2 β2 + . . . + aim βm , for i = 1, . . . , l.
Therefore α1 , . . . , αl ∈ L(β1 , . . . , βm ). Since L(α1 , . . . , αl ) is the smallest space containing α1 , . . . , αl ,
we obtain L(α1 , . . . , αl ) ⊂ L(β1 , . . . , βm ). Hence R(AB) ⊂ R(B) ⊂ Fn . 
Lemma 7. Let A be an invertible m × m matrix over F and B be an m × n matrix over the same F.
Then R(AB) = R(B) and row rank of AB = row rank of B.
Proof. By Lemma 6, we get R(AB) ⊂ R(B). We observe that B = A−1 (AB). Therefore by the same
lemma we obtain R(B) ⊂ R(AB). Thus we deduce R(AB) = R(B). Comparing dimension of both
sides, we obtain row rank of AB = row rank of B. 
The following two lemmas are left to the reader.
Lemma 8. Let A be an l × m matrix over F and B be an m × n matrix over the same F. Then
C(AB) ⊂ C(A) ⊂ Fl .
Lemma 9. Let A be an l × m matrix over F and B be an invertible m × m matrix over the same F.
Then C(AB) = C(A) and column rank of AB = column rank of A.
Lemma 10. Row equivalent matrices have the same row space and the same row rank.
Proof. Let A, B be row equivalent matrices. Then we obtain a product E of elementary matrices
such that B = EA. Now E is invertible, so the lemma follows by Lemma 7. 
The following is left as an exercise.
Lemma 11. Column equivalent matrices have the same column space and the same column rank.
Example 12. The column space of a matrix does not ! remain invariant under elementary row oper-
1 1
ations. For example consider the matrix A = . We observe that
0 0
!
R02 =R2 +R1 1 1
A −−−−−−−→ = B.
1 1
! !
1 1
The column space of A is C(A) = {c : c ∈ R} and that of B is C(B) = {c : c ∈ R}. Clearly
0 1
C(A) , C(B).

4
LECTURE - 29 ANJAN GUPTA
!
1 0
We consider another example. Let A = . The row rank of A is 1 and the row rank of
0 0
!
1 0
B = I2 = is 2. Therefore row rank of AB = A is not equal to row rank of B. Thus Lemma 7
0 1
fails to be true if the matrix A is not invertible.
Lemma 13. Let A be an m × n matrix over F and S be the solution space of the system of equations
AX = 0. Then
dim(S ) + Column rank of A = n.
Proof. Let Fm , Fn be vector spaces of column vectors of lengths m, n respectively. We consider the
linear transformation T : Fn → Fm given by T (α) = Aα. Then ker T = S Let e j denote the j-th
column vector of the identity matrix In of size n. Then the j-th coordinate of e j is 1 and the rest are
zero. If β j denotes the j-th column vector of A, then we have Ae j = β j for j = 1, . . . , n. Now
Im T = {Aα : α ∈ Fn } = {A(a1 e1 + . . . + an en ) : ai ∈ F for i = 1, . . . , n}
= {a1 Ae1 + . . . + an Aen : ai ∈ F for i = 1, . . . , n}
= {a1 β1 + . . . + an βn : ai ∈ F for i = 1, . . . , n}
= L(β1 , . . . , βn ) = C(A).
The result follows by the Rank-nullity theorem. 
Proposition 14. Let A be an invertible m × m matrix over F and B be an m × n matrix over the same
F. Then column rank of AB = column rank of B.
Proof. Let S , S 0 denote the solution spaces of the system of equations BX = θ and (AB)X = θ
respectively. Now by Lemma 13, we get
dim(S ) + Column rank of B = n
dim(S 0 ) + Column rank of AB = n.
Since A is invertible, we have S = S 0 . Therefore column rank of AB = column rank of B. 
The following is an obvious consequence.
Lemma 15. Row equivalent matrices have the same column rank.
Theorem 16. Let A be matrix of size m × n over F. Then
Row rank of A = Column rank of A = rank A.
Proof. The matrix A can be reduced to a row reduced echelon matrix by elementary row operations.
The row rank, column rank and rank of a matrix are invariant under elementary row operations by
Lemmas 10, 15 and Theorem 7, Lecture 16. Therefore the theorem follows from Lemma 5. 
Corollary 17. If A is a nonsingular square matrix, then row vectors of A are linearly independent
and so are the column vectors.

Department of Mathematics, Indian Institute of Science Education and Research Bhopal, Bhopal Bypass Road,
Bhopal, Madhya Pradesh, India. Pin - 462066.
E-mail address: [email protected]