Mathematics

for Computer Scientists 2

Mark Groves
Tobias Mai
Summer 2025

June 6, 2025
Contents

1 Algebraic structures 3
1.1 Binary operations and groups . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3 Working with complex numbers . . . . . . . . . . . . . . . . . . . . 19
1.4 Working with modular arithmetic . . . . . . . . . . . . . . . . . . . 31

2 Vector spaces 42
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.2 Elementary vector-space theory . . . . . . . . . . . . . . . . . . . . . 47
2.3 Subspaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

3 Linear transformations and matrices 67

3.1 Linear transformations . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.2 Matrix representations of linear transformations . . . . . . . . . . . 72
3.3 Matrix algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

4 Matrix algebra and systems of linear equations 85

4.1 On the columns and rows of a matrix . . . . . . . . . . . . . . . . . . 85
4.2 The inverse of a matrix . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.3 Systems of linear equations . . . . . . . . . . . . . . . . . . . . . . . 99
4.4 Determinants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

5 Eigenvalues and eigenvectors 120

 1 Algebraic structures

1 Algebraic structures

1.1 Binary operations and groups

Definition

A binary operation on a non-empty set X is a function X × X → X.

Usually we use the notation x1 · x2 instead of ·( x1 ,x2 ).

Examples

1. Addition and multiplication are binary operations on the set N: for all n1 ,
n2 ∈ N the quantities n1 + n2 and n1 .n2 are also elements of N. They are
also binary operations on the sets Z, Q and R.

2. Let M be a non-empty set and X be the set of all functions M → M. Compo-

sition of functions is a binary operation on X: for all functions f , g : M → M
we have that f ◦ g is also a function M → M.

Definitions

A binary operation · on a non-empty set X is

(i) associative if (a · b) · c = a · (b · c) for all a,b,c ∈ X,

(ii) commutative if a · b = b · a for all a, b ∈ X.

Examples

1. Addition and multiplication are associative, commutative binary operations

on N, Z, Q and R.

2. Let X be the set of all functions R → R.

Suppose that f , g, h ∈ M. By the definition ( F1 ◦ F2 )(y) = F1 ( F2 (y)) we

find that

3
 1.1 Binary operations and groups

(( f ◦ g) ◦ h)( x ) = ( f ◦ g)(h( x )) = f ( g(h( x ))

and
( f ◦ ( g ◦ h))( x ) = f (( g ◦ h)( x )) = f ( g(h( x )),
so that
(( f ◦ g) ◦ h)( x ) = ( f ◦ ( g ◦ h))( x )
for all x ∈ X. Hence ( f ◦ g) ◦ h = f ◦ ( g ◦ h), that is ◦ is associative.

Define f ( x ) = sin x and g( x ) = 2x. It follows that f ( g( π2 )) = sin(π ) = 0

but g( f ( π2 )) = 2 sin( π2 ) = 2. Hence f ◦ g 6= g ◦ f , that is ◦ is not commu-
tative.

Definition

A group is a non-empty set X equipped with a binary operation · which has the
following properties.

• The binary operation · is associative. (Associativity)

• There exists an element i ∈ X such that (Existence of
x · i = i · x = x for all x ∈ X. a neutral element)
• For every x ∈ X there exists an element (Existence of inverses)
x −1 ∈ X such that x · x −1 = x −1 · x = i.

If the binary operation is commutative the group is called an abelian group.

We usually use the notation (X,·) for a group.

Examples

1. (Z,+) is an abelian group,where

the neutral element is 0, since 0 + x = x + 0 = x for all x ∈ Z,

the inverse of x is the number − x, since x + (− x ) = (− x ) + x = 0 for all

x ∈ Z.

(Q,+) and (R,+) are also abelian groups.

4
 1.1 Binary operations and groups

2. (Q \ {0}, .) is an abelian group,where

the neutral element is 1, since 1.x = x.1 = x for all x ∈ Q \ {0},

the inverse of x is the number x1 , since x. 1x = x1 .x = 1 for all x ∈ Q \ {0}.

(R \ {0},.) is also an abelian group.

3. Let X be a non-empty set. The set of all bijections X → X forms a group

with respect to composition of functions:

The neutral element is the identity mapping I : X → X with I ( x ) = x for

all x ∈ X, since f ◦ I = I ◦ f = f for every bijection f : X → X.

The inverse of f : X → X is the inverse function f −1 : X → X, since f ◦

f −1 = f −1 ◦ f = I for every bijection f : X → X.

If X is a finite set we refer to this group as the symmetric group S(X ) and its
elements as permutations. For a permutation σ of the set { x1 ,x2 , . . . ,xn } we
often use the notation
 
x1 x2 · · · x n
,
σ ( x1 ) σ ( x2 ) · · · σ ( x n )

so that e.g.  
1 2 3
1 3 2
denotes the permutation 1 7→ 1, 2 7→ 3, 3 7→ 2 of the set {1,2,3}. This permu-
tation can also be written in cyclic notation as (1)(23) or simply (23). The
notation S({1, . . . ,n}) is often abbreviated to Sn ab, so that e.g.

S3 = {(),(123),(321),(23),(13),(12)}.

4. A symmetry of a planar geometric figure is an angle- and distance-preserving

bijection of the plane onto itself which also maps this figure onto itself.

The diagram shows the six symmetries of an equiliateral triangle: the iden-
tity I, rotations R±2π/3 through ± 2π
3 and reflections T1 , T2 , T3 in the angle
bisectors.

5
 1.1 Binary operations and groups

2 3

I 1 R2π/3 3 R−2π/3 2

2 3 1 2 3 1

T1 T2 T3 2
1 3

3 2 2 1 1 3

The set of symmetries of an equilateral triangle is a group with respect to

composition of functions:

The neutral element is I.

−1
One finds that I −1 = I, R2π/3 = R−2π/3 , R− 1 −1
−2π/3 = R2π/3 , T1 = T1 ,
T2−1 = T2 and T3−1 = T3 .

The group is however not abelian, since T1 R2π/3 = T2 but R2π/3 T1 = T3 .

We can represent the symmetries as permutations of the set {1,2,3}:

R2π/3 = (123), R−2π/3 = (321), T1 = (23), T2 = (13), T3 = (12).

We can thus identify the symmetry group of an equilaterial triangle with

the symmetric group S3 .

6
 1.1 Binary operations and groups

5. Let m, n be natural numbers. An (m × n) (real) matrix is an mn-tuple

A = (aij ) i=1,...,m, of real numbers. We write
j =1,...,n

 
a11 a12 · · · a1n  
 a21 a22 · · · a2n  

 
A = (aij ) =  .. .. ..  m rows,
 . . . 

am1 am2 · · · amn 
| {z }
n columns
i.e. we arrange the mn-tuple (aij ) i=1,...,m, in a rectangular array with m rows
j =1,...,n
and n columns. The coefficient aij is found in the i-th row and j-th column.

The set R m×n of (m × n) real matrices is an abelian group with respect to

pointwise addition of the coefficients:

(aij ) + (bij ) := (aij + bij ),

since

the neutral element is the zero matrix (0) whose coefficients are all zero,

the inverse of (aij ) is the matrix (− aij ).

Proposition

Let (X,·) be a group.

1. The neutral element i is unique.

2. The inverse of each element x ∈ X is unique

Proof

1. Suppose that i1 and i2 are neutral elements, so that in particular

x = x · i1 , y = i2 · y

for all x, y ∈ X. This holds in particular for x = i2 and y = i1 , so that

i2 = i2 · i1 , i1 = i2 · i1 .

Hence i2 = i1 .

7
 1.1 Binary operations and groups

2. Let y and z be two inverses of x, so that

y · x = x · y = i, z · x = x · z = i.

It follows that

y = i · y = (z · x ) · y = z · ( x · y) = z · i = z. 

Mappings between groups which preserve the group structure are particularly
important.

Definition

Let (G,· G ) and ( H, · H ) be groups. A function f : G → H with the property that

f ( x · G y) = f ( x ) · H f (y) for all x,y ∈ G

is called a (group) homomorphism. If f is also a bijection it is called a (group)

isomorphism.

Proposition

Let (G,· G ) and ( H, · H ) be groups and f : G → H be a group homomorphism. It

follows that

(i) f (i G ) = i H , where i G and i H are the neutral elements of G and H,

(ii) f ( x −1 ) = f ( x )−1 for all x ∈ G.

Proof

We write both · G and · H as ·, since it is clear from the context which is meant.

(i) Note that

f (i G · i G ) = f (i G ) · f (i G ),
| {z }
= iG
that is
f (i G ) = f (i G ) · f (i G ).

8
 1.1 Binary operations and groups

It follows that

i H = f (i G ) − 1 · f (i G )


= f (i G ) − 1 · f (i G ) · f (i G )


= f (i G ) − 1 · f (i G ) · f (i G )
= i H · f (i G )
= f (i G ).

(ii) Note that

i H = f (i G ) = f ( x − 1 · x ) = f ( x − 1 ) · f ( x )
and
i H = f (i G ) = f ( x · x − 1 ) = f ( x ) · f ( x − 1 ).
The equality f ( x )−1 = f ( x −1 ) now follows from the uniqueness of inverses.

Examples

1. The exponential function exp : (R,+) → (R \ {0},.) is a group homomor-

phism because of the identity
exp( x + y) = exp( x ). exp (y), x,y ∈ R.

2. In an earlier example we identified the symmetry group ST of the equi-

liateral triangle with the symmetric group S3 . This identification can be
formally expressed using the mapping S : ST → S3 with
S( I ) = (), S( R2π/3 ) = (123), S( R−2π/3 ) = (321),
S(T1 ) = (23), S(T2 ) = (13), S(T3 ) = (12).
Thus S : ST → S3 is a group homomorphism.

3. Let (G,·) be a group. An automorphism is an isomorphism G → G. The set

Aut(G ) of all automorphisms of G forms a group with respect to composi-
tion of functions.

Definition

A subgroup (Y,·) of a group (X, ·) is a subset Y of X which is itself a group with

respect to the binary operation ·.

9
 1.2 Fields

Proposition

Let (X,·) be a group and Y a non-empty subset of X. Furthermore, suppose that

1. i ∈ Y,

2. Y is closed with respect to ·, that is x1 , x2 ∈ Y ⇒ x1 · x2 ∈ Y,

3. Y is closed with respect to inverses, that is x ∈ Y ⇒ x −1 ∈ Y.

It follows that (Y,·) is a subgroup of (X,·).

Example

The subgroups of S3 are

{()} and S3 (trivially),

{(),(12)}, {(),(13)} and {(),(23)} (each a reflection of an equilateral triangle),

{(),(123),(321)} (rotation of an equilateral triangle through ± 2π

3 ).

1.2 Fields

Definition

A field is a non-empty set X equipped with two binary operations + (‘addition’)

and . (’multiplication’) which have the following properties:

(A1) x + (y + z) = ( x + y) + z ∀ x,y,z ∈ X (associativity of

addition)
(A2) x+y=y+x ∀ x,y ∈ X (commutativity of
addition)
(A3) There exists an element (existence of
0 ∈ X such that additive identity)
x + 0 = 0 + x = x ∀x ∈ X
(A4) For every x ∈ X there exists (existence of
an element − x ∈ X such that additive inverses)
x + (− x ) = − x + x = 0

10
 1.2 Fields

(A5) x.(y.z) = ( x.y).z ∀ x,y,z ∈ X (associativity of

multiplication)
(A6) x.y = y.x ∀ x,y ∈ X (commutativity of
multiplication)
(A7) There exists an element (existence of
1 ∈ X \ {0} such that multiplicative identity)
1.x = x.1 = x ∀ x ∈ X
(A8) For every x ∈ X \{0} there exists (existence of
an element x −1 ∈ X such that multiplicative inverses)
x.x −1 = x −1 .x = 1
(A9) x.(y + z) = x.y + x.z ∀ x,y,z ∈ X (distributive law)

We usually denote a field by (X, + ,.).

Remarks

1. (A1)-(A4) state that (X,+) is an abelian group, while (A5)-(A8) state that
(X \ {0},.) is an abelian group.

2. These properties, which we also call the axioms of arithmetic, state simply
that the ‘usual’ rules of arithemetic apply to a field.

Lemma (Rules of arithmetic)

Let x,y,z,w be elements of a field (K, + ,.). It follows that

1. x + y = x + z ⇒ y = z

2. −(− x ) = x and − x = (−1).x

3. x.(−y) = −( x.y)

4. x.y = x.z, x 6= 0 ⇒ y = z

5. x.y = 0 ⇒ x = 0 or y = 0

6. the equation x + y = z for x has the unique solution x = z + (−y)

7. the equation y.x = z, y 6= 0 for x has the unique solution x = y−1 .z

8. x.y−1 + z.w−1 = ( x.w + z.y)(y.w)−1 , y, w 6= 0

11
 1.2 Fields

We now present some examples of fields.

• The real numbers

The set of real numbers forms a field with respect to addition and multiplication.
We usually write a.b, a.b−1 and a + (−b) as respectively ab, a/b and a − b.

• The rational numbers

Definition

The set nm o
Q= : m ∈ Z, n ∈ Z \{0}
n
of rational numbers forms a field with respect to addition and multiplication. We
establish this fact using the following definition.

Definition

A subfield (Y, + ,.) of a field (X, + ,.) is a subset Y of X which is itself a field with
respect to the binary operations + and ..

Proposition

Let X be a field with respect to +, . and Y be a subset of X. Suppose furthermore

that

1. Y is closed with respect to + and .;

2. 0 ∈ Y and 1 ∈ Y;

3. Y is closed with respect to inverses:

a ∈ Y ⇒ − a ∈ Y, a ∈ Y \{0} ⇒ a−1 ∈ Y.

Under these conditions (Y, + ,.) is a subfield of (X, + ,.).

12
 1.2 Fields

Theorem

Q is a field with respect to the addition and multiplication of the real numbers.

Proof

We consider Q as a subset of R and verify the criteria in the above proposition.

1. Q is closed with respect to the addition and multiplication of the real

numbers: For a, c ∈ Z and b, d ∈ Z \{0} we have that

a c a.d + b.c
+ = ∈ Q,
b d b.d
a c a.c
. = ∈ Q.
b d b.d

2. For every n ∈ Z \{0} we have that

0 n
0= , 1= ,
n n
so that 0, 1 ∈ Q.

3. For m ∈ Z, n ∈ Z \{0} we have that

m (−m)
− = ∈Q
n n
and for m, n ∈ Z \{0} we have that
 m −1 n
= ∈ Q.
n m
Q is therefore closed with respect to inverses.

• Prime fields

Let p be a prime number and ∼ the equivalence relation

a∼b ⇔ a ≡ b (mod p), i.e. p|(b − a)

13
 1.2 Fields

on Z. The set
Z p := {[0], . . . ,[ p − 1]}
of equivalence classes forms a field (Z p , ⊕ ,⊙) with

[ a ] ⊕ [ b ] : = [ a + b ], [ a] ⊙ [b] := [ ab].

(We usually abbreviate [ a] ⊕ [b] and [ a] ⊙ [b] to respectively [ a] + [b] and [ a][b].)

We first show that ⊕ and ⊙ are well-defined operations, i.e.

[ a ] = [ c ], [ b ] = [ d ] ⇒ [ a + b] = [c + d], [ ab] = [cd].

[ a] = [c] and [b] = [d] means

a − c = q1 p, b − d = q2 p

for some integers q1 and q2 . It follows that

(a + b) − (c + d) = (q1 + q2 ) p, ab − cd = (bq1 + cq2 ) p,

i.e. [ a + b] = [c + d] and [ ab] = [cd].

Axioms (A1)-(A7) and (A9) are readily verified (the additive and multiplica-
tive identities are respectively [0] and [1] and the additive inverse to [ a] is
−[ a] := [− a]).

We will prove the following assertion later:

Every [ a] with a ∈ {1, . . . , p − 1} has a

multiplicative inverse element
⇔
p is a prime number

• Complex numbers

Definitions

A complex number is an ordered pair (a,b), where a and b are real numbers.

The set of all complex numbers is denoted by C.

14
 1.2 Fields

Lemma

The formulae

(a,b) + (c,d) := (a + c,b + d), (addition)

(a,b).(c,d) := (ac − bd, ad + bc) (multiplication )

define binary operations on C with respect to which it is a field. The additive and
multiplicative identities are given by respectively (0,0) and (1,0), and
 
−1 a b
−(a,b) = (− a, − b), (a,b) = ,− 2 .
a2 + b2 a + b2

Lemma

1. The subset X = {(a,0) : a ∈ R } of C is a subfield of (C, + ,.)

2. The function ψ : (a,0) 7→ a is a field isomorphism, that is a bijection X → R

with the properties that
ψ ( x 1 + x 2 ) = ψ ( x 1 ) + ψ ( x 2 ), ψ( x1 .x2 ) = ψ( x1 )ψ( x2 )
for all x1 , x2 ∈ X.

Proof

1. We show that:

(i) The identities belong to X. This is trivial: (0,0), (1,0) ∈ X.

(ii) X is closed with respect to + and .. This follows from

(a,0) + (b,0) = (a + b,0), (1)

(a,0).(b,0) = (ab,0) (2)

(iii) X is closed with respect to inverses. This follows from

−(a,0) = (− a,0),
(a,0)−1 = (a−1 ,0), a 6= 0.

2. This follows from (1) and (2). 

15
 1.2 Fields

Remark

The above notation for complex numbers is a little unwieldy and is usual simpli-
fied as follows.

We identify the subset X of C with R and write (a,0) as a.

We define i = (0,1) and note that

i.i = (0,1).(0,1)
= (−1,0)
= −1.

It follows that

(a,b) = (a,0) + (0,b)

= (a,0) + (0,1).(b,0)
= a + ib.

We therefore write (a,b) as a + ib and work with the usual rules of arithmetic and
the additional identity i2 = −1.

Example

Let z1 = ( x1 ,y1 ), z2 = ( x2 ,y2 ), so that

z1 z2 = ( x1 ,y1 ).( x2 ,y2 )

= ( x1 x2 − y1 y2 ,x1 y2 + x2 y1 ).

Write z1 = x1 + iy1 , z2 = x2 + iy2 . We have that

z1 z2 = ( x1 + iy1 )( x2 + iy2 )
= x 1 x 2 + x 1 y 2 i + x 2 y 1 i + y 1 y 2 i2
= x1 x2 − y1 y2 + ( x1 y2 + x2 y1 )i.

16
 1.2 Fields

Definition

A field (K, + ,·) is called ordered if there is a relation ≺ on K with the following
properties:

(O1) For all x,y ∈ K exactly one of the relations

x = y, x ≺ y, y≺x
holds (trichotomy)
(O2) For all x,y,z ∈ K we have that
x ≺ y, y ≺ z ⇒ x ≺ z
(transitivity)

(O3) For all x,y,z ∈ K we have that 

x≺y ⇒ x+z≺y+z 

monotony
(O4) For all x,y,z ∈ K we have that 



x ≺ y, 0 ≺ z ⇒ x.z ≺ y.z

Remark

(O1) and (O2) are equivalent to the statement that  a total order, i.e. it is reflex-
ive, antisymmetric, transitive and connex. (O3) and (O4) assert that the order is
compatible with the axioms of arithmetic (A1)–(A9).

Notation

1. If 0 ≺ c we say: “c is positive”.
If c ≺ 0 we say: “c is negative”.

2. Often we write b ≻ a instead of a ≺ b.

Lemma

Let x,y,z,w be elements of an ordered field (K, + ,.).

1. x + z ≺ y + z ⇒ x ≺ y

2. x ≺ y, z  w ⇒ x + z ≺ y + w

3. x ≻ 0, y  0 ⇒ x + y ≻ 0

17
 1.2 Fields

4. x ≺ y ⇒ −y ≺ − x

5. x ≺ y, z  w ⇒ x + (−w) ≺ y + (−z)

6. x ≺ y, z ≺ 0 ⇒ y.z ≺ x.z

7. 0 ≺ x ≺ y ⇒ 0 ≺ y−1 ≺ x −1

8. x2 ≻ 0 for each x 6= 0

9. 1 ≻ 0

10. x.y ≻ 0 for all x,y ≺ 0

11. 0 ≺ x ≺ 1 ⇒ x2 ≺ x

12. x ≻ 1 ⇒ x2 ≻ x

Results 1-6 remain correct if we replace “≺” by “” everywhere.

Examples

1. R is an ordered field with respect to the relation

a≺b ⇔ a<b

2. Q is an ordered field since it inherits the ordering of the real numbers.

3. Let p be a prime number. There is no relation ≺ on Z p with respect to which

Z p is an ordered field.

Suppose ≺ is such a relation. We know that [0] ≺ [1]. It follows by monotony

that
[0] ⊕ [1] ≺ [1] ⊕ [1] , [1] ⊕ [1] ≺ [2] ⊕ [1] , . . . , [ p − 2] ⊕ [1] ≺ [ p − 1] ⊕ [1]
| {z } | {z } | {z } | {z } | {z } | {z }
= [1] = [2] = [2] = [3] = [ p − 1] = [ p]
and hence by transitivity that
[1] ≺ [ p ] = [0].
This contradicts trichotomy.

4. Similarly, there is no relation ≺ on C with respect to which C is an ordered

field. The existence of this relation implies that precisely one of the asser-
tions 0 ≺ i and i ≺ 0 holds, but both lead to the contradiction 1 ≺ 0.

18
 1.3 Working with complex numbers

1.3 Working with complex numbers

Definitions

Consider the complex number z = x + iy.

1. x is the real part of z. We write x = Re z.

2. y is the imaginary part of z. We write y = Im z.

3. x − iy is the complex conjugate of z. We write x − iy = z̄.

p p
4. x2 + y2 is the modulus or magnitude of z. We write x 2 + y2 = | z |.

z is called real if Im z = 0 and imaginary if Re z = 0.

Proposition

Every complex number z satisfies

1. z + z̄ = 2Re z,

2. z − z̄ = 2i Im z,

3. zz̄ = |z|2 .

Proposition

All complex numbers z1 , z2 obey the rules

1. z1 + z2 = z̄1 + z̄2 ,

2. z1 z2 = z̄1 z̄2 ,

1 1
3. = if z1 6= 0,
z1 z̄1

4. z̄1 = z1 ,

5. |z̄1 | = |z1 |.

19
 1.3 Working with complex numbers

Geometric representations of complex numbers

Since a complex number is an ordered pair ( x,y) of real numbers we can represent
it as a point in the coordinate plane (the complex plane):

Im

y z = x + iy

Im z
z̄ is related to z
by a reflection in
x Re the real axis.
Re z

−y z̄ = x − iy

It is also helpful to use polar coordinates.

Im
Here
y z = x + iy x = r cos θ,
y = r sin θ,
r q
r = x 2 + y2 ,
y
θ tan θ = .
x
x Re

Note in particular that

r = | z |.
The angle θ is an argument of the complex number z = x + iy. It is not unique
(determined only up to integer multiples of 2π). This nonuniquness is eliminated
by restricting to values of θ in the interval (−π,π ]. In this setting θ is called the
principal argument of z und denoted by arg z:

20
 1.3 Working with complex numbers

Im Im

z z
|z| arg z
|z|
arg z
Re Re

Im Im

Re Re
arg z |z|
arg z
|z|
z z

We usually define arg 0 = 0.

Example

Describe the following sets geometrically.

(i) {z ∈ C : |z − (1 + i)| = 3}

(ii) {z ∈ C : 1 < |z| ≤ 3}

(iii) {z ∈ C : |z − i| < |z + i|}

(iv) {z ∈ C : −π/4 < arg z ≤ π/4}

21
 1.3 Working with complex numbers

Solution

(i) {z ∈ C : |z − (1 + i)| = 3} is the set of all complex numbers whose distance to

the complex number 1 + i is equal to 3. It is therefore a circle with centre 1 + i
and radius 3:

Im

1+ i
1
3

1 Re

(ii) {z ∈ C : 1 < |z| ≤ 3} is the set of all complex numbers whose distance to the
origin is bigger than 1 or smaller than or equal to 3. It is therefore an annulus
with centre 0 and radii 1 and 3:

Im

1
Re
3

22
 1.3 Working with complex numbers

(iii) {z ∈ C : |z − i| < |z + i|} is the set of all complex numbers which are closer
to i than −i. It is therefore the upper half-plane:

Im

Re

−1

(iv) {z ∈ C : −π/4 < arg z ≤ π/4} is the set of all complex numbers which make
an angle with the real axis between −π/4 and π/4. It is therefore a sector:

Im

π
4

π Re
4

Lemma

The complex numbers

z1 = x1 + iy1 = r1 cos θ1 + ir1 sin θ1 ,

z2 = x2 + iy2 = r2 cos θ2 + ir2 sin θ2

obey the rules

23
 1.3 Working with complex numbers

z1 z2 = r1 r2 cos(θ1 + θ2 ) + ir1 r2 sin(θ1 + θ2 ) (i)

und

z1n = r1n cos nθ1 + ir1n sin nθ1 , n = 1,2,3, . . . . (de Moivre’s theorem) (ii)

Proof

(i) Note that

z1 z2 = (r1 cos θ1 + ir1 sin θ1 )(r2 cos θ2 + ir2 sin θ2 )

= r1 r2 cos θ1 cos θ2 + i2 r1 r2 sin θ1 sin θ2 + ir1 r2 cos θ1 sin θ2 + ir1 r2 sin θ1 cos θ2
= r1 r2 (cos θ1 cos θ2 − sin θ1 sin θ2 ) + ir1 r2 (cos θ1 sin θ2 + sin θ1 cos θ2 )
= r1 r2 cos(θ1 + θ2 ) + ir1 r2 sin(θ1 + θ2 ).

(ii) This result follows inductively from (i). 

Remark

We can interpret (i) geometrically: the distance from the point z1 z2 to the origin
is the product of the distances from the points z1 and z2 to the origin, while the
angle with the real axis made by the point z1 z2 is the sum of the angles with the
real axis made by the points z1 and z2 :

Im

z1 z 2
z2

r1 r2 r2

z1
θ1+θ2 r1
θ2
θ1
Re

24
 1.3 Working with complex numbers

Functions of a complex variable

We begin with the definitions of the exponential, trigonometic and hypberbolic

functions.

Definitions

For z ∈ C we define

ez := ex (cos y + i sin y), where x = Re z, y = Im z,

1
sin z := (eiz − e−iz ),
2i
1 iz
cos z := (e + e−iz) ,
2
1
sinh z := (ez − e−z ),
2
1 z
cosh z := (e + e−z ).
2

Further special functions such as tan(·), cot(·), etc. are defined from the above
functions in the usual fashion.

Proposition

We have that

cosh z = cos(iz), z ∈ C,
sinh z = −i sin(iz), z ∈ C,
eiθ = cos θ + i sin θ, θ ∈ R, (Euler’s formula)

and the basic identities for the relationships between the exponential, trigono-
metric and hypberolic functions remain valid.

Remark

We can write the complex number z in polar coordinates as z = r cos θ + ir sin θ

and hence as z = reiθ using Euler’s formula.

25
 1.3 Working with complex numbers

Remark (zeros)

1. The calculation

|ez | = |ex (cos y + i sin y)|

= |ex || cos y + i sin y|
= ex

for z = x + iy with x,y ∈ R shows that

|ez | = eRe z .
In particular, the exponential function has no zeros in the complex plane.

2. We have that

sin z = 0 ⇔ z = 0, ±π, ±2π, . . . ,

π 3π 5π
cos z = 0 ⇔ z = ± ,± ,± ,...
2 2 2

(see below).

3. The formulae in the previous proposition show that

sinh z = 0 ⇔ z = 0, ±πi, ±2πi, . . . ,

πi 3πi 5πi
cosh z = 0 ⇔ z = ± ,± ,± ,....
2 2 2

Example

Find the zeros of the function cos(·) : C → C.

Solution

We have that

cos z = 0
1
⇔ (eiz + e−iz ) = 0
2
⇔ eiz = −e−iz
⇔ e2iz = −1.

26
 1.3 Working with complex numbers

Write z = x + iy and note that −1 = 1eiπ :

Im

π | − 1| = 1
−1 arg(−1) = π
1 Re

It follows that

e2iz = −1
⇔ e2i (x+iy) = −1
⇔ e−2y e2ix = 1eiπ
⇔ e−2y = 1, 2x = π + 2nπ, n = 0, ± 1, ± 2, . . .
π
⇔ y = 0, x = + nπ, n = 0, ± 1, ± 2, . . . .
2

We conclude that
π 3π 5π
cos z = 0 ⇔ z = ± ,± ,± ,....
2 2 2

Definition

Let n ∈ N0 . A complex polynomial of degree n is an expression of the form

a n z n + a n − 1 z n − 1 + . . . + a1 z + a0 ,
where a0 , . . . , an are constant complex numbers and an 6= 0.

Theorem (Fundamental theorem of algebra)

Let n ∈ N. Every complex polynomial of degree n has at least one complex zero.

Corollary

Let n ∈ N. Every polynomial

z n + a n − 1 z n − 1 + . . . + a1 z + a0

27
 1.3 Working with complex numbers

of degree n has n complex zeros z1 ,. . . ,zn (which are not necessarily distinct) and
can therefore be factorised as

zn + an−1 zn−1 + . . . + a1 z + a0 = (z − z1 )(z − z2 ) . . . (z − zn ),

where (without loss of generality) we have set an = 1.

Remark

If the coefficients a0 , . . . , an are real, all zeros occur in complex-conjugate pairs (as
a special case they can be real). Suppose that z⋆ is a zero. It follows that

a n ( z ⋆ ) n + a n − 1 ( z ⋆ ) n − 1 + . . . + a1 ( z ⋆ ) + a0 = 0
⇒ a n ( z ⋆ ) n + a n − 1 ( z ⋆ ) n − 1 + . . . + a1 ( z ⋆ ) + a0 = 0
⇒ ān (z̄⋆ )n + ān−1 (z̄⋆ )n−1 + . . . + ā1 z̄⋆ + ā0 = 0
⇒ an (z̄⋆ )n + an−1 (z̄⋆ )n−1 + . . . + a1 z̄⋆ + a0 = 0, (since a0 , . . . , an are real)

so that z̄⋆ is also a zero.

Definition

An n-th root of a complex number is a solution of the equation

zn = a.

Example

Find all fifth roots of unity and interpret the result geometrically.

Solution

We are asked to determine all solutions of the equation

z5 = 1.

Write z = reiθ and note that 1 = 1ei0:

28
 1.3 Working with complex numbers

Im
|1| = 1
1 arg(1) = 0
1 Re

It follows that

z5 = 1
⇔ r5 e5iθ = 1ei0
⇔ r5 = 1, 5θ = 0 + 2nπ, n = 0, ± 1, ± 2, . . .
2nπ
⇔ r = 1, θ= , n = 0, ± 1, ± 2, . . . .
| {z 5 }
The five values of this expression
in (−π,π ] are 0, 2π 4π 2π 4π
5 , 5 ,− 5 ,− 5

The fifth roots of unity are therefore

2πi 4πi 2πi 4πi
ei0 , e 5 , e 5 , e− 5 , e− 5 .
|{z}
=1

Im
2πi
e 5

4πi
e 5

The roots are uniformly distributed

2π points on a circle with centre 0 and
5 radius 1.
1 Re
They are the corners of a regu-
lar pentagon.
4πi
e− 5

2πi
e− 5

The following lemma is proved using the method in the above example.

29
 1.3 Working with complex numbers

Lemma

Let n be a natural number. Every non-zero complex number a has precisely n

distinct n-th roots.

For n ≥ 2 they are uniformly distributed points on a circle with centre 0 and radius
| a|1/n . They are the corners of a regular n-gon.

Proposition

Let a, b, c be constant complex numbers with a 6= 0. The quadratic polynomial

az2 + bz + c

has

precisely one zero −b/(2a) if b2 − 4ac = 0,

precisely two zeros −b/(2a) + z1 , −b/(2a) + z2 where z1 , z2 are the two

square roots of (b2 − 4ac)/(4a2 ), if b2 − 4ac 6= 0.

Proof

Completing the square yields

 2  
2 b b2
az + bz + c = a z + − −c . 
2a 4a

Remark

Since z2 = −z1 we often abuse the notation and summarise the above proposition
as r
b b2 − 4ac
z=− ± ,
2a 4a2
p
where (b2 − 4ac)/4a2 stands for either z1 or z2 .

30
 1.4 Working with modular arithmetic

Example

Find all complex zeros of the polynomial

z3 − 3z2 + 4z − 2.

Solution

1 is obviously a zero, so that (z − 1) is a factor of this polynomial. The calculation

z2 − 2z + 2


z − 1 z3 − 3z2 + 4z − 2
z3 − z2
−2z2 + 4z − 2
−2z2 + 2z
2z − 2
2z − 2
0

shows that
z3 − 3z2 + 4z − 2 = (z − 1)(z2 − 2z + 2).
The zeros of the polynomial
z2 − 2z + 2
are found using the formula from the previous proposition. They are
r
2 4 − 4.1.2 √
± = 1 ± −1 = 1 ± i.
2 4

It follows that
z3 − 3z2 + 4z − 2 = (z − 1)(z − 1 − i)(z − 1 + i).

1.4 Working with modular arithmetic

Definition

Let a and b be natural numbers. The natural number d is a greatest common

denominator of a and b if it has the following properties:

31
 1.4 Working with modular arithmetic

(i) d| a and d|b.

(ii) Any other natural number c with c| a and c|b satisfies c|d.

Remark

The greatest common denominator of a and b is unique. Supose that d1 and d2 are
greatest common denominators of a and b. By part (ii) of the definition we have
d1 |d2 and d2 |d1 , so that d1 = d2 .

We denote the greatest common denominator of a and b by (a,b).

Proposition

Suppose that a and b are natural numbers with a ≥ b, and q ∈ N0 , r ∈ {1, . . . ,b − 1}

satisfy
a = bq + r.
It follows that
(a,b) = (b,r ).
(Obviously (a,b) = b in the case r = 0.)

Proof

We show that d := (b,r ) has the defining properties of (a,b).

Obviously d|b and d|r, so that d| (qb + r ) and d|b.

| {z }
=a

Suppose c ∈ N satisfies c| a and c|b. It follows that c|b and c| (a − bq), so that
| {z }
=r
c|d (since d = (b,r )). 

Example

Calculate (2406,654).

32
 1.4 Working with modular arithmetic

Solution

We apply the above proposition iteratively:

2406 = 654.3 + 444 (2406,654) = (654,444)

654 = 444.1 + 210 = (444,210)
444 = 210.2 + 24 = (210,24)
210 = 24.8 + 18 = (24,18)
24 = 18.1 + 6 = (18,6)
18 = 6.3 =6

Remark

This is the euclidian algorithm for computing (a,b), which can be summarised
as follows:

a = bq1 + r1 , (0 ≤ r 1 < b ) (a,b) = (b,r1 )

b = r 1 q2 + r 2 , (0 ≤ r 2 < r 1 ) (b,r1 ) = (r1 ,r2 )
r 1 = r 2 q3 + r 3 , (0 ≤ r 3 < r 2 ) (r1 ,r2 ) = (r2 ,r3 )
.. .. ..
. . .
r k−3 = r k−2 qk−1 + r k−1 (0 ≤ r k − 1 < r k − 2 ) (rk−3 ,rk−2 ) = (rk−2 ,rk−1 )
r k−2 = r k−1 qk + r k (rk−2 ,rk−1 ) = rk−1
|{z}
=0
The algorithm terminates after at most b steps since the remainders become strictly
smaller at each step and each is smaller than or equal to b.

Proposition

Let a and b be natural numbers and d = (a,b). There are integers u, v such that

d = ua + vb.

Proof

We can compute u and v using the extended euclidean algorithm:

33
 1.4 Working with modular arithmetic

Let u0 = 1, u1 = 0 and v0 = 0, v1 = 1.

For i = 1, . . . , k let

ui + 1 = ui − 1 − q i ui , v i + 1 = v i − 1 − qi v i .

By induction we can easily show that

ri −1 = ui a + vi b, i = 2, . . . ,k,

and in particular we have that

d = uk a + vk b. 

Remark

The integers u and v are called Bézout coefficients.

Example

Let u0 = 1, u1 = 0 and v0 = 0, v1 = 1, and

2406 = 654.3 + 444, u2 = u0 − 3u1 = 1, v2 = v0 − 3v1 = −3,

654 = 444.1 + 210, u3 = u1 − u2 = −1, v3 = v1 − v2 = 4,
444 = 210.2 + 24, u4 = u2 − 2u3 = 3, v4 = v2 − 2v3 = −11,
210 = 24.8 + 18, u5 = u3 − 8u4 = −25, v5 = v3 − 8v4 = 92,
24 = 18.1 + 6, u6 = u4 − u5 = 28, v6 = v4 − v5 = −103,
18 = 6.3

It follows that
6 = 28.2406 − 103.654

Proposition

Suppose that n ≥ 2 and a ∈ N with (n,a) = 1. There exists v ∈ N with

va ≡ 1 (mod n).

34
 1.4 Working with modular arithmetic

Proof

Since (n,a) = 1 there are Bézout coefficients u, v such that

un + va = 1.

It follows that
va ≡ 1 (mod n). 

Lemma

Suppose that p ≥ 2. It follows that Z p is a field if and only if p is a prime number.

Proof

Suppose that p is a prime number. We show that [ a] with a ∈ {1, . . . , p − 1}

has a multiplicative inverse in Z p . (We already know that the remaining field
axioms are satisfied.)

Since ( p,a) = 1 there exists v ∈ N with

va ≡ 1 (mod p).

It follows that
[v][ a] = [1]
in Z p , i.e [v] is the multiplicative inverse of [ a].

Suppose that p is not a prime number. There exist b1 , b2 ∈ {2, . . . , p − 1} with

p = b1 b2 and hence
[b1 ][b2 ] = [0].
It follows that Z p is not a field, since the product of two non-zero numbers in
a field is itself non-zero. 

Example

Find [6533]−1 in Z7039 .

35
 1.4 Working with modular arithmetic

Solution

We have that
581.7039 − 626.6533 = 1,
so that
[6533]−1 = [−626] = [6413]
in Z7039 .

Proposition (Chinese remainder theorem)

Let m, n be natural numbers with m,n ≥ 2 and (n,m) = 1 and a, b ∈ Z. The

simultaneous equations

x ≡ a (mod n),
x ≡ b (mod m)

are solvable and the solution

x = avm + bun,

where u, v are Bézout coefficients with

un + vm = 1,

is unique up to an additive integer multiple of nm.

Proof

Since (m,n) = 1 there are Bézout coefficients u, v such that

un + vm = 1.

It follows that
x = avm + bun
is a solution to the given equations because

x = a + (− a + b)un ⇒ x ≡ a (mod n),

x = b + (a − b)vm ⇒ x ≡ b (mod m).

Suppose x is a solution. Clearly x + pnm is also a solution for each p ∈ Z.

36
 1.4 Working with modular arithmetic

Suppose x1 and x2 are solutions. It follows that

x1 ≡ x2 (mod n), x1 ≡ x2 (mod m),

i.e. n|( x1 − x2 ) und m|( x1 − x2 ). Since n and m have no common divisors this
implies that nm|( x1 − x2 ) (see below). 

Example

Two cog wheels with respectively 13 and 17 teeth fit together. The teeth and gaps
are numbered from respectively 1 to 13 and 1 to 17 (in the direction of rotation).
Tooth 1 sits in gap 1. After how many clicks does tooth 4 of the small wheel sit in
gap 6 of the larger wheel?

Solution

Let x − 1 be the number of clicks, so that

x ≡ 4 (mod 13),
x ≡ 6 (mod 17)

We have that (13,17) = 1 because 13 and 17 are prime numbers and

4.13 − 3.17 = 1,

so that
x = −4.3.17 + 6.4.13 = 108
solves the above equations.

The equations are equivalent to

x ≡ 108 (mod 221),

so that the wheels are in the shown positions after 108-1, 329-1, 550-1, . . . clicks.

37
 1.4 Working with modular arithmetic

Prime numbers

Proposition

Suppose that a and b are natural numbers and p is a prime number such that p| ab.
It follows that p| a or p|b.

Proof

Suppose that p6 | a, so that ( p,a) = 1. There exist Bézout coefficients u and v with

1 = pu + av,

and it follows that

b = bpu + abv . 
| {z }
divisible by p

Remark

This result does not hold if p is not a prime number. For example 4|2.6 but 46 |2
and 46 |6.

Corollary

A prime number which divides a product a1 · · · am of natural numbers divides

one of the factors.

Corollary

Suppose that (a,n) = 1 and (b,n) = 1. It follows that (ab,n) = 1.

Proof

If ab and n have a common factor, then they have a common prime factor p. But
p| ab implies p| a or p|b. 

38
 1.4 Working with modular arithmetic

Definition

A composite number is a natural number which is bigger than 1 and not a prime
number.

Remark

The expression “composite number" alludes to the fact that every natural num-
ber n > 1 which is not prime can be reduced to a product of (powers of) prime
numbers. For example:

360 = 3.120
= 3.30.4
= 3.3.10.2.2.
= 2.2.2.3.3.5
| {z }
prime decomposition
of 360

Theorem (prime decomposition theorem)

Every natural number n > 1 has a unique prime decomposition.

Proof

Suppose there is a natural number n > 1 without a prime decomposition.

There exists a smallest such natural number N.

Note that N is not a prime number. There therefore exist natural numbers a,
b > 1 such that
N = ab.
Since a,b < N they both have prime decompositions, so that N = ab also has
a prime decomposition. This is a contradiction.

Suppose there is a natural number n > 1 with two different prime decompo-
sitions. There exists a smallest such natural number N.

39
 1.4 Working with modular arithmetic

The two prime decompositions of N have no common prime factor p (for

otherwise N/p would be a smaller natural number with two different prime
decompositions). There are therefore prime numbers p, q and natural num-
bers a, b such that
N = pa = qb
but q6 | a and p6 |b. It follows that q| pa but q6 | p and q6 | a, and similarly p|qb but
p6 |q and p6 |b. This is a contradiction. 

Definition

Let n ≥ 2 and φ(n) be the number of elements a in {1, . . . ,n}, which have no
common factors with n, i.e. (a,n) = 1. We call φ : {2,3, . . . } → N Euler’s totient
function.

Example

φ(6) = 2 since 1, 5 have no common factors with 6.

Proposition (properties of Euler’s totient function)

1. If p is a prime number and k ≥ 1, then φ( pk ) = pk (1 − 1p ).

2. If p, q are prime numbers, then φ( pq) = ( p − 1)(q − 1).

Proof

1. Among the numbers 1, 2, . . . , pk only p, 2p, 3p, . . . , pk−1 p have common

factors with pk . It follows that

φ( pk ) = pk − pk−1 = pk (1 − 1p ).

2. Among the numbers 1, 2, . . . , pk only p, 2p, 3p, . . . , qp and q, 2q, 3q, . . . , pq

have common factors with pq (note that we have counted the number pq
twice). It follows that

φ( pq) = pq − p − q + 1 = ( p − 1)(q − 1). 

40
 1.4 Working with modular arithmetic

Lemma (Euler’s theorem)

Let n ≥ 2 and a ∈ N with (a,n) = 1. It follows that

aφ(n) ≡ 1 (mod n).

Proof

Consider the set S1 = {m1 ,m2 , . . . ,mφ(n) } of numbers in {1, . . . ,n} which have no
factors in common with n. The elements of the set S2 = { am1 ,am2 , . . . ,amφ(n) } also
have no factors in common with n. Calculating modulo n, we conclude that the
sets S1 und S2 are identical. It follows that

m1 m2 · · · mφ(n) ≡ aφ(n) m1 m2 · · · mφ(n) (mod n).

Since m j and n have no common factor there exists M j with

M j m j ≡ 1 (mod n),

so that we can cancel m1 , . . . , mφ(n) in the previous equation using M1 , . . . , Mφ(n) .

We thus find that
1 ≡ aφ(n) (mod n). 

Corollary (Fermat’s little theorem)

Let p be a prime number and a ∈ N with (a,p) = 1. It follows that

a p−1 ≡ 1 (mod p).

Proof

This result follows from Euler’s theorem since φ( p) = p − 1. 

41
 2 Vector spaces

2 Vector spaces

2.1 Introduction

In elementary mathematics one thinks of a vector as an arrow in the plane or in

space. The arrow has a direction and a length.

x The vectors x und y repre-

sent translations.

0 The zero vector 0 is an

y arrow with length 0 (‘no
translation’).

x Two vectors x and y are added

by performing the translations
x+y y one after another.
y y+x
Obviously x + 0 = x
x (and 0 + x = x).

x
The order in which the translations
y are performed is irrelevant, so that
y
x + y = y + x.

x x+y For three vectors x, y, z we also

= have that (x + y) + z = x + (y + z).
y+x

42
 2.1 Introduction

x
For every vector x there is a vector − x
which represents the opposite transla-
tion.

−x Obviously x + (−− x) = 0
− x) + x = 0).
(and (−

We can also scale these vectors:

3
2
x The vector αx has the same
x direction as x and α times
the length.

− 32 x For α < 0 the vector αx

has the opposite direction to
x and |α| times the length.

We find the following rules for scalar multiplication. For all vectors x, y and all
scalars α and β we have that

(α + β)x = αx + βx,

(α · β)x = α( βx),

1x = x,

α(x + y) = αx + αy.

Now we use these properties of vectors as an abstract definition of these objects.

Definition

Suppose that V is a set equipped with a binary operation + which has the follow-
ing properties.

43
 2.1 Introduction

(V1) (u + v) + w = u + (v + w) ∀u, v, w ∈ V,
(V2) v+ w = w+ v ∀v, w ∈ V,
(V3) There exists an element 0 ∈ V (‘the zero vector’) such that
v + 0 = 0 + v = v ∀v ∈ V,
(V4) For every v ∈ V there exists a unique
Element − v ∈ V (‘the inverse vector to ’) such that
− v) = 0.
− v + v = v + (−

Now let (K, + ,·) be a field with identities 0 and 1. We say that V is a vector space
over K if there exists a mapping
K × V → V, (α, v) 7→ αv (⋆)
with the properties

(S1) (α + β)v = αv + βv ∀α, β ∈ K, v ∈ V,

(S2) (α · β)v = α( βv) ∀α, β ∈ K, v ∈ V,
(S3) 1v = v ∀v ∈ V,
(S4) α(v + w) = αv + αw ∀α ∈ K, v, w ∈ V.

The elements of V and K are respectively vectors and scalars, the binary opera-
tion + is vector addition and the mapping (⋆) is scalar multiplication.

In the special cases K = R and K = C we refer to V as a real or complex vector

space.

Proposition

Let V be a vector space over a field K. It follows that

0v = 0
and
(−1)v = − v
for all v ∈ V.

Proof

We have that
0v = (0 + 0)v
= 0v + 0v (S1)

44
 2.1 Introduction

and therefore

0 = − 0v + 0v (definition of − 0v)
= − 0v + (0v + 0v)
− 0v + 0v) + 0v
= (− (V1)
= 0 + 0v (definition of − 0v)
= 0v (V3)

Moreover
v + (−1)v = 1v + (−1)v (S3)
= (1 − 1 ) v (S1)
= 0v
= 0,
so that
− v = (−1)v
(because of the uniqueness of inverse vectors). 

Remark

We usually write + and − as + and − since the meaning is clear from the context.
In most examples we do not write the vectors in bold.

Examples

1. The set
R n = {x = ( x1 , . . . ,xn ) : x1 , . . . , xn ∈ R }
is a real vector space with componentwise vector addition

x + y : = ( x1 + y1 , . . . , x n + y n )

and scalar multiplication

αx := (αx1 , . . . , αxn ).

The zero vector is

0 = (0, . . . ,0)
and the inverse vector to x is

−x = (− x1 , . . . , − xn ).

45
 2.1 Introduction

In the cases n = 2 and n = 3 we can represent such vectors as arrows in the

plane or in space.



The vector x = ( x1 ,x2 ) represents



x x2 a translation through a distance

 x1 in the horizontal direction and
through x2 in the vertical direc-


 tion.








x1

2. Let K be a field. The set K n is a vector space over K with componentwise

vector addition and scalar multiplication.

In linear algebra we usually write the n-tuple (k1 , . . . , kn ) in the vector space
K n as a column vector  
k1
 .. 
 . ,
kn
and we now adopt this notation.

3. The set {0} (with trivial vector addition and scalar multiplication) is a vec-
tor space over every field.

4. The set R m×n of all (m × n) real matrices is a real vector space with compo-
nentwise vector addition

(aij ) + (bij ) := (aij + bij ),

i.e.
   
a11 a12 · · · a1n b11 b12 · · · b1n
 a a22 · · · a2n   b b22 · · · b2n 
 21   21 
 .. .. ..  +  .. .. .. 
 . . .   . . . 
am1 am2 · · · amn bm1 bm2 · · · bmn
 
a11 + b11 a12 + b12 · · · a1n + b1n
a +b
21 a22 + b22 · · · a2n + b2n 

 21
:=  .. .. .. 
 . . . 
am1 + bm1 am2 + bm2 · · · amn + bmn

and scalar multiplication

α(aij ) := (αaij ),

46
 2.2 Elementary vector-space theory

i.e.    
a11 a12 · · · a1n αa11 αa12 · · · αa1n
 a a22 · · · a2n   αa αa22 · · · αa2n 
 21   21 
α  .. .. ..  : =  .. .. ..  .
 . . .   . . . 
am1 am2 · · · amn αam1 αam2 · · · αamn
The zero vector is the zero matrix (0) whose coefficients are all 0 and the
inverse vector to (aij ) is the matrix (− aij ).

5. The set K m×n of all (m × n) matrices with entries from the field K is a vector
space K with componentwise vector addition and scalar multiplication.

Note that K m×1 = K m .

6. The set of all functions f : R → R ( f : C → C) ist a real (complex) vector

space with pointwise vector addition
( f + g)( x ) := f ( x ) + g( x ), x ∈ R (C )
and scalar multiplication
(α f )( x ) := α f ( x ), x ∈ R (C ).

7. The set
R [ x ] = {αn x n + αn−1 x n−1 + · · · + α1 x + α0 : n ∈ N0 , α0 , . . . ,αn ∈ R }
of all real polynomials is a real vector space with pointwise vector addition
and scalar multiplication.

8. Suppose that n ∈ N0 . The set

R n [ x ] = {αn x n + αn−1 x n−1 + · · · + α1 x + α0 : α0 , . . . ,αn ∈ R }
of all real polynomials with degree less than or equal to n is a real vector
space with pointwise vector addition and scalar multiplication.

2.2 Elementary vector-space theory

Definition

Let V be a vector space over a field K and v1 , v2 , . . . , vn be vectors in V. A linear

combination of v1 , v2 , . . . , vn is a vector of the form
k 1 v1 + k 2 v2 + . . . k n v n ,
where k1 , k2 , . . . , kn are scalars.

47
 2.2 Elementary vector-space theory

Examples

     
1 1 0
1. In R2the vector is a linear combination of the vectors und
1 0 1
2
in R since      
1 1 0
=1 +1 .
1 0 1

2. In R [ x ] the polynomial x4 + x2 + 2 is a linear combination of the polynomi-

als x4 + 21 x2 , x2 + 1 and 1, since
x4 + x2 + 2 = 1( x4 + 21 x2 ) + 12 ( x2 + 1) + 23 (1).

Definition

Let V be a vector space over a field K and v1 , v2 , . . . , vn be vectors in V. The span

of these vectors is the set
h v1 , . . . , v n i = { k 1 v1 + k 2 v2 + · · · k n v n : k 1 , k 2 , . . . , k n ∈ K }
of all possible linear combinations of v1 , v2 , . . . , vn .

Example

1. In R [ x ] we have that
h1, x, x2 i = {α0 + α1 x + α2 x2 : α0 , α1 , α2 ∈ R } = P2 (R ).

2. Let v1 , v2 6= 0 be two non-parallel vectors in R3 (i.e. we cannot write v2 =

λv1 for any λ ∈ R). We find that
h v1 , v2 i = { x = α 1 v1 + α 2 v2 : α 1 , α 2 ∈ R }
is a plane through the origin:

v2
v1

48
 2.2 Elementary vector-space theory

Remark

We usually define h∅i = {0}.

Definition

A vector space V is said to be finite-dimensional if there is a finite subset S of V

such that hSi = V. Otherwise V is infinite-dimensional.

Examples

1. The vector space R2 is finite-dimensional since

  
2 α1
R = : α1 , α2 ∈ R
α2
     
1 0
= α1 + α2 : α1 , α2 ∈ R
0 1
   
1 0
= , .
0 1

2. The vector space R [ x ] is infinite-dimensional. Suppose that S is a nonempty

finite subset of R [ x ]. Set m = max{deg p : p ∈ S}. It follows that x m+1 6∈ hSi,
so that R [ x ] 6= hSi.

3. The vector space {0} (over any field) is finite-dimensional since {0} = h∅i.

Definition

Let V be a vector space over a field K. The vectors v1 , v2 , . . . , vn in V are said to

be linearly independent if

α1 v1 + · · · + α n v n = 0 ⇒ α1 = 0, . . . , αn = 0.

Otherwise they are linearly dependent.

49
 2.2 Elementary vector-space theory

Remark

If we consider
α1 v1 + · · · + α n v n = 0
as an equation for α1 , . . . , αn , we always have the solution α1 = 0, . . . , αn = 0. The
linear independence of v1 , v2 , . . . , vn is the statement that this is the only solution.

Examples

     
1 0 0
3
1. In R the vectors 0 , 0 , 1 are linearly independent: it follows
    
0 1 0
from        
1 0 0 0
α 1 0  + α 2  1  + α 3  0  = 0 
0 0 1 0
| {z
  }
α1
= α2 

α3
that α1 = 0, α2 = 0, α3 = 0.
   
1 1
2. In R3 the vectors 1, −1 are linearly independent: it follows from
0 0
     
1 1 0
α 1  1  + α 2  − 1  =  0
0 0 0
|  {z  }
α1 + α2
=  α1 − α2 
0

that α1 + α2 = 0, α1 − α2 = 0 and hence α1 = 0, α2 = 0.

3. In F (R ) the functions sin(·) und cos(·) are linearly independent: it follows

from
α1 sin x + α2 cos x = 0, for all x ∈ R
that α1 = 0 (set x = π2 ) and α2 = 0 (set x = 0).

50
 2.2 Elementary vector-space theory
     
1 1 0
4. In R2 the vectors , , are linearly dependent since
1 0 1
       
1 1 0 0
1 −1 −1 = .
1 0 1 0

5. In R [ x ] the polynomials x4 + x2 + 2, x4 + 21 x2 , x2 + 1, 1 are linearly depen-

dent since

1( x4 + x2 + 2) − 1( x4 + 12 x2 ) − 21 ( x2 + 1) − 32 (1) = 0.

6. {0} is a linearly dependent subset of any vector space since

α0 = 0

for all scalars α, and in particular for α = 1.

Remark

We usually regard ∅ als linearly independent.

Proposition

Let V be a vector space. The vectors v1 , . . . , vn are linearly independent if and

only if no vector vi can be written as a linear combination of the vectors v1 , . . . ,
vi −1 .

Proof

Suppose that vi can be written as a linear combination of the vectors v1 , . . . ,

vi −1 , i.e. there exist scalars k1 , . . . , ki −1 such that

vi = k 1 v1 + · · · + k i −1 vi −1 .

It follows that

k1 v1 + · · · + ki −1 vi −1 − vi + 0vi +1 + · · · + 0vn = 0,

so that v1 , . . . vn are linearly dependent.

51
 2.2 Elementary vector-space theory

Now suppose that v1 , . . . vn are linearly dependent. There exist scalars k1 , . . . ,

kn – not all zero – such that
k1 v1 + · · · + kn vn = 0.
Let i = max{ j : k j 6= 0} (so that k j+1 , . . . , kn = 0). It follows that
k1 k
vi = − v1 − · · · − i −1 vi −1 ,
ki ki
so that vi can be written as a linear combination of the vectors v1 , . . . , vi −1. 

Remark

We can similarly show that v1 , . . . , vn are linearly independent if and only if no

vector vi can be written as a linear combination of the other vectors v1 , . . . , vi −1 ,
vi +1 , . . . , v n .

Proposition

Let V be a vector space, v1 , . . . , vn be linearly independent vectors and

v ∈ hv1 , . . . , vn i. It follows that the coefficients α1 , . . . , αn in the formula
v = α1 v1 + · · · + α n v n
are uniquely determined.

Proof

Let k1 , . . . , kn and ℓ1 , . . . ℓn be scalars with

v = k 1 v1 + · · · k n v n , v = ℓ1 v 1 + · · · + ℓ n v n .
It follows that
(k1 − ℓ1 )v1 + · · · + (kn − ℓn )vn = 0,
so that k1 − ℓ1 = 0, . . . , kn − ℓn = 0, i.e. k1 = ℓ1 , . . . , kn = ℓn . 

We henceforth restrict ourselves to finite-dimensional vector spaces.

Definition

Let V be a finite-dimensional vector space. The set {v1 , . . . , vn } is a basis for V if

v1 , . . . , vn are linearly independent and hv1 , . . . , vn i = V.

52
 2.2 Elementary vector-space theory

Examples

1. Let K be a field. The set {e1 , . . . en } with

 
0
 .. 
.
 
 0
 
e j = 1 ← j-th component
 
 0
.
 .. 
0
is the canonical basis for K n .

2. The set {1, x, . . . , x n } is the canonical basis for R n [ x ].

3. Let K be a field. The set { Eij : i = 1, . . . ,m, j = 1, . . . ,n} with

 
0 0 ... 0
0 1 . . . 0 ← i-th row
 
Eij =  .. .. .. 
. . .
0 0 ... 0
↑
j-th column
is the canonical basis for K m×n .

Remark

Let {v1 , . . . , vn } be a basis for a finite-dimensional vector space V. Every vector

v ∈ V has a unique representation
v = α1 v1 + · · · + α n v n
as a linear combination of v1 , . . . , vn .

Now we show that every finite-dimensional vector space has a basis.

Lemma

1. A nonempty finite subset {v1 , . . . ,vn } of a finite-dimensional vector space

V is a basis if and only if it is a maximal linearly independent set (i.e. no
proper superset of {v1 , . . . ,vn } is linearly independent).

53
 2.2 Elementary vector-space theory

2. A nonempty finite subset {v1 , . . . ,vn } of a finite-dimensional vector space

V is a basis if and only if it is a minimal spanning set for V (i.e. no proper
subset of {v1 , . . . ,vn } spans V).

Theorem

Every finite-dimensional vector space V has a basis.

Proof

Let E0 be a finite spanning set for V. If E0 is minimal, it is a basis for V. Otherwise

there is a vector v0 ∈ E0 such that E1 := E0 \ {v0 } is a finite spanning set for V.

Now we proceed iteratively. If E j is minimal, it is a basis for V. Otherwise there

is a vector v j ∈ E j such that E j+1 := E j \ {v j } is a finite spanning set for V.

Since E0 is finite this procedure yields a minimal spanning set for V after finitely
many steps. 

Remark

The basis for a nontrivial finite-dimensional vector space V is not unique. If

{v1 , . . . , vn } is a basis for V then {v1 − v2 ,v2 − v3 , . . . ,vn−1 − vn ,2vn } is also a basis
for V.

Next we show that all bases for a given finite-dimensional vector space at least
have the same number of elements.

Lemma (Steinitz exchange theorem)

Let V be a finite-dimensional vector space, S = {v1 , . . . ,vn } a spanning set for V

and T = {w1 , . . . ,wm } a linearly independent subset of V. It follows that m ≤ n.

54
 2.2 Elementary vector-space theory

Proof

Since hSi = V we can write w1 as a linear combination of v1 , . . . , vn . The set

{w1 ,v1 , . . . , vn }
is therefore linearly dependent. Hence there is a vector in the sequence w1 , v1 ,
. . . , vn which can be written as a linear combination of the previous vectors.
This element is v j1 for some j1 . The set
S1 = {w1 ,v1 , . . . ,v j1 −1 ,v j1 +1 , . . . ,vn }
also spans V. We have exchanged v j1 for w1 .

Since hS1 i = V we can write w2 as a linear combination of w1 , v1 , . . . , v j1 −1 ,

v j1 +1 , . . . , vn . The set
{w1 ,w2 ,v1 , . . . ,v j1 −1 ,v j1 +1 , . . . ,vn }
is therefore linearly dependent. Hence there is a vector in the sequence w1 ,
w2 , v1 , . . . , v j1 −1 , v j1 +1 , . . . , vn , which can be written as a linear combination
of the previous vectors. This element is v j2 for some j2 . The set
S2 = {w1 ,w2 ,v1 , . . . ,v j1 −1 ,v j1 +1 , . . . ,v j2 −1 ,v j2 +1 , . . . ,vn }
also spans V. We have exchanged v j2 for w2 .

Now we proceed iteratively. We can exchange v j1 , . . . , v jm for w1 , . . . , wm . In

particular, we have that n ≥ m. 

Theorem

All bases for a given finite-dimensional vector space have the same number of
elements.

Definition

The dimension of a nontrivial finite-dimensional vector space is the number of

elements in a basis and is denoted by dim V.

Remark

Since ∅ is a basis for the trivial vector space {0} we have dim{0} = 0.

55
 2.3 Subspaces

Examples

1. dim K n = n for an arbitrary field K.

2. dim R n [ x ] = n + 1.

3. dim K m×n = mn for an arbitrary field K.

Corollary

Let V be a finite-dimensional vector space with dim V = n.

1. Every linearly independent finite subset of V has at most n elements. If it

has exactly n elements it is a basis for V.

2. Every finite spanning subset of V has at least n elements. If it has exactly n

elements it is a basis for V.

2.3 Subspaces

Definition

Let V be a vector space over a field K. A subset W of V is a subspace of V if W is

a vector space over the same field K.

Remark

W 6= ∅ is a subspace of V if and only if

(i) W is closed under scalar multplication, i.e. αw ∈ W for all w ∈ W und α ∈ K.

(ii) W is closed under vector addition, i.e. w1 + w2 ∈ W for all w1 , w2 ∈ W,

((i) implies 0 ∈ W and −w ∈ W for all w ∈ W, so that (V1)–(V4) follow from (i),(ii).)

56
 2.3 Subspaces

Examples

1. Let K be a field and λ1 , . . . , λn fixed scalars in K which do not all vanish.

The hyperplane

E = {k = (k1 , . . . ,kn ) : λ1 k1 + · · · + λn kn = 0},

which obviously contains 0, is a subspace of K n :

Let k = (k1 , . . . ,kn ) ∈ E, so that λ1 k1 + · · · + λn kn = 0, and α ∈ K. It

follows that αk = (αk1 , . . . , αkn ) satisfies the equation

λ1 (αk1 ) + · · · + λn (αkn ) = 0,

so that αk ∈ E.

Let k = (k1 , . . . ,kn ), ℓ = (ℓ1 , . . . ,ℓn ) ∈ E, so that λ1 k1 + · · · + λn k1 = 0 and

λ1 ℓ1 + · · · + λn ℓn = 0. It follows that k + ℓ = (k1 + ℓ1 , . . . , kn + ℓn ) satis-
fies the equation

λ1 (k1 + ℓ1 ) + · · · + λn (kn + ℓn ) = 0,

so that k + ℓ ∈ E.

Now let K = R.

E In the case n = 2 the subspace

E = {( x,y) ∈ R2 : λ1 x + λ2 y = 0}
x
is a line through the origin.

In the case n = 3 the subspace

E
E = {( x,y,z) ∈ R3 : λ1 x + λ2 y + λ3 z = 0}
y
is a plane through the origin.

57
 2.3 Subspaces

2. Let F (R ) be the vector space of all real functions R → R. The set

R [ x ] = {αn x n + αn−1 x n−1 + · · · + α1 x + α0 : n ∈ N0 , α0 , . . . ,αn ∈ R }

of all real polynomials is a subspace of F (R ).

3. Let n ∈ N0 . The set

R n [ x ] = {αn x n + αn−1 x n−1 + · · · + α1 x + α0 : α0 , . . . ,αn ∈ R }

of all real polynomials with degree at most n is a subspace of R [ x ].

Lemma

Let V be a vector space over a field K and v1 , v2 , . . . , vn be vectors in V. The span

hv1 , . . . , vn i of v1 , . . . , vn is a subspace of V.

Proof

Obviously 0 ∈ hv1 , . . . , vn i.

Let α ∈ K and v ∈ hv1 , . . . , vn i, so that there exist k1 , . . . , kn ∈ K with

v = k 1 v1 + · · · + k n v n .

It follows that
αv = αk1 v1 + · · · + αkn vn ∈ hv1 , . . . , vn i.

Let v, w ∈ hv1 , . . . , vn i, so that there exist k1 , . . . , kn , ℓ1 , . . . , ℓn ∈ K with

v = k 1 v1 + · · · + k n v n , w = ℓ1 v 1 + · · · + ℓ n v n .

It follows that

v + w = ( k 1 + ℓ1 ) v 1 + · · · + ( k n + ℓ n ) v n ∈ h v 1 , . . . , v n i. 

Proposition

Let V be a vector space over a field K and W1 , W2 subspaces of V. It follows that

W = W1 ∩ W2 is a subspace of V.

58
 2.3 Subspaces

Proof

It follows from 0 ∈ W1 and 0 ∈ W2 that 0 ∈ W1 ∩ W2 .

Let v ∈ W and α ∈ K. It follows from v ∈ W1 ∩ W2 that v ∈ W1 and v ∈ W2 .

Since W1 and W2 are subspaces of V, one has that αv ∈ W1 , αv ∈ W2 and hence
αv ∈ W1 ∩ W2

Let v1 , v2 ∈ W. It follows from v1 , v2 ∈ W1 ∩ W2 that v1 , v2 ∈ W1 and v1 , v2 ∈ W2 .

Since W1 and W2 are subspaces of V, one has that v1 + v2 ∈ W1 , v1 + v2 ∈ W2
and hence v1 + v2 ∈ W1 ∩ W2 . 

Remark

The union of two susbspaces of a vector space is in general not a vector space.

y
E2
(1,1) The sets E1 = {( x,0) : x ∈ R } and
E2 = {(0,y) : y ∈ R } are subspaces
of R2 .
E1 x
Noe that (1,0) ∈ E1 , (0,1) ∈ E2
but (1,1) = (1,0) + (0,1) 6∈ E1 ∪ E2 .

Lemma

Let V be a finite-dimensional vector space. Every subspace W of V is finite-

dimensional with dim W ≤ dim V.

Furthermore dim W = dim V if and only if W = V.

Proof

Let n = dim V.

Let v1 , . . . , vm be linearly independent vectors in W and hence in V. Note that

59
 2.3 Subspaces

m ≤ n. It follows that there is a largest value m⋆ of m, for which we can find

m linearly independent vectors in W. These m⋆ vectors form a maximal linearly
independent subset of W and hence a basis for W. We conclude that dim W =
m⋆ ≤ n.

If m⋆ = n, then {v1 , . . . ,vm⋆ } is a basis for V. It follows that V = hv1 , . . . ,vm⋆ i =

W. 

W = V trivially implies that m⋆ = n. 

Example

The proper subspaces of R3 have dimension 1 or 2.

The subspaces of dimension 1 have the form hvi for some nontrivial vector v.
They are lines through the origin.

The subspaces of dimension 2 have the form hv1 , v2 i for two linearly inde-
pendent vectors v1 , v2 . They are planes through the origin.

v2
v1

60
 2.3 Subspaces

Lemma (Basis extension theorem)

Let V be a finite-dimensional vector space with dim V = n, W a subspace of V

and {v1 , . . . ,vm } a basis for W. There exist vectors vm+1 , . . . , vn ∈ V \ W such that
{v1 , . . . , vn } is a basis for V.

Proof

T := {v1 , . . . ,vm } is a linearly independent subset of V. If m < n, then it is not a

basis for V and hence not maximal. It follows that there exists vm+1 ∈ V such
that T1 := {v1 , . . . ,vm+1 } is linearly independent. We note that vm+1 is not a
linear combination of v1 , . . . , vm i.e. vm+1 6∈ hT i = W.

We continue iteratively. Tj := {v1 , . . . ,vm+ j } is a linearly independent subset

of V. If m + j < n, then it is not a basis for V and hence not maximal. It fol-
lows that there exists vm+ j+1 ∈ V such that Tj+1 := {v1 , . . . ,vm+ j+1 } is linearly
independent. We note that vm+ j+1 is not a linear combination of v1 , . . . , vm+ j
i.e. vm+ j+1 6∈ hTj i ⊇ W.

The set Tn−m = {v1 , . . . ,vn } is linearly independent and hence a basis for V.

Let W1 , W2 be subspaces of a finite-dimensional vector space V, so that W1 ∩ W2

is also a (finite-dimensional) subspace of V. Now we ask: what is dim(W1 ∩ W2 )?

Example

Let W1 and W2 be respectively a line and a plane through the origin in R3 .

61
 2.3 Subspaces

W1
W1

W2 W2

If the line W1 lies in the plane W2 , If the line W1 intersects the plane W2
then only at the origin, then

dim(W1 ∩ W2 ) = dim W1 = 1. dim(W1 ∩ W2 ) = dim{0} = 0.

The dimension of W1 ∩ W2 therefore depends not only upon dim W1 and dim W2
ab.

Definition

Let W1 , W2 be subspaces of a vector space V. The sum of W1 and W2 is the subset

W1 + W2 := {w1 + w2 : w1 ∈ W1 , w2 ∈ W2 }

of V.

Proposition

The sum of two subspaces of a vector space V is itself a subspace of V.

Theorem

Let W1 , W2 be subspaces of a finite-dimensional vector space V. It follows that

dim(W1 ∩ W2 ) + dim(W1 + W2 ) = dim W1 + dim W2 .

62
 2.3 Subspaces

Proof

First note that W1 ∩ W2 is a subspace of W1 and a subspace of W2 . Let {v1 , . . . , vr }

be a basis for W1 ∩ W2 . We extend this basis to a basis {v1 , . . . , vr , w1 , . . . , ws } for
W1 and to a basis {v1 , . . . ,vr ,z1 , . . . , zt } for W2 .

It follows that {v1 , . . . , vr , w1 , . . . ,ws ,z1 , . . . , zt } is a basis for W1 + W2 :

We have that

hv1 , . . . , vr , w1 , . . . ,ws ,z1 , . . . , zt i = hv1 , . . . , vr , w1 , . . . , ws i + hv1 , . . . , vr , z1 , . . . , zt i

= W1 + W2 .

Let α1 , . . . , αr , β 1 , . . . β s , γ1 , . . . , γt be scalars with

α1 v1 + · · · + αr vr + β 1 w1 + · · · + β s ws + γ1 z1 + · · · + γt zt = 0. (⋆)

Obviously γ1 z1 + · · · + γt zt ∈ W2 , and it follows from

γ1 z 1 + · · · + γ t z t = − α 1 v 1 − · · · − α r v r − β 1 w 1 − · · · − β s w s

that γ1 z1 + · · · + γt zt ∈ W1 . Hence γ1 z1 + · · · + γt zt lies in W1 ∩ W2 and can

therefore be written as a linear combination

γ1 z 1 + · · · + γ t z t = λ 1 v 1 + · · · + λ r v r

of v1 , . . . , vr . It follows from

γ1 z 1 + · · · + γ t z t − λ 1 v 1 − · · · − λ r v r = 0

and the linear independence of v1 , . . . , vr , z1 , . . . , zt that γ1 = 0, . . . , γt = 0

(and also λ1 = 0, . . . , λr = 0). Finally (⋆) and the linear independence of v1 ,
. . . , vr , w1 , . . . , vs imply that α1 = 0, . . . , αr = 0, β 1 = 0, . . . , β s = 0.

The vectors v1 , . . . , vr , w1 , . . . ,ws ,z1 , . . . , zt are therefore linearly independent.

Hence

dim(W1 + W2 )
=r+s+t
= dim(W1 ∩ W2 ) + (dim W1 − dim(W1 ∩ W2 )) + (dim W2 − dim(W1 ∩ W2 ))
= dim W1 + dim W2 − dim(W1 ∩ W2 ). 

63
 2.3 Subspaces

Definition

Let W1 , W2 be subspaces of a vector space V. The sum of W1 and W2 is called the

direct sum of W1 and W2 if each w ∈ W1 + W2 has a unique representation

w = w1 + w2 , w1 ∈ W1 , w2 ∈ W2 .

In this case we write the sum as W1 ⊕ W2 .

Lemma

Let W1 , W2 be subspaces of a vector space V. The sum of W1 and W2 is a direct

sum if and only if W1 ∩ W2 = {0}.

Proof

Suppose that the sum of W1 and W2 is direct. Choose w ∈ W1 ∩ W2 . We can

write
w = |{z} 0 = |{z}
w + |{z} 0 + |{z}
w ,
∈ W1 ∈ W2 ∈ W1 ∈ W2
and the uniqueness of the representation implies that w = 0.

Now suppose that W1 ∩ W2 = {0}. Suppose that v ∈ V can be written as

v = y1 + y2 and v = z1 + z2 with y1 , z1 ∈ W1 and y2 , z2 ∈ W2 . It follows that

y 1 + y 2 = z1 + z2 ⇒ y −z =z −y ,
| 1 {z }1 |2 {z }2
∈ W1 ∈ W2

Hence y1 − z1 and z2 − y2 lie in W1 ∩ W2 and are therefore both zero. We

conclude that y1 = z1 , y2 = z2 and the representation of v is unique, i.e. the
sum is direct. 

Remarks

dim(W1 ⊕ W2 ) = dim W1 + dim W2

B is a basis for W1 ⊕ W2 if and only if it is the union of a basis B1 for W1 and

a basis B2 for W2 .

64
 2.3 Subspaces

Definition

Let V be a vector space. Two proper subspaces W1 and W2 of V are complemen-

tary if V = W1 ⊕ W2 .

Lemma

Let V be a finite-dimensional vector space. Every propert subspace W1 of V has

a complement.

Proof

Let {v1 , . . . ,vm } a basis for W1 . Extend it to a basis {v1 , . . . ,vn } for V. Note that
W1 = hv1 , . . . ,vm i and set W2 = hvm+1 , . . . ,vn i.

Obviously

V = hv1 , . . . , vn i = hv1 , . . . ,vm i + hvm+1 , . . . , vn i = W1 + W2 .

Let v ∈ W1 ∩ W2 . There are scalars α1 , . . . , αm and αm+1 , . . . , αn such that

v = α1 v1 + · · · + α m v m , v = α m+1 v m+1 + · · · + α n v n .

It follows that

α1 v1 + · · · + αm vm − αm+1 vm+1 − · · · − αn vn = 0,

and the linear independence of v1 , . . . , vn implies that α1 = 0, . . . , αn = 0. We

conclude that v = 0. Hence W1 ∩ W2 = {0}. 

Remark

The complement of a proper subspace of a finite-dimensional vector space is not

unique.The ‘horizontal’ line E1 = he1 i in R2 is e.g. complemented by every other
line E2 = hvi (where v and e1 are linearly independent).

65
 2.3 Subspaces

E2

We have that
v
R2 = E1 ⊕ E2
e1 E1
with E1 = he1 i, E2 = hvi, where v is
not parallel to e1 .

66
 3 Linear transformations and matrices

3 Linear transformations and matrices

3.1 Linear transformations

Mappings between vector spaces which are compatible with the vector-space
structure are particularly important.

Definitions

Let V and W be vector spaces over a field K. A mapping T : V → W is said to be

a linear transformation if

T ( v1 + v2 ) = T ( v1 ) + T ( v2 ) for all v1 , v2 ∈ V

and
T (αv) = αT (v) for all v ∈ V and α ∈ K.

Examples

1. Let V and W be vector spaces over a field K. The zero transformation

T : V → W with T (v) = 0 for all v ∈ V is linear. If V is a suspace of W the
identity transformation I : V → W with I (v) = v is also linear.

2. The projections P1 , P2 : R2 → R2 given by the formulae

       
x x x 0
P1 = , P2 =
y 0 y y

are linear transformations

P2 [U] U The transformations P1 and P2

project the shadows of a set
U respectively vertically and
horizontally.

P1 [U]

67
 3.1 Linear transformations

3. Let w 6= 0 be a fixed complex number. The combined rotation and scaling

T : C → C with T (z) = wz is a linear transformation.

Im

wz

z
Rr
r

θ +φ
θ

Re

T is a combined rotation and scaling with scale factor R = |w| and

angle of rotation φ = arg w. (Here r = |z| and θ = arg z.)

4. The formula T ( p) = p′ defines a linear transformation R [ x ] → R [ x ] since

( p1 + p2 )′ = p1′ + p2′ and (αp)′ = αp′ .

Proposition

Let V and W be vector spaces over a field K. The set of all linear transformations
T : V → W is a vector space over K with pointwise addition
(T1 + T2 )(v) := T1 (v) + T2 (v), v∈V
and scalar multiplication
(αT )(v) := αT (v), v ∈ V.
(We denote this vector space by Hom(V,W ).)

Definitions

Let V and W be vector spaces over a field K and T : V → W be a linear transfor-

mation.

1. The kernel of T is the subset

ker T := {v ∈ V : T (v) = 0}
of V.

68
 3.1 Linear transformations

2. The image of T is the subset

Im T = { T (v) : v ∈ V }

of W.

Proposition

Let V and W be vector spaces over a field K and T : V → W be a linear transfor-

mation.

1. ker T is a subspace of V.

2. Im T is a subspace of W.

Proof

1. ker T is not empty since 0 ∈ ker T. Suppose that v1 , v2 ∈ ker T, so that

T (v1 ) = 0, T (v2 ) = 0. It follows that T (v1 + v2 ) = T (v1 ) + T (v2 ) = 0, so
that v1 + v2 ∈ ker T. Now suppose that α ∈ K, v ∈ ker T. It follows that
T (αv) = αT (v) = 0, so that αv ∈ ker T.

2. Im T iis not empty since 0 = T (0) ∈ Im T. Suppose that w1 , w2 ∈ Im T. There

therefore exist v1 , v2 ∈ V with w1 = T (v1 ), w2 = T (v2 ). It follows that w1 +
w2 = T (v1 ) + T (v2 ) = T (v1 + v2 ) ∈ Im T. Now suppose that α ∈ K, w ∈ Im T.
It follows from w = T (v) for some v ∈ V that αw = αT (v) = T (αv) ∈ Im T.

Lemma

Let V and W be vector spaces over a field K. A linear transformation T : V → W

is injective if and only if ker T = {0}.

Proof

Suppose that ker T = {0}.

69
 3.1 Linear transformations

Note that

T ( x ) = T (y) ⇒ T ( x ) − T (y) = 0
⇒ T ( x − y) = 0
⇒ x − y ∈ ker T,

so that x − y = 0, i.e. x = y. It follows that T is injective.

Now suppose that T is injective, i.e. T ( x ) = T (y) implies x = y. Hence z ∈

ker T, so that T (z) = T (0) = 0, implies that z = 0. It follows that ker T = {0}.

Proposition

Let V and W be vector spaces over a field K und T : V → W be a linear transforma-

tion. Suppose that {e1 , . . . , en } is a basis for V. It follows that Im T = hT (e1 ), . . . ,T (en )i.

Proof

Suppose that w ∈ Im T. There therefore exists v ∈ V with T (v) = w. We can

write v as a linear combination

v = α1 e1 + · · · + α n e n ,

so that

w = T (v) = α1 T (e1 ) + · · · + αn T (en ) ∈ hT (e1 ), . . . ,T (en )i.

Hence Im T ⊆ hT (e1 ), . . . ,T (en )i.

On the other hand, it follows from w ∈ hT (e1 ), . . . ,T (en )i that w can be written
as a linear combination

w = β 1 T (e1 ) + · · · + β n T (en ) = T ( β 1 e1 + · · · + β n en ) ∈ Im T.

Hence hT (e1 ), . . . ,T (en )i ⊆ Im T. 

Definitions

Let V and W be vector spaces over a field K and T : V → W be a linear transfor-

mation.

70
 3.1 Linear transformations

1. The rank of T (rank T) is the dimension of the image Im T of T.

2. The nullity of T (null T) is the dimension of the kernel ker T of T.

Lemma (rank-nullity formula)

Let V and W be vector spaces over a field K and T : V → W be a linear transfor-

mation. Furthermore, suppose that V is finite-dimensional. It follows that Im T is
finite-dimensional and
dim V = rank T + null T.

Proof

Let {v1 , . . . ,vm } be a basis for ker T and extend it to a basis {v1 , . . . ,vn } for V. We
show that { T (vm+1 ), . . . ,T (vn )} is a basis for Im T. It follows that Im T is finite-
dimensional with dim Im T = n − m = dim V − dim ker T.

Note that

hT (vm+1 ), . . . ,T (vn )i = hT (v1 ), . . . , T (vm ),T (vm+1 ), . . . ,T (vn )i = Im T.

Now we show that { T (vm+1 ), . . . ,T (vn )} is linearly independent. Let αm+1 ,

. . . , αn be scalars with

αm+1 T (vm+1 ) + · · · + αn T (vn ) = 0,

so that
T (αm+1 vm+1 + · · · + αn vn ) = 0.
Hence αm+1 vm+1 + · · · + αn vn ∈ ker T and can therefore be written as a linear
combination of {v1 , . . . ,vm }. There therefore exist scalars α1 , . . . , αm such that

α m +1 v m +1 + · · · + α n v n = α1 v1 + · · · + α m v m ,

and it follows from

α1 v1 + · · · + α m v m − α m +1 v m +1 − · · · − α n v n = 0

and the linear independence of {v1 , . . . ,vn } that α1 = 0, . . . , αn = 0.

71
 3.2 Matrix representations of linear transformations

Corollary

Let V and W be finite-dimensional vector spaces over a field K with dim V =

dim W and T : V → W be a linear transformation.

T is injective if and only if it is surjective.

Proof

T ist surjective ⇔ Im T = W
⇔ rank T = dim W (Im T is a subspace of W)
⇔ null T = 0
⇔ ker T = {0}
⇔ T is injective. 

3.2 Matrix representations of linear transformations

In this section V und W are finite-dimensional vector spaces over a field K and T :
V → W is a linear transformation. Let {e1 , . . . , en } be a basis for V and { f 1 , . . . , f m }
be a basis for W (so that n = dim V and m = dimW).

Every element v ∈ V can be written as a unique linear combination of the basis

vectors e1 , . . . , en : there are unique scalars α1 , . . . , αn such that

v = α1 e1 + α2 e2 + · · · + α n e n . (1 )

It follows that
T ( v ) = α 1 T ( e 1 ) + α 2 T ( e 2 ) + · · · + α n T ( e n ). (2 )

Remark

The n vectors T (e1 ), . . . , T (en ) determine the linear transformation T uniquely:

with this information we can reconstruct T (v) for every v using (1) and (2).

72
 3.2 Matrix representations of linear transformations

On the other hand T (e1 ), . . . , T (en ) are elements of W and can therefore be written
as unique linear combinations of the basis vectors f 1 , . . . f m : there are unique
scalars αij , i = 1, . . . , m, j = 1, . . . ,n such that
T (e1 ) = α11 f 1 + α21 f 2 + · · · + αm1 f m ,
T (e2 ) = α12 f 1 + α22 f 2 + · · · + αm2 f m ,
.. .. ..
. . .
T (en ) = α1n f 1 + α2n f 2 + · · · + αmn f m .

Remark

The scalars αij , i = 1, . . . , m, j = 1, . . . ,n determine the linear transformation T

uniquely: with this information we can reconstruct T (e1 ), . . . , T (en ) und hence
T (v) for each v.

Definition

The m × n matrix  
α11 α12 · · · α1n
 α α22 · · · α2n 
 21 
A = (αij ) =  .. .. .. 
 . . . 
αm1 αm2 · · · αmn
is the matrix of T : V → W with respect to the basis {e1 , . . . ,en } for V and { f 1 , . . . , f m }
for W.

Remark

To compute the matrix A of T, we write the image T (ei ) of ei as a linear combi-

nation of the basis vectors f 1 , . . . , f m . The coefficients are the elements of the i-th
column of A.

Example

Find the matrix of the linear transformation T : R3 → R2 with

 
x  
x + y − z
T y =
2x + z
z

73
 3.2 Matrix representations of linear transformations

with respect to

(a) the standard bases for R3 and R2 ,

     
 1 1 0     
1 1
(b) the bases  0  , 1 , 0 and , for respectively R3 and
  1 0
−1 1 1
2
R .

Solution

(a) Note that

 
1      
1 1 0
T 0 =
  =1 +2 ,
2 0 1
0
 
0      
1 1 0
T 1 =
  =1 +0 ,
0 0 1
0
 
0      
− 1 1 0
T  0 = = −1 +1 .
1 0 1
1
The matrix of T with respect to the given bases is therefore
 
1 1 −1
.
2 0 1

(b) Note that

 
1      
2 1 1
T 0  = =1 +1 ,
1 1 0
−1
 
1      
1 1 1
T  1 = =3 −2 ,
3 1 0
1
 
0      
− 1 1 1
T  0 = =1 −2 .
1 1 0
1
The matrix of T with respect to the given bases is therefore
 
1 3 1
.
1 −2 −2

74
 3.3 Matrix algebra

Example

Let n ∈ N. Find the matrix of the linear transformation T : R n [ x ] → R n [ x ] with

T ( p) = p′ with respect to the standard basis for R n [ x ].

Solution

Note that

T (1) = 0 = 0.1 + 0x + 0x2 + · · · + 0x n ,

T ( x ) = 1 = 1.1 + 0x + 0x2 + · · · + 0x n ,
T ( x2 ) = 2x = 0.1 + 2x + 0x2 + · · · + 0x n ,
T ( x3 ) = 3x2 = 0.1 + 0x + 3x2 + · · · + 0x n ,
.. ..
. .
T ( x n ) = nx n−1 = 0.1 + 0x + 0x2 + · · · + nx n−1 + 0x n .

The matrix of T with respect to the standard basis for R n [ x ] is therefore

 
0 1 0 0 ··· 0
0 0 2 0 · · · 0 
 
0 0 0 3 · · · 0 
 
 .. .. .. .. ..  .
. . . . .
 
0 0 0 0 · · · n 
0 0 0 0 ··· 0

3.3 Matrix algebra

In this section X, Y and Z are finite-dimensional vector spaces over a field K and
S : X → Y, T : Y → Z are linear transformations. Furthermore X = { x1 , . . . , xn },
Y = {y1 , . . . ,ym } and Z = {z1 , . . . , zℓ } are bases for X, Y and Z.

Denote the matrices of S and T with respect to the given bases for X, Y and Z by
respectively

75
 3.3 Matrix algebra
 
s11 s12 · · · s1n
 s21 s22 · · · s2n 
X  
MY (S) = (sij ) =  .. .. .. 
 . . . 
sm1 sm2 · · · smn
and  
t11 t12 · · · t1m
Y
t21
 t22 · · · t2m  
MZ (T ) = (tij ) =  .. .. ..  ,
 . . . 
t ℓ1 t ℓ2 · · · t ℓ m
so that

S( x1 ) = s11 y1 + s21 y2 + · · · + sm1 ym ,

S( x2 ) = s12 y1 + s22 y2 + · · · + sm2 ym ,
.. .. ..
. . .
S( xn ) = s1n y1 + s2n y2 + · · · + smn ym ,

i.e.
m
S( x j ) = ∑ skj yk , j = 1, . . . ,n,
k=1
and

T (y1 ) = t11 z1 + t21 z2 + · · · + tℓ1 zℓ ,

T (y2 ) = t12 z1 + t22 z2 + · · · + tℓ2 zℓ ,
.. .. ..
. . .
T (ym ) = t1m z1 + t2m z2 + · · · + tℓm zℓ ,

i.e.
ℓ
T (y j ) = ∑ tkj zk , j = 1, . . . ,m.
k=1

Now consider the function T ◦ S, which is a linear transformation X → Z. Let

X ( T ◦ S) be its matrix with respect to the given bases for X and Z.
MZ

Now we ask the following question: can we define the product of two matrices
in such a way that the formula
X Y
MZ ( T ◦ S ) = MZ (T ).MYX (S) (⋆)

holds?

76
 3.3 Matrix algebra

Lemma

X ( T ◦ S) is the ℓ × n Matrix whose entry in row i and column j equals

MZ
m
∑ tik skj .
k=1

Proof

Let U = T ◦ S and
 
u11 u12 · · · u1n
u u22 · · · u2n 
X  21 
MZ (U ) = (uij ) =  .. .. .. 
 . . . 
u ℓ1 u ℓ2 · · · u ℓ n

be the matrix of U with respect to the given bases for X and Z, so that
ℓ
U (x j ) = ∑ uij zi , j = 1, . . . ,n. (1 )
i =1

Note that

U ( x j ) = T (S( x j ))
!
m
=T ∑ skj yk
k=1
m
= ∑ skj T (yk )
k=1
m ℓ
= ∑ skj ∑ tik zi
k=1 i =1
ℓ m
= ∑ ∑ tik skj zi . (2 )
i =1 k=1

Comparing (1) and (2) yields

m
uij = ∑ tik skj . 
k=1

The formula (⋆) is therefore correct with the following definition.

77
 3.3 Matrix algebra

Definition

Let K be a field and A ∈ K ℓ×m , B ∈ K m×n . The product AB is the matrix C = (cij )
in K ℓ×n with
m
cij = ∑ aik bkj .
k=1

Remark

We compute cij by multiplying the elements in row i of A by the elements in

column j of B and summing the results:

  
↓
 ↓ 
→ → · · · → 
i-th row 
..

 
 .


↓
j-th column

Example

      
1 2 6 5 4 3 6+4 5+2 4+0 3−2 10 7 4 1
 
3 4 2 1 0 −1 =  18 + 8 25 + 4 12 + 0 9 − 4  = 26 19 12 5 .
5 6 30 + 12 15 + 6 20 + 0 15 − 6 42 31 20 9

Remarks

1. Matrix multiplication is associative and distributive over addition, but not

commutative. For example:
    
1 2 5 6 19 22
=
3 4 7 8 43 50

but     
5 6 1 2 23 34
= .
7 8 3 4 31 46

78
 3.3 Matrix algebra

2. The set K n×n is a ring with multiplicative identity

 
1 0 ··· 0
 0 1 · · · 0
 
 .. . . ..  .
. . .
0 0 ··· 1

Lemma

Let v be a vector in X and a1 , . . . , an , b1 , . . . , bm be the coefficients in the repre-

sentations of v and S(v) as linear combinations of the bases X = { x1 , . . . ,xn } for
X and Y = {y1 , . . . ,ym } for Y, i.e.

v = a1 x 1 + · · · + a n x n ,
S(v) = b1 y1 + · · · + bm ym .

It follows that
X
b = MY (S)a,
where    
a1 b1
 .. 
b =  ...  .
 
a =  . ,
an bm

Proof

Note that
!
n
Sv = S ∑ ak xk
k=1
n
= ∑ ak S( x k )
k=1
n m
= ∑ ∑ak sik yi
k=1 i =1
m n
= ∑ ∑ sik ak yi ,
i =1 k=1

so that
n
bi = ∑ sik ak , i = 1, . . . ,m.
k=1

79
 3.3 Matrix algebra

X (S)a is an m × 1 matrix whose

Considering a as a n × 1 matrix, we find that MY
entry in row i, column 1 is
n
∑ sik ak = bi . 
k=1

Example

Let n ∈ N. The matrix of the linear transformation T : R n [ x ] → R n [ x ] with T ( p) =

p′ with respect to the standard basis for R n [ x ] is
 
0 1 0 0 ··· 0
0 0 2 0 · · · 0 
 
0 0 0 3 · · · 0 
 
 .. .. .. .. ..  .
. . . . .
 
0 0 0 0 · · · n 
0 0 0 0 ··· 0

Using this matrix we can compute p′ for each p ∈ R n [ x ]. We represent the poly-
nomial
p = α0 + α1 x + α2 x 2 + · · · + α n −1 x n −1 + α n x n
as the column vector  
α0

 α1 


 α2 

 .. ,

 . 

α n −1 
αn
and the polynomial p′ is represented by
    
0 1 0 0 ··· 0 α0 α1
0 0 2 0 · · · 0  α1   2α2 
    
0 0 0 3 · · · 0  α2   3α3 
    
 .. .. .. .. ..  ..  =  ..  .
. . . . .  .   . 
  
 
0 0 0 0 · · · nαn−1  nαn 
0 0 0 0 ··· 0 αn 0

Hence
p′ = α1 + 2α2 x + 3α3 x2 + · · · + nαn x n−1.

80
 3.3 Matrix algebra

Notation

Since S, T and v can be regarded as matrices, we usually write T (v) as Tv and

S ◦ T as ST. This notation is also used for infinite-dimensional vector spaces.

Now let X ′ = { x1′ , . . . , xn′ }, Y ′ = {y1′ , . . . ,y′m } be further bases for respectively X
and Y.

Definition

The change of basis matrix BXX from X to X ′ is the matrix of the identity trans-
′
formation I : X → X with respect to X and X ′ , i.e.
X X
BX ′ = M X ′ ( I ).

Example

Compute the change of basis matrix from the basis B1 = {1,x,x2 , . . . , x n } to the
basis B2 = {1,1 + x,1 + x + x2 , . . . , 1 + x + x2 + · · · + x n } for R n [ x ].

Remark

We write the old basis vectors als linear combinations of the new basis vectors:

1 = 1,
x = − 1 + (1 + x ),
x2 = −(1 + x ) + (1 + x + x2 ),
.. ..
. .
x n = −(1 + x + · · · + x n−1 ) + (1 + x + · · · + x n ).

Hence  
1 −1 0 0 ···
0
0
 1 −1 0 ···
0 
0 0 1 −1 ···
0
BBB21
 
= 0 0 0 1 ···
0
 
 .. .. .. .. .. 
. . . . .
0 0 0 0 ··· 1

81
 3.3 Matrix algebra

Lemma

X ′ Y X X ′
MY ′ ( S ) = BY ′ M Y ( S ) BX .

Proof

′
X (S) and M X (S) by respectively (s ) and (s′ ), so that
Denote MY Y′ ij ij
m
S( x j ) = ∑ skj yk , j = 1, . . . ,n
k=1

and
m
S( x ′j ) = ∑ s′kj y′k , j = 1, . . . ,n. (⋆)
k=1

′
X and BY by respectively (b ) and c , so that
Denote BX Y′ ij ij
n
x ′j = ∑ bkj xk , j = 1, . . . ,n
k=1

and
m
yj = ∑ ckj y′k , j = 1, . . . ,m
k=1

Note that
!
n
S( x ′j ) = S ∑ bij xi
i =1
n
= ∑ bij S( xi )
i =1
n m
= ∑ bij ∑ ski yk
i =1 k=1
n m m
= ∑ bij ∑ ski ∑ cℓk y′ℓ
i =1 k=1 ℓ=1
m m n
= ∑ ∑ cℓk ∑ ski bij y′ℓ
ℓ=1 k=1 i =1
m m
′
= ∑ ∑ cℓk ( MYX (S)BXX )kj y′ℓ
ℓ=1 k=1
m
X′
= ∑ ( BYY ′ MYX (S) BX )ℓ j y′ℓ
ℓ=1

82
 3.3 Matrix algebra

It follows that
Y ′
s′ℓ j = ( BY X X
′ M Y ( S ) BX ) ℓ j

(see (⋆)), i.e.

X ′ Y X X ′
MY ′ ( S ) = BY ′ M Y ( S ) BX . 

Example

Verify the formula

B′ B′
MB3′ (T ) = BBB2′ MBB23 (T ) BB33
2 2

for the linear transformation T : → R3 R2

with
 
x  
x + y − z
T y = ,
2x + z
z
where

B3 , B2 are the standard bases for respectively R3 and R2 ,

      
 1 1 0     
′ ′ 1 1
B3 , B2 are the bases  0  ,  1 ,  0  and , for
  1 0
−1 1 1
3 2
respectively R and R .

Solution

We have already computed

   
B3 1 1 −1 B′ 1 3 1
MB2 ( T ) = , MB3′ (T ) = .
2 0 1 2 1 −2 −2

Note that
       
1 1 0 0
 0  = 1 0  + 0 1  − 1  0  ,
−1 0 0 1
       
1 1 0 0
 1 = 1  0  + 1  1  + 1  0  ,
1 0 0 1
       
0 1 0 0
 0 = 0  0  + 0  1  + 1  0  ,
1 0 0 1

83
 3.3 Matrix algebra

so that  
1 1 0
B3′
B B3 =  0 1 0 .
−1 1 1

Note that
     
1 1 1
=0 +1 ,
0 1 0
     
0 1 1
=1 −1 ,
1 1 0

so that  
0 1
BBB2′ = .
2 1 −1

Hence
   
0 1 1 1 −1 1 1 0
B B3′
BBB2′ MB23 (T ) B B3 = 1 −1 2 0 1  0 1 0
2
−1 1 1
  
2 0 1 1 1 0
= −1 1 −2  0 1 0
−1 1 1
 
1 3 1
= .
1 −2 −2

84
 4 Matrix algebra and systems of linear equations

4 Matrix algebra and systems of linear equations

4.1 On the columns and rows of a matrix

Definition

Let K be a field and A ∈ K m×n . Write A = (aij ) i=1,...,m, . The matrix AT ∈ K n×m
j =1,...,n
defined by the formula
aTij := a ji
is the transpose of A.

Remarks

1. We obtain AT from A by exchaning the rows and columns, so that the trans-
pose of the matrix
 
a11 a12 · · · a1n  
 a a22 · · · a2n   
 21 
A =  .. .. ..  m rows,
 . . . 

am1 am2 · · · amn 
| {z }
n columns
is the matrix  
a11 a21 · · · am1  
 a a22 · · · am2   
T  12 
A =  .. .. ..  n rows.
 . . . 

a1n a2n · · · amn 
| {z }
m columns

2. The transpose of the column vector

 
v1
v =  ... 
 
vn

is the row vector

vT = (v1 , . . . ,vn )
and vice versa.

85
 4.1 On the columns and rows of a matrix

Notation

Let K be a field and A ∈ K m×n . Denote the columns and rows of A by c1A , . . . , cnA
bzw. r1A , . . . , rm
A , so dass

 
r1A
   
 
A A A
 r2A 
A = c1 c2 · · · cn  =  .
 .. 

 . 

A
rm

In this notation  
a1j
 a2j 
c jA =  .  ,
 
j = 1, . . . ,n
 .. 
amj
and
riA = (ai1 ,ai2 , . . . ,ain ), i = 1, . . . ,m.

Remark

It follows from the formula

m
cij = ∑ aik bkj
k=1

for C = AB with A ∈ K ℓ×m , B ∈ K m×n (so that C ∈ K ℓ×n ) that

m
cCj = ∑ bkj ckA , j = 1, . . . ,n
k=1

and
m
riC = ∑ aik rkB , i = 1, . . . ,ℓ.
k=1
In particular one finds that

the columns of AB are linear combinations of the columns of A,

the rows of AB are linear combinations of the rows of B.

86
 4.1 On the columns and rows of a matrix

Definitions

Let K be a field and A ∈ K m×n .

1. The column space of A is the subspace hc1A , . . . , cnA i of K m . Its dimension is

the column rank of A.

2. The row space of A is the subspace h(r1A )T , . . . , (rm

A )T i of K n . Its dimension

is the row rank of A.

Remark

The column rank (row rank) of A is the maximum number of linearly indepen-
dent columns (rows) of A.

Example

The row rank of the matrix

 
1 −2 0 4
 3 1 1 0 
 
 −1 −5 −1 8 
3 8 2 −12

is 2, since r1 and r2 are linearly independent and r3 = 2r1 − r2 , r4 = −3r1 + 2r2 .

Theorem

Let K be a field. The formula column rank A = row rank A holds for each A ∈
K m× n .

Proof

Let c = column rank A and {e1 , . . . , ec } be a basis for the column space of A. Every
column c jA of A can be written as a linear combination of e1 , . . . , ec , i.e. there are
scalars rkj such that
c
c jA = ∑ rkj ek , j = 1, . . . ,n
k=1

87
 4.1 On the columns and rows of a matrix

It follows that A = ER, where R = (rkj ) k=1,...,c, and

j =1,...,n

 

E = e1 e2 · · · ec .

Hence each row of A is a linear combination of the c rows of R, so that

row rank A ≤ c = column rank A.

The same argument yields

T T
row
| {z A } ≤ column
rank | {zrank A } .
= column rank A = row rank A
Hence column rank A = row rank A. 

In view of the previous lemma we speak only of ‘the rank’ of a matrix. Now we
turn to the question of computing it.

Proposition

Let A be a matrix with m rows whose first r elements on the diagonal are non-
zero, but whose sub-diagonal elements and last m − r rows vanish, i.e.
 
a11
 0 a22 
 .
 . .. ⋆ 

 . . 
 
 0 · · · 0 arr 
A=  (⋆)
 
 
 

 0 0 


This matrix satisfies rang A = r.

88
 4.1 On the columns and rows of a matrix

Definition

An elementary row operation on a matrix is one of the following procedures.

(i) Swap two rows;

(ii) Multiply one row by a scalar λ 6= 0;

(iii) Add an arbitrary multiple λ of a row to another row.

Elementary column operations are defined analogously.

Proposition

Elementary row (columnn) operations do not change the row (column) space of
a matrix and hence do not change its rank.

We can therefore determine the rank of a matrix by converting it into the form (⋆)
by means of elementary row and column operations.

Example

Determine the rank of the matrix

 
2 −1 3
1 0 1 
 
0 2 −1
1 1 4

Solution

Using elementary row and column operations we generate a sequence of matrices

with the same rank:

89
 4.1 On the columns and rows of a matrix

   
2 −1 3 1 0 1
1 0 1 
  2 −1 3 
 −→   (r 1 ↔ r 2 )
0 2 −1 0 2 −1
1 1 4 1 1 4
 
1 0 1
0 −1 1 
−→   (r2 → r2 − 2r1 , r4 → r4 − r1 )
0 2 −1
0 1 3
 
1 0 1
0 −1 1
−→ 
0
 (r3 → r3 + 2r2 , r4 → r4 + r2 )
0 1
0 0 4
 
1 0 1
0 −1 1
−→ 
0
 (r4 → r4 − 4r3 )
0 1
0 0 0

The rank of the last – and hence also the first – matrix is therefore 3.

Definition

An m × n matrix A is in echelon form if it satisfies the following conditions.

(i) The last m − r rows are identically zero, but the first r rows are not identically
zero. (Here 0 ≤ r ≤ m.)

(ii) Supposing that the first non-vanishing element in row i lies in columnn ji , we
have that
1 ≤ j1 < j2 < . . . < jr ≤ n.
 
0 · · · 0 a1j1 ⋆ ⋆ ⋆ ⋆ ⋆ ⋆ ⋆ ⋆
0
 ··· 0 0 · · · 0 a2j2 ⋆ ⋆ ⋆ ⋆ ⋆ 
. .. .. .. .. .. .. .. .. 
 .. . . . . . . . .
 
A=
0 ··· 0 0 ··· 0 0 · · · 0 arjr ⋆ ⋆ 
0
 ··· 0 0 ··· 0 0 ··· 0 0 ··· 0 
 .. .. .. .. .. .. .. .. 
. . . . . . . .
0 ··· 0 0 ··· 0 0 ··· 0 0 ··· 0

90
 4.1 On the columns and rows of a matrix

Proposition

The rank of a matrix in echelon form is the number of rows which are not identi-
cally zero.

Lemma (Gauß algorithm)

Any matrix can be converted into echelon form using elementary row operations.

Proof

We begin with an m × n matrix A. The proof is complete if the columns are all
identically zero.

Otherwise let j1 be the smallest index for which the column c jA1 does not vanish.

By swapping rows we can arrange that the first entry of c jA1 is not zero. With
a misuse of notation we continue to denote the entries of c jA1 by a1j1 , . . . , amj1 .

We can convert the remaining entries of this column to zero using the elemen-
tary column operations
aij1
ci → ci − c1 , i = 2, . . . ,m.
a1j1

We then obtain a matrix of the form

 
0 · · · 0 a1j1 ⋆ · · · ⋆
0 ··· 0 0 
 
 .. .. .. ′ ,
. . . A 
0 ··· 0 0

where A′ is an (m − 1) × (n − j1 ) matrix.

We obtain the stated result by iteration. 

91
 4.1 On the columns and rows of a matrix

Remark

We can convert a matrix in echelon form into the form (⋆) by swapping columns.

Remark

Let K be a field.

We know that Hom(K n ,K m ) ∼ = Km×n (the isomorphism maps a linear transfor-

mation K n → K m to its matrix with respect to the standard bases for K n and K m ).

In particular, every matrix A ∈ K m×n defines a linear transformation T : K n → K m

by the formula Tx = Ax. Moreover, the matrix of T with respect to the standard
bases for K n and K m is given by A.

1. In view of this remark we do not distinguish between T and A and write

ker A instead of ker T, Im A instead of Im T etc.

2. In this notation we have that rank A := dim Im A. It follows from

Im A = hAe1 , . . . , Aeni
| {z }
the columns of A
that dim Im A is the dimension of the column space of A. This definition
coincides with our previous definition of rank A.

3. By converting A into echelon form we find that rank A is the number of

rows which are not identically zero. According to the rank-nullity formula
null A = n − rank A.

4. A is surjective if and only if rank A = m and injective if and only if

null A = 0.

92
 4.2 The inverse of a matrix

4.2 The inverse of a matrix

Definitions

Let K be a field.

1. The n × n matrix  
1

In :=  .. 
. 
1
is the identity matrix. It has the property that

MIn = In M = M

for all M ∈ K n×n .

2. The matrix A ∈ K n×n is said to be invertible oder regular if there is a matrix

B ∈ K n×n with
AB = BA = In .
In this case B is the inverse of A and is denoted by A−1 .

Proposition

Let K be a field.

(i) The inverse of an invertible matrix in K n×n is unique.

(ii) The product of two invertible matrices A, B in K n×n is invertible with ( AB)−1 =
B −1 A −1 .

Definition

Let K be a field. The set of all invertible matrices in K n×n forms a group with
respect to matrix multiplication called the general linear group GL(n,K ).

93
 4.2 The inverse of a matrix

Example

  
a b
GL(2,R ) = : a,b,c,d ∈ R, ad 6= bc
c d
and   −1  
a b 1 d −b
= , ad 6= bc.
c d ad − bc −c a

Remark

Let n ∈ N and V be an n-dimensional vector space over a field K.. The identity
matrix In ∈ K n×n is the matrix of the identity transformation V → V, x 7→ x (with
respect to an arbitrary basis for V).

Proposition

Let n ∈ N and V be an n-dimensional vector space over a field K. and

MBB21 (T ) ∈ K n×n be the matrix of a linear transformation T : V → V with respect
to bases B1 , B2 for V.

MBB21 (T ) is invertible if and only if T is bijective. In this case we have that

MBB21 (T )−1 = MBB2 (T −1 ).
1

Proof

Suppose that T is invertible, so that TT −1 = T −1 T = I and hence

MBB22 (TT −1 ) = In , MBB1 (T −1 T ) = In ,
| {z } | 1 {z }
B1 B2 B2 B1
= MB2 ( T ) MB ( T − 1 )
1
−1
= MB ( T ) MB2 ( T )
1

It follows that MBB21 (T ) is invertible with MBB21 (T )−1 = MBB2 (T −1 ).

1

B B
Suppose that MB21 (T ) is invertible. Its inverse MB21 (T )−1 defines a linear
B
transformation S : V → V witht MBB12 (S) = MB21 (T )−1 . The matrices of the
linear transformations ST : V → V (with respect to B1 ) and TS : V → V (with
respect to B2 ) are respectively
B B B B
MB11 (ST ) = MBB12 (S) MB21 (T ) = MB21 (T )−1 MB21 (T ) = In ,

94
 4.2 The inverse of a matrix

and
MBB22 (TS) = MBB21 (T ) MBB2 (S) = MBB21 (T ) MBB21 (T )−1 = In ,
1

so that ST = TS = I. It follows that T is invertible with T = S−1 . 

Corollary

Let B1 and B2 be two bases for a finite-dimensional vector space V over a field
K.. It follows that BBB21 is invertible with ( BBB21 )−1 = BBB2 .
1

Proposition

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

1. A ist invertible.

2. A ist injective.

3. A ist surjective.

4. rank A = n.

5. AT is invertible.

Lemma

An n × n matrix A is invertible if and only if it can be converted into the identity

matrix using elementary row operations.

Proof

A is invertible if and only if rank A = n, i.e. if it can be converted into the echelon
form  
a11
 0 a22 
 
 .. .. 
 . . 
0 · · · 0 ann
using elementary row operations.

95
 4.2 The inverse of a matrix

Furthermore every matrix A can be converted into the above form using elemen-
tary row operations if and only if it can be converted into the identity matrix
using elementary row operations. 

Now we present a method to compute the inverse of a matrix with maximal rank.

Definition

An n × n elementary matrix is the result of an elementary operation on In .

The elementary operation ri ↔ r j or ci ↔ c j yields the matrix

 
.. ..
1 0 . . 
 .. .. .. 
0 . . . 
 . 

 1 .. 0 

· · · · · · · · · 0 · · · · · · 0 1 0 · · · · · ·  ← i-th row
 
 .. 

 . 1 0 


Vij =  .
. . . .
. 
. . . 
 
 .. 
 0 1 . 
 
· · · · · · 0 1 0 · · · · · · 0 · · · · · · · · ·  ← j-th row

 .. 

 0 . 1 

 .
.. .
.. . ..


 
.. ..
. . 1
↑ ↑
i-th column j-th column

The elementary operation ri → λri oder ci → λci with λ 6= 0 yields the matrix
 
1
 .. 
 . 
 
 1 
Miλ = 
 
λ  ← i-th row
 
 1 
 .. 
 . 
1
↑
i-th column

96
 4.2 The inverse of a matrix

The elementary operation ri → ri + λr j oder c j → c j + λci yields the matrix

 
.. ..
1 0 . . 
 .. .. .. 
0 . . . 
 .. 
 1 . 0 
 
· · · · · · · · · 1 · · · · · · 0 λ 0 · · · · · ·  ← i-th row
 
 .
.. 1


 0 

Aijλ = 
 .. .. .. 
. . . 
 
 .. .. 
 . 1 . 
 
· · · · · · · · · · · · · · · · · · 1 · · · · · · · · ·  ← j-th row

 .. .. 

 . . 1 
 .. .. .. 

 . . . 

.. ..
. . 1
↑ ↑
i-th column j-th column

Proposition

1. The result of an elementary row operation on an m × n matrix A is obtained

by multiplying by the corresponding m × m elementary matrix on the left.

2. The result of an elementary column operation on an m × n matrix A is ob-

tained by multiplying by the corresponding n × n elementary matrix on the
right.

Corollary

The elementary matrices are invertible with Vij−1 = Vij , ( Miλ )−1 = Mi1/λ and
( Aijλ )−1 = Aij−λ .

Corollary

Let A be an invertible n × n matrix, so that A can be converted into the identity

matrix by a sequence of elementary row operations with corresponding elemen-
tary matrices E1 , . . . , E p . It follows that
A−1 = E p E p−1 · · · E1 .

97
 4.2 The inverse of a matrix

Proof

This result follows from the equation

E p E p−1 · · · E1 A = In . 

Remarks

1. A can therefore be written as a product

A = E1−1 · · · E−
p
1

of elementary matrices.

2. The equation
A−1 = E p E p−1 · · · E1 In ,
shows that the elementary row operations which convert A into In also con-
vert In into A−1 .

Example

Compute the inverse of the matrix

 
2 −1 3
1 0 1  ∈ GL(3,R ).
0 2 −1

98
 4.3 Systems of linear equations

Solution

The calculation
   
2 −1 3 1 0 0 1 0 1 010
 1 0 1 0 1 0  −→  2 −1 3 1 0 0  (r 1 ↔ r 2 )
0 2 −1 0 0 1 0 2 −1 0 0 1
 
1 0 1 0 1 0
−→  0 −1 1 1 −2 0  (r2 → r2 − 2r1 )
0 2 −1 0 0 1
 
1 0 10 1 0
−→  0 −1 1 1 −2 0  (r3 → r3 + 2r2 )
0 0 1 2 −4 1
 
1 0 0 −2 5 −1
−→  0 −1 0 −1 2 −1  (r 1 → r 1 − r 3 , r 2 → r 2 − r 3 )
0 0 1 2 −4 1
 
1 0 0 −2 5 −1
−→  0 1 0 1 −2 1  (r 2 → − r 2 )
0 0 1 2 −4 1

shows that  −1  

2 −1 3 −2 5 −1
 1 0 1  =  1 −2 1  .
0 2 −1 2 −4 1

4.3 Systems of linear equations

In this section we consider the following problem.

Consider the system of linear equations


a11 x1 + a12 x2 + · · · + a1n xn = b1 , 




a21 x1 + a22 x2 + · · · + a2n xn = b2 , 
.. .. .. (⋆)
. . . 




am1 x1 + am2 x2 + · · · + amn xn = bm , 

where the coefficients aij and the right-hand sides bi are given. Do there exist
value of the unknowns x1 , . . . , xn which solve these equations? (Here we work
in a field K..) And how can we describe the set of solutions?

99
 4.3 Systems of linear equations

Example

The real equations

a11 x1 + a12 x2 + a13 x3 = b1 ,

a21 x1 + a22 x2 + a23 x3 = b2 ,
a31 x1 + a32 x2 + a33 x3 = b3

describe three planes, so that the solution set M of this system of equations con-
sists of the points of intersection of the planes. There are several possible geomet-
ric configurations.

All three planes coincide.

v2
v1 M = {a + λ1 v1 + λ2 v2 : λ1 ,λ2 ∈ R }
a

consists of the points of this plane and

is given by two parameters λ1 , λ2 .

Two planes coincide, the third is paral-

lel to these. M = ∅.

Two planes coincide, the third inter-

sects the common plane in a line
v

a M = {a + λv : λ ∈ R }

which is given by a parameter λ.

The planes are parallel to each other.

M = ∅.

100
 4.3 Systems of linear equations

Two planes are parallel to each other,

the other intersects each of the others
in a line. M = ∅.

The planes are all different and each

pair intersects each other in a different
line. M = ∅.

The planes are all different and inter-

sect each other in a line
v

a M = {a + λv : λ ∈ R },

which is given by a parameter λ.

The planes are all different and inter-

sect each other in a point. M consists
of this point.

Now we consider the system (⋆) more systematically. We can write it in the form

Ax = b

where A ∈ K m×n and b ∈ K m are given and the unknown vector x ∈ K n is to be

foand.

101
 4.3 Systems of linear equations

Definitions

Let
Ax = b
be a system of linear equations for the unknown x ∈ K n .

The matrix A ∈ K m×n is its matrix of coefficients and the matrix ( A|b) ∈ K m×(n+1)
is its extended matrix of coefficients.

The system is called homogeneous if b = 0; otherwise it is called inhomoge-

neous.

Proposition

The system of linear equations

Ax = b
has a solution if and only if rank ( A|b) = rank A.

Proof

First notice that rank ( A|b) ≥ rank A with equality if and only if b lies in the
column space of A.

Writing the equation

Ax = b
as
x1 c1A + · · · + xn cnA = b,
we see that it has a solution if and only if b can be written as a linear combination
of the columns of A. However this is the case if and only if b lies in the column
space of A. 

Definition

If the system of linear equations

Ax = b
has no solution, i.e. rank ( A|b) > rank A, we say that it is inconsistent.

102
 4.3 Systems of linear equations

Proposition

1. The solution set of the homogeneous system

Ax = 0

is the subspace ker A of K n and has dimension n − rank| A.

2. Suppose that the system

Ax = b
has a solution x⋆ . Its solution set is

x⋆ + ker A.

Proposition

Suppose that ( A|b) can be converted into ( A′ |b′ ) by an elementary row opera-
tion. The solution sets of the systems

Ax = b

and
A′ x = b′
are identical.

Remark

The elementary column swap ci ↔ c j in the matrix A converts the system of linear
equations
x1 c1A + · · · + xi ciA + · · · + x j c jA + · · · + xn cnA = b
into
x1 c1A + · · · + xi c jA + · · · + x j ciA + · · · + xn cnA = b
and thus corresponds to a relabelling of the variables xi and x j .

103
 4.3 Systems of linear equations

Example

Suppose that ( A|b) ∈ K m×(n+1) is in echelon form with rank ( A|b) = rank A. By
relabelling the variables we may assume that ( A|b) has the form
 
a11 b1
 0 a22 b2 
 .
 . .. A2 .. 

 . . . 
 
 0 · · · 0 arr br 
( A|b) =  
 
 
 

 0 0 0 

with a11 , . . . arr 6= 0 and A2 ∈ Kr×(n−r) . It follows that Ax = b is equivalent to

      
a11 x1 xr +1 b1
 0 a22   x2   xr+2  b2 
      
 .. . .   ..  + A2  ..  =  ..  ,
 . .  .   .  .
0 · · · 0 arr xr xn br
| {z }
: = A1

and the solution set of this system is

  
 x1 
 

  ..        


  .  x b x



  1 1 r + 1 

x
 r  n  ..  −1  ..  −1  .. 
  ∈ K :  .  = A1  .  − A1 A2  .  .

  xr +1  


  . 
. x r b r x n 




 .  


 
xn

Note that the solution set is given by n − r parameters xr+1, . . . , xn .

Example

Solve the real system of linear equations

x + 3y + 2z = k,
2x + ky + 4z = 12,
kx + 2y − 4z = 20.

104
 4.3 Systems of linear equations

Solution

We consider the extended matrix of coefficients

 
13 2 k
 2 k 4 12  .
k 2 −4 20

We have that
   
13 2 k 1 3 2 k
 2 k 4 12  −→  0 k − 6 0 12 − 2k  (r2 → r2 − 2r1 , r3 → r3 − kr1 )
k 2 −4 20 0 2 − 3k −4 − 2k 20 − k2

For k = 6 we find that

   
13 2 k 1 3 2 6
 2 k 4 12  −→  0 0 0 0  (r2 → r2 − 2r1 , r3 → r3 − 6r1 )
k 2 −4 20 0 −16 −16 −16
 
1326
1
−→  0 1 1 1  (r2 ↔ r3 , r2 → − 16 r 2 ),
0000

and the solution set of the system

x + 3y + 2z = 6,
y+z=1

is     
 x 3+λ 
 y  = 1 − λ  : λ ∈ R .
 
z λ

For k 6= 6 we find that

 
13 2 k
 2 k 4 12 
k 2 −4 20
 
1 3 2 k
(r2 → r2 − 2r1 , r3 → r3 − kr1 ,
−→  0 1 0 −2 
r2 → k−1 6 r2 )
0 2 − 3k −2(2 + k) 20 − k2
 
1 3 2 k
−→  0 1 0 −2  (r3 → r3 − (2 − 3k)r2 ).
0 0 −2(2 + k) 24 − 6k − k 2

105
 4.4 Determinants

The system is therefore inconsistent if k = −2. On the other hand, if k 6= −2

the system

x + 3y + 2z = k,
y = −2,
−2(2 + k)z = 24 − 6k − k2

has the unique solution

 
2(18 + k)
  
x  2+k


 
y = 
 − 2 .

z  2
 −24 + 6k + k 


2 (2 + k )

4.4 Determinants

Definition

Let K be a field. A mapping det : K n×n → K is called a determinant function if

it has the following properties.

1. It is linear in each row of its argument, i.e.

     
r1 r1 r1
     
 ..   ..   .. 

 . 

 .




 . 

det  α1 r j + α2 r j  = α1 det  r1j
 1 2 r2j
    
 + α2 det  
     
 ..   ..   .. 

 . 

 .




 . 

rn rn rn

for all r1 , . . . , r j−1 , r j+1 , . . . , rn , r1j , r2j ∈ K1×n , α1 , α2 ∈ K and j = 1, . . . ,n.

2. If A is not invertible (i.e. rank A < n), then det A = 0.

106
 4.4 Determinants

Remarks

1. The zero mapping is the trivial determinant function.

2. In particular, det A = 0 if two rows of A are identical or A has a row which

vanishes identically.

Lemma

Let K be a field and det : K n×n → K a determinant function.

1. Converting the matrix A into the matrix A′ by swapping two rows yields
det A′ = − det A.

2. Converting the matrix A into the matrix A′ by multiplying a row with λ ∈

K \ {0} yields det A′ = λ det A.

3. Converting the matrix A into the matrix A′ by addition of a multiple of one

row to another yields det A′ = det A.

Proof

1. Suppose that A′ is obtained from A by the elementary row operation

ri ↔ r j . It follows that

107
 4.4 Determinants
 
r1A
 
 .. 

 . 

riA + r jA
 
 ← i-th row
 

0 = det  .. 
 . 

riA + r jA
 
 ← j-th row
 
 .. 

 . 

rnA

   
.. ..
 .   . 
riA r jA
   
   
   

= det  ..   .. 
.  + det  . 
   
riA + r jA riA + r jA
   
   
   
.. ..
. .
       
.. .. .. ..
 .   .   .   . 
 A  
riA
 
r jA
 
r jA

 ri       
       
= det  ...  + det  .. .. ..
       
.  + det  .  + det  . ,
       

r A  
r jA
 
riA
 
r jA

 i       
       
.. .. .. ..
. . . .
| {z } | {z }
=0 =0
so that
0 = det A + det A′ .

2. This follows directly from the linearity of the determinant function in each
row.

3. Suppose that A′ is obtained from A by the elementary row operation

ri → ri + λr j . It follows that

108
 4.4 Determinants
 
r1A
 
 .. 

 . 

riA + λr jA
 
  ← i-th row
 

det  .. 
 . 

r jA
 
  ← j-th row
 
 .. 

 . 

rnA

   
.. ..
 .   . 
riA r jA
   
   
   

= det  ..   .. 
.  + λ det  . ,
   
r jA r jA
   
   
   
.. ..
. .
| {z }
=0

so that
det A′ = det A. 

Remark

If A is invertibe, then det A = µ A det In , where µ A 6= 0 is a constant which depends

only upon A (since A can be converted into In by elementary row operations).

Lemma

Let K be a field. There is at most one normalised determinant function K n×n →

K, i.e. a determinant function with det In = 1).

Proof

Let det1 and det2 be normalised determinant functions K n×n → K.

If A is not invertible, then det1 A = det2 A = 0.

109
 4.4 Determinants

Suppose that A is invertible. It follows that

det 1 A = µ A det 1 In = µ A , det 2 A = µ A det 2 In = µ A ,

so that det 1 A = det 2 A. 

Proposition

Let K be a field and det : K n×n → K a normalised determinant function. It follows

that det A = 0 if and only if A is not invertible.

Proof

By definition det A = 0 if A is not invertible. If on the other hand A is invertible,

then det A = µ A det In = µ A with µ A 6= 0. 

Finally we confirm that a normalised determinant function actually exists.

Theorem

Let K be a field. The inductively defined functions


a , n = 1,
 11

 (Expansion
rdeti A = n i = 1, . . . ,n, alongh the i-th
i+ j ′
 ∑ (−1) aij det Aij , n > 1,

 row)
j =1

and

 a11 , n = 1,

 (Expansion
cdet j A = n j = 1, . . . ,n, along the j-th
i+ j ′
 ∑ (−1) aij det Aij , n > 1,

 column)
i =1

where  
a11 · · · a1j · · · a1n
 .. .. 
 . . 
′
  (n −1)×(n −1)
 ai1 · · · aij · · · ain  ∈ K
Aij =  
 .. .. 
 . . 
an1 · · · anj · · · ann

110
 4.4 Determinants

is the matrix obtained from A by deleting its i-th row and j-th column, are nor-
malised determinant functions K n×n → K (and are therefore all identical).

We henceforth write the unique normalised determinant function als det.

Remark

The signs in the above formulae form a ‘chess board’ pattern:

 
+ − + − + ...
− + − + − . . .
 
+ − + − + . . .
 
− + − + − . . .
 
.. .. .. .. ..
. . . . .

Examples

1. We have that
     
a b a b a b
det = a det − b det = ad − bc,
c d c d c d
so that the matrix  
a b
c d
is invertible if and only if ad 6= bc.

2. We have that
 
a11 a12 a13
det  a21 a22 a23 
a31 a32 a33
     
a11 a12 a13 a11 a12 a13 a11 a12 a13
= a11 det  a21 a22 a23  − a12 det  a21 a22 a23  + a13 det  a21 a22 a23 
a31 a32 a33 a31 a32 a33 a31 a32 a33
     
a a a a a a
= a11 det 22 23 − a12 det 21 23 + a13 det 21 22
a32 a33 a31 a33 a31 a32
= a11 (a22 a33 − a32 a23 ) − a12 (a21 a33 − a31 a23 ) + a13 (a21 a32 − a31 a22 )
= a11 a22 a33 + a12 a23 a31 + a13 a21 a32 − a31 a22 a13 − a32 a23 a11 − a33 a21 a12 .

(This is Sarrus’ rule.)

111
 4.4 Determinants

3. Expanding along the first column yields

   
a11 a11
 0 a22   0 a22 
   
det  .. ..  = a 11 det  .. .. 
 . .   . . 
0 · · · 0 arr 0 · · · 0 arr
 
a22
= a11 det  ... . . .
 
.
0 · · · arr

Continuing iteratively, we obtain the formula

det A = a11 a22 . . . arr

for an upper triangular matrix

 
a11
 0 a22 
 
A =  .. . . .
 . . 
0 · · · 0 arr

4. Note that
 
1 a2 + bc a3
det 1 b2 + ac b3  = (a − b)(a − c)(b − c)(a2 + b2 + c2 ),
1 c2 + ab c3

because
 
1 a2 + bc a3
det 1 b2 + ac b3 
1 c2 + ab c3
 
1 a2 + bc a3
= det 0 (b − a)(a + b − c) (b − a)(b2 + ba + a2 )
0 (c − a)(a + c − b) (c − a)(c2 + ca + a2 )
(r 2 → r 2 − r 1 , r 3 → r 3 − r 1 )
 
1 a2 + bc a3
= (b − a)(c − a) det 0 a + b − c b2 + ba + a2 
0 a + c − b c2 + ca + a2

112
 4.4 Determinants
 
a + b − c b2 + ba + a2
= (b − a)(c − a) det
a + c − b c2 + ca + a2
(Expansion along the first column)
 
a+b−c b2 + ba + a2
= (b − a)(c − a) det
2(c − b) (c − b)(a + b + c)
(r 2 → r 2 − r 1 )
 
a + b − c b2 + ba + a2
= (b − a)(c − a)(c − b) det .
2 a+b+c
| {z }
2 2 2
= −a − b − c

Lemma

Let K be a field. The formula det AB = det A det B holds for all A,B ∈ K n×n .

Proof

Choose B ∈ K n×n and define Det : K n×n → K by the formula

Det A = det AB.

We show that Det is a determinant function.

It follows from
     
r1 r1 r1 B
     
 ..   ..   .. 
Det  .  = det  .  B = det  . 
     
rn rn rn B

that

113
 4.4 Determinants
   
r1 r1 B
   
 ..   .. 

 . 


 . 

 1 2   1 2 
Det  α1 r j + α2 r j  = det  α1 r j B + α2 r j B 
   
 ..   .. 

 . 


 . 

rn rn B
   
r1 B r1 B
   
 ..   .. 

 . 


 . 

 1   2
= α1 det  r j B  + α2 det  r j B 

   
 ..   .. 

 . 


 . 

rn B rn B
   
r1 r1
   
 ..   .. 
 .   . 
   
 1  
= α1 Det  r j  + α2 Det  r j  ,2 
   
 ..   .. 
 .   . 
   
rn rn

so that Det is linear in each row of its argument.

Since the columns of AB are linear combinations of the columnds of A, we

have that Rang AB ≤ Rang A. It follows that

Rang A < n ⇒ Rang AB < n

⇒ det AB = 0
⇒ Det A = 0.

Hence
| {z A} = µ A Det
Det I ,
| {z n}
det A = µ A det In ,
| {z }
= det AB = det B =1
so that
det AB = det A det B. 

Corollary

If A ∈ K n×n is invertible, then det A−1 = 1/ det A.

114
 4.4 Determinants

Proof

It follows from A−1 A = In that

det A−1 det A = det A−1 A = det In = 1. 

Lemma

Let K be a field. det A = det AT for every A ∈ K n×n .

Proof

If A is not invertible, then AT is also not invertible. Hence det A = det AT = 0.

Now suppose that A is invertible, so that it can be written as a product

A = E1 · · · E p

of elementary matrices. Every elementary matrix E satisfies det ET = det E:

Note that VijT = Vij and ( Miλ )T = Miλ , so that det VijT = det Vij and det( Miλ )T =
det Miλ .

Furthermore ( Aijλ )T = Aλji and det Aijλ = det In = 1, det Aλji = det In = 1, so that
det( Aijλ )T = det Aλji .

It now follows from

AT = ETp · · · E1T
that

det AT = det ETp · · · det E1T = det E1 · · · det E p = det E1 · · · E p = det A. 

115
 4.4 Determinants

Corollary

Let K be a field. The function det : K n×n → K is linear in each column of its
argument, i.e.
     
det c1 · · · α1 c1j + α2 c2j · · · cn = α1 det c1 · · · c1j · · · cn + α2 det c1 · · · c2j · · · cn

for all c1 , . . . , c j−1 , c j+1 , . . . , cn , c1j , c2j ∈ K n , α1 , α2 ∈ K and j = 1, . . . ,n.

Corollary

Let K be a field and A ∈ K n×n .

1. Converting the matrix A into the matrix A′ by swapping two columns yields
det A′ = − det A.

2. Converting the matrix A into the matrix A′ by multiplying a column with

λ ∈ K \ {0} yields det A′ = λ det A.

3. Converting the matrix A into the matrix A′ by addition of a multiple of one

column to another yields det A′ = det A.

Now we present methods for inverting a regular matrix and inverting a system
of linear equations which are of academic rather than pratical interest.

Definition

♯
Let K be a field and A ∈ K n×n . Die Matrix A♯ = (aij ) ∈ K n×n with

aij♯ = (−1) j+i det A′ji

is the complementary matrix to A.

116
 4.4 Determinants

Remark

Note that

aij♯ = (−1) j+i det A′ji

 
a11 · · · a1i · · · a1n
 .. .. 
 . . 
j +i  
= (−1) det  a j1 · · · a ji · · · a jn 
 . .. 
 .. . 
an1 · · · ani · · · ann
 
a11 · · · a1,i −1 0 a1,i +1 · · · a1n
 .. .. .. .. .. 
 . . . . . 
 
 a j−1,1 · · · a j−1,i −1 0 a j−1,i +1 · · · a j−1,n 
 
= (−1) j+i det 
 a j,1 · · · a j,i −1 1 a j,i +1 · · · a jn  
 a j+1,1 · · · a j+1,i −1 0 a j+1,i +1 · · · a j+1,n 
 
 .. .. .. .. .. 
 . . . . . 
an1 · · · an,i −1 0 an,i +1 · · · ann
 
a11 · · · a1,i −1 0 a1,i +1 · · · a1n
 .. .. .. .. .. 
 . . . . . 
 
a j−1,1
 · · · a j−1,i −1 0 a j−1,i +1 · · · a j−1,n 

= det 
 a j,1 · · · a j,i −1 1 a j,i +1 · · · a jn  
a j+1,1
 · · · a j+1,i −1 0 a j+1,i +1 · · · a j+1,n 

 .. .. .. .. .. 
 . . . . . 
an1 · · · an,i −1 0 an,i +1 · · · ann


= det c1A · · · ciA−1 e j ciA+1 · · · cnA .

Proposition

Let K be a field A ∈ K n×n and A♯ the complementary matrix to A. It follows that

A♯ A = AA♯ = (det A) In .

117
 4.4 Determinants

Proof

Note that
n
♯ ♯
( A A)ij = ∑ aik akj
k=1
n

= ∑ akj det c1A · · · ciA−1 ek ciA+1 · · · cnA
k=1
!
n
= det c1A ··· ciA−1 ∑ akj ek ciA+1 ··· cnA
k=1
 
= det ··· c1A ciA−1 c jA ciA+1 ··· cnA

det A, i = j,
=
0, i 6= j
= (det A)δij .

AA♯ is calculated in a similar fashion. 

Corollary

Let K be a field and A ∈ GL(n,K ). It follows that

1
A −1 = A♯ .
det A

Lemma (Cramer’s rule)

Let K be a field, A ∈ GL(n,K ) and b ∈ K n . The unique solution of the system of

linear equations
Ax = b
is given by the formula


det c1A · · · ciA−1 b ciA+1 · · · cnA
xi = , i = 1, . . . ,n.
det A

118
 4.4 Determinants

Proof

Note that
1
x = A −1 b = A♯ b,
det A
so that
1
xi = ( A ♯ b )i
det A
n
1
aij♯ b j
det A j∑
=
=1
n
1

= ∑ b j det c1A · · · ciA−1 e j ciA+1 · · · cnA
det A j=1
n
!
1 A A A A
= det c1 · · · ci −1 ∑ b j e j ci +1 · · · cn
det A j =1
1

= det c1A · · · ciA−1 b ciA+1 · · · cnA . 
det A

119
 5 Eigenvalues and eigenvectors

5 Eigenvalues and eigenvectors

Definition

Let V be a vector space over a field K and T : V → V a linear transformation. The

subspace U of V is invariant under T if T [U ] ⊆ U.

Remark

If U is invariant under T, then T [U ] is a subspace of U.

Examples

1. Let V be a vector space over a field K. {0}, V, ker T and Im T invariant

subspaces of every linear transformation T : V → V.

2. Let n ∈ N. The subspace R n [ x ] of R [ x ] is invariant under the linear trans-

formation T : p 7→ p′ .

Definitions

Let V be a vector space over a field K and T : V → V a linear transformation.

Suppose there exists a scalar λ and a vector v 6= 0 with

Tv = λv.

We say that λ is an eigenvalue of T with corresponding eigenvector v.

Let λ be an eigenvalue of T. The subspace

Eλ = {v : Tv = λv} = ker(T − λI )

of V is the eigenspace corresponding to λ. Its dimension is the geometric multi-

plicity of λ.

120
 5 Eigenvalues and eigenvectors

Remark

Eλ is a subspace of V which is invariant under T since it is the kernel of the linear

transformation T − λI : V → V.

Examples

1. Let V be a vector space over a field K.

The identity transformation I : V → V has the eigenvalue 1 and E1 = V

(so that all nontrivial vectors are eigenvectors with eigenvalue 1).

Every linear transformation T : V → V which is not injective has the

eigenvalue 0 and E0 = ker T (so (so that all nontrivial vectors in ker T
are eigenvectors with eigenvalue 0).

2. Now we consider three geometric examples in R2 .

v Tv
v-1 v1 θ v
Rv1 Sv v1
Tv1
Rv-1

A reflection R : R2 → R2 has the eigenvalues 1 and −1: E1 is the axis of

reflection, E−1 is the perpendicular line through the origin.

A rotation S : R2 → R2 through the angle θ ∈ (0,π ) ∪ (π,2π ) has no

eigenvalues.

A simple shear T : R2 → R2 has the eigenvalue 1: E1 is the fixed axis.

3. Ley C∞ (R ) be the space of all infinitely differentiable functions R → R. The

linear transformation T : C∞ (R ) → C∞ (R ) with T ( f ) = f ′ has every real
number λ as an eigenvalue since T (eλx ) = λeλx .

121
 5 Eigenvalues and eigenvectors

Proposition

Let V be a finite-dimensional vector space over a field K and T : V → V a linear

transformation. A scalar λ is an eigenvalue of T if and only if det(T − λI ) = 0.

Proof

Note that

Tv − λv = 0 for some v 6= 0
⇔(T − λI )v = 0
⇔ ker(T − λI ) 6= {0}
⇔ dim ker(T − λI ) > 0
⇔ dim Im (T − λI ) < dim V
| {z }
= rank (T − λI )
⇔ det(T − λI ) = 0. 

Definition

Let V be an n-dimensional vector space over a field K and T : V → V a linear

transfoirmation. The function c : K → K with
 
a11 − λ a12 a13 · · · a1n
 21 a22 − λ a23 · · · a2n 
 a 
c(λ) = det( A − λIn ) = det  a31
 a32 a33 − λ · · · a3n  ,
 .. .. .. .. 
 . . . . 
an1 an2 an3 · · · ann − λ

where A = (aij ) is the matrix of T with respect to a basis for V, is called the
characteristic function of T.

Proposition

Let V be an n-dimensional vector space over a field K. The characteristic function

of a linear transformation T : V → V is a polynomial of degree n with coefficients
in K.

122
 5 Eigenvalues and eigenvectors

Proof

Note that
A − λIn = (c1A − λe1 | · · · |cnA − λen ).
Since the determinant is linear in each colum of its argument, we find that

det( A − λIn ) = (−λ)n det(e1 | · · · |en )

n
+ (−λ)n−1 ∑ det(e1 | · · · |ciA | · · · |en )
i =1
n −2
+ (−λ) ∑ det(e1 | · · · |ciA | · · · |c jA | · · · |en )
1≤ i < j ≤ n
+ ···
n
+ (−λ) ∑ det(c1A | · · · |ei | · · · |cnA )
i =1
A
+ det(c1 | · · · |cnA ),

so that c(λ) = det( A − λIn ) is a polynomial with deg c ≤ n. Moreover, the coeffi-
cient of λn is (−1)n det(e1 | · · · |en ) = (−1)n det In = (−1)n 6= 0, so that deg c = n.

Remarks

1. We usually refer to c as the characteristic polynomial of T.

2. In particular T can have at most n distinct eigenvalues.

Example

Compute the eigenvalues of the real matrix

 
3 2 4
A= 2 0
 2
4 2 3

together with the corresponding eigenspaces.

123
 5 Eigenvalues and eigenvectors

Solution

The eigenvalues of A are the zeros of the characteristic polynomial

c(λ) = det( A − λI3 )

 
3−λ 2 4
= det  2 −λ 2 
4 2 3−λ
 
−1 − λ 2 4
= det  0 −λ 2  (c1 → c1 − c3 )
1+λ 2 3−λ
 
−1 2 4
= (1 + λ) det  0 −λ 2 
1 2 3−λ
 
−1 2 4
= (1 + λ) det  0 −λ 2  (r 3 → r 3 + r 1 )
0 4 7−λ
 
−λ 2
= −(1 + λ) det
4 7−λ

= −(1 + λ)(λ2 − 7λ − 8)

= −(1 + λ)2 (λ − 8),

i.e. λ = −1 and λ = 8.

We have that

E−1 = {x ∈ R3 : ( A + I3 )x = 0}
      
 x 4 2 4 x 0 
=  y : 2 1 2 y = 0 .
     
 
z 4 2 4 z 0

We therefore study the extended matrix of coefficients

 
4240
2 1 2 0
4240
 
4240
→0 0 0 0 (r2 → r2 − 12 r1 , r3 → r3 − r1 )
0000

It follows that

124
 5 Eigenvalues and eigenvectors
  
 x 
E−1 =  y  : 4x + 2y + 4z = 0
 
z
 1  
 −2µ − λ 
=  µ  : λ, µ ∈ R
 
λ
*  1   −1 +
−2
=  1 , 0  ,
0 1
and the geometric multiplicity of this eigenvalue is
dim E−1 = dim ker( A + I3 ) = 2.

We have that
E8 = {x ∈ R3 : ( A − 8I3 )x = 0}
      
 x −5 2 4 x 0 
=  y  :  2 −8 2  y  = 0 .
 
z 4 2 −5 z 0
We therefore study the extended matrix of coefficients
 
−5 2 4 0
 2 −8 2 0 
4 2 −5 0
 
−1 4 −1 0
→  2 −8 2 0  (r 1 → r 1 + r 3 )
4 2 −5 0
 
−1 4 −1 0
→ 0 0 0 0 (r2 → r2 + 2r1 , r3 → r3 + 4r1 )
0 18 −9 0
 
−1 4 −1 0
→  0 2 −1 0  (r3 → 91 r3 , r3 ↔ r2 )
0 0 0 0
It follows that
  
 x 
− x + 4y − z = 0,
E8 =  y :

 2y − z = 0 
z
  
 λ 
= 1
 λ :λ∈R

 2 
λ
*  1 +
= 1 ,
2
1

125
 5 Eigenvalues and eigenvectors

and the geometric multiplicity of this eigenvalue is

dim E8 = dim ker( A − 8I3 ) = 1.

Remark

A linear transformation may have no eigenvalues.

The matrix  
0 2
∈ R 2× 2
1 0
√
has eigenvalues ± 2, but  
0 2
∈ Q 2× 2
1 0
has no eigenvalues
√ since the characteristic polynomial c(λ) = λ2 − 2 has the
real roots ± 2 but no rational roots.

The matrix  
0 4
∈ C 2× 2
−1 0
has eigenvalues ±2i, but  
0 4
∈ R 2× 2
−1 0
has no eigenvalues since the characteristic polynomial c(λ) = λ2 + 4 has the
complex roots ±2i but no real roots.

Lemma

Let V be a nontrivial, finite-dimensional, complex vector space and T : V → V a

linear transformation. It follows that T has an eigenvalue.

Proof

Let n = dim V, so that the characteristic polynomial c of T is a polynomial of

degree n ≥ 1 with complex coefficients. According to the fundamental theorem
of algebra c has at least one complex root. 

126
 5 Eigenvalues and eigenvectors

Definitions

Let K be a field.

1. A matrix M ∈ K n×n is called diagonal if it has the form

 
m11 0

 m22 

 . . .
 . 
0 mnn

We write this matrix as

diag (m11 , . . . ,mnn ).

2. Let V be a finite-dimensional vector space over K. A linear transformation

T : V → V is called diagonalisable if there is a basis for V with respect to
which the matrix of T is diagonal.

Proposition

Let V be a finite-dimensional vector space over a field K. A linear transforma-

tion T : V → V is diagonalisable if and only if there is a basis for V consisting of
eigenvectors of T.

Proof

T is diagonalisable
There is a basis {v1 , . . . ,vn } for V with respect to which the matrix
⇔
of T has the form diag(λ1 , . . . ,λn )
There is a basis {v1 , . . . ,vn } for V such that Tvi = λi vi for i =
⇔
1, . . . , n
⇔ There is a basis for V consisting of eigenvectors of T 

127
 5 Eigenvalues and eigenvectors

Definition

Let K be a field. A matrix A ∈ K n×n is diagonalisable if it is similar to a diagonal

matrix, i.e. if there is a matrix P ∈ GL(n,K ) such that P−1 AP is a diagonal matrix.

Remark

A matrix A ∈ K n×n is diagonalisable if and only if it is the matrix of a diagonalis-

able linear transformation T : K n → K n .

It follows that A ∈ K n×n is diagonalisable if and only if there is a basis for K n

consisting of eigenvectors of A.

Example

The real matrix  

3 2 4
A =  2 0 2
4 2 3
has the eigenvalues −1 and 8 with corresponding eigenspaces
*  1   −1 +
−2
E− 1 =  1 , 0  ,
0 1

* 1 +

E8 = 1 .
2
1
 1    
−2 −1 1
The eigenvectors  1 ,  0 ,  21  are linearly independent and therefore
0 1 1
3
form a basis B for R .

The linear transformation R3 → R3 , x 7→ Ax has the matrix A with respect to the

usual basis      
 1 0 0 
A =  0 ,  1 ,  0
 
0 0 1
and the matrix diag(−1, − 1,8) with respect to the matrix B . With
 1 
− 2 −1 1
B
P = BA =  1 0 12 
0 1 1

128
 5 Eigenvalues and eigenvectors

we have that
P−1 AP = diag(−1, − 1,8).

Lemma

Let V be a vector space over a field K and λ1 , . . . , λm be distinct eigenvalues of the

linear transformation T : V → V with corresponding eigenvectors v1 , . . . , vm . It
follows that v1 , . . . , vm are linearly independent.

Proof

We prove this result by induction over m. The result for m = 1 is trivial.

Now suppose that the result holds for m = k. Let α1 , . . . , αk+1 be scalars such that
α1 v1 + · · · + αk+1 vk+1 = 0. (⋆)
Mutiplying (⋆) by λ1 yields
λ1 α1 v1 + λ1 α2 v2 + · · · + λ1 αk+1 vk+1 = 0.
Applying T to (⋆) yields
λ1 α1 v1 + λ2 α2 v2 + · · · + λk+1 αk+1 vk+1 = 0.
It follows that
(λ2 − λ1 )α2 v2 + · · · + (λk+1 − λ1 )αk+1 vk+1 = 0.
However by the inductive hypothesis v2 , . . . , vk+1 are linearly independent, so
that
(λ − λ ) α = 0, . . . , (λk+1 − λ1 ) αk+1 = 0,
| 2 {z 1} 2 | {z }
6= 0 6= 0
⇒ α2 = 0, . . . , αk+1 = 0.
It then follows from (⋆) that
α1 v1 = 0,
so that α1 also vanishes. Hence v1 , . . . , vk+1 are linearly independent. 

Corollary

Let V be a vector space over a field K and λ1 , . . . , λm be distinct eigenvalues of the

m
linear transformation T : V → V. It follows that ∑ Eλi is direct.
i =1

129
 5 Eigenvalues and eigenvectors

Corollary

Let V be a finite-dimensional vector space over a field K and T : V → V be a linear

transformation whose eigenvalues are λ1 , . . . , λm . It follows that
m
∑ dim Eλi ≤ dim V
i =1

with equality if and only if T is diagonalisable.

Proof

m
L
Eλi is a subspace of V, and the union of bases for Eλ1 , . . . , Eλm is a basis for
i =1
Lm
Eλi . Hence
i =1 !
m
M m
dim V ≥ dim Eλ i = ∑ dim Eλi
i =1 i =1
m
L
with equality if and only if Eλi = V.
i =1

m
L
If Eλi = V, then the union of bases for Eλ1 , . . . , Eλm is a basis for V consisting
i =1
of eigenvectors of T.

m
L
If V has a basis B consisting of eigenvectors of T, then B ⊆ Eλi implies that
i =1
m
L
Eλi = V. 
i =1

Examples

1. The real matrix  

3 2 4
A 1 =  2 0 2
4 2 3

130
 5 Eigenvalues and eigenvectors

has eigenvalues −1 and 8 with corresponding eigenspaces

*  1   −1 +
−2
E− 1 =  1 , 0  ,
0 1

* 1 + 

E8 =  12  .
1

It follows from dim E−1 + dim E8 = 2 + 1 = 3 that A1 is diagonalisable.

2. The real matrix  

1 1 0
A 2 =  0 1 0
0 0 2
has eigenvalues 1 and 2 with corresponding eigenspaces
* 1  +
E1 = 0 ,
0
* 0  +
E2 = 0 .
1

It follows from dim E1 + dim E2 = 1 + 1 < 3 that A2 is not diagonalisable.

Remark

Let V be an n-dimensional vector space over a field K. Suppose that the char-
acteristic polynomial c of a linear transformation T : V → V can be decomposed
into distinct linear factors, i.e.

c(λ) = (−1)n (λ − λ1 ) . . . (λ − λn ),

where λ1 , . . . λn are distinct scalars. It follows that T diagonalisable since in this

case dim Eλi = 1, i = 1, . . . ,n.

131