DEPARTMENT OF MATHEMATICS

UNIVERSITY OF MORATUWA
SRI LANKA

MA(101) Mathematics
Lecture Notes for level 1, Semester 1
2005

3. Inverse Matrices

by

T.M.J.A.Cooray

25th April 2005

 3 Inverses of a Matrix
How does one define division of matrices? .This is equivalent to asking
1
how one can define or A-1 for a matrix A.
A

Let In be the n × n identity matrix. Then AI n = I n A = A for any n × n

matrix A.
As I n acts like the number 1 in multiplication of numbers, our question
is equivalent to asking whether there exists a matrix B such that AB
= BA = I n For this to make sense, both A and B (if B exists) have to be
n × n matrices.

Definition 3.1
Let A be an n×n matrix. An n×n matrix B is called an inverse of A if
AB = BA = I n .

If A has an inverse (denoted by A −1 ), it is said to be invertible or non-

singular. Otherwise, it is said to be non-invertible or singular.

Example 3.1

 3 4  7 − 4   1 0 
(i)   =  and
 5 7  − 5 3   0 1 
 7 − 4  3 4   1 0 
  = 
 − 5 3  5 7   0 1 
 3 4  7 − 4
So   is invertible and   is its inverse.
5 7 −5 3 

 1 1  a b   a + c b + d  1 1
(ii)   =  So   is non-
 0 0  c d   0 0  0 0
invertible

Theorem 3.1

a b 
1. A 2 × 2 matrix A=  is invertible if and only if ad - bc ≠ 0.
c d 
1  d − b
2. If ad - bc ≠ , then A =
−1
  .
ad − bc  − c a 

Example 3.2

3 4
For the matrix A=  in example 3.1 (i),
5 7
ad - bc = 3× 7 - 4 ×5 = 1. So A is indeed invertible and

1  d − b  7 − 4
A =
−1
 = 
ad − bc  − c a   − 5 3 

Theorem 3.2.
Let A be an n × n matrix. If A is invertible, then the inverse of A is
unique.
Proof. Suppose P and Q are both inverses of A.
Then P = PI = P(AQ) = (PA )Q = IQ = Q .

If A is a matrix such that AQ = I and QA = I,

then A = P.

Example 3.3 Consider the product

2 3 − 4 1 − 1 1   1 0 0 
    
 0 1 − 1  1 2 2 =
  0 1 0 
 − 1 − 2 3 1 1 2   0 0 1 
144244314243 14243
P Q I

Check that QP = I too.

More Facts about Inverses.

Let A and B be n × n invertible matrices.
1. AB is invertible and (AB) −1 = B −1A −1 .

2. AT is invertible and (A T ) −1 = (A −1 ) T .
3. If AB = I n , then BA = I n .
Fact 3 says that it suffices to find the “right inverse”
of A. It will be a “two-sided inverse” of A.
Maple command for Inverse matrix of A (see example 3.4)

 2 0 −1 
 
A =  1 1 − 2
0 1 0 
 

with(LinearAlgebra):
A := < <2,1,0> | <0,1,1> |<-1,-2,0> >;
 2 0 -1
 
A := 1 1 -2
 
0 1 0

> B:=A^(-1);
 2 -1 1 
 
 3 3 3 
B :=  0 0 1 

 1 -2 2 
 
 3 3 3 

Then B= A-1

Algorithm (Finding Matrix Inverse)

Let B = (b ij ) be an n × n matrix, where the n columns are denoted by B1,

B2, . ,….Bn.

b  b  b 
 11
  12
  1n

b 21  b 22  b 2n 
B = b
1  B = b
2  ………… B = b
n 
 31
  32
  3n

 ..   ..   .. 
b n1  b n2  b nn 
Also denote the n columns of In by E1, E2, · · · , En .Finding B such that
AB=In is equivalent to solving the following n systems of linear
equations:
AB1 = E1 , AB 2 = E 2 , ......ABn = E n .

Here Bi consists of the unknown variables b1i , b 2i ,....., b ni , for the ith system
ABi =Ei.
To solve each of these systems, we would form the augmented matrix
(A | E i ) and transform it to the reduced row-echelon form. From the last

column of the reduced row-echelon form, we read off the unique

solution of Bi. Observe that all the n systems have the same coefficient
matrix A. As such, we may as well solve all of them in one procedure by
putting all the columns, Ei’s together. In other words, we augment A by
In to form the bigger augmented matrix (A|In). Now, transform this big
augmented matrix (A|In) into reduced row-echelon form by elementary
row operations. Then In is automatically transformed into B, the inverse
of A.
The elementary row operations (ERO) on a matrix are

I. Interchange two rows (which we denote by) Ri ↔R j

II. Multiply one row by a non-zero number R → αR

j j
.

III. Add a multiple of one row to another row R → αR + R

j i j

Every matrix can be brought to echelon form by a sequence of

elementary row operations.
This is achieved using the Gaussian algorithm, one version of which
is as follows:

1. If the matrix consists entirely of zeros, stop (it is in echelon form).

2. Otherwise, find the first column with a non-zero entry (say a) and
use a row interchange to bring that entry to the top row.

3. Subtract multiples of the top row from the rows below it so that each
entry below the pivot a becomes zero. (This completes the first row. All
subsequent operations are carried out on the rows below it.)

4. Repeat steps 1 - 4 on the remaining rows.

  2 0 −1 
 
Example 3. 4 Find the inverse of A = 1 1 − 2
0 1 0 
 

Solution: Define Augmented matrix (A |I ), then apply (ERO)’s

 2 0 −1 1 0 0  1 1 − 2 0 1 0
   
 1 1 − 2 0 1 0  R ↔ R  2 0 −1 1 0 0
1 2

0 1 0 0 0 1 0 1 0 0 0 1
   

1 1 − 2 0 1 0
 
0 − 2 3 1 − 2 0
R → −2 R + R 
2
0 1
1
0 0 0 1 
2

1 1 − 2 0 1 0
 
R ↔ R 0 1
2 3
0 0 0 1
0 − 2 3 1 − 2 0
 

 1 0 − 2 0 1 − 1
 
R → −R + R  0 1 0 0 0
1 2 1
1
R → 2 R + R  0 0 3 1 − 2 2 
3 2 3
 1 0 − 2 0 1 −1 
1  
R → R 0 1 0 0 0 1 
3 
3 3


 0 0 1 1 / 3 − 2 / 3 2 / 3

 1 0 0 2 / 3 −1 / 3 1 / 3 
 
 0 1 0 0 0 1 
R → 2R + R  
 0 0 1 1 / 3 − 2 / 3 2 / 3
1 3 1

2 / 3 −1/ 3 1/ 3 
 
Hence, A = 0
−1
0 1 
 1 / 3 − 2 / 3 2 / 3
 
If A is invertible, its inverse is unique. Therefore, there should be a unique solution to
each of the systems AB1 = E1 ……, ..., ABn = En. This means that when we transform
(A | E i ) into the reduced row-echelon form, A should be transformed into the identity

matrix In, and Ei is transformed into the ith column Bi of the inverse of A.
Remark
A is invertible if and only if the reduced row-echelon form of A is the identity matrix.
Maple command for Elementry row operations
with(LinearAlgebra):
A := < <2,1,0> | <0,1,1> |<-1,-2,0> >;

2 0 -1

A := 1 1 -2
 
0 1 0

B:=<<1,0,0>|<0,1,0>|<0,0,1>>;
 1 0 0
 
B :=  0 1 0
 
0 0 1

C:=<A|B>;
2 0 -1 1 0 0

C := 1 1 -2 0 1 0
 
0 1 0 0 0 1

C1 := RowOperation( C,[1,2] );
1 1 -2 0 1 0

C1 := 2 0 -1 1 0 0
 
0 1 0 0 0 1

C2 := RowOperation( C1,[2,1],-2 );
1 1 -2 0 1 0

C2 := 0 -2 3 1 -2 0
 
0 1 0 0 0 1

C3 := RowOperation( C2,[2,3] );
1 1 -2 0 1 0

C3 := 0 1 0 0 0 1
 
0 -2 3 1 -2 0

C4 := RowOperation( C3,3,1/3 );
 1 1 -2 0 1 0

0 1 0 0 0 1
C4 :=  
 -2 1 -2
0 1 0
 3 3 3 
C5 := RowOperation( C4,[3,2],2/3 );
 1 1 -2 0 1 0
0 
1 0 0 0 1 
C5 :=  
 1 -2 2 
0 0 1
3 3 3 

C6 := RowOperation( C5,[1,3],2 );
 1 2 -1 4 
1 0
 3 3 3 
C6 :=  0 1 0 0 0 1 
 
 1 -2 2 
 0 0 1 
 3 3 3 
C7 := RowOperation( C6,[1,2],-1 );
  1 2 -1 1 
 0 0
 3 3 3 
C7 :=  0 1 0 0 0 1 


 0 1 -2 2 
 0 1 
 3 3 3 

2 / 3 −1/ 3 1/ 3 
 
Hence A = 0
−1
0 1 
 1 / 3 − 2 / 3 2 / 3
 

20th April 2005