≻≻≻≻—≻≻≻≻—Lecture Eight —≺≺≺≺—≺≺≺≺

Chapter Three: Cofactor Expansion

 
Recall. Let A = aij be an n × n matrix. We define the determinant of A by det(A)=
|A| = Σσ∈Sn ǫ(σ)a1σ(1)a2σ(2) · · · anσ(n), where Sn is the collection of all permutations of Jn =
{1, 2, . . . , n}. Furthermore, we proved
(a) the determinants of a matrix and its transpose are equal.
(b) if B is obtained from A by multiplying a row (column) of A by a real number c, then
det(B)= c det(A).
(c) if matrix B results from matrix A by interchanging two rows (columns) of A, then
det(B)= −det(A).
(d) if B is obtained from A by adding to the sth row (column) of A a constant c times the
rth row (column), r 6= s, of A, then det(B)=det(A).
(e) if a matrix is upper triangular (lower triangular, diagonal), then its determinant is the
product of the elements on the main diagonal.

Linear Algebra lecture 8−1

Theorem 3.1.8. The determinant of a product of two matrices is the product of their
determinants; that is det(AB) = det(A)det(B).

◦
     
Proof. 1 Denote An×n = aij , Bn×n = bij and AB = Cn×n = cij .
Then cij = Σnk=1aik bkj for all 1 ≤ i, j ≤ n.
2◦ Let Sn be the collection of all permutations of Jn = {1, 2, . . . , n}.
By definition of determinant, we have
det (C) = Σσ∈Sn ǫ(σ)c1σ(1)c2σ(2) · · · cnσ(n)
= Σσ∈Sn ǫ(σ)(Σnk1=1a1k1 bk1σ(1))(Σnk2=1a2k2 bk2σ(2)) · · · (Σnkn=1ankn bknσ(n))
= Σσ∈Sn ǫ(σ)Σnk1,k2,...,kn=1(a1k1 bk1σ(1))(a2k2 bk2σ(2)) · · · (ankn bknσ(n))
= Σnk1,k2,...,kn=1Σσ∈Sn ǫ(σ)(a1k1 a2k2 · · · ankn )(bk1σ(1)bk2σ(2) · · · bknσ(n))
= Σnk1,k2,...,kn=1(a1k1 a2k2 · · · ankn )(Σσ∈Sn ǫ(σ)(bk1σ(1)bk2σ(2) · · · bknσ(n))).
Linear Algebra lecture 8−2
  
Rowk1 (B)
 Row (B)
k2
Proof. 3◦ For each fixed k1, k2, . . . , kn, let D = dij = 
 
. Then the term
 
 .
. 
Rowkn (B)
Σσ∈Sn ǫ(σ)(bk1σ(1)bk2σ(2) · · · bknσ(n)) = Σσ∈Sn ǫ(σ)(d1σ(1)d2σ(2) · · · dnσ(n)) = detD.
When any two of numbers k1, k2, . . . , kn are equal,
one has det(D)= 0 as D has two equal rows.
Furthermore, when all k1, k2, . . . , kn are not equal,
then k1 = τ (1), k2 = τ (2), . . . , kn = τ (n) for some τ ∈ Sn.
It follows that det(D)= ǫ(τ )det(B)
because the determinant changes sign whenever two rows are exchanged
and ǫ(τ ) tells even or odd transpositions needed to obtain the permutation τ .
4◦ By 2◦ and 3◦, one gets
det (C) = Στ ∈Sn (a1τ (1)a2τ (2) · · · anτ (n))(ǫ(τ )det(B))
= det(B)Στ ∈Sn ǫ(τ )(a1τ (1)a2τ (2) · · · anτ (n))
= det(B)det(A).
We complete the proof.
Linear Algebra lecture 8−3
 1
Corollary 3.1.2. If A is nonsingular, then det(A)6= 0 and det(A−1 )= .
det(A)

Proof. Skip.

Linear Algebra lecture 8−4

  
Definition. Let A = aij be an n × n matrix. For 1 ≤ i, j ≤ n, let Mij be the
(n − 1) × (n − 1) submatrix of A obtained by deleting the ith row and jth column of A.
The determinant det(Mij ) is called the minor of aij and the cofactor Aij of aij is defined as
Aij = (−1)i+j det(Mij ).

 
3 −1 2
Example 3.2.1. Let A = 4 5 6. Then, for example,
 
7 1 2
4 6 3 −1 −1 2
     
(1) the submatrices M12 = , M23 = , and M31 = .
7 2 7 1 5 6
(2) The minor of a12 is det(M12)= −34, the minor of a23 is det(M23)= 10, and the minor of
a31 is det(M31)= −16.
(3) Also the cofactor of a12 is A12 = (−1)1+2det(M12) = (−1)(−34) = 34, the cofactor of
a23 is A23 = (−1)2+3det(M23) = (−1)(10) = −10, and the cofactor of a31 is A31 =
(−1)3+1det(M31) = (+1)(−16) = −16.
Linear Algebra lecture 8−5
  
Theorem 3.2.9. (Cofactor expansion, Laplace expansion) Let A = aij be an n × n
matrix. Then
(a) (expansion of det(A) along the rth row)
for each 1 ≤ r ≤ n, det(A) = ar1Ar1 + ar2Ar2 + · · · + arnArn.
(b) (expansion of det(A) along the sth column)
for each 1 ≤ s ≤ n, det(A) = a1sA1s + a2sA2s + · · · + ansAns.

Linear Algebra lecture 8−6

  
1 2 −3 4
 −4 2 1 3
Example 3.2.2. Please evaluate the determinant of A =   using the
 
 3 0 0 −3
2 0 −2 3
cofactor expansion.

" #! " #!
2 −3 4 1 2 −3
det(A) = 3 det 2 1 3 − (−3) det −4 2 1 ←− expansion along 3rd row
0 −2 3 2 0 −2
    
= 3 −(−2) det 22 43 + 3 det 22 −31 ←− expansion along 3rd row
    
+3 −2 det −4 1 + 2 det 1 −3
2 −2 2 −2
↑ expansion along 2nd column

= 3 [2 × (−2) + 3 × 8] + 3 [(−2) × 6 + 2 × 4] = 48.

Linear Algebra lecture 8−7

  
1 2 −3 4
 −4 2 1 3
Example 3.2.3. Please evaluate the determinant of A =  .
 
 3 0 0 −3
2 0 −2 3
    
1 2 −3 4 ×(−1) 1 2 −3 4
det (A) = det  −43 20 10 −3 3 ← − +  = det −5 0 4 −1

   3 0 0 −3
2 0 −2 3 2 0 −2 3
" # ! " #!
−5 4 −1 ×(−3) ×3 −5 4 −1
= −2 det 3 0 −3 ← −+ = −2 det 18 −12 0
2 −2 3 ←−−−−− + −13 10 0
 
18 −12 = (−2) × (−1) × 24 = 48.
= (−2) × (−1) det −13 10

Linear Algebra lecture 8−8

  
a11 a12 a13
Note. Consider a 3 × 3 matrix A = a21 a22 a23. Then
 
a31 a32 a33
det(A) = a11a22a33 − a12a21a33 − a13a22a31 − a11a23a32 + a12a23a31 + a13a21a32
(1) : expansion along the second row
= −a21(a12a33 − a13a32) + a22(a11a33 − a13a31) − a23(a11a32 − a12a31)
= a21A21 + a22A22 + a23A23
(2) : expansion along the third column
= +a13(a21a32 − a22a31) − a23(a11a32 − a12a31) + a33(a11a22 − a12a21)
= a13A13 + a23A23 + a33A33.

Linear Algebra lecture 8−9

Proof. 1◦ The second formula follows from the first by the fact that det(A) = det(AT ).
Thus we only give the proof for first expansion.
◦
 
2 Denote An×n = aij .
Given any fixed 1 ≤ r ≤ n, we express
det(A) = Σσ∈Sn ǫ(σ)a1σ(1) · · · arσ(r) · · · anσ(n)
n


= Σk=1 Σ{σ∈Sn|σ(r)=k}ǫ(σ)a1σ(1) · · · ark · · · anσ(n)
n


= Σk=1ark Σ{σ∈Sn|σ(r)=k}ǫ(σ)a1σ(1) · · · a(r−1)σ((r−1))a(r+1)σ((r+1)) · · · anσ(n) .

Linear Algebra lecture 8−10

Proof. 3◦ Now for each fixed k and σ ∈ Sn such that σ(r) = k,
 
we denote D(n−1)×(n−1) = dij = Mrk resulting from A
by deleting the rth row and kth column of A
 a related permutation τ ∈ Sn−1 by
and define


 σ(i) if σ(i) < k for i = 1, . . . , (r − 1)

 σ(i) − 1 if σ(i) > k for i = 1, . . . , (r − 1)
τ (i) = .


 σ(i + 1) if σ(i + 1) < k for i = r, . . . , (n − 1)
 σ(i + 1) − 1 if σ(i + 1) > k for i = r, . . . , (n − 1)


Since σ is a permutation of Jn = {1, . . . , n} and σ(r) = k,

τ is a permutation on Jn−1 = {1, . . . , (n − 1)}.
Also note ǫ(σ) = (−1)|k−r|ǫ(τ ) because σ can be obtained by
first performing |k − r| transposition that move k to location r
and then ignoring k while applying τ to other elements.
On the other hand, given any τ ∈ Sn−1, there exists uniquely σ ∈ Sn
satisfying σ(r) = k and the above relationship between σ and τ .
Thus, The correspondence between σ and τ gives a bijection
between the sets {σ ∈ Sn|σ(r) = k} and Sn−1.

Linear Algebra lecture 8−11

Proof. 3◦ Furthermore, note a1σ(1) · · · a(r−1)σ((r−1))a(r+1)σ((r+1)) · · · anσ(n)
= d1τ (1) · · · d(r−1)τ ((r−1))drτ (r) · · · d(n−1)τ ((n−1)).
Consequently, Σ{σ∈Sn|σ(r)=k}ǫ(σ)a1σ(1) · · · a(r−1)σ((r−1))a(r+1)σ((r+1)) · · · anσ(n)
= Στ ∈Sn−1 (−1)|k−r|ǫ(τ )d1τ (1) · · · d(n−1)τ ((n−1))
= (−1)r+k det(D) = Ark .
4◦ From 2◦ and 3◦, we obtain


det(A) = Σnk=1ark Σ{σ∈Sn|σ(r)=k}a1σ(1) · · · a(r−1)σ((r−1))a(r+1)σ((r+1)) · · · anσ(n) .
= Σnk=1ark Ark

The proof is complete.

Linear Algebra lecture 8−12

  
Definition. Let A = aij be an n × n matrix. The n × n matrix adj A, called the adjoint
of A, is the matrix whose i, jth element is the cofactor Aji of aji. Namely,
 
A11 A21 · · · An1
 12 A22 · · · An2 
A 
adj A =  . .. ..  .
 . 
A1n A2n · · · Ann  
3 −1 2
Example 3.2.1. Consider A = 4 5 6. Then, for example,
 
7 1 2
4 6 −1 2
   
(1) the submatrices M12 = and M31 = .
7 2 5 6
(2) The minor of a12 is det(M12)= −34 and the minor of a31 is det(M31)= −16.
(3) Also the cofactor of a12 is A12 = (−1)1+2(−34) = 34 and the cofactor of a31 is A31 =
(−1)3+1(−16) = −16.  
4 4 −16
Then the adjoint of A is adj A =  34 −8 −10.
 
−31 −10 19
Linear Algebra lecture 8−13
  
Lemma. Given an n × n matrix A = aij , denote Aij the cofactor of aij for 1 ≤ i, j ≤ n.
Let c1, c2, . . . , cn be real number. Then,
(a) for 1 ≤ r ≤ n, c1Ar1 + c2Ar2 + · · · + cnArn = det(C),
 
a11 a12 · · · a1n
 . .. ... .. 
 . 
   
where C = cij =  c1 c2 · · · cn 

 ←rth row.
 .. .. ... .. 
 
an1 an2 · · · ann

(b) for 1 ≤ s ≤ n, c1A1s + c2A2s + · · · + cnAns = det(D),

↓sth column 

a11 · · · c1 · · · a1n
    a21 · · · c2 · · · a2n 
where D = dij =  . .

 . . . . .. ... .. 

an1 · · · cn · · · ann

Linear Algebra lecture 8−14

Proof of (a). 1◦ Note for 1 ≤ j ≤ n,
the cofactor Crj of crj for matrix C and
the cofactor Arj of arj for matrix A are equal
because the corresponding entries of A and C are equal
except those in the rth row.
Therefore, along with the cofactor expansion for determinant,
det(C) = c1Cr1 + c2Cr2 + · · · + cnCrn
= c1Ar1 + c2Ar2 + · · · + cnArn .

Linear Algebra lecture 8−15

  
Theorem 3.2.10. If A = aij is an n × n matrix and denote Aij the cofactor of aij ,
then
(
det(A) if i = r
(a) for 1 ≤ s ≤ n, ai1Ar1 + ai2Ar2 + · · · + ainArn = .
0 if i 6= r
(
det(A) if j = s
(b) for 1 ≤ r ≤ n, a1j A1s + a2j A2s + · · · + anj ans = .
0 if j 6= s

Proof. 1◦ This theorem simply follows the lemma above,

the cofactor expansion of determinant, and the fact that
the determinant of a matrix having two equal rows (columns) is 0.

Linear Algebra lecture 8−16

  
Theorem 3.2.11. If A = aij is an n × n matrix, then A(adj A) = (adj A)A =
det(A)In .

Proof. 1◦ Note the i, j element of

 
a11 a12 · · · a1n
 
 21 a22 · · · a2n  A11 A21 · · · Aj1 · · · An1
a 
 . .. .. 
 .  A12 A22 · · · Aj2 · · · An2 
A(adj A) =   
. .. .. .. 
· · · in   .
a a a 
 i1 i2 
 . . .
 . . .  A1n A2n · · · Ajn · · · Ann


an1 an2 · · · ann

is ai1Aj1 + ai2Aj2 + · · · + ainAjn
which is 0 if i 6= j and det(A) if i = j by the previous theorem.
It follows A(adj A) = det(A)In .
2◦ Similar argument gives (adj A)A = det(A)In.

Linear Algebra lecture 8−17

Corollary 3.2.3. If A is an n × n matrix and det (A) 6= 0, then A is nonsingular and
1
A−1 = det(A) (adj A).

Theorem. Let A be an n × n matrix. The following statements are equivalent.

(a) A is nonsingular.
(b) x = 0 is the only solution to Ax = 0.
(c) A is row equivalent to In .
(d) The linear system Ax = b has a unique solution for every n-vector b.
(e) det(A) 6= 0.
Proof. The equivalences between statements (a)-(d), please see lecture 4.
As to (a)⇔(e), it follows Corollary 3.1.2 and Corollary 3.2.3.

Linear Algebra lecture 8−18

Theorem 3.2.13. (Cramer’s Rule) Consider the following linear system
a11x1 + a12x2 + · · · + a1nxn = b1
a21x1 + a22x2 + · · · + a2nxn = b2
.. .. .. .. ..
an1x1 + an2x2 + · · · + annxn = bn
 
of n equations in n unknowns. If the coefficient matrix An×n = aij has a nonzero
determinant, then the system has the unique solution
det(A1 ) det(A2) det(An)
x1 = det(A) , x2 = det(A) , . . . , xn = det(A) ,
 
b1
where Aj is the matrix obtained from A by replacing the jth column of A by b =  .. .
 
bn

Linear Algebra lecture 8−19

 1
Proof. 1◦ If det(A)6= 0, then A is nonsingular and A−1 = det(A)
(adj A).
 
x1
Consequently, the solution for the linear system Ax = b, where x =  .. ,
 
xn
1
is uniquely derived to be x = A−1b = det(A) (adj A)b.
1
Note then for each 1 ≤ j ≤ n, xj = det(A)
(A1j b1 + A2j b2 + · · · + Anj bn),
det(Aj )
which equals det(A) by lemma on page 14.

Remark. The Cramer’s rule is applicable only to the case where we have the same number
of equations and unknowns as well as the determinant of the coefficient matrix is nonzero.
Furthermore, it is much less efficient than the Gaussian elimination or the Gauss-Jordan
reduction. However, the Cramer’s rule yields a formula for finding the solution sometimes
which may be useful for further manipulating the solution.

Linear Algebra lecture 8−20