Linear Algebra 2: Lecture 16

Irena Penev

Summer 2023

Throughout this lecture, F is a fixed field.

Remark on notation: For a scalar α ∈ F \ {0}, we sometimes use the

notation α1 instead of α−1 (the multiplicative inverse of α in the field F). For
instance, in Z3 , we have 11 = 1−1 = 1 and 12 = 2−1 = 2 (because in Z3 , we
have that 2 · 2 = 1). In a similar vein, for scalars a, b ∈ F such that b ̸= 0,
we sometimes write ab instead of b−1 a. For example, in Z5 , we have that
3−1 = 2 (because 3 · 2 = 1), and so 43 = 3−1 · 4 = 2 · 4 = 3.

1 The multiplicative property of determinants

In general, for matrices A, B ∈ Fn×n and a scalar α ∈ F, we have that

det(A + B) Z
= det(A) + det(B)
 and det(αA) Z
= αdet(A).


However, as we shall see (see Theorem 1.3 below), we do have that

det(AB) = det(A)det(B).

We begin with a couple of useful propositions.

Proposition 1.1. Let A, B ∈ Fn×n . Then AB is invertible if and only if A

and B are both invertible.

Proof. If A and B are invertible, then it is clear that AB is invertible with

inverse B −1 A−1 . Suppose now thatnAB is invertible. Then
o rank(AB) = n.1
But we know that rank(AB) ≤ min rank(A), rank(B) ,2 and it follows that
rank(A) = rank(B) = n. But now A and B are both invertible.3

Proposition 1.2. Let A, E ∈ Fn×n , and assume that E is an elementary

matrix. Then det(EA) = det(E)det(A).
1
This follows from Theorem 4.1 from Lecture Notes 7.
2
This was Problem 1(b) from HW#9 from Linear Algebra 1 (winter 2022).
3
Again, this follows from Theorem 4.1 from Lecture Notes 7.

1
Proof. Let R be an elementary row operation that corresponds to the ele-
mentary matrix E, so that both the following hold:
1. E is the matrix obtained by performing R on In ;4
2. EA is the matrix obtained by performing R on A.
By Theorem 4.2 from Lecture Notes 15, there exists some scalar α ∈ F \ {0}
such that for any matrix M ∈ Fn×n , the determinant of the matrix obtained
by performing the elementary row operation R on M is αdet(M ). So,
1. det(E) = αdet(In ) = α;
2. det(EA) = αdet(A).
It follows that
det(EA) = αdet(A) = det(E)det(A),
which is what we needed to show.
We are now ready to prove the multiplicative property of determinants.
Theorem 1.3. Let A, B ∈ Fn×n . Then det(AB) = det(A)det(B).
Proof. Suppose first that at least one of A, B is non-invertible. Then by
Proposition 1.1, AB is also non-invertible. But by Theorem 5.1 from Lecture
Notes 15, non-invertible matrices have determinant zero, and so det(AB) =
0 = det(A)det(B).5
From now on, we assume that A and B are both invertible. Therefore, they
can each be written as a product of elementary matrices,6 say A = E1A . . . EpA
and B = E1B . . . EqB , where E1A , . . . , EpA , E1B , . . . , EqB are elementary matrices.
So, AB = E1A . . . EpA E1B . . . EqB . By repeatedly applying Proposition 1.2, we
get that
ˆ det(A) = det(E1A ) . . . det(EpA );7
4
Here, it is possible that E = In . In this case, we can take R to be the multiplication
of the first row by the scalar 1.
5
If A is non-invertible, then det(A) = 0, and if B is non-invertible, then det(B) = 0. In
either case det(A)det(B) = 0.
6
This follows from Corollary 5.1 of Lecture Notes 4.
7
Indeed, by repeatedly applying Proposition 1.2, we see that
det(A) = det(E1A . . . EpA )

= det(E1A )det(E2A . . . EpA )

= det(E1A )det(E2A )det(E3A . . . EpA )

..
.

= det(E1A ) . . . det(EpA ).

2
 ˆ det(B) = det(E1B ) . . . det(EqB );

ˆ det(AB) = det(E1A ) . . . det(EpA )det(E1B ) . . . det(EqB ).

But now

det(AB) = det(E1A ) . . . det(EpA ) det(E1B ) . . . det(EqB )

| {z }| {z }
=det(A) =det(B)

= det(A)det(B),

which is what we needed to show.

Corollary 1.4. Let A ∈ Fn×n be an invertible matrix. Then

det(A−1 ) = 1
det(A) .

Proof. Since AA−1 = In , we see that

det(A)det(A−1 ) = det(AA−1 ) by Theorem 1.3

= det(In )

= 1,

and consequently, det(A−1 ) = 1

det(A) .

Corollary 1.5. Let A be an orthogonal matrix in Rn×n . Then det(A) = ±1

(i.e. det(A) is either +1 or −1).

Proof. Since A is orthogonal, it satisfies AT A = In (by definition). Therefore,

(∗) (∗∗)
1 = det(In ) = det(AT A) = det(AT )det(A) = det(A)2 ,

where (*) follows from Theorem 1.3, and (**) follows from Theorem 2.2 of
Lecture Notes 15. But now we see that det(A) = ±1, which is what we
needed to show.

Warning: The converse of Corollary 1.5 is false, i.e. matrices whosedetermi-


1 1
nant is ±1 need not be orthogonal. For example, the matrix A =
2 3
satisfies det(A) = 1, but A is not orthogonal. More generally, suppose that A
is any invertible matrix in Rn×n . Then by Theorem 5.1 of Lecture Notes 15,
we have that det(A) ̸= 0. We now form the matrix B by multiplying one
1
row or one column of A by det(A) , and we see that det(B) = 1. However, B
need not be orthogonal.

3
2 Laplace expansion
Our goal in this section is to prove a formula (“Laplace explansion”) for the
determinant of a square matrix in terms of determinants of smaller matrices.
We begin with a technical proposition.
Proposition 2.1. Let A ∈ F(n−1)×(n−1) (where n ≥ 2) and a ∈ Fn−1 . Then
 A 0  
det = det(A).
aT 1 n×n
 
A 0
Proof. First, set = [ai,j ]n×n , so that all the following hold:
aT 1 n×n
ˆ A = [ai,j ](n−1)×(n−1) ;
ˆ an,n = 1;
ˆ for all i ∈ {1, . . . , n − 1}, ai,n = 0 ;
ˆ for all j ∈ {1, . . . , n − 1}, an,j is the j-th entry of the vector a.
Next, let Sn∗ be the set of all permutations σ ∈ Sn such that σ(n) = n, and for
all σ ∈ Sn∗ , let σ ∗ : {1, . . . , n − 1} → {1, . . . , n − 1} be given by σ ∗ (i) = σ(i)
for all i ∈ {1, . . . , n − 1}. So, for all σ ∈ Sn∗ , we have that σ ∗ ∈ Sn−1 , and
moreover, we have that sgn(σ ∗ ) = sgn(σ).8 Furthermore, it is clear that the
mapping σ 7→ σ ∗ is a bijection from Sn∗ to Sn−1 . We now compute:
 A 0   P
det T = sgn(σ)a1,σ(1) . . . an−1,σ(n−1) an,σ(n)
a 1 n×n σ∈Sn

(∗) P
= sgn(σ)a1,σ(1) . . . an−1,σ(n−1) an,σ(n)
∗
σ∈Sn | {z }
=1

P
= sgn(σ)a1,σ(1) . . . an−1,σ(n−1)
∗
σ∈Sn

sgn(σ ∗ )a1,σ∗ (1) . . . an−1,σ∗ (n−1)

P
=
∗
σ∈Sn

P
= sgn(π)a1,π(1) . . . an−1,π(n−1)
π∈Sn−1

= det(A),
This follows from the definition of the sign of a permutation. Indeed, fix any σ ∈ Sn∗ ,
8

and suppose that the disjoint cycle decomposition of σ has k cycles (when all cycles of
length one are included); clearly, one of those cycles is (n). Moreover, the disjoint cycle
decomposition of σ ∗ is obtained from the disjoint cycle decomposition of σ by deleting the
cycle (n); so, the disjoint cycle decomposition of σ ∗ has k − 1 cycles (when all cycles of
length one are included). But now sgn(σ) = (−1)n−k = (−1)(n−1)−(k−1) = sgn(σ ∗ ).

4
where (*) follows from the fact that for all σ ∈ Sn \ Sn∗ , we have that
a1,σ(1) . . . an−1,σ(n−1) an,σ(n) = 0.9

We now introduce some terminology and notation. Given a matrix

A = [ai,j ]n×n in Fn×n (where n ≥ 2) and indices p, q ∈ {1, . . . , n}, Ap,q is the
(n − 1) × (n − 1) matrix obtained from A by deleting the p-th row and q-th
column. For example, for the matrix
 
1 2 3
A =  4 5 6 ,
7 8 9

we have
     
5 6 2 3 2 3
ˆ A1,1 = ; ˆ A2,1 = ; ˆ A3,1 = ;
8 9 8 9 5 6
     
4 6 1 3 1 3
ˆ A1,2 = ; ˆ A2,2 = ; ˆ A3,2 = ;
7 9 7 9 4 6
     
4 5 1 2 1 2
ˆ A1,3 = ; ˆ A2,3 = ; ˆ A3,3 = .
7 8 7 8 4 5

For a matrix A ∈ Fn×n (with n ≥ 2), the determinants det(Ai,j ) (with

i, j ∈ {1, . . . , n}) are referred to as the first minors of A, whereas num-
bers Ci,j := (−1)i+j det(Ai,j ) (with i, j ∈ {1, . . . , n}) are referred to as the
cofactors of A.
We now prove a recursive formula for computing determinants in terms
of minors or cofactors. It allows us to compute the determinant of a square
matrix in terms of determinants of smaller square matrices. This formula is
called “Laplace expansion” or “cofactor expansion.”

Laplace expansion. Let A = [ai,j ]n×n (where n ≥ 2) be a matrix in Fn×n .

Then both the following hold:

(a) [expansion along the i-th row] for all i ∈ {1, . . . , n}, we have that
n
(−1)i+j ai,j det(Ai,j );
P
det(A) =
j=1

(b) [expansion along the j-th column] for all j ∈ {1, . . . , n}, we have that
n
(−1)i+j ai,j det(Ai,j ).
P
det(A) =
i=1
9
This is because for any σ ∈ Sn \ Sn∗ , there exists some i ∈ {1, . . . , n − 1} such that
σ(i) = n, and then we have that ai,σ(i) = ai,n = 0.

5
Remark: If we write Ci,j := (−1)i+j det(Ai,j ) for all i, j ∈ {1, . . . , n}
(so, the Ci,j ’s are the cofactors of A), then the formula from (a) becomes
Pn n
P
det(A) = ai,j Ci,j , and the formula from (b) becomes det(A) = ai,j Ci,j .
j=1 i=1
This is why Laplace expansion is also referred to as “cofactor expansion.”

Proof. In view of Theorem 2.2 from Lecture Notes 15, it is enough to prove (b).
Fix j ∈ {1, . . . , n}. We must show that
n
(−1)i+j ai,j det(Ai,j ).
P
det(A) =
i=1

  n
P
First, set A = a1 . . . an . Then aj = ai,j ei , and so
i=1
 
det(A) = det a1 . . . aj−1 aj aj+1 an

 n

P
= det a1 . . . aj−1 ai,j ei aj+1 an
i=1

(∗) n
P  
= ai,j det a1 . . . aj−1 ei aj+1 an ,
i=1

where (*) follows from Proposition 3.1(a) from Lecture Notes 15. Fix an
arbitrary index i ∈ {1, . . . , n}. To complete the proof, it now suffices to show
that
 
det a1 . . . aj−1 ei aj+1 an = (−1)i+j det(Ai,j ).

By iteratively performing n − j column swaps on the matrix

 
Bi := a1 . . . aj−1 ei aj+1 an ,

we can obtain the matrix

 
Ci := a1 . . . aj−1 aj+1 an ei .

By iteratively performing n − i row swaps on the matrix Ci , we can obtain

the matrix  
Ai,j 0
,
aT 1
where aT is the row vector of length n − 1 obtained from the i-th row of A
by deleting its j-th entry.10 Since swapping two rows or two columns has
10
T
So, aT = , and in particular Fn−1 .

ai,1 ... ai,j−1 ai,j+1 ... ai,n

6
the effect of changing the sign of the determinant, we see that

det(Bi ) = (−1)n−j det(Ci )

 A 0

i,j
= (−1)n−j (−1)n−i det
aT 1

(∗)
= (−1)2n−i−j det(Ai,j )

= (−1)i+j det(Ai,j )

where (*) follows from Proposition 2.1. This completes the argument.

Example 2.2. Consider the matrix

 
2 0 1
A =  3 4 5 ,
7 0 8

with entries understood to be in R. Compute det(A) in two ways:

(a) via Laplace expansion along the third row;

(b) via Laplace expansion along the second column.

Solution. (a) We compute:

2 0 1
det(A) = 3 4 5
7 0 8

0 1 2 1 2 0
= (−1)3+1 7 + (−1)3+2 0 + (−1)3+3 8
4 5 3 5 3 4

0 1 2 0
= 7 +8
4 5 3 4
| {z } | {z }
=−4 =8

= 36.

7
(b) We compute:

2 0 1
det(A) = 3 4 5
7 0 8

3 5 2 1 2 1
= (−1)1+2 0 + (−1)2+2 4 + (−1)3+2 0
7 8 7 8 3 5

2 1
= 4
7 8
| {z }
=9

= 36.

Exercise 2.3. Compute the determinant of the matrix

 
1 2 0 −1 −2
 3 −4 0 −2 −1 
 
A =   1 2 2 0 1 ,

 1 0 0 0 2 
2 −1 0 1 3

with entries understood to be in R.

Solution. As a general rule, it is best to expand along a row or column that

has a lot of zeros (if such a row or column exists), since that reduces the
amount of calculation that we need to perform. In the calculation below, the
row or column along which we are about to expand is in red (to facilitate

8
reading).

1 2 0 −1 −2
3 −4 0 −2 −1
det(A) = 1 2 2 0 1
1 0 0 0 2
2 −1 0 1 3

1 2 −1 −2
expansion
3 −4 −2 −1
= (−1)3+3 2 along 3rd
1 0 0 2
column
2 −1 1 3

1 2 −1 −2
3 −4 −2 −1
= 2
1 0 0 2
2 −1 1 3

 2 −1 −2  expansion
= 2 (−1)3+1 1
−4 −2 −1 + along 3rd
−1 1 3 row
1 2 −1 
+(−1)3+4 2 3 −4 −2
2 −1 1

 2 −1 −2 1 2 −1 
= 2 −4 −2 −1 −2 3 −4 −2
−1 1 3 2 −1 1
| {z } | {z }
=−11 −25

= 78,

where the determinants of the two 3 × 3 matrices from the second-to-last line
can be obtained in various ways: Laplace expansion, elementary row/column
operations, or our diagram for computing determinants of 3 × 3 matrices
(described in section 2 of Lecture Notes 15).

Example 2.4. Compute the determinant of the matrix

 
1 −1 2
A =  −2 4 1 ,
3 −3 5

with entries understood to be in R.

9
Solution. We combine various methods for computing determinants, as fol-
lows:
1 −1 2
det(A) = −2 4 1
3 −3 5

1 0 2
C2 →C2 +C1
= −2 2 1
3 0 5

expansion
1 2
= (−1)2+2 2 along 2nd
3 5
| {z } column
=−1

= −2.

3 Cramer’s rule
Before stating Cramer’s rule, we set up some notation. For a matrix A ∈ Fn×n ,
a vector b ∈ Fn , and an index j ∈ {1, . . . , n}, we denote by Aj (b) the matrix
obtained from A by replacing the j-th column of A with b. For example, if
   
1 1 1 4
A =  0 2 2  and b =  5 ,
0 0 3 6

then
 
4 1 1
ˆ A1 (b) = 5 2 2
 ;
6 0 3
 
1 4 1
ˆ A2 (b) =  0 5 2 ;
0 6 3
 
1 1 4
ˆ A3 (b) =  0 2 5 .
0 0 6

Cramer’s rule. Let A be an invertible matrix in Fn×n , and let b ∈ Fn .

Then the matrix-vector equation Ax = b has a unique solution, namely
h iT
det(A1 (b)) det(A2 (b)) det(An (b))
x = det(A) det(A) . . . det(A) .

10
Proof. Since A is invertible, we know that the matrix-vector equation Ax = b
has a unique solution, namely, x = A−1 b. Now, for this solution x, we set
 T
x = x1 . . . x n . Our goal is to show that
h iT
det(A1 (b)) det(A2 (b)) det(An (b))
x = det(A) det(A) ... det(A) .

Fix an index j ∈ {1, . . . , n}. We must show that

det(Aj (b))
xj = det(A) .
 
Set A =Pn a1 . . . an . Using the fact that Ax = b and the fact that
Ax = i=1 xi ai (by the definition of matrix-vector multiplication), we
compute:
 
det(Aj (b)) = det a1 . . . aj−1 b aj+1 . . . an
 
= det a1 . . . aj−1 Ax aj+1 . . . an

 n

P
= det a1 . . . aj−1 xi ai aj+1 . . . an
i=1

(∗) n
P  
= xi det a1 . . . aj−1 ai aj+1 . . . an
i=1

(∗∗)
 
= xj det a1 . . . aj−1 aj aj+1 . . . an

= xj det(A),

where (*) follows from Proposition 3.1(a) of Lecture Notes 15, and (**)
follows from the fact that for all i ∈ {1, . . . , n} \ {j}, the matrix
 
a1 . . . aj−1 ai aj+1 . . . an

has two identical columns and therefore (by Proposition 2.4 of Lecture
Notes 15) has determinant zero. We have now shown that

det(Aj (b)) = xj det(A).

Since A is invertible, Theorem 5.1 of Lecture Notes 15 guarantees that

det(A) ̸= 0. So, we can divide both sides of the equality above by det(A) to
obtain
det(Aj (b))
xj = det(A) .

This completes the argument.

11
Example 3.1. Let
  

2 1 0 1
A =  0 2 2  and b =  1 ,
1 1 1 0

with entries understood to be in Z3 . Solve the matrix-vector equation Ax = b.

Solution. Note that det(A) = 2, and in particular, A is invertible (by Theo-

rem 5.1 from Lecture Notes 15). So, Cramer’s rule applies. We compute:

1 1 0
ˆ det(A1 (b)) = 1 2 2 = 2;
0 1 1

2 1 0
ˆ det(A2 (b)) = 0 1 2 = 1;
1 0 1

2 1 1
ˆ det(A3 (b)) = 0 2 1 = 0.
1 1 0
By Cramer’s rule, Ax = b has a unique solution, namely
h iT
det(A1 (b)) det(A2 (b)) det(A3 (b))
x = det(A) det(A) det(A)

 2 1 0
T
= 2 2 2

 T
= 1 2 0 .

4 The adjugate matrix

Given a matrix A ∈ Fn×n (with n ≥ 2), with cofactors Ci,j = (−1)i+j det(Ai,j )
(for i, j ∈ {1, . . . , n}), the cofactor matrix of A is the matrix [Ci,j ]n×n . The
adjugate matrix (also called the classical adjoint) of A, denoted by adj(A), is
the transponse of the cofactor matrix of A, i.e.

adj(A) := [Ci,j ]n×n T .

So, the i, j-th entry of adj(A) is the cofactor Cj,i (note the swapping of the
indices).

12
Example 4.1. Consider the matrix
 
1 1 1
A =  0 2 2 ,
0 0 3
with entries understood to be in R. Compute the cofactor and adjugate
matrices of A.
Solution. For all i, j ∈ {1, 2, 3}, we let Ci,j = (−1)i+j det(Ai,j ). (So, the
Ci,j ’s are the cofactors of A.) We compute:
2 2
ˆ C1,1 = (−1)1+1 = 6;
0 3

0 2
ˆ C1,2 = (−1)1+2 = 0;
0 3

0 2
ˆ C1,3 = (−1)1+3 = 0;
0 0

1 1
ˆ C2,1 = (−1)2+1 = −3;
0 3

1 1
ˆ C2,2 = (−1)2+2 = 3;
0 3

1 1
ˆ C2,3 = (−1)2+3 = 0;
0 0

1 1
ˆ C3,1 = (−1)3+1 = 0;
2 2

1 1
ˆ C3,2 = (−1)3+2 = −2;
0 2

1 1
ˆ C3,3 = (−1)3+3 = 2.
0 2
So, the cofactor matrix of A is
   
C1,1 C1,2 C1,3 6 0 0
 C2,1 C2,2 C2,3  =  −3 3 0 .
C3,1 C3,2 C3,3 0 −2 2
The adjugate matrix of A is the transpose of the cofactor matrix, i.e.
 
6 −3 0
adj(A) =  0 3 −2  .
0 0 2

13
Theorem 4.2. Let A be an invertible matrix in Fn×n . Then
A−1 = 1
det(A) adj(A).

Proof. Since A is invertible, Theorem 5.1 of Lecture Notes 15 guarantees

1
that det(A) ̸= 0, and in particular, the expression det(A) adj(A) is defined.
−1
We will prove the theorem by showing that matrices A and det(A) 1
adj(A)
have the same corresponding entries. Fix indices i, j ∈ {1, . . . , n}. By the
1
definition of adj(A), we see that the i, j-th entry of the matrix det(A) adj(A)
is
(−1)j+i det(Aj,i )
det(A) .
We will use Cramer’s rule to show that this is also the i, j-th entry of the
matrix A−1 . 
Set A−1 = a∗1 . . . a∗n . Since AA−1 = In , we have that


A a∗1 . . . a∗n
   
= e1 . . . en ,

and consequently (by the definition of matrix-vector multiplication), that

Aa∗1 . . . Aa∗n
   
= e1 . . . en .

In particular, the two matrices above have the same j-th column, and so
Aa∗j = ej , i.e. a∗j is the solution of the equation Ax = ej (this solution is
unique because A is invertible). So, by Cramer’s rule, we have that
det(An (ej )) T
h i
det(A1 (ej ))
a∗j = det(A) . . . det(A)
.

The i-th entry of a∗j is

det(Ai (ej ))
det(A) .
By Laplace explansion along the i-th column, we get that

det(Ai (ej )) = (−1)j+i det(Aj,i ).

So, the i-th entry of a∗j (which is precisely the i, j-th entry of A−1 ) is
(−1)j+i det(Aj,i )
det(A) ,

which is what we needed to show.

Example 4.3. Show that the matrix

 
1 1 1
A =  0 2 2 ,
0 0 3

(with entries understood to be in R) is invertible, and using Theorem 4.2,

find its inverse A−1 .

14
Solution. The matrix A is upper triangular, and so its determinant can be
computed by multiplying the entries along the main diagonal. So, det(A) =
1 · 2 · 3 = 6. Since det(A) ̸= 0, Theorem 5.1 from Lecture Notes 15 guarantees
that A is invertible. In Example 4.1, we compute the adjugate matrix of A:
 
6 −3 0
adj(A) =  0 3 −2  .
0 0 2

So, by Theorem 4.2, we have that

A−1 = 1
det(A) adj(A)

 
6 −3 0
1
= 6 0 3 −2 
0 0 2
 
1 −1/2 0
=  0 1/2 −1/3  .
0 0 1/3

Corollary 4.4. Let a, b, c, d ∈ F. Then the matrix

 
a b
A =
c d

is invertible if and only if ad ̸= bc, and in this case, the inverse of A is given
by the formula  
−1 1 d −b
A = ad−bc .
−c a
Proof. By Theorem 5.1 from Lecture Notes 15, we know that A is invertible
if and only if det(A) ̸= 0. Clearly, det(A) = ad − bc, and it follows that A is
invertible if and only if ad − bc ̸= 0, i.e. if and only if ad ̸= bc.
Now, assume that A is invertible, so that ad ̸= bc. We first compute the
cofactors Ci,j of A:

ˆ C1,1 = (−1)1+1 det(A1,1 ) = d;

ˆ C1,2 = (−1)1+2 det(A1,2 ) = −c;

ˆ C2,1 = (−1)2+1 det(A2,1 ) = −b;

ˆ C2,2 = (−1)2+2 det(A2,2 ) = a.

15
The cofactor matrix of A is
   
C1,1 C1,2 d −c
= .
C2,1 C2,2 −b a

The adjugate matrix of A is the transpose of the cofactor matrix, i.e.

 
d −b
adj(A) = .
−c a

By Theorem 4.2, we now have that

 
d −b
A−1 = 1
det(A) adj(A) = 1
ad−bc ,
−c a

which is what we needed to show.

16