MATH4018 Introduction to Linear Algebra (I) NTU 113-1

Instructor : Kwok-Wing Tsoi

Homework 6 (Suggested Solutions)

Section A
Exercise A1 :
(a) What does it mean by the kernel, image, nullity and rank of a matrix A ?
(b) How do row operations operations correspond to multiplication of elementary row matrices ?
(c) What does it mean for a matrix to be in its REF ? How do we tell the rank and the nullity (correctly)
of a matrix in its REF ?
(d) State QAP Theorem. Use it to prove that rank(A) = rank(At ). Hence, write down as many quantities
as possible that equal rank(A).
(a) Let a matrix A ∈ Mm×n (F ). Consider the linear transformation fA : F n → F m defined by v 7→ Av.
The kernel, image, nullity and rank of the matrix A refers to that of the linear map fA .
(b) Each row operation corresponds to left multiplication by an elementary matrix.
(c) See lecture notes for the definition of REF. If A is in its REF, then
— the rank of A is the number of non-zero rows in its REF,
— the nullity of A equals the number of columns of A minus the rank of A (as a consequence of the
Rank-Nullity Theorem).
(d) For every matrix A ∈ Mm×n (F ), there are invertible matrices P , Q such that QAP is in its Smith
normal form. See lecture notes for a proof of rank(A) = rank(At ). The following quantities are equal :
rank(A), rank(At ), number of non-zero rows in REF, number of linearly independent rows of A,
number of linearly independent columns of A.

Section B
Exercise B1 : Consider the matrices
   
1 2 3 3 3 3
A = 4 5 6 and A′ = 1 2 3
7 8 9 7 8 9

Find a matrix Q such that QA = A′ .

By row operations,
     
1 2 3 4 5 6 3 3 3
r ↔r r →r −r
A = 4 5 6 1→ 2 1 2 3 1 →1 2 1 2 3 = A′
7 8 9 7 8 9 7 8 9
III
Therefore, in terms of elementary matrices, we have E−1,2,1 I
E1,2 A = A′ . Hence, we can take
    
1 −1 0 0 1 0 −1 1 0
III I
Q = E−1,2,1 E1,2 = 0 1 0 1 0 0 =  1 0 0
0 0 1 0 0 1 0 0 1

Exercise B2 : For each of the following matrices A ∈ Mm×n

 (R) : 
    1 2 0 1 1
2 1 1 2 1 2 4 1 3 0
(a) A = −1 2  (b) A = 2 1 −1 (c) A =  3

6 2 5 1
2 −1 1 5 4
−4 −8 1 3 1

(i) Transform the matrix into REF.

(ii) Write down the rank and nullity of A.
(iii) Are the rows of A linearly independent in Rn ?
(iv) Are the columns of A linearly independent in Rm ?
(v) Write down a basis for im(A).
  
2 1
(a) (i) A REF is given by 0 2.5
0 0
(ii) rank(A) = 2, nullity(A) = 0
(iii) Since rank(A) = 2 < 3, the rows of A are not linearly independent.
(iv) Since rank(A) = 2, the columns of A are linearly independent.
(v) After sifting the columns, a basis of im(A) = {(2, −1, 2)t , (1, 2, −1)t }.
 
1 2 1
(b) (i) A REF is given by 0 −3 −3
0 0 0
(ii) rank(A) = 2, nullity(A) = 1
(iii) Since rank(A) = 2 < 3, the rows of A are not linearly independent.
(iv) Since rank(A) = 2 < 3, the columns of A are not linearly independent.
(v) After sifting the columns, a basis of im(A) = {(1, 2, 1)t , (2, 1, 5)t }.
 
1 2 0 1 1
0 0 1 1 −2
(c) (i) A REF is given by  0 0 0 6 7 


0 0 0 0 2
(ii) rank(A) = 4, nullity(A) = 1
(iii) Since rank(A) = 4, the rows of A are linearly independent.
(iv) Since rank(A) = 4 < 5, the columns of A are not linearly independent.
(v) After sifting the columns, a basis of im(A) = {(1, 2, 3, −4)t , (0, 1, 2, 1)t , (1, 3, 5, 3)t , (1, 0, 1, 1)t }.

Exercise B3 : Let A be a matrix with n columns and P is an invertible n × n matrix. Prove that
rank(AP ) = rank(A). Use this to deduce that column operations do not change the rank of a ma-
trix.

Suppose P is invertible (i.e. P −1 exists). Then note that by Rank inequalities,

rank(AP ) ≤ rank(A) and rank(A) = rank((AP )P −1 ) ≤ rank(AP ).

Thus, we have rank(AP ) = rank(A).

Exercise B4 : Let A, B ∈ Mm×n (F ). Prove that rank(A) = rank(B) if and only if there exists invertible
matrices P and Q such that B = QAP .

(⇒) Suppose rank(A) = rank(B). Then A and B have the same Smith normal form. Therefore, by
QAP Theorem, there exists invertible matrices P1 , P2 , Q1 , Q2 such that

Q1 AP1 = Q2 BP2 ⇒ Q−1 −1

2 Q1 AP1 P2 = B.

Therefore, we may take Q = Q−1 −1

2 Q1 and P = P1 P2 (which are invertible).

(⇐) Suppose there exists invertible matrices P and Q such that B = QAP . Therefore, we have rank(B) =
rank(QAP ). Since multiplying by invertible matrices does not change the rank, we have

rank(B) = rank(QAP ) = rank(AP ) = rank(A).

Section C
Exercise C1 : Let A, B be two m × n matrices over a field F .
(a) Prove that im(A + B) ⊆ im(A) + im(B).
(b) Deduce that rank(A + B) ≤ rank(A) + rank(B).

(a) Let w ∈ im(A + B). By definition, it means w = (A + B)(v) for some v. Therefore,
w = (A + B)v = |{z} Bv ∈ im(A) + im(B).
Av + |{z}
∈im(A) ∈im(B)

This proves im(A + B) ⊆ im(A) + im(B).

(b) By (a), we have
dim(im(A + B)) ≤ dim(im(A) + im(B))
= dim(im(A)) + dim(im(B)) − dim(im(A) ∩ im(B))
≤ dim(im(A)) + dim(im(B)).
This proves rank(A + B) ≤ rank(A) + rank(B).
Exercise C2 : Suppose A is a square matrix such that
rank(A2 ) = rank(A).
Prove that im(A) ∩ ker(A) = {0}.
(
Av = 0
Let v ∈ im(A) ∩ ker(A). Then we have .
v = Au for some u
Combining these we have A2 u = 0 so u ∈ ker(A2 ).

Since we are given rank(A2 ) = rank(A), by Rank-Nullity Theorem, we have nullity(A2 ) = nullity(A)
as well. Observe that ker(A) ⊆ ker(A2 ). The equality on dimension asserts that ker(A) = ker(A2 ).

Therefore, we have u ∈ ker(A2 ) = ker(A) and thus Au = 0. Hence, v = Au = 0.

This proves that im(A) ∩ ker(A) = {0}.

An alternative proof. (Cheap but required deeper understanding in various concepts)

A
Consider the linear map im(A) −→ im(A2 ). By HW4 Exercise B3, this linear map is surjective. Since
we are given that dim(im(A)) = dim(im(A2 )), TFAE implies that this linear map is an isomorphism. In
particular, the only element in im(A) that gets sent to 0 under A is just 0. This is equivalent to saying
that im(A) ∩ ker(A) = {0}.
 
1 0 2 0 −2
Exercise C3 : Suppose the REF of a matrix A ∈ M3×5 (R) is 0 1 −5 0 −3. It is known that the
0 0 0 1 6
     
1 0 1
first column, second column and fourth column of A are −1, −1, −2. Find the two missing
3 1 0
columns of A. (Hint. If done correctly, this can be done in one-line with minimum calculations)
 
1 0 1
Let B = −1 −1 −2 be the matrix formed by the 1st, 2nd, 4th columns of A and it becomes
3 1 0
the identity matrix after reducing to REF. Since row operations correspond to multiplying invertible
matrices to the left, the operations involved in this case correspond to multiplying B −1 . Therefore, we
can ‘undo’ these operations by multiplying B to the left and hence the missing 3rd and 5th columns are
         
1 0 1 2 2 1 0 1 −2 4
−1 −1 −2 −5 = 3 and −1 −1 −2 −3 = −7
3 1 0 0 1 3 1 0 6 −9
respectively.