Lecture IV.

Determinants

4-1. Basic Definition

a b 
For 2  2 matrix A    , the determinant of A is defined to be ad-bc. To see the definition of the

c d
 

determinant for general n  n matrices, let us first consider the 3  3 case.

a  1 5 3 
 11 a12 a13  
 
Example 4.1.1: For A  a21 a22 a23   2 4 1 , we have
   
a 31 a 32 a 33   0 2 0 
 
 4 1 2 1 2 4 
det A  1  det    5  det 

  3  det 
0 0 

 0 2   2 .
 2 0
     

Actually, the above equation can be expressed as

a22 a23  a21 a23  a21 a22 

det A  (1)11  a11  det a   (1)12  a12  det    (1)1 3  a13  det    2
 32 a 33  a 31 a 33  a 31 a 32 

Hence, the determinant of a 3  3 matrix can be computed based on the determinants of its 2  2
sub-matrices. □

A recursive definition of the determinant of a general n  n n  2 matrix follows.

Definition 4.1.2: For n  2 , the determinant of a square matrix A  aij   nn , denoted by det A, is

the sum of n terms of the form 11 j a1 j det A 1 | j  , where A 1 | j   n 1n 1 is obtained by

deleting the first row and the jth column of A. More precisely,
n
det A=  11 ja1 j det A 1 | j  (4.1.1)
j 1

Definition 4.1.3: 11 j det A 1 | j  is called the (1,j)th cofactor of A. In general, 1i  j det A i | j  ,

where A i | j   n 1n 1 is formed by deleting the ith row and jth column of A, is called the (i,j)th

cofactor of A, which hereafter is denoted by cij. □

1
Theorem 4.1.4: Let A  nn , then

n n
det A   akjckj k th row expansion   ailcil l th column expansion .
j 1 i 1

That is, det A can be evaluated by expanding along any row or any column. □

Remark: The formula (4.1.1) is simply the expansion along the first row. □
1 0 1
0
 
3 1 2 2 

Example 4.1.5: Consider A    . To find det A, perhaps the easiest way is to expand
1 0 2 1 
 
2 0 0 1
 
along the 2nd column, which contains the most zero entries. □

Example 4.1.6: Check that

1 2 51
 
 4 2 2 
0
A   1  4  (2)  (6) .
0 0 2 1 
 
0 0 0 6 
 
The determinant of a lower (or upper) triangular matrix equals the product of the main diagonal entries
(exercise). □

4-2. Important Properties of Determinants

Property 4.2.1: Let A  nn . Then det A  det AT .

[Proof]: Expanding det A by 1st row of A is the same as expanding det AT by the 1st column of AT . □

Property 4.2.2: If any row or any column of A is zero, then det A  0 . □

Property 4.2.3: Suppose a   . Then det aA  an det A for A  nn .

[Proof]: By induction. For n=1, the assertion is true. Suppose it is true for n  k  1 , i.e., for any

B  k 1k 1 det aB  ak 1 det B . Then for n=k,

2
 k
det aA   aa1 j  1, j  - cofactor of aA 
j 1
k
  aa1 j  11 j det 
aA 1 | j 

(by definition of cofactor)
j 1
dimension k -1k -1
k
  aa1 j  11 j ak -1 det A 1 | j  (by assumption made about n  k  1)
j 1
k k
  aka1 j  11 j det A 1 | j   ak  a1 j  11 j det A 1 | j 
j 1 j 1
k
 a det A .

Property 4.2.4: Let A  nn and r  r1 r2  rn  be a row vector. Also let
 

 M be the matrix obtained by replacing the ith row of A by r.

 B be the matrix obtained by adding r to the ith row of A.

Then det B  det A  det M .

n
[Proof]: det B   aij  rij cij cij is the i, j  th cofactor of A
j 1
n n
  aijcij   rijcij  det A  det M . □
j 1 j 1

 1 7 5  1 7 5  1 7 5 
     
     
Example: Verify that det  2 0 3   det 2 0 3  det 2 0 3  . □
     
1  0 4  1 7  (1) 1 4 7   0 1 1
     

Property 4.2.5: If A has two rows (or columns) that are identical, then det A=0.

a b 
[Proof]: The proof is done by induction. The statement is true for n=2, since det    0 . Suppose

a b 

that the claim holds for n  k  1 . Then for n=k, expand det A along any row or column other than the

two equal rows or columns gives det A=0. This is because every cofactor involves the determinant of an

(k  1)  (k  1) matrix which contains two identical rows (or columns) and is zero by assumption. The

statement is thus true for n=k. □

3
  1 1 0 7
 
2 3 4 5

Example: Consider A    . Let us evaluate det A via expansion along the second row:
 1 1 0 7
 
1 5 4 3
 

 1 1 0 7
 
2 3 4 5

det A   
 1 1 0 7
 
1 5 4 3
 
1 0 7   1 0 7  1 1 7   1 1 0 
      
       
 (2)  det 1 0 7   3  det  1 0 7   4  det  1 1 7   5  det  1 1 0 .
       
5 4 3 1 4 3 1 5 3 1 5 4
       
□

Property 4.2.6: If B is obtained from A by interchanging two rows (or columns) of A, then we have

det B   det A .

[Proof]: Suppose we interchange the ith row ri and the jth row rj of A to get B, i  j . Denote by h, g

the determinant of the matrix obtained from A by replacing the ith row of A by h and jth row of A by g.

Therefore det A  ri , rj and det B  rj , ri . We know that h, h  0 for all h (from Property (4.2.5))

and that
h  p, g  h, g  p, g 

 from Property 4.2.4 .
h, p  g  h, p  h, g 


Hence
0  ri  rj , ri  rj  ri , ri  rj  rj , ri  rj  ri , ri  ri , rj  rj , ri  rj , rj
   

 0 

0 

 ri , rj  rj , ri  det A  det B,

which implies det B   det A . □

Property 4.2.7: If B is obtained from A by multiplying the ith row of A by a scalar a , then

det B  a det A .

[Proof]: Expand det B along the ith row.

4
Property 4.2.8: If B is obtained from A by adding a multiple of one row (or one column) of A to another

row (or column) of A, then det B  det A .

[Proof]: Suppose that the ith row ri of A is replaced by ri  a rj to get B. Then we have

det B  det A  det M , where M is obtained from A by replacing ri by arj (Prop. 4.2.4). Since

 
   
arj   rj 
   
det M  det     a  det   0,
 rj   rj 
   
 
   

a matrix with two identical rows.

we have det B  det A . □
a   a 
 11 a12 a13   11 a12 a13 
Example: Consider A  a21 a22 a23  , B  aa  a
 11 21 aa12  a22 aa13  a23  .
   
a 31 a 32 a 33   a 31 a 32 a 33 

a a12 a13  a 
 11  11 a12 a13 

Then det B  det A  det aa11 aa12 aa13   det A  a det a11 a12 a13   det A . □
   
 a 31 a 32 a 33 
 a 31 a 32 a 33 

0

For A, B  nn , a celebrated formula regarding the determinant of product is det AB  det A det B ,
the proof of which is far from trivial, even though nothing is of surprise as ab  det ab   det a det b
when specialized to the scalar case of n  1 . Yet we are not empty-handed when extending to n  1 ,
bearing in mind that oftentimes we would be able to say something using elementary matrices. Hence let’s
begin with elementary matrices of three kinds; afterwards the general case will follow.

Determinants of elementary matrices: Recall three kinds of elementary matrices.

Pij : interchange the ith and the jth rows of I.

Si : multiply the ith row of I by a scalar a .

Tij : add a scalar multiple of the ith row of I to the jth row of I.

We have

5
 det Pij   det I  1, from Property 4.2.6
det Si  a det I  a, from Property 4.2.7
det Tij  det I  1, from Property 4.2.8

More importantly, for any A  nn ,


det Pij A   det A  det Pij det A
 Prop. 4.2.6
det S A  a det A  det S det A
i i
 Prop. 4.2.7
det T A  det A  det T det A
 ij ij
 Prop. 4.2.8

We thus conclude that

det EA  det E det A for any elementary matrix E !

That is, the determinant of the product of an elementary matrix (of any kinds) and a square matrix
A  nn equals to the product of respective determinants.

Property 4.2.9: A is nonsingular if and only if det A  0 (A is singular iff det A  0 ).

[Proof]: (  ) Since A is nonsingular, A  E1E2  El , where Ei is an elementary matrix (why ? see

Lecture 3). We then have

det A  det E1E2  El  det E1 det E2  det El  0 .

(  ) If A is singular, (  AX  I has no solution), in G H , the reduced row echelon form of  A I ,
   

G must has at least one zero row. Since G H  E1E2  Ek  A I , we have G  E1E2  Ek A , and
   
0  det G  det E1E2  Ek A  det E1  det Ek det A .

0

This implies det A  0 . □

Property 4.2.10: Let A, B  nn . Then det AB  det A det B .

[Proof]: If A is nonsingular, it follows A  E1E2  Ek and

det AB  det E1E2  Ek B  det E1  det Ek det B  det E 2  Ek det B  det A det B .
1E

A

If A is singular, by following the previous arguments we can put A  EG , where E is the product of a

sequence of elementary matrices (hence det E  0 ) and G contains zero rows at its bottom. This implies

AB  EGB , and the square matrix GB has zero rows at the bottom (why). We then have

6
 det AB  det EGB  det E det GB  0

0
det A= det E det
 G  0,  det A det B  0  det AB . □
0
1
Corollary 4.2.11: If A is nonsingular, then det A1  .
det A
[Proof]:  AA1  I, det AA1  det A det A1  1 . □

Note: In general,
det M1  M2   det M1  det M2 , for arbitrary M1 and M2 .

1 2   3 1
Ex., Take M1    and M  
 2

1 3 . Then det M1  det M2  9  det M1  M2   23 . □
2 5   

4-3. Adjoint of Matrices and Representation of Inverse

Motivation: For a square matrix A, we can use Gauss elimination (or elementary row operations) to solve

the linear equation Ax  b or to find the inverse of A (if it exists). A closed-form expression of the

solution A1b or the inverse A1 can be obtained through elementary matrix multiplication. The
determinant can be used for providing alternative closed-form expressions for A1b or A1 , directly
in terms of the entries of A . This is the main focus of this section.

A. Representation of Inverse

To proceed, suppose A  aij   nn and let 1  i  n be fixed. Let B be obtained by replacing the

jth row of A by the ith row of A ( i  j ):

a11   a1n  a11   a1n 

   
     
   
 ai 1   ain   ai 1   ain 
 
   
A   , B     
   
a j 1   a jn   ai 1   a in 
   
       
 
a   ann  a   ann 
 n 1   n 1 

7
Then B has two identical rows and hence det B  0 (by Property 4.2.5). Let us evaluate det B by

expanding along the jth row:

c j 1 
n n  
0  det B   bjkc jk   aikc jk  ai1  ain     , for i  j , (4.3.1)
k 1 k 1
  
c jn 

where c jk be the (jk)th cofactor of A. Equation (4.3.1) says that the sum of the products of the entries of

one row of A and cofactors of another row of A equals zero.

Example:
 1 1 0 7  1 1 0 7
   
2 3 4 5 2 3 4 5
 
A  B   (replacing the 3 row of A by the 1 row)
rd st
 1 2 3 6  1 1 0 7
   
1 5 4 3 1 5 4 3
   
4
0  det B   b3kc3k (expanding along the 3rd row)
k 1
n
 1  c31  
a11
-1  c32  0  c33  7  c34 
a13 a14
 aikc jk □
a12 k 1

On the other hand, by definition of determinant we have

ci1 
n  
det A   aikcik  ai1  ain     . (4.3.2)
k 1
  
cin 

Note that, for a fixed 1  i  n , (4.3.1) defines n  1 equations, which can be combined with equation
(4.3.2) to get

c11  ci 1,1 ci,1 ci 1,1  cn,1 

   
a  a         0  0 det A 0  0 

i1 in  
    
    the i th position 
ith row of A c1n  ci 1,n ci,n ci 1,1  cn ,n 
 

:= adjoint of A

 
 i th row of A   adjoint of A  det A   0  0
 1 0  0

 the i th position 
:Ai  
:ei , i th row of the identity matrix In

(4.3.3)

8
By collecting equation (4.3.3) for all 1  i  n we have the key relation: Denoting ri  1n the ith

row of A ,

 r1   e1 
   
    adjoint of A  det A    
    (4.3.4)
   
r
 n   en 
 
A I

Remarks:

(a) We often use the shorthand adj(A) to denote the adjoint of A.

(b) Since A  adjA  det A  I (from (4.3.4), we have

1
A1  adjA whenever det A  0 . (4.3.5)
det A
(c) Equation (4.3.5) provides an elegant representation of the inverse. However, (4.3.5) is seldom used in
practice since it involves the computations of determinants of matrices (this is not preferred in
numerical computations).

B. Representation of Solution to Ax  b : Cramer’s Rule

Suppose A  nn is nonsingular, i.e. det A  0 . Equation (4.3.5) implies that the solution to
Ax  b is
1
x  A1b  adjA b . (4.3.6)
det A
Let us write
 x1  b 
   1 c11  cn 1 
x2  b   
 2     .
x    , b    , and adjA   
 
   
x  bn  c1n  cnn 
 n   
With (4.3.6) we immediately have
 n 
  c j 1bj 
 j 1 
 x1  n 
   
x2    c j 2bj 
   det A  j 1 
1
 
    
x   
 n  n 
 c b 
  jn j 
 j 1 

9
In particular,
b a  a1n 
 1 12
1
n
1  
x1  det A  c j 1bj  det A det A1 , where A1       
j 1  
bn an 2  ann 
 
1
In general, we have x k  det A det Ak , where Ak is obtained from A by replacing the kth column by

b. Hence the solution is given by (the so-called Cramer’s rule)

 det A  a 
 1  11  a1,i 1 b1 a1,i 1  a1,n 
1    
x  det A    , where Ai         , 1  i  n.
   
 det An  an 1  an,i 1 bn an,i 1  an,n 
  
replace the i th column of A by b

1 0 2  x1   6 
  
    
Example: Solve 3 4 6 x 2  = 30 by the Cramer’s rule.
    
1 2 3 x 3   8 
    
  
A b

 6 0 2 1 6 2 1 0 6 
   
     
det A  44  0 , A1   30 4 6 , A2  3 30 6 , A3  3 4 30
     
 8 2 3  1 8 3   1 2 8 
     
det A1 40 10 det A2 72 18 det A3 152 38
x1    , x2    , x3    . □
det A 44 11 det A 44 11 det A 44 11

The following conditions are equivalent for a square matrix A  nn to be nonsingular:
(1) The inverse of A exists.
(2) Ax  0 iff x  0 .
(3) Ax  b has a unique solution for every b .
(4) The reduced row echelon form of A is the identity matrix.
(5) A can be expressed as a product of elementary matrices.
(6) det A  0 .

10
 4-4. Geometric Interpretations of Determinants

a 0  0 a b b 
 ad    ad  
0 d d  0 d d 

a  a 
   
 0   0 

a b b 
 ad  bc  
c d d 

a 
 
c 

c d

a b
The area of the parallelogram equals

 (a  b)  c   (c  d )  b 
(a  b)  (c  d )  2     2    ad  bc .
 2   2 
a b 
In general, the absolute value of the determinant of the 2  2 matrix   is equal to the area of the

c d 

a  b 
parallelogram determined by column vectors c  and   .
  d
 

a b  a b  a b  a ka 
Note: If   is singular, then det 


c d   0 . In this case we must have
   
c d   c kc  , and hence
c d     
 

the two column vectors are along the same line in the two-dimensional plane.

11