D ETERMINANTS

To every square matrix A, we can associate a scalar det A,

called the determinant of A.

1/63
D ETERMINANTS

To every square matrix A, we can associate a scalar det A,

called the determinant of A.
Determinants play an important and useful role in linear
algebra. They admit a nice geometric interpretation in
terms of volume (more on this later). As such, they arise
naturally and are useful in multivariable calculus.
Determinants of 2 × 2 or 3 × 3 matrices are easy to define:
   
a b 
  = det a b = ad − bc.
c d  c d
 
a11 a12 a13       
a21 a22 a23  = a11 a22 a23 −a12 a21 a23 +a13 a21 a22  .
       
  a32 a33  a31 a33  a31 a23 
a31 a32 a33 

Of course for a 1 × 1 matrix A = [a], one sets det A := a.

1/63
D ETERMINANTS

To every square matrix A, we can associate a scalar det A,

called the determinant of A.
Determinants play an important and useful role in linear
algebra. They admit a nice geometric interpretation in
terms of volume (more on this later). As such, they arise
naturally and are useful in multivariable calculus.
Determinants of 2 × 2 or 3 × 3 matrices are easy to define:
   
a b 
  = det a b = ad − bc.
c d  c d
 
a11 a12 a13       
a21 a22 a23  = a11 a22 a23 −a12 a21 a23 +a13 a21 a22  .
       
  a32 a33  a31 a33  a31 a23 
a31 a32 a33 

Of course for a 1 × 1 matrix A = [a], one sets det A := a.

Question: How to define det A for any n × n matrix A?
1/63
G ENERAL D EFINITION OF D ETERMINANT
There are several ways to define the determinant of an arbitrary
square matrix.
Axiomatic Approach: Determine some basic properties
that the determinant function is expected to satisfy and
then show that there exists a unique function satisfying
these basic properties.
Recursive Method: Define the determinant for a 1 × 1
matrix first (this is trivial!). Then for n > 1, define the
determinant of a n × n matrix in terms of an expression
involving determinants of (n − 1) × (n − 1) matrices. We
see recursively that the determinant of a n × n matrix
A = (aij ) will be a sum of n! terms, each consisting of a
product of n entries of the matrix and each assigned a
positive or negative sign.
Permutation Expansion: The determinant of a n × n matrix
is defined at once as an expnasion involving n! terms
indexed by the n! permutations σ of {1, 2, . . . , n}.
2/63
Axiomatic Approach
In what follows we shall use the symbol K to denote the set of
scalars. Thus K is either the set R of real numbers or the set C
of complex numbers. Recall that the rows of a n × n matrix
A = (aij ) are denoted by A1 , . . . , An . In the axiomatic approach,
we think of det A as a function D(A1 , . . . , An ) of the rows of A.
The basic properties we require of this function are as follows.
1. [Multilinearity] D is linear in each row, i.e., for any
i = 1, . . . , n, and any α, β ∈ K,

D(A1 , . . . , Ai−1 , αB + βC, Ai+1 , . . . , An )

3/63
Axiomatic Approach
In what follows we shall use the symbol K to denote the set of
scalars. Thus K is either the set R of real numbers or the set C
of complex numbers. Recall that the rows of a n × n matrix
A = (aij ) are denoted by A1 , . . . , An . In the axiomatic approach,
we think of det A as a function D(A1 , . . . , An ) of the rows of A.
The basic properties we require of this function are as follows.
1. [Multilinearity] D is linear in each row, i.e., for any
i = 1, . . . , n, and any α, β ∈ K,

D(A1 , . . . , Ai−1 , αB + βC, Ai+1 , . . . , An )

= αD(A1 , . . . , Ai−1 , B, Ai+1 , . . . , An )
+ βD(A1 , . . . , Ai−1 , C, Ai+1 , . . . , An )

2. [Alternating] If Ai = Aj for some i 6= j, then

D(A1 , . . . , An ) = 0.
3. [Normalization] D(e1 , . . . , en ) = 1. In other words,
determinant of the identity matrix should be 1.
3/63
Consequences of Multilinearity and Alternating Property

Suppose D : Mn (K) → K is any multilinear and alternating

function (of rows), i.e., D satisfies properties 1 and 2 above.
Then we can easily see that:
If A ∈ Mn (K) has a row of zeros, i.e., if Ai = 0 for some i,
then D(A) = 0, i.e., D(A1 , . . . , An ) = 0.
[Proof: Write 0 = 0 + 0 in the i th coordinate and use
multilinearity.]
If A ∈ Mn (K) and B is obtained from A by interchanging
two rows, say the i th and j th rows, i.e., if Ai = 0 for some i,
then D(B) = −D(A).
[Proof: Consider the matrix obtained from A by replacing
both i th and j th rows by Ai + Aj , i.e., consider
D(. . . , Ai + Aj , . . . , Ai + Aj , . . . ) and use properties 1 and 2.]
The last observation implies that if B is obtained from
A ∈ Mn (K) by k row interchanges, then D(B) = (−1)k D(A).

4/63
Determinant function

One has the following remarkable result.

Theorem
There exists a unique function det : Mn (K) → K satisfying
properties 1, 2 and 3 above. Moreover, if D : Mn (K) → K is any
multilinear alternating function (i.e., D satisfies 1 and 2), then

D(A) = D(e1 , . . . , en ) det A.

We omit the proof (but will be happy to explain to any interested

students outside regular class hour).
The existence part of the above thereom is usually proved by
exhibiting one explicit function that satisfies the desired
properties and here one takes recourse to the permutation
expansion. But first we need to review some facts about
permutation and in particular discuss the notion of sign of a
permutation.
5/63
Permutations
Let n be a positive integer.
Definition
A permutation of the set {1, . . . , n} of the set of first n positive
integers is a one-one onto map of the set {1, . . . , n} onto itself.
We denote by Sn the set of all permutations of {1, . . . , n}.

Permutations can be written in one-line or two-line

notation; for example,
   
1 2 3 4 5 1 2 3 4 5
43125, ,
4 3 1 2 5 1 4 3 2 5
are examples of permutations of {1, 2, 3, 4, 5}. The first
two represent the same permutation, while the third is a
“transposition” that interchanges 2 and 4 (and leaves othes
fixed).
Product of permutations is defined by composition:
στ := σ ◦ τ . Note that the product of permutations is also a
permutation. 6/63
Sign of a Permutation
To every σ ∈ Sn , we can associate a number sgn(σ) which
is either 1 or −1, called the sign of σ,and defined as
follows.
sgn(σ) := (−1)inv(σ) ,
where inv(σ) denotes the number of inversions in σ, i.e., it
is the number of pairs (i, j) such that:i < j, but σ(i) > σ(j).
Another way to describe sgn(σ) is by using the
corresponding permutation matrix Pσ . Namely,
sgn(σ) = (−1)k if Pσ is obtained from In by k row
interchanges. Equivalently,
sgn(σ) = (−1)k if σ is a product of k transpositions.
For any permutations σ, τ ∈ Sn , we have
sgn(στ ) = sgn(σ)sgn(τ ).
If D : Mn (K) → K is any multilinear alternating function,
then we see easily that
D(Aσ(1) , . . . , Aσ(n) ) = sgn(σ)D(A1 , . . . , An ) for any σ ∈ Sn .
7/63
Existence of Determinant function
As remaked earlier, the existence of a normalized, multilinear,
alternating function det : Mn (K) → K can be shown by using
the following ‘definition’.
We may then use the uniqueness to show the equivalence of
these definitions and also to derive various properties.
Definition (Permutation Expansion)
For any n × n matrix A = (aij ), the determinant of A is the scalar
X
detA = sgn(σ) a1σ(1) a2σ(2) . . . anσ(n)
σ∈Sn

where the summation is over all permutations σ of {1, . . . , n}.

It is easy to deduce from this definition is that

det A = det AT for any A ∈ Mn (K).
As a consequence, det A is multilinear and alternating in the
columns of A as well.
8/63
Properties of Determinants
Let A = (aij ) be a n × n matrix. For any i, j = 1, . . . , n, we let Aij
denote the (n − 1) × (n − 1) submatrix of A obtained from A by
removing the i th row and j th column. We define the (i, j)th
cofactor of A to be cofij (A) = (−1)i+j det Aij .
(Laplace Expansion Formula for Determinant) For any
i ∈ {1, . . . , n}, we have expansion along i th row:
n
X
det A = aij cofij (A),
j=1

and for any j ∈ {1, . . . , n}, we have expansion along j th

column:
Xn
det A = aij cofij (A).
i=1
(Product Formula for Determinant) det AB = (det A)(det B)
for any A, B ∈ Mn (K). Consequently, if A is invertible, then
det A 6= 0.
9/63
Properties of Determinants
Let A = (aij ) be a n × n matrix. For any i, j = 1, . . . , n, we let Aij
denote the (n − 1) × (n − 1) submatrix of A obtained from A by
removing the i th row and j th column. We define the (i, j)th
cofactor of A to be cofij (A) = (−1)i+j det Aij .
(Laplace Expansion Formula for Determinant) For any
i ∈ {1, . . . , n}, we have expansion along i th row:
n
X
det A = aij cofij (A),
j=1

and for any j ∈ {1, . . . , n}, we have expansion along j th

column:
Xn
det A = aij cofij (A).
i=1
(Product Formula for Determinant) det AB = (det A)(det B)
for any A, B ∈ Mn (K). Consequently, if A is invertible, then
det A 6= 0.
9/63
Adjoints and Inverses and Cramer’s Rule
The adjoint of a matrix A, denoted adj(A) is defined to be the
transpose of the cofactor matrix of A.
Lemma
For any A ∈ Mn (K),

(adj A)A = A(adj A) = diag(det A, . . . , det A) = (det A)In .

10/63
Adjoints and Inverses and Cramer’s Rule
The adjoint of a matrix A, denoted adj(A) is defined to be the
transpose of the cofactor matrix of A.
Lemma
For any A ∈ Mn (K),

(adj A)A = A(adj A) = diag(det A, . . . , det A) = (det A)In .

Corollary
If det A 6= 0, then A is invertible and A−1 = (det A)−1 (adj A).

Sketch of Proof of Lemma: Given any square matrix A, let Mik

denote the matrix obtained by replacing the i th row of A by its
k th row.

10/63
Adjoints and Inverses and Cramer’s Rule
The adjoint of a matrix A, denoted adj(A) is defined to be the
transpose of the cofactor matrix of A.
Lemma
For any A ∈ Mn (K),

(adj A)A = A(adj A) = diag(det A, . . . , det A) = (det A)In .

Corollary
If det A 6= 0, then A is invertible and A−1 = (det A)−1 (adj A).

Sketch of Proof of Lemma: Given any square matrix A, let Mik

denote the matrix obtained by replacing the i th row of A by its
k th row. Clearly det Mik = 0 for i 6= k and det Mik = det A if
the other hand, expanding by i th row we see that
i = k . On P
det Mik = nj=1 akj cofij (A).

10/63
Adjoints and Inverses and Cramer’s Rule
The adjoint of a matrix A, denoted adj(A) is defined to be the
transpose of the cofactor matrix of A.
Lemma
For any A ∈ Mn (K),

(adj A)A = A(adj A) = diag(det A, . . . , det A) = (det A)In .

Corollary
If det A 6= 0, then A is invertible and A−1 = (det A)−1 (adj A).

Sketch of Proof of Lemma: Given any square matrix A, let Mik

denote the matrix obtained by replacing the i th row of A by its
k th row. Clearly det Mik = 0 for i 6= k and det Mik = det A if
i = k . On Pthe other hand, expanding th
n Pnby i row we see that
det Mik = j=1 akj cofij (A). Hence j=1 akj cofij (A) = det(A)δik ,
i.e., A adj(A) = det(A)In . Similarly, we can show that
adj(A)A = det(A)In . 10/63
Cramer’s Rule

11/63
Cramer’s Rule

Theorem (Cramer’s Rule)

Given a system of linear equation Ax = b with an invertible
coefficient matrix A,

11/63
Cramer’s Rule

Theorem (Cramer’s Rule)

Given a system of linear equation Ax = b with an invertible
coefficient matrix A, let Mj be the matrix obtained from A by
replacing the j th column by the column b.

11/63
Cramer’s Rule

Theorem (Cramer’s Rule)

Given a system of linear equation Ax = b with an invertible
coefficient matrix A, let Mj be the matrix obtained from A by
replacing the j th column by the column b. Then

det Mj
xj = for j = 1, . . . , n.
det A
Sketch of Proof: Since A is invertible, the system Ax = b has
the unique solution x = A−1 b and det A 6= 0. From the last
corollary, A−1 = (det A)−1 adj(A), and so we see that
n n
1 X 1 X 1
xj = adj(A)jk bk = bk cofkj (A) = det Mj
det A det A det A
k =1 k =1

where the last step follows from the expansion of det Mj along
the j th column.
11/63
GEM Applied to Compute the Determinant
One of the most efficient ways of calculating determinants is
using a variant of GEM.
First observe if the effect of an elementary row operation
on the determinant of a square matrix A of size n × n.
Ri ↔Rj
I. A −→ A0 ⇒ det A0 = − det A
cR
II. A −→i A0 ⇒ det A0 = c det A
Ri +c Rj
III. A −→ A0 ⇒ det A0 = det A

Thus if A0 is a REF of A and it is obtained by elementary

row operations of type I and III only (as is usually the
case), then det A = (−1)k det A0 where k is the total
number of row interchanges made.
In particular, if a REF A0 of A has a row of zeros, then
det A = 0. Otherwise A0 is upper triangular with the pivots
p1 , . . . , pn on the diagonal. In this case det A0 = p1 p2 · · · pn
and det A = (−1)k p1 p2 · · · pn . 12/63
More on computations of determinants
We already used the following fact for upper triangular matrices.
Fact: If A is a (lower or upper) triangular square matrix, then
det A is the product of its diagonal entries.
A proof of this is easily obtained using the recursive formula
(Laplace expansion) or the permutation expansion. More
generally, for block triangular matrices, we have the following:

13/63
More on computations of determinants
We already used the following fact for upper triangular matrices.
Fact: If A is a (lower or upper) triangular square matrix, then
det A is the product of its diagonal entries.
A proof of this is easily obtained using the recursive formula
(Laplace expansion) or the permutation expansion. More
generally, for block triangular matrices, we have the following:
Theorem
 
A B
If M = is a square matrix in block upper triangular
0 D
form, where A, B, D are submatrices of appropriate sizes, then

det M = (det A)(det D).

Taking transpose, we obtain a similar result for block lower

triangular
 matrices.
 However, even when A, B, C, D are square
A B
blocks,   need not equal (det A)(det D) − (det B)(det C).
C D
13/63
Determinants and Area or Volume
The following facts may be taken as definitions or proved as
exercises using previously known definitions of area / volume.
If a = (a1 , a2 ) and b = (b1 , b2 ) are vectors in the plane R2 ,
then the area of the parallelogram with  vertices  at 0, a, b is
a1 a2
the absolute value of det A, where A = .
b1 b2
Likewise. if a = (a1 , a2 , a3 ) and b = (b1 , b2 , b3 ) and
c = (c1 , c2 , c3 ) are vectors in the 3-space R3 , then the
volume of the parallelopiped with vectors a, b, c
representing edges, is given by
 
a1 a2 a3
| det A| where A = b1 b2 b3  .
c1 c2 c3
In general, n × n determinants can be interpreted as the
hypervolume in the n-space Rn of a n-dimensional
parallelotope.
14/63