Linear Algebra II

Richard Earl

Hilary Term 2024

0. INTRODUCTION AND PRELIMINARY MA-
TERIAL

0.1 Syllabus

Introduction to determinant of a square matrix: existence and uniqueness. Proof of existence

by induction. Proof of uniqueness by deriving explicit formula from the properties of the
determinant. Permutation matrices. (No general discussion of permutations). Basic properties
of determinant, relation to volume. Multiplicativity of the determinant, computation by row
operations. [2]

Determinants and linear transformations: de…nition of the determinant of a linear transforma-

tion, multiplicativity, invertibility and the determinant. [0.5]

Eigenvectors and eigenvalues, the characteristic polynomial, trace. Eigenvectors for distinct
eigenvalues are linearly independent. Discussion of diagonalization. Examples. Eigenspaces,
geometric and algebraic multiplicity of eigenvalues. Eigenspaces form a direct sum. [2.5]

Gram-Schmidt procedure. Spectral theorem for real symmetric matrices. Quadratic forms and
real symmetric matrices. Application of the spectral theorem to putting quadrics into normal
form by orthogonal transformations and translations. [3]

0.2 Reading list

(1) T. S. Blyth and E. F. Robertson, Basic Linear Algebra (Springer, London, 2nd edition
2002).
(2) C. W. Curtis, Linear Algebra – An Introductory Approach (Springer, New York, 4th edition,
reprinted 1994).
(3) R. B. J. T. Allenby, Linear Algebra (Arnold, London, 1995).
(4) D. A. Towers, A Guide to Linear Algebra (Macmillan, Basingstoke 1988).
(5) S. Lang, Linear Algebra (Springer, London, Third Edition, 1987).
(6) R. Earl, Towards Higher Mathematics - A Companion (Cambridge University Press, Cam-
bridge, 2017)

INTRODUCTION AND PRELIMINARY MATERIAL 1

0.3 Introduction

Towards the end of the Linear Algebra I course, it was explained how a linear map T : V ! V
can be represented, with respect to a choice of basis, by a square n £n matrix where n = dim V.
When we make a choice of basis fe1 , . . . , en g for V, then a vector v 2 V becomes represented
by a unique co-ordinate vector (c1 , . . . , cn ) 2 Rn such that
v = c1 e1 + ¢ ¢ ¢ + cn en
and T becomes represented by the matrix A = (aij ) where
T ei = a1i e1 + ¢ ¢ ¢ + ani en .
Note that the co-ordinates of T ei are the entries of the ith column of A.
Given the same linear map T can be represented by in…nitely many di¤erent matrices, at
least two questions arise:
² What do these di¤erent matrices have in common, given they represent the same linear
map?
² Is there a best matrix representative – for purposes of computation or comprehension –
amongst all these di¤erent matrices?
The second question will lead us to a discussion of eigenvectors and diagonalizability. Should
we be able to …nd a matrix representative that is diagonal, then many calculations will be
considerably simpler. If this is possible, and it is an ‘if’, then the linear map is said to be
diagonalizable and the vectors in the basis are called eigenvectors. An eigenvector is a non-zero
vector v such that T v = λv for some scalar λ known as the eigenvalue of v.
Returning to the …rst question, we shall …nd that all the algebraic properties of T apply to
each of its matrix representatives. If A and B are two matrices representing T then there is an
invertible matrix P such that
A = P ¡1 BP.
We can then show that each matrix representative has the same determinant, trace, rank,
nullity, eigenvalues and functional properties – e.g. T is self-inverse. Any calculation we make,
pertaining to the algebra of T , reassuring yields the same answer. The matrix P is a change of
basis matrix providing an invertible change of variable.
However, the same cannot be said of geometric properties of T. In general, an invertible
change of variable will alter lengths, angles, areas, volumes etc.. If, say, we wished to change
variables to show a curve that isn’t in normal form – such as
x2 + xy + y 2 = 1
– is in fact an ellipse, and determine its area, then we need to ensure that the area remains
invariant under the change of co-ordinates. The matrices that preserve the scalar product –
and so preserve angle, distance, area – are the orthogonal matrices. That is, P ¡1 = P T . It is an
easy check to see that the only matrices which might be diagonalized by an orthogonal change
of variable are the symmetric matrices. At the end of the course we meet the important spectral
theorem which shows the converse: symmetric matrices can be diagonalized by an orthogonal
change of variable.

INTRODUCTION 2
0.4 Notation

(v1 j v2 j ¢ ¢ ¢ j vn ) denotes the m £ n matrix with columns v1 , v2 , . . . , vn 2 Rm col

(r1 / r2 / ¢ ¢ ¢ / rm ) denotes the m £ n matrix with rows r1 , r2 , . . . , rm 2 Rn
e1 , e2 , . . . , en denotes the canonical basis for Rn .
LA denotes, for an m £ n matrix A, the map Rncol ! Rm col given by x 7! Ax.
Mi (λ) denotes the ERO that multiplies the ith row by λ 6= 0.
Sij denotes the ERO that swaps the ith and jth rows.
Aij (λ) denotes the ERO that adds λ £ (row i) to row j.
diag(α1 , . . . , αn ) denotes the diagonal n £ n matrix with entries α1 , . . . , αn .
[A]ij denotes the (i, j)th entry of a matrix A.

NOTATION 3
1. DETERMINANTS.

1.1 De…nitions

A square matrix has a number associated with it called its determinant. There are various
di¤erent ways of introducing determinants, each of which has its advantages but none of which
is wholly ideal as will become clearer below. The de…nition we shall use is an inductive one,
de…ning the determinant of an n £ n matrix in terms of (n ¡ 1) £ (n ¡ 1) determinants. Quite
what the determinant of a matrix signi…es will be discussed shortly in Remark 11.

Notation 1 Given a square matrix A and 1 6 I, J 6 n, we write AIJ for the (n ¡ 1) £ (n ¡ 1)

matrix formed by removing the Ith row and the Jth column from A.

Example 2 Let 0 1
1 ¡3 2
A=@ 0 7 1 A.
¡5 1 3
Then (a) removing the 2nd row and 3rd column or (b) removing the 3rd row and 1st column,
we get µ ¶ µ ¶
1 ¡3 ¡3 2
(a) A23 = ; (b) A31 = .
¡5 1 7 1

Our inductive de…nition of a determinant is then:

De…nition 3 The determinant of a 1 £ 1 matrix (a11 ) is simply a11 itself. The determinant
det A of an n £ n matrix A = (aij ) is then given by

det A = a11 det A11 ¡ a21 det A21 + a31 det A31 ¡ ¢ ¢ ¢ + (¡1)n+1 an1 det An1 .

Notation 4 The determinant of a square matrix A is denoted as det A and also sometimes as
jAj . So we may also write the determinant of the matrix A in Example 2 as
¯ ¯
¯ 1 ¡3 2 ¯
¯ ¯
¯ 0 7 1 ¯¯.
¯
¯ ¡5 1 3 ¯

Proposition 5 The determinants of 2 £ 2 and 3 £ 3 matrices are given by the following for-
mulae.
(a) For 2 £ 2 matrices µ ¶
a11 a12
det = a11 a22 ¡ a12 a21 .
a21 a22

DETERMINANTS. 4
(b) For 3 £ 3 matrices
0 1
a11 a12 a13
det @ a21 a22 a23 A = a11 a22 a33 + a12 a23 a31 + a13 a21 a32 ¡ a12 a21 a33 ¡ a13 a22 a31 ¡ a11 a23 a32 .
a31 a32 a33
(1.1)

Proof. (a) Applying the above inductive de…nition, we have det A11 = det(a22 ) = a22 and
det A21 = det(a12 ) = a12 , so that
µ ¶
a11 a12
det = a11 det A11 ¡ a21 det A21 = a11 a22 ¡ a12 a21 .
a21 a22

(b) For the 3 £ 3 case

¯ ¯ ¯ ¯ ¯ ¯
¯ a a ¯ ¯ a a ¯ ¯ a a ¯
det A11 = ¯¯ 22 23 ¯,
¯ det A21 = ¯¯ 12 13 ¯,
¯ det A31 = ¯¯ 12 13 ¯,
¯
a32 a33 a32 a33 a22 a23

so that
0 1
a11 a12 a13 ¯ ¯ ¯ ¯ ¯ ¯
¯ a22 a23 ¯ ¯ a12 a13 ¯ ¯ a12 a13 ¯
det @ a21 a22 a23 A = a11 ¯¯ ¯ ¡ a21 ¯ ¯ ¯
¯ a32 a33 ¯ + a31 ¯ a22 a23 ¯
¯
a32 a33 ¯
a31 a32 a33
= a11 (a22 a33 ¡ a23 a32 ) ¡ a21 (a12 a33 ¡ a13 a32 ) + a31 (a12 a23 ¡ a13 a22 )

using the formula for 2 £ 2 determinants. This rearranges to (1.1).

Example 6 Let Rθ and Sθ be the rotation and re‡ection matrices

µ ¶ µ ¶
cos θ ¡ sin θ cos 2θ sin 2θ
Rθ = , Sθ = .
sin θ cos θ sin 2θ ¡ cos 2θ

Rθ represents rotation by θ anti-clockwise about the origin and Sθ represents re‡ection in y =

tan θ. Note, for any θ, that

det Rθ = cos2 θ + sin2 θ = 1, det Sθ = ¡ cos2 2θ ¡ sin2 2θ = ¡1.

Example 7 Returning to the matrix from Example 2, we have

¯ ¯
¯ 1 ¡3 2 ¯¯
¯
¯ 0 7 1 ¯¯ = 1| £{z7 £ 3} + (¡3) £ 1 £ (¡5) + 2| £{z0 £ 1}
¯ | {z }
¯ ¡5 1 3 ¯ 21 0
15
¡1| £{z 1 £ 1} ¡ (¡3) £ 0 £ 3 ¡ 2 £ 7 £ (¡5)
| {z } | {z }
1 0 ¡70
= 105.

DEFINITIONS 5
Remark 8 In the 2 £ 2 and 3 £ 3 cases, but only in these cases, there is a simple way to
remember the determinant formula. The 2 £ 2 formula is the product of entries on the left-to-
right diagonal minus the product of those on the right-to-left diagonals. If, in the 3 £ 3 case,
we allow diagonals to ‘wrap around’ the vertical sides of the matrix – for example as below
0 1 0 1
& .
@ & A, @ . A,
& .

– then from this point of view a 3 £ 3 matrix has three left-to-right diagonals and three right-to-
left. A 3 £ 3 determinant then equals the sum of the products of entries on the three left-to-right
diagonals minus the products from the three right-to-left diagonals. This method of calculation
does not apply to n £ n determinants when n > 4.

De…nition 9 Let A be an n £ n matrix. Given 1 6 I, J 6 n the (I, J)th cofactor of A,

denoted CIJ (A) or just CIJ , is de…ned as CIJ = (¡1)I+J det AIJ and so the determinant det A
can be rewritten as
det A = a11 C11 + a21 C21 + ¢ ¢ ¢ + an1 Cn1 .

Proposition 10 Let A be a triangular matrix. Then det A equals the product of the diagonal
entries of A. In particular it follows that det In = 1 for any n.

Proof. This is left to Sheet 1, S3.

Remark 11 (Summary of Determinant’s Properties) As commented earlier, there are

di¤erent ways to introduce determinants, each with their own particular advantages and disad-
vantages.

² With De…nition 3, the determinant of an n £ n matrix is at least unambiguously and

relatively straightforwardly given. There are other (arguably more natural) de…nitions
which require some initial work to show that they’re well-de…ned. For example, we shall
see that det has the following algebraic properties
(i) det is linear in the rows (or columns) of a matrix (see Theorem 13(A)).
(ii) if a matrix has two equal rows then its determinant is zero (see Theorem 13(B)).
(iii) det In = 1.
In fact, these three algebraic properties uniquely characterize a function det which assigns
a number to each n £ n matrix (Proposition 26). As a consequence of this uniqueness it
also follows that
(¤) det AT = det A for any square matrix A (see Corollary 20).
The problem with the above approach is that the existence and uniqueness of such a func-
tion are still moot.

DEFINITIONS 6
 ² Using De…nition 3 we avoid these issues, but unfortunately we currently have no real
sense of what the determinant might convey about a matrix. The determinant of a 2 £ 2
matrix is uniquely characterized by the two following geometric properties. Given a 2 £ 2
matrix A, with associated map LA , it is then the case that
(a) for any region S of the xy-plane, we have
area of LA (S) = jdet Aj £ (area of S). (1.2)

(b) The sense of any angle under LA is reversed when det A < 0 but remains the same
when det A > 0.
We can demonstrate (a) and (b) by noting that the Jacobian of
µ ¶ µ ¶µ ¶ µ ¶
x a b x ax + by
LA = =
y c d y cx + dy
equals ¯ ¯
∂(f1 , f2 ) ¯¯ a b ¯¯
=¯ .
∂ (x, y) c d ¯
These two properties best show the signi…cance of determinants. Thinking along these
lines, the following properties should seem natural enough:
(α) det AB = det A det B (Corollary 19).
(β) a square matrix is singular if and only it has zero determinant (Corollary 18).
However, whilst these geometric properties might better motivate the importance of de-
terminants, they would be less useful in calculating determinants. Their meaning would
also be less clear if we were working in more than three dimensions (at least until we had
de…ned volume and sense/orientation in higher dimensions) or if we were dealing with
matrices with complex numbers as entries.
² The current de…nition appears to lend some importance to the …rst column; De…nition
3 is sometimes referred to as expansion along the …rst column. From Sheet 1, P1 one
might (rightly) surmise that determinants can be calculated by expanding along any row
or column (Theorem 28).
² Finally, calculation is di¢cult and ine¢cient using De…nition 3. (For example, the for-
mula for an n £ n determinant involves the sum of n! separate products (Propositions 26
and 27(b)). We shall, in due course, see that a much better way to calculate determinants
is via EROs. This method works well with speci…c examples but less well in general as too
many special cases arise; if we chose to de…ne determinants this way, even determining
the general formulae for 2 £ 2 and 3 £ 3 determinants would become something of a chore.
In the following we rigorously develop the theory of determinants. These proofs are often
technical and not particularly illuminating and only a selection of the proofs will be covered
in lectures. I’d suggest the signi…cant properties of determinants are (i), (ii), (iii), (¤), (a),
(b), (α), (β) above and these should be committed to memory. The next signi…cant result
(or method) appears in Remark 21 where we begin the discussion of calculating determinants
e¢ciently.

DEFINITIONS 7
Notation 12 (a) We shall write (r1 / ¢ ¢ ¢ /rn ) for the n £ n matrix with rows r1 , ¢ ¢ ¢ , rn 2 Rn .
(b) We shall write (v1 j ¢ ¢ ¢ jvn ) for the n £ n matrix with columns v1 , ¢ ¢ ¢ , vn 2 Rncol .
(c) We shall write e1 , . . . , en for the standard basis of Rn .

Theorem 13 The map det de…ned in De…nition 3 has the following properties.
(A) det is linear in each row. That is, det C = λ det A + µ det B where

A = (r1 / ¢ ¢ ¢ /ri¡1 /ri /ri+1 / ¢ ¢ ¢ /rn ),

B = (r1 / ¢ ¢ ¢ /ri¡1 /v/ri+1 / ¢ ¢ ¢ /rn ),
C = (r1 / . . . /ri¡1 /λri + µv/ri+1 / . . . /rn ).

(B) If A = (r1 / ¢ ¢ ¢ /rn ) with ri = rj for some i 6= j, then det A = 0.

(B’) If the matrix B is produced by swapping two di¤erent rows of A then det B = ¡ det A.

Before proceeding to the main proof we will …rst prove the following.

Lemma 14 Together, properties (A) and (B) are equivalent to properties (A) and (B’).

Proof. Suppose that det has properties (A), (B). Let A = (r1 / ¢ ¢ ¢ /rn ) and B be produced by
swapping rows i and j where i < j. Then

0 = det(r1 / ¢ ¢ ¢ /ri + rj / ¢ ¢ ¢ /ri + rj / ¢ ¢ ¢ /rn ) [by (B)]

= det(r1 / ¢ ¢ ¢ /ri / ¢ ¢ ¢ /ri + rj / ¢ ¢ ¢ /rn ) + det(r1 / ¢ ¢ ¢ /rj / ¢ ¢ ¢ /ri + rj / ¢ ¢ ¢ /rn ) [by (A)]
= fdet(r1 / ¢ ¢ ¢ /ri / ¢ ¢ ¢ /ri / ¢ ¢ ¢ /rn ) + det(r1 / ¢ ¢ ¢ /ri / ¢ ¢ ¢ /rj / ¢ ¢ ¢ /rn )g
+ fdet(r1 / ¢ ¢ ¢ /rj / ¢ ¢ ¢ /ri / ¢ ¢ ¢ /rn ) + det(r1 / ¢ ¢ ¢ /rj / ¢ ¢ ¢ /rj / ¢ ¢ ¢ /rn )g [by (A)]
= f0 + det Ag + fdet B + 0g [by (B)]
= det A + det B

and so property (B’) follows. Conversely, if det has properties (A), (B’) and ri = rj for i 6= j
then
det(r1 / ¢ ¢ ¢ /ri / ¢ ¢ ¢ /rj / ¢ ¢ ¢ /rn ) = det(r1 / ¢ ¢ ¢ /rj / ¢ ¢ ¢ /ri / ¢ ¢ ¢ /rn ),
as the two matrices are equal, but by property (B’)

det(r1 / ¢ ¢ ¢ /ri / ¢ ¢ ¢ /rj / ¢ ¢ ¢ /rn ) = ¡ det(r1 / ¢ ¢ ¢ /rj / ¢ ¢ ¢ /ri / ¢ ¢ ¢ /rn ),

so that both determinants are in fact zero.

We continue now with the proof of Theorem 13.
Proof. (A) If n = 1 then (A) equates to the identity (λa11 + µv1 ) = λ(a11 ) + µ(v1 ). As
an inductive hypothesis, suppose that (A) is true for (n ¡ 1) £ (n ¡ 1) matrices. We are
looking to show that the n £ n determinant function is linear in the ith row. Note, for j 6= i,
that Cj1 (C) = λCj1 (A) + µCj1 (B) by our inductive hypothesis as these cofactors relate to

DEFINITIONS 8
(n ¡ 1) £ (n ¡ 1) determinants. Also Ci1 (C) = Ci1 (A) = Ci1 (B) as Ci1 is independent of the
ith row. Hence

det C = a11 C11 (C) + ¢ ¢ ¢ + (λai1 + µv1 )Ci1 (C) + ¢ ¢ ¢ + an1 Cn1 (C)
= a11 (λC11 (A) + µC11 (B)) + ¢ ¢ ¢ + λai1 Ci1 (A) + µv1 Ci1 (B) + ¢ ¢ ¢ + an1 (λCn1 (A) + µCn1 (B))
= λfa11 C11 (A) + ¢ ¢ ¢ + an1 Cn1 (A)g + µfa11 C11 (B) + ¢ ¢ ¢ + v1 Ci1 (B) + ¢ ¢ ¢ + an1 Cn1 (B)g
= λ det A + µ det B.

We have therefore proved (A) for all square matrices. In what follows, note that if (B) is
true of certain matrices then so is (B’) as we have shown that (A) and (B) are equivalent to
(A) and (B’).
(B) For a 2 £ 2 matrix, if r1 = r2 then
µ ¶
a11 a12
det A = = a11 a12 ¡ a12 a11 = 0.
a11 a12

So (B) (and hence (B’)) hold for 2 £ 2 matrices. Assume now that (B) (or equivalently (B’)) is
true for (n ¡ 1) £ (n ¡ 1) matrices. Let A = (r1 / ¢ ¢ ¢ /rn ) with ri = rj where i < j. Then

det A = a11 C11 (A) + ¢ ¢ ¢ + an1 Cn1 (A) = ai1 Ci1 (A) + aj1 Cj1 (A)

by the inductive hypothesis as Ak1 has two equal rows when k 6= i, j. Note that as ri = rj , with
one copy of each being removed from Ai1 and Aj1 , then the rows of Ai1 are the same as the
rows of Aj1 but come in a di¤erent order. The rows of Ai1 and Aji can be reordered to be the
same as follows: what remains of rj in Ai1 can be moved up to the position of ri ’s remainder in
Aj1 by swapping it j ¡ i ¡ 1 times, each time with the next row above. (Note that we cannot
simply swap the rows ri and rj in A to show det A = 0 as this would be assuming (B’) for n £ n
matrices which is equivalent to what we’re trying to prove.) By our inductive hypothesis

det A = ai1 Ci1 (A) + aj1 Cj1 (A)

= (¡1)1+i ai1 det Ai1 + (¡1)1+j aj1 det Aj1 [by de…nition of cofactors]
1+i j¡i¡1 1+j
= (¡1) ai1 (¡1) det Aj1 + (¡1) aj1 det Aj1 [by j ¡ i ¡ 1 uses of (B’)]
j
= (¡1) (ai1 ¡ aj1 ) det Aj1 = 0 [as ai1 = aj1 because ri = rj ].

Hence (B) is true for n £ n determinants and the result follows by induction.

Corollary 15 Let A be an n £ n matrix and λ a real number.

(a) If the matrix B is formed by multiplying a row of A by λ then det B = λ det A.
(b) det(λA) = λn det A.
(c) If any row of A is zero then det A = 0.

Proof. (a) This follows from the fact that det is linear in its rows, and then if (a) is applied
consecutively to each of the n rows part (b) follows. Finally if ri = 0 for some i, then ri = 0ri
and so (c) follows from part (a).

DEFINITIONS 9
Notation 16 We will denote the three EROs as:
(a) Mi (λ) denotes multiplication of the ith row by λ 6= 0.
(b) Sij denotes swapping the ith and jth rows.
(c) Aij (λ) denotes adding λ £ (row i) to row j.

Lemma 17 (a) The determinants of the elementary matrices are

det Mi (λ) = λ; det Sij = ¡1; det Aij (λ) = 1.

In particular, elementary matrices have non-zero determinants.

(b) If E, A are n £ n matrices and E is elementary then det EA = det E det A.
(c) If E is an elementary matrix then det E T = det E.

Proof. We shall prove (a) and (b) together. If E = Mi (λ) and then det EA = λ det A
by Corollary 15(a). If we choose A = In then we …nd det Mi (λ) = λ and so we also have
det EA = det E det A when E = Mi (λ).
If E = Sij then by Theorem 13(B’) det EA = ¡ det A. If we take A = In then we see
det Sij = ¡1 and then we also have det EA = det E det A when E = Sij .
If E = Aij (λ) and A = (r1 / ¢ ¢ ¢ /rn ) then

det(EA) = det(r1 / ¢ ¢ ¢ /ri + λrj / ¢ ¢ ¢ /rj / ¢ ¢ ¢ /rn )

= det(r1 / ¢ ¢ ¢ /ri / ¢ ¢ ¢ /rj / ¢ ¢ ¢ /rn) + λ det(r1 / ¢ ¢ ¢ /rj / ¢ ¢ ¢ /rj / ¢ ¢ ¢ /rn ) [by Theorem 13 (A)]
= det A + 0 [by Theorem 13 (B)]
= det A.

If we take A = In then det Aij (λ) = 1 and so det EA = det E det A also follows when E =
Aij (λ).
(c) Note that Mi (λ) and Sij are symmetric and so there is nothing to prove in these cases.
Finally
det(Aij (λ)T ) = det Aji (λ) = 1 = det Aij (λ).

Corollary 18 (Criterion for Invertibility) A square matrix A is invertible if and only if

det A 6= 0, in which case det(A¡1 ) = (det A)¡1 .

Proof. If A is invertible then it row-reduces to the identity; that is, there are elementary
matrices E1 , . . . , Ek such that Ek ¢ ¢ ¢ E1 A = I. Hence, by repeated use of Lemma 17(b),

1 = det I = det Ek £ ¢ ¢ ¢ £ det E1 £ det A.

In particular det A 6= 0. Further as Ek ¢ ¢ ¢ E1 = A¡1 then det(A¡1 ) = (det A)¡1 . If, however, A
is singular then A reduces to a matrix R with at least one zero row so that det R = 0. So as
before
det Ek £ ¢ ¢ ¢ £ det E1 £ det A = det R = 0
for some elementary matrices Ei . As det Ei 6= 0 for each i, it follows that det A = 0.

DEFINITIONS 10
Corollary 19 (Product Rule for Determinants) Let A, B be n£n matrices. Then det AB =
det A det B.

Proof. If A and B are invertible then they can be written as products of elementary matrices;
say A = E1 . . . Ek and B = F1 . . . Fl . Then

det AB = det E1 £ ¢ ¢ ¢ £ det Ek £ det F1 £ ¢ ¢ ¢ £ det Fl = det A det B

by Lemma 17(b). Otherwise one (or both) of A or B is singular. Then AB is singular and so
det AB = 0. But, also det A £ det B = 0 as one or both of A, B is singular.

Corollary 20 (Transpose Rule for Determinants) Let A be a square matrix. Then

det AT = det A.

Proof. A is invertible if and only if AT is invertible. If A is invertible then A = E1 . . . Ek for

some elementary matrices Ei . Now AT = EkT ¢ ¢ ¢ E1T by the product rule for transposes and so,
by Lemma 17(c) and the product rule above,

det AT = det EkT £ ¢ ¢ ¢ £ det E1T = det Ek £ ¢ ¢ ¢ £ det E1 = det A.

If A is singular then so is AT and so det A = 0 = det AT .

Remark 21 Corollaries 18, 19, 20, represent the most important algebraic properties of deter-
minants. However we are still lumbered with a very ine¢cient way of calculating determinants
in De…nition 3. That de…nition is practicable up to 3 £ 3 matrices but rapidly becomes laborious
after that. A much more e¢cient way to calculate determinants is using EROs and ECOs, and
we have been in a position to do this since showing det EA = det E £ det A for elementary E.
An ECO involves postmultiplication by an elementary matrix but the product rule shows they
will have the same e¤ects on the determinant. Spelling this out:

² Adding a multiple of a row (resp. column) to another row (resp. column) has no e¤ect
on a determinant.

² Multiplying a row or column of the determinant by a scalar λ multiplies the determinant

by λ.

² Swapping two rows or two columns of a determinant multiplies the determinant by ¡1.

The following examples will hopefully make clear how to e¢ciently calculate determinants using
EROs and ECOs.

Example 22 Use EROs and ECOs to calculate the following 4 £ 4 determinants.

¯ ¯ ¯ ¯
¯ 1 2 0 3 ¯¯ ¯ 2 2 1 ¡3 ¯
¯ ¯ ¯
¯ 4 ¡3 1 0 ¯¯ ¯ 0 6 ¡2 1 ¯
¯ , ¯ ¯.
¯ 0 2 5 ¡1 ¯¯ ¯ 3 2 1 1 ¯
¯ ¯ ¯
¯ 2 3 1 2 ¯ ¯ 4 2 ¡1 2 ¯

DEFINITIONS 11
Solution.
¯ ¯ ¯ ¯
¯ 1 2 0 3 ¯¯ ¯ 1 2 0 3 ¯ ¯ ¯
¯ ¯ ¯ ¯ ¯
¯ 4 ¡3 1 0 ¯¯ ¯ 0 ¡11 1 ¡12 ¯ ¯ ¡11 1 ¡12 ¯
¯ = ¯¯ ¯=¯ 2 5 ¡1 ¯
¯ 0 2 5 ¡1 ¯¯ ¯ ¯ ¯
¯ ¯ 0 2 5 ¡1 ¯ ¯ ¡1 1 ¡4 ¯
¯ 2 3 1 2 ¯ ¯ 0 ¡1 1 ¡4 ¯
¯ ¯
¯ 0 ¡10 32 ¯ ¯ ¯
¯ ¯ ¯ ¡10 32 ¯¯
¯
= ¯ 0 7 ¡9 ¯ = ¡1 £ ¯¯
¯ = 134,
¯ ¡1 1 ¡4 ¯ 7 ¡9 ¯

where, in order, we (i) add appropriate multiples of the row 1 to lower rows to clear the rest of
column 1, (ii) expand along column 1, (iii) add appropriate multiples of row 3 to rows 1 and 2
to clear the rest of column 1, (iv) expand along column 1 and (v) employ the 2 £ 2 determinant
formula.
¯ ¯ ¯ ¯
¯ 2 2 1 ¡3 ¯ ¯ 2 2 1 ¡3 ¯ ¯ ¯
¯ ¯ ¯ ¯ ¯ 6 ¡2 1 ¯
¯ 0 6 ¡2 1 ¯ ¯ 1 ¯¯ ¯ ¯
¯ ¯ = ¯ 0 6 ¡21 11 = 2 ¯¯ ¡1 ¡ 12 11 ¯
¯ 3 2 1 ¯ ¯ ¯ ¯
¯ 0 ¡1 ¡ 2 2 ¯
1 ¯ 2
¯ ¯ ¡2 ¡3 8 ¯
¯ 4 2 ¡1 2 ¯ ¯ 0 ¡2 ¡3 8 ¯
¯ ¯
¯ 0 ¡5 34 ¯ ¯ ¯
¯ ¯ ¯ ¡5 34 ¯
¯ 1 11 ¯
= 2 ¯ ¡1 ¡ 2 2 ¯ = 2 ¯ ¯ ¯ = 166,
¡2 ¡3 ¯
¯ 0 ¡2 ¡3 ¯

where, in order, we (i) add appropriate multiples of row 1 to lower rows to clear the rest of
column 1, (ii) expand along column 1, (iii) add appropriate multiples of row 2 to rows 1 and 3
to clear the rest of column 1, (iv) expand along column 1 and (v) employ the 2 £ 2 determinant
formula.
Alternatively, for this second determinant, it may have made more sense to column-reduce as
the third column has a helpful leading 1 and we could have instead calculated the determinant
as follows.
¯ ¯ ¯ ¯
¯ 2 2 1 ¡3 ¯ ¯ 0 0 1 0 ¯ ¯ ¯
¯ ¯ ¯ ¯ ¯ ¯
¯ 0 6 ¡2 1 ¯ ¯ 4 10 ¡2 ¡5 ¯ ¯ 4 10 ¡5 ¯
¯ ¯ = ¯ ¯=¯ 1 0 4 ¯
¯ 3 2 1 1 ¯¯ ¯ 1 0 1 4 ¯¯ ¯¯ ¯
¯ ¯ 6 4 ¡1 ¯
¯ 4 2 ¡1 2 ¯ ¯ 6 4 ¡1 ¡1 ¯
¯ ¯
¯ 0 10 ¡21 ¯ ¯ ¯
¯ ¯ ¯ 10 ¡21 ¯
= ¯¯ 1 0 4 ¯¯ = ¡ ¯¯ ¯ = 166.
¯ 0 4 ¡25 ¯ 4 ¡25 ¯

where, in order, we (i) add appropriate multiples of column 3 to other columns to clear the rest
of row 1, (ii) expand along row 1 (iii) add appropriate multiples of row 2 to rows 1 and 3 to
clear the rest of column 1, (iv) expand along column 1 and (v) employ the 2 £ 2 determinant
formula.

² We will demonstrate in Theorem 28 the as-yet-unproven equivalence of expanding along

any row or column.

DEFINITIONS 12
Example 23 Let a, x be real numbers. Determine the following 3 £ 3 and n £ n determinants.
¯ ¯
¯ ¯ ¯ x 1 ¢¢¢ 1 ¯
¯ x a a ¯ ¯ ¯
¯ ¯ ¯ 1 x ¢¢¢ 1 ¯
¯ ¯
(a) ¯¯ x x a ¯¯, (b) ¯ .. .. . . .. ¯.
¯ x x x ¯ ¯ . . . . ¯
¯ ¯
¯ 1 1 ¢¢¢ x ¯

Solution. (a) Subtracting row 3 from the other rows, and expanding along column 1, we obtain
¯ ¯ ¯ ¯
¯ x a a ¯ ¯ 0 a¡x a¡x ¯ ¯ ¯
¯ ¯ ¯ ¯ ¯ a¡x a¡x ¯
¯ x x a ¯=¯ 0 0 a ¡ x ¯¯ = x ¯¯ ¯ = x(a ¡ x)2 .
¯ ¯ ¯ 0 a ¡ x ¯
¯ x x x ¯ ¯ x x x ¯
Similarly for (b) if we note that the sum of each column is the same and then add the bottom
n ¡ 1 rows to the …rst row (which won’t a¤ect the determinant), we see it equals
¯ ¯
¯ x + n ¡ 1 x + n ¡ 1 ¢¢¢ x + n ¡ 1 ¯
¯ ¯
¯ 1 x ¢ ¢ ¢ 1 ¯
¯ ¯
¯ .. .. ... .. ¯
¯ . . . ¯
¯ ¯
¯ 1 1 ¢¢¢ x ¯
¯ ¯
¯ 1 1 ¢¢¢ 1 ¯
¯ ¯
¯ 1 x ¢¢¢ 1 ¯
¯ ¯
= (x + n ¡ 1) ¯ .. .. . . . ¯
¯ . . . .. ¯
¯ ¯
¯ 1 1 ¢¢¢ x ¯
¯ ¯
¯ 1 1 ¢¢¢ 1 ¯¯
¯
¯ 0 x ¡ 1 ¢¢¢ 0 ¯¯
¯
= (x + n ¡ 1) ¯ .. .. ... .. ¯
¯ . . . ¯¯
¯
¯ 0 0 ¢¢¢ x ¡ 1 ¯
where, in order, we (i) take the common factor of x + n ¡ 1 out of the …rst row, (ii) subtract the
…rst row from each of the other rows, (iii) note the determinant is upper triangular to …nally
obtain a result of (x + n ¡ 1)(x ¡ 1)n¡1 .
We conclude this section by de…ning the Vandermonde 1 determinant useful in interpolation.

Example 24 (Vandermonde Matrix) For n > 2 and real numbers x1 , . . . , xn we de…ne

0 1
1 x1 x21 ¢ ¢ ¢ xn¡1
1
B 1 x2 x2 ¢ ¢ ¢ xn¡1 C
B 2 2 C Y
B 2 n¡1 C
Vn = B 1 x3 x3 ¢ ¢ ¢ x3 C and then det Vn = (xi ¡ xj ).
B .. .. .. . . .. C
@ . . . . . A i>j
2 n¡1
1 xn xn ¢ ¢ ¢ xn
In particular, Vn is invertible if and only if the xi are distinct.
Solution. This is left to Sheet 1, Exercise 5.
1
After the French mathematician Alexandre-Théophile Vandermonde (1735-1796).

DEFINITIONS 13
1.2 Permutation Matrices

It was claimed in Remark 11 that the determinant function for n £ n matrices is entirely
determined by certain algebraic properties. In light of Lemma 14, these properties are equivalent
to

(i) det is linear in the rows of a matrix.

(ii) if a matrix has two equal rows then its determinant is zero.

(ii)’ if the matrix B is produced by swapping two of the rows of A then det B = ¡ det A.

(iii) det In = 1.

To see why these properties determine det, we …rst consider the n = 2 case. Given a 2 £ 2
matrix A = (aij ), we can calculate its determinant as follows. As det is linear in row 1 then
¯ ¯ ¯ ¯ ¯ ¯
¯ a11 a12 ¯ ¯ a11 0 ¯ ¯ 0 a12 ¯
¯ ¯=¯ ¯+¯ ¯
¯ a21 a22 ¯ ¯ a21 a22 ¯ ¯ a21 a22 ¯

which, as det is linear in row 2, equals

½¯ ¯ ¯ ¯¾ ½¯ ¯ ¯ ¯¾
¯ a11 0 ¯ ¯ a11 0 ¯ ¯ 0 a12 ¯ ¯ 0 a12 ¯
¯ ¯+¯ ¯ ¯ ¯+¯ ¯ .
¯ 0 a22 ¯ ¯ a21 0 ¯ + ¯ a21 0 ¯ ¯ 0 a22 ¯

Again as det is linear in rows the above equals

¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯
¯ 1 0 ¯ ¯ 1 0 ¯ ¯ ¯ ¯ ¯
a11 a22 ¯¯ ¯ + a11 a21 ¯ ¯ + a12 a21 ¯ 0 1 ¯ + a12 a22 ¯ 0 1 ¯.
0 1 ¯ ¯ 1 0 ¯ ¯ 1 0 ¯ ¯ 0 1 ¯

Then, using (ii), this equals

¯ ¯ ¯ ¯
¯ 1 0 ¯ ¯ ¯
a11 a22 ¯¯ ¯ + a12 a21 ¯ 0 1 ¯
0 1 ¯ ¯ 1 0 ¯

which, using (ii)’, equals ¯ ¯ ¯ ¯

¯ 1 0 ¯ ¯ ¯
a11 a22 ¯¯ ¯ ¡ a12 a21 ¯ 1 0 ¯.
0 1 ¯ ¯ 0 1 ¯
Finally, using (iii), we’ve shown
¯ ¯
¯ a11 a12 ¯
¯ ¯ == a11 a22 ¡ a12 a21
¯ a21 a22 ¯

If we were to argue similarly for a 3 £ 3 matrix A = (aij ), we could …rst use linearity to expand
the determinant into a linear combination of 33 = 27 determinants, with entries 1 and 0, each
multiplied by a monomial a1i a2j a3k . But we can ignore those cases where i, j, k involves some

PERMUTATION MATRICES 14
repetition as the corresponding determinant is zero. There would, in fact, be only 3! = 6
non-zero contributions giving us the formula
¯ ¯ ¯ ¯ ¯ ¯
¯ 1 0 0¯ ¯ 0 1 0¯ ¯ 0 0 1¯
¯ ¯ ¯ ¯ ¯ ¯
a11 a22 a23 ¯¯ 0 1 0¯¯ + a12 a23 a31 ¯¯ 0 0 1¯¯ + a13 a21 a32 ¯¯ 1 0 0¯¯
¯ 0 0 1¯ ¯ 1 0 0¯ ¯ 0 1 0¯
¯ ¯ ¯ ¯ ¯ ¯
¯ 0 1 0¯ ¯ 0 0 1¯ ¯ 1 0 0¯
¯ ¯ ¯ ¯ ¯ ¯
+a12 a21 a33 ¯¯ 1 0 0¯¯ + a13 a22 a31 ¯¯ 0 1 0¯¯ + a11 a23 a32 ¯¯ 0 0 1¯¯.
¯ 0 0 1¯ ¯ 1 0 0¯ ¯ 0 1 0¯

The …rst determinant here is det I3 which we know to be 1. The other determinants all have
the same rows (1, 0, 0), (0, 1, 0), (0, 0, 1) as I3 but appearing in some other order. In each case,
it is possible (if necessary) to swap (1, 0, 0) – which appears as some row of the determinant –
with the …rst row, so that it is now in the correct place. Likewise the second row can be moved
(if necessary) so it is in the right place. By a process of elimination the third row is now in the
right place and we have transformed the determinant into det I3 . We know what the e¤ect of
each such swap is, namely multiplying by ¡1, and so the six determinants above have values 1
or ¡1. For example,
¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯
¯ 0 1 0 ¯ ¯ 1 0 0 ¯ ¯ 1 0 0 ¯ ¯ 0 0 1 ¯ ¯ 1 0 0 ¯
¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯
¯ 0 0 1 ¯ = ¡ ¯ 0 0 1 ¯ = ¯ 0 1 0 ¯ = 1, ¯ 0 1 0 ¯ = ¡ ¯ 0 1 0 ¯ = ¡1.
¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯
¯ 1 0 0 ¯ ¯ 0 1 0 ¯ ¯ 0 0 1 ¯ ¯ 1 0 0 ¯ ¯ 0 0 1 ¯

So …nally we have, as we found in Proposition 5(b), that

¯ ¯
¯ a11 a12 a13 ¯¯
¯
¯ a21 a22 a23 ¯¯ = a11 a22 a23 + a12 a23 a31 + a13 a21 a32 ¡ a12 a21 a33 ¡ a13 a22 a31 ¡ a11 a23 a32 .
¯
¯ a31 a23 a33 ¯

The general situation is hopefully now clear for an n £ n matrix A = (aij ). Using linearity
to expand along each row in turn, det A can be written as the sum of nn terms
X
det Pi1 ¢¢¢in a1i1 ¢ ¢ ¢ anin

where Pi1 ¢¢¢in is the matrix whose rows are ei1 , . . . , ein – that is the entries of Pi1 ¢¢¢in are all
zero except entries (1, i1 ), . . . , (n, in ) which are all 1. At the moment each of i1 , . . . , in can
independently take a value between 1 and n, but most such choices lead to the determinant
det Pi1 ¢¢¢in being zero as some of the rows ei1 , . . . , ein are repeated. In fact, det Pi1 ¢¢¢in can only
be non-zero when
fi1 , . . . , in g = f1, . . . , ng.
That is i1 , . . . , in are 1, . . . n in some order or equivalently the rows of Pi1 ¢¢¢in are e1 , . . . , en in
some order.

De…nition 25 An n£n matrix P is said to be a permutation matrix if its rows are e1 , . . . , en

in some order. This is equivalent to saying that each row and column contains a single entry 1
with all other entries being zero.

PERMUTATION MATRICES 15
 Thus we have shown:

Proposition 26 The function det is entirely determined by the three algebraic properties (i),
(ii) and (iii). Further, the determinant det A of an n £ n matrix A = (aij ) equals
X
det A = det Pi1 ¢¢¢in a1i1 ¢ ¢ ¢ anin (1.3)

where the sum is taken over all permutation matrices Pi1 ¢¢¢in = (ei1 / ¢ ¢ ¢ /ein ).

We further note:

Proposition 27 (a) The columns of a permutation matrix are eT1 , . . . , eTn in some order.
(b) The number of n £ n permutation matrices is n!.
(c) A permutation matrix has determinant 1 or ¡1.
(d) When n > 2, half the permutation matrices have determinant 1 and half have determi-
nant ¡1.

Proof. (a) The entries in the …rst column of a permutation matrix P are the …rst entries of
e1 , e2 , . . . en in some order and so are 1, 0, . . . 0 in some order – that is the …rst column is eTi for
some i. Likewise each column of P is eTi for some i. If any of the columns of P were the same
then this would mean that a row of P had two non-zero entries which cannot occur. So the
columns are all distinct. As there are n columns then each of eT1 , . . . , eTn appears exactly once.
(b) This is equal to the number of bijections from the set f1, 2, . . . , ng to itself.
(c) The rows of a permutation matrix P are e1 , . . . , en in some order. We know that
swapping two rows of a matrix has the e¤ect of multiplying the determinant by ¡1. We can
create a (possibly new) matrix P1 by swapping the …rst row of P with the row e1 (which
appears somewhere); of course no swap may be needed. The matrix P1 has e1 as its …rst row
and det P1 = § det P depending on whether a swap was necessary or not. We can continue in
this fashion producing matrices P1 , . . . , Pn such that the …rst k rows of Pk are e1 , . . . , ek in that
order and det Pk = § det Pk¡1 in each case, depending on whether or not we needed to make
any swap to get ek to the kth row. Eventually then Pn = In and det P = § det Pn = 1 or ¡1
depending on whether an even or odd number of swaps had to be made to turn P into In .
(d) Let n > 2 and let S12 be the elementary n £ n matrix associated with swapping the …rst
and second rows of a matrix. If P is a permutation matrix then S12 P is also a permutation
matrix as its rows are still e1 , . . . , en in some order; further

det S12 P = det S12 £ det P = ¡ det P.

For each permutation matrix P with det P = 1, we have S12 P being a permutation matrix
with det(S12 P ) = ¡1; conversely for every permutation matrix P~ with det P~ = ¡1 we have
S12 P~ being a permutation matrix with det(S12 P~ ) = 1 As these processes are inverses of one
another, because S12 (S12 P ) = P , there are equal numbers of determinant 1 and determinant
¡1 permutation matrices, each separately numbering 12 n!.
We now prove a result already mentioned in Remark 11. Our inductive de…nition of the
determinant began by expanding down the …rst column. In fact it is the case that we will arrive
at the same answer, the determinant, whichever column or row we expand along.

PERMUTATION MATRICES 16
Theorem 28 (Equality of determinant expansions 2 ) Let A = (aij ) be an n £ n matrix
and let Cij denote the (i, j)th cofactor of A. Then the determinant det A may be calculated by
expanding along any column or row of A. So, for any 1 6 i 6 n, we have

det A = a1i C1i + a2i C2i + ¢ ¢ ¢ + ani Cni [this is expansion along the ith column] (1.4)
= ai1 Ci1 + ai2 Ci2 + ¢ ¢ ¢ + ain Cin [this is expansion along the ith row]. (1.5)

Proof. We showed in Theorem 13 and Proposition 10 that det has properties (i), (ii), (iii),
and have just shown in Proposition 26 that these properties uniquely determine the function
det. Making the obvious changes to Theorem 13 and Proposition 10 it can similarly be shown,
for any i, that the function which assigns

a1i C1i + a2i C2i + ¢ ¢ ¢ + ani Cni (1.6)

to the matrix A = (aij ) also has properties (i), (ii), (iii). By uniqueness it follows that (1.6)
also equals det A. That is, expanding down any column also leads to the same answer of det A.
Then
det A = det AT = [AT ]1i C1i (AT ) + [AT ]2i C2i (AT ) + ¢ ¢ ¢ + [AT ]ni Cni (AT )
= ai1 Ci1 + ai2 Ci2 + ¢ ¢ ¢ + ain Cin

by expanding down the ith column of AT , but this is the same sum found when expanding
along the ith row of A.
In practical terms, however, Laplace’s result isn’t that helpful. We have already discounted
repeated expansion along rows and columns of hard-to-calculate cofactors as a hugely ine¢cient
means to …nd determinants (see Remarks 11 and 21). However, it does lead us to the following
theorem of interest.

Theorem 29 (Existence of the Adjugate) Let A be an n £ n matrix. Let Cij denote the
(i, j)th cofactor of A and let C = (Cij ) be the matrix of cofactors. Then

C T A = AC T = det A £ In .

In particular, if A is invertible, then

CT
A¡1 = (1.7)
det A
Proof. Note n n
X X
T T
[C A]ij = [C ]ik [A]kj = Cki akj .
k=1 k=1

When i = j then
n
X
T
[C A]ii = aki Cki = det A
k=1

2
This was proved by Pierre-Simon Laplace (1749-1827) in 1772, though Leibniz had been aware of this result
a century earlier.

PERMUTATION MATRICES 17
by Theorem 28 as this is the determinant calculated by expanding along the ith column. On
the other hand, if i 6= j, then consider the matrix B which has the same columns as A except
for the ith column of B which is a copy of A’s jth column. As the ith and jth columns of B
are equal then det B is zero. Note that the (k, i)th cofactor of B equals Cki as A and B agree
except in the ith column; so if expanding det B along its ith column we see
n
X n
X
0 = det B = bki Cki = akj Cki = [C T A]ij .
k=1 k=1

So C T A = det A £ In . That AC T = det A £ In similarly follows. Finally if A is invertible, then

det A 6= 0, and (1.7) follows.
De…nition 30 With notation as in Theorem 29 the matrix C T is called the adjugate of A (or
sometimes the adjoint of A) and is written adjA.
Corollary 31 (Cramer’s Rule 3 ) Let A be an n £ n matrix, b in Rncol and consider the linear
system (Ajb). The system has a unique solution if and only if det A 6= 0, which is given by
CT b
x= .
det A
Proof. There is a solution if and only if LA is onto which is then unique if and only if the
kernel is trivial. That is A is invertible and the result follows from the previous theorem and
Corollary 18.
Remark 32 Writing A = (aij ) and b = (b1 , . . . , bn )T then Cramer’s Rule with n = 2 expressly
reads as
b1 a22 ¡ b2 a12 b2 a11 ¡ b1 a21
x1 = , x2 = ,
det A det A
where det A = a11 a22 ¡ a12 a21 . When n = 3 Cramer’s rule reads as
¯ ¯ ¯ ¯ ¯ ¯
¯ a22 a23 ¯ ¯ a12 a13 ¯ ¯ a12 a13 ¯
b1 ¯¯ ¯ ¡ b2 ¯ ¯ ¯
¯ a32 a33 ¯ + b3 ¯ a22 a23 ¯
¯
a32 a33 ¯
x1 = ,
det A
¯ ¯ ¯ ¯ ¯ ¯
¯ a21 a23 ¯ ¯ a11 a13 ¯ ¯ a11 a13 ¯
¡b1 ¯¯ ¯ + b2 ¯ ¯ ¯
¯ a31 a33 ¯ ¡ b3 ¯ a21 a23 ¯
¯
a31 a33 ¯
x2 = ,
¯ ¯ ¯ det A ¯ ¯ ¯
¯ a a ¯ ¯ a a ¯ ¯ a a ¯
b1 ¯¯ 21 22 ¯¯ ¡ b2 ¯¯ 11 12 ¯¯ + b3 ¯¯ 11 12 ¯¯
a31 a32 a31 a32 a21 a22
x3 = ,
det A
where det A = a11 a22 a33 + a12 a23 a31 + a13 a21 a32 ¡ a12 a21 a33 ¡ a13 a22 a31 ¡ a11 a23 a32 .

Cramer’s rule though is a seriously limited and impractical means of solving linear systems.
The rule only applies when the matrix A is square and invertible, and the computational power
required to calculate so many cofactors and det A make it substantially more onerous than
row-reduction.
3
Named after the Swiss mathematician, Gabriel Cramer (1704-1752), who discovered this result in 1750.

PERMUTATION MATRICES 18
1.3 Determinants of Linear Maps

De…nition 33 Let T : V ! V be a linear map of a …nite dimensional vector space V. Then

the determinant of T is de…ned by
det T = det A
where A is a matrix representing T with respect to some basis for V.

Proposition 34 (a) The determinant of a linear map is well-de…ned.

(b) If S : V ! V is a second linear map then

det(ST ) = det S £ det T.

(c) T : V ! V is invertible if and only if det T 6= 0. If T is invertible then
¡ ¢ 1
det T ¡1 = .
det T
Proof. (a) As T may have many di¤erent matrix representatives, it is possible that di¤erent
representatives might have di¤erent determinants. However any two representatives, A and B,
of T are similar matrices so that A = P ¡1 BP for some invertible matrix P. Speci…cally if

A = E TE and B= F TF

then A = P ¡1 BP where P = F IE . By the product rule for determinants we have

¡ ¢ 1
det A = det P ¡1 BP = £ det B £ det P = det B,
det P
and hence each matrix representative has the same determinant.
(b) Say that S and T are represented by A and B wrt the same basis for V. Then ST is
represented by AB wrt the same basis. So, by the product rule,
det (ST ) = det (AB) = det A £ det B = det S £ det T.

(c) Say that T is invertible. Then there is a linear map S : V ! V such that ST = I = T S.
So
1 = det I = det S £ det T,
showing det T 6= 0. Conversely say that det T 6= 0. Let A be a matrix representating T wrt
some basis. So det A 6= 0 and A is an invertible matrix. Let S : V ! V be the linear map
represented by A¡1 wrt the same basis. Then ST is represented by A¡1 A = I wrt this basis,
and T S is represented by AA¡1 = I wrt this basis. But the identity matrix represents the
identity map wrt all bases and so
ST = I = T S.
Thus S = T ¡1 and T is invertible. Finally, when T is invertible, we have
¡ ¢ ¡ ¢
det T ¡1 (det T ) = det T ¡1 T = det I = 1,
and the result follows.

DETERMINANTS OF LINEAR MAPS 19

Example 35 Let V = h1, x, x2 i be the space of real polynomials in x of degree at most 2. De…ne
D, T : V ! V by
(Df) (x) = f 0 (x), (T f ) (x) = f(x + 1).
Evaluate det D and det T.

Solution. As D(1) = 0 then D is not invertible and so det D = 0. Alternatively the matrix
for D wrt f1, x, x2 g is 0 1
0 1 0
D= @ 0 0 2 A,
0 0 0
and det D = 03 = 0 as this is an upper triangular matrix.
Now the matrix for T wrt the same basis is
0 1
1 1 1
T =@ 0 1 2 A
0 0 1

and so det T = 13 = 1.

DETERMINANTS OF LINEAR MAPS 20