Linear Algebra

Differential Geometry - Dynamical Systems
Monographs # 1
Constantin Udriste
LINEAR ALGEBRA
Geometry Balkan Press

Bucharest, Romania
Ioana Boca
Linear Algebra
Monographs # 2
Differential Geometry - Dynamical Systems * Monographs

Editor-in-Chief Prof.Dr. Constantin Udriste
Politehnica University of Bucharest
Analytic and Differential Geometry

Constantin Udriste, Ioana Boca. - Bucharest:
Differential Geometry - Dynamical Systems * Monographs, 2000
Includes bibliographical references.
c Balkan Society of Geometers, Differential Geometry - Dynamical Systems
* Monographs, 2000
Neither the book nor any part may be reproduced or transmitted in any form
or by any means, electronic or mechanical, including photocopying, microfilming
or by any information storage and retrieval system, without the permission in
writing of the publisher.
Preface
This textbook covers the standard linear algebra material taught at the University
Politehnica of Bucharest, and is designed for a 1-semester course.
The prerequisites are highschool algebra and geometry.
Chapters 14 are intended to introduce first year students to the basic notions
of vector space, linear transformation, eigenvectors and eigenvalues, bilinear and
quadratic forms, and to the usual linear algebra techniques.
The linear algebra language is used in Chapters 5, 6, 7 to present some notions and
results on vectors, straight lines and planes, transformations of coordinate systems.
We end with some exam samples. Each sample involves facts from two or more
chapters.
The topics treated in this book and the presentation of the material are similar
to those in several of the first authors previous works [19][25]. Parts of some linear
algebra sections follow [1]. The selection of topics and problems relies on the teaching
experience of the authors at the University Politehnica of Bucharest, including lectures
and seminars taught in English at the Department of Engineering Sciences.
The publication of this volume was supported by MEN Grant #21815, 28.09.98,
CNCSU-31; this support provided the oportunity to include the present textbook in
the University Lectures Series published by the Editorial House of Balkan Society of
Geometers.
We wish to thank our colleagues for helpful discussions on the problems and topics
treated in this book and on our teaching activities. Any further suggestions will be
greatly appreciated.
The authors
July 12, 2000
iii
Contents
1 Vector Spaces
1
Vector Spaces . . . . . . . . . . . . . . . .
2
Vector Subspaces . . . . . . . . . . . . . .
3
Linear Dependence. Linear Independence
4
Bases and Dimension . . . . . . . . . . . .
5
Coordinates.
Isomorphisms. Change of Coordinates . .
6
Euclidean Vector Spaces . . . . . . . . . .
7
Orthogonality . . . . . . . . . . . . . . . .
8
Problems . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1
1
4
7
9
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
11
15
18
22
2 Linear Transformations
1
General Properties . . . . . . . . . . . . . .
2
Kernel and Image . . . . . . . . . . . . . . .
3
The Matrix of a Linear Transformation . .
4
Particular Endomorphisms . . . . . . . . . .
5
Endomorphisms of Euclidean Vector Spaces
6
Isometries . . . . . . . . . . . . . . . . . . .
7
Problems . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
25
25
29
31
34
37
41
43
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
Euclidean Spaces
. . . . . . . . . .
. . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
45
45
47
51
54
61
63
66
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
67
67
70
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
72
76
79
3 Eigenvectors and Eigenvalues

1
General Properties . . . . . . . . . .
2
The Characteristic Polynomial . . .
3
The Diagonal Form . . . . . . . . . .
4
The Canonical Jordan Form . . . . .
5
The Spectrum of Endomorphisms on
6
Polynomials and Series of Matrices .
7
Problems . . . . . . . . . . . . . . .
4 Bilinear Forms. Quadratic Forms

1
Bilinear Forms . . . . . . . . . . . . . .
2
Quadratic Forms . . . . . . . . . . . . .
3
Reduction of Quadratic Forms to
Canonical Expression . . . . . . . . . . .
4
The Signature of a Real Quadratic Form
5
Problems . . . . . . . . . . . . . . . . .
v
vi
5 Free Vectors
1
Free Vectors . . . . . . . . . .
2
Addition of Free Vectors . . .
3
Multiplication by Scalars . .
4
Collinearity and Coplanarity
5
Inner Product in V3 . . . . .
6
Vector (cross) Product in V3
7
Mixed Product . . . . . . . .
8
Problems . . . . . . . . . . .
CONTENTS
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
81
81
83
84
85
87
90
92
94
6 Straight Lines and Planes in Space

1
Cartesian Frames . . . . . . . . . . . . .
2
Equations of Straight Lines in Space . .
3
Equations of Planes in Space . . . . . .
4
The Intersection of Two Planes . . . . .
5
Orientation of Straight Lines and Planes
6
Angles in Space . . . . . . . . . . . . . .
7
Distances in Space . . . . . . . . . . . .
8
Problems . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
95
95
96
97
100
101
102
104
107
7 Transformations of Coordinate Systems

1
Translations of Cartesian Frames . . . .
2
Rotations of Cartesian Frames . . . . .
3
Cylindrical Coordinates . . . . . . . . .
4
Spherical Coordinates . . . . . . . . . .
5
Problems . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
109
109
110
113
115
116
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Exam Samples
119
Bibliography
131
Chapter 1
Vector Spaces
1
Vector Spaces
The vector space structure is one of the most important algebraic structures.
The basic models for (real) vector spaces are the spaces of ndimensional row or
column matrices:
M1,n (R) = v = [a1 , . . . , an ] ; aj R, j = 1, n
a1
..
Mn,1 (R) = v = . ; aj R, j = 1, n .
an
We will identify Rn with either one of M1,n (R) or Mn,1 (R). A row (column)
matrix is also called a row (column) vector.
The definition of matrix multiplication makes the use of column vectors more
convenient for us. We will also write a column vector in the form t [a1 , . . . , an ] or
t
(a1 , . . . , an ), as the transpose of a row vector, in order to save space.
An abstract vector space is endowed with two operations: addition of vectors
and multiplication by scalars, where the scalars will be the elements of a field. For
the examples above, these are just the usual matrix addition and multiplication of a
matrix by a real number:
a1
b1
a1 + b1
.. ..
..
. + . =
.
an
bn
an + bn
a1
ka1
k ... = ... ,
an
kan
Let us first recall the definition of a field.
1
where k R .
CHAPTER 1. VECTOR SPACES
DEFINITION 1.1 A field K is a set endowed with two laws of composition:

K K K,
K K K,
(a, b) a + b
(a, b) ab ,
called addition and respectively multiplication, satisfying the following axioms :

i) (K, +) is an abelian group. Its identity element is denoted by 0.
ii) (K? , ) is an abelian group, where K? = K \ {0}. Its identity element is denoted
by 1.
iii) Distributive law: (a + b)c = ac + bc, a, b, c K.
Examples of fields
(a) K = R, the field of real numbers;
(b) K = C, the field of complex numbers;
(c) K = Q, the field of rational numbers;
(d) K = Q[ 2] = {a + b 2 ; a, b Q}.
DEFINITION 1.2 A vector space V over a field K (or a Kvector space) is a set
endowed with two laws of composition:
(a) addition : V V V, (v, w) v + w
(b) multiplication by scalars : K V V, (k, v) kv
satisfying the following axioms:
(i) (V, +) is an abelian group
(ii) multiplication by scalars is associative with respect to multiplication in K:
k(lv) = (kl)v,
k, l K, v V
(iii) the element 1 K acts as identity:

1v = v,
v V
(iv) the distributive laws hold:

k(v + w) = kv + kw,
(k + l)v = kv + lv,
k, l K, v, w V.
The elements of K are usually called scalars and the elements of V vectors. A
vector space over C is called a complex vector space; a vector space over R is called
a real vector space. When K is not specified, we understand K = R or K = C.
1. VECTOR SPACES
Examples of vector spaces

(i) K as a vector space over itself, with addition and multiplication in K.
(ii) Kn is a Kvector space.
(iii) Mm,n (K) with usual matrix addition and multiplication by scalars is a Kvector
space.
(iv) The set of space vectors V3 is a real vector space with addition given by the
parallelogram law and scalar multiplication given as follows:
if k R, v V3 , then kv V3 and:
the length of kv is the length of v multiplied by | k |;
the direction of kv is the direction of v;
the sense of kv is the sense of v if k > 0, and the opposite sense of v if k < 0.
(v) If S is a nonempty set and W is a Kvector space, then V = {f | f : S W}
becomes a Kvector space with:
(f + g)(x) = f (x) + g(x), (kf )(x) = kf (x),
for all k K, f, g V.
(vi) The solution set of an algebraic linear homogeneous system with n unknowns
and coefficients in K is a Kvector space with the operations induced from Kn .
(vii) The solution set of an ordinary linear homogeneous differential equation is a
real vector space with addition of functions and multiplication of functions by
scalars.
(viii) The set K[X] of all polynomials with coefficients in K is a Kvector space.
THEOREM 1.3 Let V be a Kvector space. Then V has the following properties:
(i) 0K v = 0V , v V
(ii) k 0V = 0V , k K
(iii) (1) v = v, v V.
Proof. To see (i) we use the distributive law to write
0K v + 0K v = (0K + 0K ) v = 0K v + 0V .
Now 0K v cancels out (in the group (V, +)), so we obtain 0K v = 0V . Similarly,
k 0V + k 0V = k(0V + 0V ) = k 0V
implies k 0V = 0V .
For (iii), v + (1) v = (1 + (1)) v = 0K v = 0V . Hence (1) v is the additive
inverse of v. QED
COROLLARY 1.4 Let V be a Kvector space. Then:

(i) (k v) = (k) v = k (v), for all k K, v V
(ii) (k l) v = k v l v, for all k K, v V
(iii) k v = 0V
v = 0V or k = 0K
(iv) k v = l v, v 6= 0V
= k = l
(v) k v = k w, k 6= 0K
= v = w
We leave the proof of the corollary as an exercise; all properties follow easily from
Theorem 1.3 and the definition of a vector space.
We will usually write 0 instead of 0K or 0V , since in general it is clear whether
we refer to the number or to the vector zero.
Vector Subspaces
Throughout this paragraph V denotes a Kvector space and S a nonempty subset of

V.
DEFINITION 2.1 A vector subspace W of V is a nonempty subset with the following properties:
(i) If u, v W, then u + w W
(ii) If u W, k K, then k u W.
REMARKS 2.2
(i) If W is a vector subspace, then 0V W.
(ii) W is a vector subspace if and only if W is a vector space with the operations
induced from V.
(iii) W is a vector subspace if and only if
k, l K, u, w V = k u + l w W .
Examples of vector subspaces
(i) {0V } and V are vector subspaces of V. Any other vector subspace is called a
proper vector subspace.
(ii) The straight lines through the origin are proper vector subspaces of R2 .
(iii) The straight lines and the planes through the origin are proper vector subspaces
of R3 .
(iv) The solution set of an algebraic linear homogeneous system with n unknowns is
a vector subspace of Kn .
2. VECTOR SUBSPACES
(v) The set of odd functions and the set of even functions are vector subspaces of
the space of real functions defined on the interval (a, a), a R?+ .
DEFINITION 2.3 (i) Let v1 , . . . , vp V. A linear combination of the vectors
v1 , . . . , vp is a vector of the form
w = k1 v1 + . . . + kp vp ,
kj K .
(ii) Let S be a nonempty subset (possibly infinite) of V. A vector of the form

w = k1 v1 + . . . + kp vp ,
p N? , kj K, vj S
is called a finite linear combination of vectors of S.

The set of all finite linear combinations of vectors of S is denoted by Span S or
L(S). By definition, Span = {0}.
THEOREM 2.4 Span S is a vector subspace of V.
Proof. Let u =
p
X
ki ui Span S and w =
Then ku + lv =
p
X
i=1
(kki ) ui +
lj wj Span S, with ki , lj
j=1
i=1
K, ui , wj S.
q
X
q
X
(llj ) wj is also a finite linear combination with
j=1
elements of S. It follows that Span S is a vector subspace, by Remark 2.2(iii). QED

Span S is called the subspace spanned by S, or the linear covering of S.
PROPOSITION 2.5 If S is a subset of V and W is a vector subspace of V such
that S W, then Span S W.
This is obvious, for any finite linear combination of vectors from S is also a finite
linear combination of vectors from W, and W is closed under addition and scalar
multiplication.
THEOREM 2.6 If W1 , W2 are two vector subspaces of V, then:
(i) W1 + W2 = {v = v1 + v2 | v1 W1 , v2 W2 } is a vector subspace, called the
sum of the vector subspaces W1 and W2 .
(ii) W1 W2 is a vector subspace
of V . More generally, if {Wi }iI is a family of
T
vector subspaces, then iI Wi is a vector subspace.
(iii) W1 W2 is a vector subspace of V if and only if W1 W2 or W2 W1 .
Proof. (i) Let u, w W1 + W2 , u = u1 + u2 , w = w1 + w2 , with u1 , w1 W1 ,
u2 , w2 W2 .
Let k, l K. Then ku + lw = (ku1 + lw1 ) + (ku2 + lw2 ) is in W1 + W2 , as
ku1 + lw1 W1 , and ku2 + lw2 W2 (using Remark 2.2 (iii) for the subspaces W1 ,
W2 ). It follows that W1 + W2 is a subspace.
(ii) Exercise !
(iii) The implication from the right to the left is obvious. For the other one, assume
by contradiction that none of the inclusions holds. Then we can take u1 W1 \ W2 ,
u2 W2 \ W1 .
Now u1 , u2 W1 W2 implies u1 + u2 W1 W2 since W1 W2 is a vector
subspace.
But either u1 + u2 W1 or u1 + u2 W2 contradicts u2
/ W1 or u1
/ W2
respectively. Therefore W1 W2 or W2 W1 . QED
REMARKS 2.7
(i) Span (W1 W2 ) = W1 + W2 .
(ii) Span (S1 S2 ) = Span S1 + Span S2 .

These are straightforward from Proposition 2.5 and Theorem 2.6
PROPOSITION 2.8 Let W1 , W2 be vector subspaces of V. Then, the decomposition
v = v1 + v2 , v1 W1 , v2 W2
is unique for each v W1 + W2 if and only if W1 W2 = {0}.
Proof. Assume the uniqueness of the decomposition and let v W1 W2
W1 + W2 . Then
v = v + 0 , v W1 , 0 W2
and
v = 0 + v , 0 W1 , v W2
represent the same decomposition, implying v = 0.
Conversely, asume W1 W2 = {0} and let v W1 + W2 , v = v1 + v2 = v10 + v20
with v1 , v10 W1 , v2 , v20 W2 . Then v1 v10 = v20 v2 W1 W2 , thus v1 v10 =
0 = v2 v20 . Consequently, v1 = v10 , v2 = v20 . QED
DEFINITION 2.9 If W1 , W2 are vector subspaces with W1 W2 = {0}, then
W1 + W2 is called the direct sum of W1 and W2 and we write W1 W2 ; W1 and
W2 are called independent vector subspaces.
If V = W1 W2 , then W1 and W2 are called supplementary subspaces; each of
W1 and W2 is the supplement of the other one.
Examples
(i) Set
V = {f | f : (a, a) R} , a > 0 ,
W1 = {f V | f is an odd function} ,
W2 = {f V | f is an even function} .
Then V = W1 W2 .
(To see this, write f (x) =
any x (a, a).)
f (x) f (x)
f (x) + f (x)
+
, for any f V and
2
2
3. LINEAR DEPENDENCE. LINEAR INDEPENDENCE
(ii) Set
D1 = {(x, y) | 2x + y = 0}
D2 = {(x, y) | x y = 0}
Then R2 = D1 D2 .
Linear Dependence. Linear Independence
Let V be a Kvector space and S V, S 6= .

DEFINITION 3.1 (i) Let S = {v1 , . . . , vp } be a finite subset of V, where vi 6= vj if
i 6= j. The set S is called linearly dependent (the vectors v1 , . . . , vp are called linearly
dependent) if there exist k1 , . . . , kp K not all zero such that
p
X
ki vi = 0 .
i=1
(ii) An infinite subset S of V is called linearly dependent if there exists a finite

subset of S which is linearly dependent.
(iii) A set which is not linearly dependent is called linearly independent.
In order to study the linear dependence or independence of the vectors v1 , . . . , vp
p
P
we usually exploit the equality
ki vi = 0. If this relation implies the existence of
i=1
(k1 , . . . , kp ) 6= (0, . . . , 0), then the vectors v1 , . . . , vp are linearly dependent; if the
relation implies (k1 , . . . , kp ) = (0, . . . , 0), then the vectors are linearly independent.
REMARKS 3.2 (i) Let S be an arbitrary nonempty subset of V. Then S is linearly
dependent if and only if S contains a linearly dependent subset.
(ii) Let S be an arbitrary nonempty subset of V. Then S is linearly independent
if and only if all finite subsets of S are linearly independent.
(iii) Let v1 , . . . , vn Km . Denote by A = [v1 , . . . , vn ] Mm,n (K), the matrix
whose columns are v1 , . . . , vn . Then v1 , . . . , vn are linearly independent if and only if
the system
x1
A ... = 0
xn
admits only the trivial solution, and this is equivalent to rank A = n.
Examples
(i) {0} is linearly dependent; if 0 S, then S is linearly dependent.
(ii) If v V, v 6= 0, then {v} is linearly independent.
(iii) A set {v1 , v2 } of two vectors is linearly dependent if and only if either v1 = 0 or
else v2 is a scalar multiple of v1 .
(iv) If f1 (x) = ex ; f2 (x) = ex ; f3 (x) = sh x, then {f1 , f2 , f3 } is linearly dependent

since f1 f2 2f3 = 0.
PROPOSITION 3.3 Let L, S be nonempty subsets of V, v V \ L and L linearly
independent. Then:
(i) S is linearly dependent if and only if there exists an element w S such that
w Span (S \ {w}).
(ii) L {v} is linearly independent if and only if v
/ Span L.
Proof. (i) Assume that S is linearly dependent. Then, there exist distinct elements
v1 , . . . , vp S and k1 , . . . , kp K not all zero, such that
p
X
ki vi = 0 .
i=1
We may assume without loss of generality that k1 6= 0. Then

v1 =
p
X
ki
vi Span (S {v1 }) .
k
i=2 1
Conversely, if w S and w Span (S \ {w}), then it can be written as w =
q
X
cj wj
j=1
for some q 1, cj K, wj S \ {w}. It follows that

w+
q
X
(cj )wj = 0 ,
j=1
therefore the set {w, w1 , . . . , wq } is linearly dependent. Then S is linearly dependent

since it contains a linearly dependent subset.
(ii) Now (i) applies for S = L {v}, S \ {v} = L. Since L is linearly independent,
it follows that L {v} is linearly dependent if and only if v Span L, or equivalently
L {v} is linearly independent if and only if v
/ Span L. QED
PROPOSITION 3.4 Let S = {v1 , . . . , vp } V be a linearly independent set and
denote by Span S the vector subspace spanned by S. Then any distinct p + 1 elements
of Span S are linearly dependent.
Proof. Let wj =
p
X
aij vi Span S, j = 1, . . . , p + 1 and k1 , . . . , kp+1 K such
i=1
that
p+1
X
(3.1)
kj wj = 0 .
j=1
We replace the vectors wj to obtain

p+1
X
j=1
kj wj =
p+1
X
j=1
kj
p
X
i=1
aij vi =
p
X
i=1
p+1
X
j=1
kj aij vi .
4. BASES AND DIMENSION
Then
p
X
i=1
(3.2)
p+1
X
kj aij vi = 0 is equivalent to
j=1
p+1
X
kj aij = 0 ,
i = 1, . . . , p ,
j=1
since v1 , . . . , vp are linearly independent.

But (3.2) is a linear homogeneous system with p equations and p + 1 unknowns
k1 , . . . , kp+1 . Such a system admits nontrivial solutions. Thus, there exist k1 , . . . , kp+1 ,
not all zero, such that (3.1) holds, which means that w1 , . . . , wp+1 are linearly dependent. QED
Bases and Dimension
V is a K vector space. Let B V be a linearly independent set. Then Span B

is a vector subspace of V. A natural question concerns the existence of B linearly
independent such that Span B = V.
DEFINITION 4.1 A subset B of V which is linearly independent and also spans
V, is called a basis of V.
It can be shown that any nonzero vector space admits a basis. We are going to
use this general fact without proof. The results we are proving in this section concern
only a certain class of vector spaces the finitely generated ones.
DEFINITION 4.2 V is called finitely generated if there exists a finite set which
spans V, or if V = {0}.
Note that not all vector spaces are finitely generated. For example, the real vector
space Rn [X] is finitely generated, while R[X] is not.
THEOREM 4.3 Let V 6= {0} be a finitely generated vector space. Any finite set
which spans V contains a basis.
Proof. Let S be a finite set such that Span S = V and set
S = {v1 , . . . , vp } ,
vi 6= vj
if i 6= j .
If S is linearly independent, then S is a basis. If S is not linearly independent,

then there exist k1 , . . . , kp K, not all zero, such that
k1 v1 + . . . + kp vp = 0 .
Assume without loss of generality kp 6= 0; then vp Span{v1 , . . . vp1 }, which
implies Span S = Span{v1 , . . . , vp1 } = V.
We repeat the above procedure for S1 = {v1 , . . . , vp1 }. Continuing this way, we
eventually obtain a subset of S which is linearly dependent but still spans V. QED
10
COROLLARY 4.4 Any nonzero finitely generated vector space admits a finite basis.
Using Prop.3.4, the next consequences of the theorem are straightforward.
COROLLARY 4.5 (i) Any linearly independent subset of a finitely generated vector
space is finite.
(ii) Any basis of a finitely generated vector space is finite.
THEOREM 4.6 Let V be a finitely generated vector space, V 6= {0}. Then any
two bases have the same number of elements.
Proof. Let B, B0 be two bases of V. Assume that B has n elements and B0 has
n0 elements. We have V = Span B, since the basis B spans V. By Proposition 3.4,
no linearly independent subset of V could have more than n elements. The basis B0
is linearly independent, therefore n0 n.
The same argument works for B0 instead of B, yielding n n0 . Therefore n = n0 .
QED
DEFINITION 4.7 Let V be finitely generated.
If V 6= {0}, then the dimension of V is the number of vectors in a basis of V. The
dimension is denoted by dim V.
If V = {0}, then dim {0} = 0 by definition.
Finitely generated vector spaces are also called finite dimensional vector spaces.
The other vector spaces, which have infinite bases are called infinite dimensional.
The dimension of a finite dimensional vector space is a natural number; dim V = 0
if and only if V = {0}.
When it is necessary to specify the field, we write dimK V instead of dim V. For
example, dimC C = 1, dimR C = 2, since C can be regarded as a complex vector
space, as well as a real vector space.
Examples
(i) e1 = (1, 0, . . . , 0);
and dim Kn = n.
e2 = (0, 1, . . . , 0); . . .
en = (0, 0, . . . , 1) form a basis of Kn ,
(ii) B = {Eij |1 i m, 1 j n} is a basis of Mm,n (K), where Eij is the matrix

whose (i, j)entry is 1 and all other entries are zero; dim Mm,n (K) = mn.
(iii) B = {1, X, . . . , X n } is a basis of Kn [X]; dim Kn [X] = n + 1.
(iv) B = {1, X, . . . , X n , . . .} is a basis of K[X]; K[X] is infinite dimensional.
(v) The solution set of a linear homogeneous system of rank r with n unknowns
and coefficients in K is a K-vector space of dimension n r.
THEOREM 4.8 Let V 6= {0} be a finitely generated vector space. Any linearly
independent subset of V is contained into a basis.
Proof. Let L be a linearly independent subset of V and S be a finite set which
spans V. Since V is finitely generated, the set L is finite, by the previous corollary.
If S Span L, then L is a basis.
If S 6 Span L, then choose v S, v
/ Span L. It follows that L {v} is linearly
independent, by Proposition 3.3. Continue until we get a basis.
QED
5. COORDINATES.
ISOMORPHISMS. CHANGE OF COORDINATES
11
COROLLARY 4.9 Let L, S be finite subsets of V such that L is linearly independent and V = Span (S). Then:
(i) card L dim V card S;
(ii) card L = dim V if and only if L is a basis;
(iii) card S = dim V if and only if S is a basis.
PROPOSITION 4.10 If W is a vector subspace of V and dim V = n, n 1, then
W is finitedimensional and dim W n. Equality holds only if W = V.
Proof. Assume W 6= {0} and let v1 W, v1 6= 0. Then {v1 } is linearly independent.
If W = Span {v1 }, then we are done. If W 6 Span {v1 }, then there exists
v2 W \ Span {v1 }. Proposition 3.3 (ii) applies for L = {v1 }, v = v2 , so {v1 , v2 } is
linearly independent.
Now either W = Span{v1 , v2 } or there exists v3 W\Span{v1 , v2 }. In the latter
case we apply again Proposition 3.3 (ii) and we continue the process. The process
must stop after at most n steps, otherwise at step n + 1 we would find vn+1 such that
{v1 , . . . , vn , vn+1 } is linearly independent, which contradicts Proposition 3.4.
By the above procedure we found a basis of W which contains at most n vectors,
thus W is finite dimensional and dim V n.
Assume that dim W = n. Then any basis of W is linearly independent in V and
contains n elements; by the previous corollary it is also a basis of V. QED
THEOREM 4.11 If U, W are finitedimensional vector subspaces of V, then U +
W and U W are finite dimensional and
dim U + dim W = dim (U + W) + dim (U W) .
Sketch of proof. The conclusion is obvious if U W = U or U W = W. If not,
assume U W 6= {0} and let {v1 , . . . , vp } be a basis of U W. Then, there exist
up+1 , . . . , up+q U and wp+1 , . . . , wp+r W such that
{v1 , . . . , vp , up+1 , . . . , up+q } is a basis of U,
{v1 , . . . , vp , wp+1 , . . . , wp+r } is a basis of W,
and show that {v1 , . . . , vp , up+1 , . . . , up+q , wp+1 , . . . , wp+r } is a basis of U + W.
The idea of proof is similar if U W = {0}. QED
Coordinates.
Isomorphisms. Change of Coordinates
Let V be a finite dimensional vector space. In this section we are going to make
explicit computations with bases. For, it will be necessary to work with (finite)
ordered sets of vectors. Consequently, a finite basis B of V has three qualities: it is
linearly independent, it spans V, and it is ordered.
12
PROPOSITION 5.1 The set B = {v1 , . . . , vn } V is a basis if and only if every

vector x V can be written in a unique way in the form
(5.1)
x = x1 v1 + . . . + xn vn ,
xj K ,
j = 1, . . . , n .
Proof. Suppose that B is a basis. Then every x V can be written in the form
(5.1) since V = Span B. If also x = x01 v1 + . . . x0n vn , then
0 = x x = (x1 x01 )v1 + . . . + (xn x0n )vn .
By the linear independence of B it follows that x1 x01 = . . . = xn x0n = 0.
Conversely, the existence of the representation (5.1) for each vector implies V =
Span B. The uniqueness applied for x = 0 = 0v1 + . . . + 0vn gives the linear independence of B. QED
DEFINITION 5.2 The scalars xj are called the coordinates of x with respect to
the basis B. The column vector
x1
X = ...
xn
is called the coordinate vector associated to x.
DEFINITION 5.3 Let V and W be two vector spaces over the same field K. A
map T : V W compatible with the vector space operations, i.e. satisfying:
(a) T (x + y) = T (x) + T (y) ,
(b) T (kx) = kT (x) ,
x, y V
k K, x V
(T is additive)
(T is homogeneous),
is called a linear transformation (or a vector space morphism).

A bijective linear transformation is called an isomorphism.
If there exists an isomorphism T : V W, then V and W are isomorphic and
we write V ' W.
REMARKS 5.4 It is straightforward that:
(i) The composite of two linear transformations is a linear transformation.
(ii) If a linear transformation T is bijective, then its inverse T 1 is a linear transformation.
(iii) ' is an equivalence relation on the class of vector spaces over the same field K
(i.e. V, V ' V; V, W, V ' W = W ' V; V, W, U, U ' W and W '
V = U ' V).
(iv) T (0) = 0, for any linear transformation T .
(v) (a) and (b) in Definition 5.3 are equivalent to:
T (kx + ly) = kT (x) + lT (y) ,
k, l K, x, y V ,
5. COORDINATES.
ISOMORPHISMS. CHANGE OF COORDINATES
13
(i.e. T is compatible with linear combinations)

PROPOSITION 5.5 Every vector space V of dimension n is isomorphic to the
space Kn .
Proof. Let B = {v1 , . . . , vn } be a basis of V. Define T : V Kn by T (x) =
(x1 , . . . , xn ), where x1 , . . . , xn are the coordinates of v with respect to the basis B.
The map T is well defined and bijective, by Proposition 5.1. It is obvious that T is
linear. QED
The linear map used in the above proof is called the coordinate system associated
to the basis B.
LEMMA 5.6 Let T : V W be a linear transformation. Then the following are
equivalent:
(i) T is injective.
(ii) T (x) = 0 = x = 0.
(iii) For any linearly independent set S in V, T (S) is linearly independent in W.
Proof. (i) = (ii) Assume T injective. If T (x) = 0, then T (x) = T (0) by Remark
5.4 (iv). Now injectivity gives x = 0.
(ii) = (i) Assume (ii) and let x, y V, T (x) = T (v). Then T (x y) = 0, so
x y = 0 by (ii). Thus x = y.
(ii) = (iii) It suffices to prove (iii) for S finite (for, see Remark 3.2). Let
S = (v1 , . . . , vp ) and k1 T (v1 ) + . . . + kp T (vp ) = 0. By the linearity of T , T (k1 v1 +
. . . + kp vp ) = 0; then k1 v1 + . . . + kp vp = 0, by (ii). S linearly independent implies
k1 = . . . = kp = 0, thus T (S) is linearly independent.
(iii) = (ii) Note that (ii) is equivalent to: T (x) 6= 0, for all x 6= 0. Let x 6= 0
and S = {x} in (iii). Then {T (x)} is linearly independent, i.e. T (x) 6= 0. QED
THEOREM 5.7 Two finite dimensional vector spaces V and W are isomorphic if
and only if dim V = dim W.
Proof. Suppose that V and W are isomorphic i.e. there exists a bijective linear
transformation T : V W. Let n = dim V and choose a basis B = (v1 , . . . , vn ) of
V. Then T (B) is linearly independent by the previous lemma and the injectivity of
T.
It remains to show that W = SpanT (B). For, let y W. Then, there exists x V
such that T (x) = y, since T is surjective. But x can be written as x = x1 v1 +. . .+xn vn
since x Span B. Then
y = T (x1 v1 + . . . + xn vn ) = x1 T (v1 ) + . . . + xn T (vn ) Span T (B) .
We proved that T (B) is a basis of W. Since card T (B) = n, it follows that
dimV = dim W.
Conversely, suppose dim V = dim W = n. Then V ' Kn and W ' Kn imply
V ' W (see Remarks 5.4). QED
We will use quite often the following lemma. (See the Linear Algebra highschool
manual for a proof.)
14
LEMMA 5.8 Let A = [aij ] 1 i m Mm,n (K). Then the rank of A is equal to the
1 j n
maximal number of linearly independent columns.

THEOREM 5.9 (the change of basis) Let B = {v1 , . . . , vn } be a basis of V and
B0 = {w1 , . . . , wn } an ordered subset of V, with
wj =
n
X
cij vi ,
j = 1, . . . , n .
i=1
Then B0 is another basis of V if and only if det [cij ] 6= 0.

n
Proof. Let T : V
K be the coordinate system associated to B. Then
c1j
..
T (vi ) = ei and T (wj ) = . . Denote C = [cij ]i,j Mn,n (K). It follows that
cnj
T (wj ) is the jth column of C. Then:
det C 6= 0 if and only if rank C = n;
rank C = n if and only if T (B0 ) is linearly independent, by Lemma 5.8;
Lemma 5.6 applies for T and T 1 , thus B0 is linearly independent if and only
if T (B0 ) is linearly independent;
B0 is a basis of W if and only if B0 is linearly independent, by Corollary 4.9,
and the conclusion follows.
QED
DEFINITION 5.10 If B0 in the previous theorem is a basis, then the matrix C is

called the matrix of change of basis from B to B0 .
Note that in this case the matrix of change from B0 to B is C 1 .
COROLLARY 5.11 (the coordinate transformation formula) Let B, B0 be
two bases of V, dim V = n, C the matrix of change from B to B0 , and x V.
If X = t (x1 , . . . , xn ) and X 0 = t (x01 , . . . , x0n ) are the coordinates of x with respect
to B and B0 respectively, then X = CX 0 .
Proof. We write
x=
n
X
j=1
x0j wj =
n
X
j=1
x0j
n
X
i=1
!
cij vi
n
X
i=1
n
X
cij xj vi .
j=1
By the uniqueness of the representation of x with respect to the basis B, we get

n
X
xi =
cij x0j , i = 1, . . . , n. This means X = CX 0 . QED
j=1
6. EUCLIDEAN VECTOR SPACES
15
Euclidean Vector Spaces
Unless otherwise specified, K will denote either one of the fields R or C. Euclidean
vector spaces are real or complex vector spaces with an additional operation that will
be used to define the length of a vector, the angle between two vectors and orthogonality in a way which generalizes the usual geometric properties of space vectors in
V3 . The Euclidean vector space V3 will make the object of a separate chapter.
DEFINITION 6.1 Let V be a real or complex vector space. An inner (scalar, dot)
product on V is a map h , i : V V K which associates to each pair (v, w) a
scalar denoted by hv, wi, satisfying the following properties:
(i) Linearity in the first variable:
hv1 + v2 , wi = hv1 , wi + hv2 , wi , v1 , v2 , w W
hkv, wi = khv, wi , k K, v, w W
(additivity)
(homogeneity)
(ii) Hermitian symmetry (symmetry in the real case):

hw, vi = hv, wi ,
where the bar denotes complex conjugation.
(iii) Positivity:
hv, vi > 0,
v 6= 0, v V .
REMARKS 6.2 (i) Note that linearity in the first variable means that if w is
fixed, then the resulting function of one variable is a linear transformation from
V into K.
(ii) If K = R, then the second equality in (ii) is equivalent to: hv, kwi = khv, wi,
implying linearity in the second variable too, and (iii) is equivalent to hw, vi =
hv, wi.
(iii) By (i) and (ii) in the definition we deduce the conjugate linearity (linearity in
the real case) in the second variable:
hv, w1 + w2 i = hv, w1 i + hv, w2 i
hv, kwi = khv, wi .
(additivity)
(conjugate homogeneity)
(iv) Additivity implies that hv, 0i = 0 = h0, wi, v, w V. In particular h0, 0i = 0.

Combining this with positivity, it follows that:
hv, vi 0 ,
(v) hv, wi = 0 ,
hv, wi = 0 ,
v V
w V = v = 0
and hv, vi = 0 if and only if v = 0 .

and
v V = w = 0 , since hv, vi = 0 = v = 0.
(vi) Hermitian symmetry implies hv, vi R ,

tivity).
v V
(without imposing posi-
16
DEFINITION 6.3 A real or a complex vector space endowed with a scalar product
is called a Euclidean vector space.
Examples of canonical Euclidean vector spaces
1) Rn with hx, yi = x1 y1 + . . . + xn yn , where x = (x1 , . . . , xn ), y = (y1 , . . . , yn ).
2) Cn with hx, yi = x1 y1 + . . . + xn yn , where x = (x1 , . . . , xn ), y = (y1 , . . . , yn ).
a, bi = k
ak kbk cos , where k
ak, kbk are the lengths of a
and b
3) V3 with h
respectively, and is the angle between a

and b.
4) V = {f | f : [a, b] R , f continuous} is a real Euclidean vector space with
the scalar product given by
Zb
hf, gi =
f (t)g(t) dt .
a
5) V = {f | f : [a, b] C , f continuous} is a complex Euclidean vector space

with the scalar product given by
Zb
hf, gi =
f (t)g(t) dt .
a
THEOREM 6.4 (the CauchySchwarz inequality) Let V be a Euclidean vector

space. Then
(6.1)
| hv, wi |2 hv, vihw, wi , v, w V .
Equality holds if and only if v, w are linearly dependent (collinear ).
Proof. The case w = 0 is obvious. Assume w 6= 0. By positivity
(6.2)
hv + w, v + wi 0 ,
Take =
K .
hv, wi
and expand to obtain (6.1).
hw, wi
If equality holds in (6.1), then equality holds in (6.2) for =
hv, wi
. Then
hw, wi
hv, wi
w = 0 by positivity. Thus v, w are linearly dependent.
hw, wi
Conversely, suppose v = w. Then
| hv, wi |2
=| hv, vi |2 =| hv, vi |2 =| |2 hv, vi2 = hv, vihv, vi

= hv, vihv, vi = hv, vihw, wi .
QED
DEFINITION 6.5 Let V be a real or complex vector space.

The function k k : V R is called a norm on V if it satisfies:
(i) kvk 0 ,
v V
and kvk = 0 if and only if v = 0
(positivity)
6. EUCLIDEAN VECTOR SPACES

(ii) kkvk =| k | kvk ,
17
k K, v V
(iii) ku + vk kuk + kvk ,
u, v V (the triangle inequality).
A vector space endowed with a norm is called a normed vector space.
PROPOSITION
6.6 Let V be a Euclidean vector space. Then the function defined
p
by kvk = hv, vi is a norm on V.
Proof. Properties (i) and (ii) in Definition 6.5 are straightforward from the positivity and (conjugate) linearity of the inner product. For (iii)
ku + vk2
hu + v, u + vi = hu, ui + 2Rehu, vi + hv, vi

kuk2 + kvk2 + 2 | hu, vi | ,
since Rez | z | for any z C
kuk2 + kvk2 + 2kuk kvk ,
by Cauchys inequality
(kuk + kvk)2 .
QED
The norm defined by an inner product as in the previous proposition is called a

Euclidean norm.
REMARK 6.7 Let (V, k k) be a (real or complex) vector space. If k k is a Euclidean
norm, then it is easy to see that:
ku + vk2 + ku vk2 = 2(kuk2 + kvk2 ) ,
u, v V
(the parallelogram law) .
Note that not all norms satisfy the parallelogram law. For example, let k k :
Rn [0, ), n 2 defined by kxk = max{| x1 |, . . . , | xn |} and take u =
(1, 0, . . . , 0), v = (1, 1, 0, . . . , 0).
DEFINITION 6.8 A vector u V with kuk = 1 is called a unit vector or versor.
Any vector v V \ {0} can be written as v = kvku, where u is a unit vector. The
1
vector u =
v is called the unit vector in the direction of v.
kvk
If v, w V \ {0} and K = R, then the Cauchy inequality is equivalent to
(6.3)
hv, wi
1.
kvk kwk
This double inequality allows the following definition.

DEFINITION 6.9 Let V be a real Euclidean vector space and v, w V \ {0}.
Then the number [0, ] defined by
cos =
is called the angle between v and w.
hv, wi
kvk kwk
18
THEOREM 6.10 Let V be a real or complex normed vector space. The function
d : V V R, defined by d(v, w) = kv wk is a distance (metric) on V, i.e. it
satisfies the following properties:
(i) d(v, w) 0, v, w V and d(v, w) = 0 if and only if v = w.
(ii) d(v, w) = d(w, v), v, w V
(iii) d(u, v) d(u, w) + d(w, v), u, v, w V.
The proof is straightforward from the properties of the norm.
A vector space endowed with a distance (i.e. a map satisfying (i), (ii), (iii) in the
previous theorem) is called a metric space. If the distance is defined by a Euclidean
norm, then it is called a Euclidean distance.
Orthogonality
Let V be a Euclidean vector space. In the last section, we defined the angle between
two nonzero vectors. The definition of orthogonality will be compatible with the
definition of the angle.
DEFINITION 7.1 (i) Two vectors v, w V are called orthogonal (or perpendicular) if hv, wi = 0. We write v w when v and w are orthogonal.
(ii) A subset S 6= is called orthogonal if its vectors are mutually orthogonal, i.e.
hv, wi = 0, v, w S, v 6= w.
(iii) A subset S 6= is called orthonormal if it is orthogonal and each vector of S
is a unit vector (i.e. v S, kvk = 1).
PROPOSITION 7.2 Any orthogonal set of nonzero vectors is linearly independent.

Proof. Let S be orthogonal. In order to show that S is linearly independent, we
may assume S finite (see Remark 3.2 (iii)), S = {v1 , . . . , vp }, where v1 , . . . , vp are
distinct. Let k1 , . . . , kp K such that k1 v1 + . . . + kp vp = 0. Right multiplication by
vj in the sense of the inner product yields
k1 hv1 , vj i + . . . + kp hvp , vj i = h0, vj i = 0 ,
j = 1, . . . , p .
By the orthogonality of S, hvi , vj i = 0 for i 6= j, thus

kj hvj , vj i = 0 ,
j = 1, . . . , p .
But vj 6= 0, since 0
/ S, so hvj , vj i 6= 0 and it follows that kj = 0, j = 1, . . . , p,
which shows that S is linearly independent. QED
Combining the above result and Corollary 4.8 (iii) we obtain the following result.
COROLLARY 7.3 If dim V = n, n 1, then any orthogonal set of n nonzero
vectors is a basis of V.
7. ORTHOGONALITY
19
THEOREM 7.4 (the GramSchmidt procedure) If dimV = n, n 2 and B =

{v1 , . . . , vn } is a basis of V, then there exists an orthonormal basis B0 = {u1 , . . . , un }
of V, such that
Span {v1 , . . . , vm } = Span {u1 , . . . , um } ,
m = 1, . . . , n .
00
Proof. Consider the set B = {w1 , . . . , wn }, whose elements are defined by

w1 = v1
w2 = v2 + k12 w1
........................................................................
wm = vm + k1m w1 + . . . + km1,m wm1
........................................................................
wn = vn + k1n w1 + . . . + kn1,n wn1 , kij K, i, j {1, . . . , n}, i < j .
It is easy to prove by induction that Span {v1 , . . . , vm } = Span {w1 , . . . , wm } and
wm =
6 0, for all m = 1, . . . , n. This is trivial for m = 1. Let 2 m n and assume
Span {v1 , . . . , vm1 } = Span {w1 , . . . , wm1 } .
Then
Span{v1 , . . . , vm }
= Span{v1 , . . . , vm1 } + Span{vm }

= Span{w1 , . . . , wm1 } + Span{vm }
(by the induction hypothesis)
= Span{w1 , . . . , wm1 , vm }
= Span{w1 , . . . , wm1 , wm }
(by the definition of wm ) .
Now suppose wm = 0 for some m 2. Then vm Span{w1 , . . . , wm1 } =

Span{v1 , . . . , vm1 } implies {v1 , . . . , vm } linearly dependent, by Prop. 3.3. This
yields a contradiction.
The scalars kij can be chosen such that B00 will become an orthogonal set as
follows.
hv2 , w1 i
hw2 , w1 i = 0 hv2 , w1 i + k12 hw1 , w1 i = 0 k12 =
.
hw1 , w1 i
Assume that kij are scalars such that {w1 , . . . , wm1 } is orthogonal. Then, using
the fact that hwj , wi i = 0, i, j {1, . . . , m 1}, i 6= j, we get for all i = 1, . . . , m 1:
hwm , wi i = 0 hvm , wi i + kim hwi , wi i = 0 kim =
hvm , wi i
.
hwi , wi i
Note that hwi , wi i 6= 0 since wi 6= 0.

We showed that B00 is an orthogonal set of nonzero vectors, hence an orthogonal
basis, where:
w1 = v1
hv2 , w1 i
hw1 , w1 i
....................................................
hvm , w1 i
hvm , wm1 i
wm = vm
w1 . . .
wm1
hw1 , w1 i
hwm1 , wm1 i
....................................................
hvn , wn1 i
hvn , w1 i
w1 . . .
wn1 ,
wn = vn
hw1 , w1 i
hwn1 , wn1 i
w2 = v2
20
and Span {v1 , . . . , vm } = Span {w1 , . . . , wm }, m = 1, . . . , n.

1
Now set uj =
wj , j = 1, . . . , n. The set B0 = {u1 , . . . , un } fulfils all the
kwj k
required properties. QED
COROLLARY 7.5 Any finite dimensional Euclidean space has an orthonormal basis.
THEOREM 7.6 If dim V = n and B = {u1 , . . . , un } is an orthogonal basis, then
the representation of x V with respect to the basis B is
x=
hx, u1 i
hx, un i
u1 + . . . +
un .
hu1 , u1 i
hun , un i
In particular, if B is an orthonormal basis, then

x = hx, u1 iu1 + . . . + hx, un iun .
Proof. The vector x has an unique representation with respect to the basis B:
x=
n
X
xi ui .
(see (5.1))
i=1
We multiply this equality by each uj , j = 1, . . . , n in the sense of the inner product.

It follows that
hx, uj i =
n
X
xi hui , uj i = xj huj , uj i ,
j = 1, . . . , n ,
i=1
thus
xj =
hx, uj i
,
huj , uj i
j = 1, . . . , n .
If B is an orthonormal basis, then huj , uj i = 1, so xj = hx, uj i.
QED
DEFINITION 7.7 Let 6= S V.

A vector v V is orthogonal to S if v w, w S.
The set of all vectors orthogonal to S is denoted by S and is called Sorthogonal.
It is easy to see that S is a vector subspace of V.
If W is a vector subspace of V, then W is called the orthogonal complement of
W.
Examples
(i) {0} = V, since hv, 0i = 0, v V.
(ii) V = {0}, since
hv, wi = 0 ,
w V = hv, vi = 0 = v = 0 .
(iii) If S = {(1, 1, 0), (1, 0, 1)} R3 , then

S = {(x, y, z) R3 ; x + y = 0, x + z = 0} .
7. ORTHOGONALITY
21
REMARKS 7.8 (i) S = (Span S) .

(ii) If W is a vector subspace, then W W = {0}. (For, let w W W . It
follows that hw, wi = 0, thus w = 0.)
THEOREM 7.9 Let W be a vector subspace of a Euclidean vector space V, dimW =
n, n N . Then V = W W . Moreover, if v = w + w , w W, w W , then
the Pythagorean theorem holds, i.e. kvk2 = kwk2 + kw k2 .
Proof. Let B = {u1 , . . . , un } be an orthonormal basis of W and v V. Take
w = hv, u1 iu1 + . . . + hv, un iun , and w = v w. Obviously w W and for any
j = 1, . . . , n we get
hw , uj i = hv w, uj i = hv, uj i
n
X
hv, ui ihui , uj i = hv, uj i hv, uj i = 0 .
i=1
We showed that w B = W , therefore V = W + W . The sum is direct

since W W = {0}. It follows that the decomposition
v = w + w ,
w W, w W
is unique. Also
kvk2 = hv, vi = hw, wi + hw, w i + hw , wi + hw , w i = kwk2 + kw k2 ,
since hw, w i = hw , wi = 0.
QED
The vector w W in the decomposition of v with respect to V = W W is

called the orthogonal projection of v onto W.
PROPOSITION 7.10 Let V be a Euclidean vector space, dim V = n, n N ,
n
n
X
X
B = {u1 , . . . , un } an orthonormal basis and x =
xj uj , y =
yj uj V. If V is
j=1
a real vector space, then hx, yi =
n
X
j=1
xj yj and kxk2 =
j=1
n
X
x2j .
j=1
If V is a complex vector space, then hx, yi =
n
X
xj yj and kxk2 =
j=1
n
X
j=1
Proof. Assume V is a complex vector space. Then

* n
+
n
X
X
n P
n
P
hx, yi =
xi ui ,
yj uj =
hxi ui , yj uj i
i=1
n X
n
X
j=1
xi yj hui , uj i =
i=1 j=1
The real case is similar.
i=1 j=1
n X
n
X
i=1 j=1
xi yj ij =
n
X
xj yj .
j=1
QED
We conclude this section with a generalization of Theorem 7.4.
| x j |2 .
22
THEOREM 7.11 (GramSchmidt infinite dimensional case) If V is infinite dimensional and L = {v1 , . . . , vk , . . .} V is a countable, infinite, linearly
independent set of distinct elements, then there exists an orthonormal set L0 =
{u1 , . . . , uk , . . .} such that
Span {v1 , . . . , vk } = Span {u1 , . . . , uk } ,
k N .
The proof is based on the construction used in Theorem 7.4, i.e.

w1 = v1
hv2 , w1 i
hw1 , w1 i
.............................................................
hvk , wk1 i
hvk , w1 i
w1 . . .
wk1 ,
k 2,
wk = vk
hw1 , w1 i
hwk1 , wk1 i
.............................................................
w2 = v2
then uj =
1
wj , k N .
kwj k
Problems
1. Let V be a vector space over the field K and S a nonempty set. We define
F = {f |f : S V},
(f + g)(x) = f (x) + g(x),
(tf )(x) = tf (x),
for all
for all
f, g F,
t K, f F.
Show that F is a vector space over K.

2. Let F be the real vector space of the functions f : R R. Show that
cos t, cos2 t, . . . , cosn t, . . .
is a linearly independent system in F .
3. Let V be the real vector space of functions obtained as the composite of a
polynomial of degree at most three and the cosine function (polynomials in cos x, of
degree at most three). Write down the transformation of coordinates corresponding to
the change of basis from {1, cos x, cos2 x, cos3 x} to the basis {1, cos x, cos 2x, cos 3x},
and find the inverse of this transformation. Generalization.
4. Let V be
real vector space of the real sequences {xn } with the property
Pthe
that the series
x2n is convergent.
Let x = {xn }, y = {yn } be two elements of V.
P
1) Show that the series
xn yn is absolutely convergent.
P
2) Prove that the function defined by hx, yi =
xn yn is a scalar product on V.
5. Let V = {x| x R, x > 0}. For any x, y V, and any s R, we define
x y = xy,
Show that (V, , ) is a real vector space.
s x = xs .
8. PROBLEMS
23
6. Is the set Kn [X] of all polynomials of degree at most n, a vector space over
K? What about the set of the polynomials of degree at least n?
7. Show that the set of all convergent sequences of real (complex) numbers is a
vector space over R (C) with respect to the usual addition of sequences and multiplication of a sequence by a number.
8. Prove that the following sets are real vector spaces with respect to the usual
addition of functions and multiplication of a function by a real number.
1) {f | f : I R, I = interval R, f differentiable}
2) {f | f : I R, I = interval R, f admits antiderivatives}
3) {f | f : [a, b] R, f integrable}.
9. Which of the following pairs of operations define a real vector space structure
on R2 ?
1) (x1 , x2 ) + (y1 , y2 ) = (x1 + x2 , x2 y2 ), k(x1 , x2 ) = (kx1 , kx2 ), k R
2) (x1 , x2 ) + (y1 , y2 ) = (x1 + y1 , x2 + y2 ), k(x1 , x2 ) = (x1 , kx2 ), k R
3) (x1 , x2 ) + (y1 , y2 ) = (x1 + y1 , x2 + y2 ), k(x1 , x2 ) = (kx1 , kx2 ), k R.
10. Let Pn be the real vector space of real polynomial functions of degree at most
n. Study which of the following subsets are vector subspaces, then determine the sum
and the intersection of the vector subspaces you found.
A = {p| p(0) = 0}, B = {p| p(0) = 1}, C = {p| p(1) + p(1) = 0}.
11. Study the linear dependence of the following sets:
1) {1, 1, 1), (0, 3, 1), (1, 2, 2)} R3 ,
0 i
0 1
i 0
1 0
M (C),
,
,
,
2)
i 0
1 0
0 i
0 1
3) {1, x, x2 }, {ex , xex , x2 ex }, {ex , ex , sinh x}, {1, cos2 x, cos 2x} C (R) = the
real vector space of C functions on R.
12. Show that the solution set of a linear homogeneous system with n unknowns
(and coefficients in K) is a vector subspace of Kn . Determine its dimension.
13. Consider V = Kn . Prove that every subspace W of V is the solution set of
some linear homogeneous system with n unknowns.
14. A straight line in R2 is identified to the solution set of a (nontrivial) linear
equation with two unknowns. Similarly, a plane in R3 is identified to the solution
set of a linear equation with three unknowns; a straight line in R3 can be viewed as
the intersection of two planes, so it may be identified to the solution set of a linear
system of rank two, with three unknowns.
(a) Prove that the only proper subspaces of R2 are the straight lines passing
through the origin.
(b) Prove that the only proper subspaces of R3 are the planes passing through the
origin, and the straight lines passing through the origin.
15. Which of the following subsets of R3 are vector subspaces?
D1 :
x
y
z
x1
y
z2
= =
, D2 :
= =
1
2
1
1
1
1
24

P1 : x + y + z = 0, P2 : x y z = 1.
Determine the intersection and the sum of the vector spaces found above.
16. In R3 , we consider the vector subspaces P : x+2yz = 0 and Q : 2xy+2z =

0. For each of them, determine a basis and a supplementary vector subspace. Find a
basis of the sum P + Q and a basis of the intersection P Q.
17. Let S be a subset of R3 made of the vectors
1) v1 = (1, 0, 1), v2 = (1, 0, 1), v3 = (3, 0, 3)
2) v1 = (1, 1, 1), v2 = (1, 1, 1), v3 = (0, 0, 1), v4 = (1, 1, 2).
In each case, determine the dimension of the subspace spanned by S and point
out a basis contained in S. What are the Cartesian equations of the subspace ?
18. In each case write down the coordinate transformation formulas corresponding
to the change of basis.
1) e1 = (1, 2, 1), e2 = (1, 2, 1), e3 = (0, 1, 1) in R3
e01 = (1, 1, 1), e02 = (0, 1, 1), e03 = (1, 1, 0).
2) e1 = 1, e2 = t, e3 = t2 , e4 = t3 in P3
e01 = 1 t, e02 = 1 + t2 , e03 = t2 t, e04 = t3 + t2 .
19. Let V be a real vector space. Show that V V is a complex vector space
with respect to the operations
(u, v) + (x, y) = (u + x, v + y)
(a + ib)(u, v) = (au bv, bu + av).
This complex vector space is called the complexification of V and is denoted by
V. Show that:
1) C Rn = Cn ;
2) if S is a linearly independent set in V, then S {0} is a linearly independent
set in C V;
3) if {e1 , . . . , en } is a basis of V, then {(e1 , 0), . . . , (en , 0)} is a basis of C V, and
dimC C V = dimR V.
20. Let V be a complex vector space. Consider the set V with the same additive
group structure, but scalar multiplication restricted to multiplication by real numbers.
Prove that the set V becomes a real vector space in this way.
Denote this real vector space by R V. Show that
1) R Cn = R2n ,
2) if dimV = n, then dimR V = 2n.
21. Explain why the following maps are not scalar products:
n
P
1) : Rn Rn R, (x, y) =
|xi yi |
i=1
3) : C 0 ([0, 1]) C 0 ([0, 1]) R, (f, g) = f (0)g(0).

22. In the canonical Euclidean space R3 , consider the vector v = (2, 1, 1) and
the vector subspace P : x y + 2z = 0. Find the orthogonal projection w of v on P
and the vector w .
23. Let R4 be the canonical Euclidean space of dimension 4. Find an orthonormal
basis for the subspace generated by the vectors
u = (0, 1, 1, 0), v = (1, 0, 0, 1), w = (1, 1, 1, 1).
Chapter 2
Linear Transformations
1
General Properties
Throughout this section V and W will be vector spaces over the same field K.
We used linear linear transformations, in particular the notion of isomorphism
for the the identification of an ndimensional vector space with Kn (see Prop. 5.5,
Chap.1). In this chapter we will study linear transformations in more detail.
Recall (Def. 5.3) that a linear transformation T from V into W is a map T :
V W satisfying
(1.1)
(1.2)
T (x + y) = T (x) + T (y) ,
T (kx) = kT (x) ,
x, y V
k K, x V
(additivity)
(homogeneity).
The definition is equivalent to

(1.3)
T (kx + ly) = kT (x) + lT (y) ,
k, l K, x, y V,
(linearity)
as we pointed out in Chap. 1, Remarks 5.4 (v).

We will write sometimes T x instead of T (x).
We are actually quite familiar with some examples of linear transformations. One
of them is the following, which is in fact the main example.
MAIN EXAMPLE: left multiplication by a matrix
Let A Mm,n (K) and consider A as an operator on column vectors. It defines a
linear transformation A : Kn Km by
x1
X AX , where X = ... ;
A(x) =t(AX) ,
xn
x = (x1 , . . . , xn ) Kn ,
X =tx .
Obviously, A is linear by the known properties of the matrix multiplication.

Note that each component of the vector A(x) is a linear combination of the components of x.
25
26
CHAPTER 2. LINEAR TRANSFORMATIONS
2 1
Consider the particular case A = 3 0 M3,2 (R) ; then
1 5
x1
x2
2x1 + x2
3x1
=
x1 + 5x2
and
A(x) = (2x1 + x2 , 3x1 , x1 + 5x2 ) .
Conversely, any map A : Kn Km with the property that each component
of A(x) is a linear combination with constant coefficients of the components of x, is
given by a matrix A Mm,n (K as above; A is the coefficient matrix of t (A(x)), i.e.
if for any x Kn , the ith component of A(x) is k1 x1 + . . . + kn xn , then the ith row
of A is (k1 , . . . , kn ).
If m = n = 1, then A = [a] for some a K, and A : K K, A(x) = ax.
Linear transformations are also called vector space homomorphisms (or morphisms),
or linear operators, or just linear maps. Their compatibility with the operations can
be regarded as a kind of transport of the algebraic structure of V to W.
Note that (1.1) says that T is a homomorphism of additive groups.
A linear map F : V K (i.e. W = K) is also called a linear form on V.
More examples of linear transformations
(i) V = Pn = the vector space of real polynomial functions of degree n, W =
Pn1 and T (p) = p0 .
(ii) V = C 1 (a, b), W = C 0 (a, b), T (f ) = f 0 .
Zb
0
(iii) V = C [a, b], W = R, T (f ) =
f (t) dt.
a
(iv) V = W, c K, T : V V, T (x) = cx, x V. (For c = 1, T is the identity

map of V)
In Chapter 1, Rem. 5.4 we mentioned that T (0) = 0, for any linear transformation
T . We will show next that more properties of V are transfered to W via T .
THEOREM 1.1 Let T : V W be a linear transformation.
(i) If U is a vector subspace of V, then T (U) is a vector subspace of W.
(ii) If v1 , . . . , vp V are linearly dependent vectors, then T (v1 ), . . . , T (vp ) are linearly dependent vectors in W.
Proof. (i) Let w1 , w2 T (U) and k, l K. Then w1 = T (v1 ), w2 = T (v2 ), for
some v1 , v2 U and
kw1 + lw2 = kT (v1 ) + lT (v2 ) = T (kv1 + lv2 ) ,
by (1.3) .
1. GENERAL PROPERTIES
27
On the other hand kv1 +lv2 U since U is a subspace of V. Therefore kw1 +lw2
T (U).
(ii) By the linear dependence of v1 , . . . , vp there exist k1 , . . . , kp , not all zero such
that k1 v1 + . . . + kp vp = 0. Then T (k1 v1 + . . . + kp vp ) = T (0) = 0. Now the linearity
of T implies that k1 T (v1 ) + . . . + kp T (vp ) = 0.
QED
REMARK 1.2 Note that if in (ii) of the previous theorem we replace dependent
by independent, the statement we obtain is no longer true. The linear independence
of vectors is preserved only by injective linear maps (see Chap. 1, Lemma 5.6).
THEOREM 1.3 Assume dim V = n and let B = {e1 , . . . , en } be a basis of V, and
w1 , . . . , wn arbitrary vectors in W.
(i) There is a unique linear transformation T : V W such that
(1.4)
T (ej ) = wj ,
j = 1, . . . , n .
(ii) The linear transformation T defined by (1.4) is injective if and only if w1 , . . . , wn

are linearly independent.
Proof. (i) If x V, then x can be represented as x =
n
X
xj ej , xj K. Define
j=1
T (x) =
n
X
also y =
xj wj W. This definition of T implies T (ej ) = wj , j = 1, . . . , n. Let
j=1
n
X
yj ej V, and k, l K. Then
j=1
T (kx + ly) =
n
X
(kxj + lyj )wj = k
j=1
n
X
xj wj + l
j=1
n
X
yj wj = kT (x) + lT (y) ,
j=1
hence T is linear. For the uniqueness, suppose T1 (ej ) = wj , j.

n
n
n
X
X
X
For any x =
xj ej V it follows that T1 (x) =
xj T1 (ej ) =
xj wj , by the
j=1
j=1
j=1
linearity of T1 . Hence T1 (x) = T (x), x V.

(ii) In Chap. 1, Lemma 5.6 we gave necessary and sufficient conditions for the
injectivity of a linear transformation. From condition (iii) of that result, here it only
remains to show that
w1 , . . . , wn
Let x =
n
X
xj ej , y =
j=1
linearly independent = T injective .

n
X
yj ej V such that T (x) = T (y). Then T (x y) =
j=1
n
X
(xj yj )wj = 0. The linear dependence of w1 , . . . , wn implies that xj = yj , j,
j=1
thus x = y.
QED
28
Operations with Linear Transformations

Denote by L(V, W) the set of all linear maps defined on V with values into
W. Addition and scalar multiplication for the elements of L(V, W) are the usual
operations for maps with values into an Kvector space. (see Examples of vector
spaces (v) Chapter 1); if S, T L(V, W) then S + T and kT are defined by
(1.5)
(S + T )(x) = S(x) + T (x) ,
(1.6)
(kT )(x) = kT (x) ,
x V
x V, k K .
It is easy to see that S + T , kT L(V, W). Moreover, L(V, W) becomes a

Kvector space with the operations defined by (1.5), (1.6).
The elements of L(V, V) are called endomorphisms of V. Another notation for
L(V, V) is End (V).
The vector space L(V, K) of all linear forms on V is called the dual of V.
Let U, V, W be vector spaces over the same field K and S L(U, V), T
L(V, W). Then the composite T S makes sense as a map from U into W. It is
straightforward that T S L(U, W). The map T S is also written as T S and is
called the product of T and S. The following properties are immediate:
(1.7) (kA + lB)C = kAC + lBC ,
k, l K, A, B L(U, V), C L(V, W) .
(1.8) C(kA + lB) = kCA + lCB ,
k, l K, A, B L(V, W), C L(U, V) .
If T End (V), denote

T0=I ,
T 2 = T T , . . . , T n = T n1 T = T T n1 ,
where n 1 and I is the identity map of V.

The product of endomorphisms is a binary operation on End (V).
We mentioned in Chap. 1, Rem. 5.4 that the inverse of a bijective linear transformation is linear too (it is an easy exercise to prove it).
An injective homomorphism is called a monomorphism.
A surjective homomorphism is called an epimorphism.
A bijective homomorphism is called an isomorphism (see also Chap. 1, Def. 5.3,
Rem. 5.4).
REMARK 1.4 Let us return to the main example. Let A L(Kn , Km ), B
L(Km , Kp ) be defined as left multiplication by A Mm,n (K) and B Mn,p (K)
respectively. Then:
(i) BA L(Kn , Kp ) is the left multiplication by the matrix AB Mm,p (K).
(ii) A is bijective m = n and A is invertible.
In this case A1 is the left multiplication by the matrix A1 .
2. KERNEL AND IMAGE
29
Kernel and Image
Let V and W be two vector spaces and T L(V, W).

DEFINITION 2.1 The set
Ker T = {x V ; T (x) = 0} V
is called the kernel of T .
The image of T is the usual
Im T = T (V) = {y W ; x V such that y = T (x)} W .
REMARKS 2.2 Recall that:

(i) T (0) = 0, so 0V Ker T and 0W Im T .
(ii) Ker T = {0} if and only if T is injective.
THEOREM 2.3 (i) Ker T is a subspace of V.

(ii) Im T is a subspace of W.
Proof. (i) Let u, v Ker T , i.e. T (u) = T (v) = 0. Then
T (ku + lv) = kT (u) + lT (v) = k0 + l0 = 0 ,
Thus ku + lv Ker T , k, l K.
(ii) This is a particular case of Thm. 1.1 (i).
3
k, l K .
QED
Example. Let T : R R , T (x) = (2x1 x2 , x1 + x2 + 3x3 ). We find

Ker T
= {(x1 , x2 , x3 ) ; 2x1 x2 = 0, x1 + x2 + 3x3 = 0 }

= {(, 2, ) ; R } = Span {(1, 2, 1)} .
Im T
(y1 , y2 ) ; the system
2x1 x2 = y1
2x1 + x2 + 3x3 = y2
is compatible
= R2 .
Note that T is surjective, but not injective. We will see later in this section that
a linear transformation between two finite dimensional spaces of the same dimension
is surjective if and only if it is injective.
DEFINITION 2.4 The dimensions of Im T and Ker T are called the rank and
nullity of T respectively.
PROPOSITION 2.5 Let y Im T . The general solution of the equation
(2.1)
T (x) = y
is the sum of the general solution of T (x) = 0 and a particular solution of (2.1).
30
Proof. Denote Sy = {x ; T (x) = y} = T 1 (y), S0 = {x ; T (x) = 0} = Ker T ,

and fix xp Sy . We must show that Sy = S0 + {xp }.
For, let x
Sy . Then x
= (
x xp ) + xp S0 + {xp }, since T (
x xp ) = T (
x)
T (xp ) = y y = 0. Conversely, let x0 S0 ; then T (x0 + xp ) = T (x0 ) + T (xp ) = y,
hence x0 + xp Sy . QED
THEOREM 2.6 (The dimension formula) Let T : V W be a linear transformation and assume that V is finite dimensional. Then
(2.2)
dim V = dim (Ker T ) + dim (Im T ) .
Proof. Say that dim V = n. Then Ker T is finite dimensional too as a subspace
of V. Let p = dim (Ker T ), 1 p n 1 and B1 = {u1 , . . . , up } a basis of Ker T
(the cases p = 0, p = n are left to the reader). Extend B1 to a basis of V (see Chap.
1, Thm. 4.8):
B = {u1 , . . . , up , v1 , . . . , vnp } .
Denote wj = T (vj ), j = 1, . . . , n p and B2 = {w1 , . . . , wnp }.
If we show that B2 is a basis of Im T , it will follow that dim (Im T ) = n p, and
the proof of the theorem will be done. For, let y Im T . Then y = T (x), for some
x V. We write x in terms of the basis B of V:
x = a1 u1 + . . . + ap up + b1 v1 + . . . + bnp vnp ,
and apply T , using T (ui ) = 0. We obtain:
y = T (x) = b1 T (v1 ) + . . . + bnp T (vnp ) = b1 w1 + . . . + bnp wnp .
This proves that Im T = Span B2 .
Now suppose that
(2.3)
k1 w1 + . . . + knp wnp = 0 ,
k1 , . . . , knp K .
From wj = T (vj ) it follows that

k1 T (v1 ) + . . . + knp T (vnp ) = T (k1 v1 + ... + knp vnp ) = 0 .
Denote v = k1 v1 + . . . + knp vnp ; thus v Ker T . So we may write v in terms
of the basis B1 of Ker T , say v = c1 u1 + . . . + cp up . Then
c1 u1 + . . . + cp up k1 v1 . . . knp vnp = 0 .
By the linear independence of B, c1 = . . . = cp = k1 = . . . = knp = 0. Therefore
all scalars in (2.3) vanish, which shows that B2 is linearly independent. Consequently,
B2 is a basis of Im T . QED
In Chapter 1, Lemma 5.6 we proved that for a linear map T there are equivalent
characterizations of injectivity. Using the dimension formula, we can now state more
such characterizations.
3. THE MATRIX OF A LINEAR TRANSFORMATION
31
COROLLARY 2.7 Let T L(V, W), dim V = n, n N . Then the following

statements are equivalent:
(i) T is injective.
(ii) dim (Im T ) = n.
(iii) If {v1 , . . . , vn } is a basis of V, then {T (v1 ), . . . , T (vn )} is a basis of Im T .
Proof. (i) = (ii) T injective = Ker T = {0} = dim (Ker T ) = {0} =
dim (Im T ) = dim V 0 = n, by the dimension formula (2.3).
(ii) = (iii) It is straightforward that {T (v1 ), . . . , T (vn )} spans Im T . Then
dim (Im T ) = n implies that {T (v1 ), . . . , T (vn )} is a basis of Im T (see Cor. 4.8,
Chap. 1).
(iii) = (i) (iii) says that Im T has a basis with n elements, so dim (Im T ) =
n = dim V. By the dimension formula, dim (Ker T ) = 0, so Ker T = {0}, which is
equivalent to (i). QED
The Matrix of a Linear Transformation
Throughout this section V and W denote two finite dimensional vector spaces over
the same field K, of dimensions n and m respectively; T : V W is a linear
transformation.
Let B = {v1 , . . . , vn } and C = {w1 , . . . , wm } be bases of V and W respectively.
DEFINITION 3.1 The matrix T Mm,n (K) whose j 0 th column is the coordinate
column vector of T (vj ), j {1, ..., n}, is called the matrix associated to T (or the
matrix of T ) with respect to the bases B and C.
m
P
So, if for each j {1, ..., n} we write T (vj ) =
tij wi , then T = [tij ] 1 i m .
i=1
1 j n
Examples
(i) T : R2 R3 , T (x) = (2x1 + x2 , 3x1 , x1 + 5x2 ).
Let B = {v1 , v2 }; C = {w1 , w2 w3 } be the canonical bases of R2 and R3 respectively. Denote by T the associated matrix.
T
T = MB
.
,C
Using the definition we find:

T (v1 ) = T (1, 0) = (2, 3, 1) = 2f1 + 3f2 f3 ;
T (v2 ) = T (0, 1) = (1, 0, 5) = f1 + 5f3 .
2
1
Thus the 1st column of T is 3 and the 2nd column of T is 0 ;
1
5
2 1
T = 3 0 .
1 5
32
(ii) V = Span {f1 , f2 } C (R); f1 (x) = ex cos x, f2 (x) = ex sin x, x R.

T : V V, T (f ) = f 0 . B = C = {f1 , f2 } is a basis of V.
T (f1 )(x) = f10 (x) = ex cos x ex sin x, x R. So, T (f1 ) = f1 f2 . Similarly,
1 1
T (f2 ) = f1 + f2 . Then T =
.
1 1
REMARKS 3.2 (i) T is uniquely determined by T and T is uniquely determined
by T . (This is immediate, since a linear transformation is uniquely determined
by its values on the basis B, and each T (vj ) has a unique representation w.r.t
the basis C.)
(ii) Note that in example (i) above, using the definition of the associated matrix
w.r.t. the canonical bases we recovered the matrix mentioned in a particular
case of the main example; if T : Kn Km , is defined by
T (x) = (
n
X
t1j xj , ...,
j=1
n
X
tmj xj ),
j=1
then, according to Definition 3.1, the matrix associated to T w.r.t. the canonical
bases of Kn and Km is obviously the same as the coefficient matrix of the column
t
(T (x)), namely T = [tij ].
So, T acts as left multiplication by T .
Moreover, the next proposition points out that any linear transformation of finite
dimensional vector spaces reduces to left multiplication by a matrix. This is why left
multiplication by a matrix was called the main example.
PROPOSITION 3.3 (the matrix of a linear transformation)
Let V, W, B, C, T , T as in Def.3.1.
n
m
P
P
Let also x V, x =
xj vj , and y W, y =
yi wi such that y = T (x).
j=1
i=1
Denote X =t [x1 , ..., xn ], Y =t [y1 , ..., yn ]. Then T can be written in the matrix form
as:
n
X
(3.4)
Y = T X, i.e. yi =
tij xj , i = 1, ..., m.
j=1
Proof. T (vj ) =
m
P
i=1
tij wi , j = 1, ..., n, since T is the matrix of T .
Then y = T (x) =
n
P
j=1
xj
m
P
i=1
tij wi =
m P
n
P
(
tij )wi .
i=1 j=1
By the uniqueness of the representation of y in the basis C, yi =

1, ..., m.
n
P
j=1
tij xj ,
i =
QED
COROLLARY 3.4 Let U be another finite dimensional vector space, of basis D,

T L(V, W), S L(W, U),
T = the matrix of T w.r.t. B and C;
S = the matrix of S w.r.t. C and D.
Then ST = the matrix of ST w.r.t. B, D.
3. THE MATRIX OF A LINEAR TRANSFORMATION
33
PROPOSITION 3.5 Let r = rank T .

Then dim (ImT ) = r and dim (KerT ) = n r.
Proof. Denote Cj =t [t1j , ..., tmj ] the jth column of T . By Lemma 5.8 in Chapter
1, we may assume without loss of generality, that C1 , ..., Cr are linearly independent
in Mm,1 ' Km and any p columns of T are linearly dependent for p r + 1 (if r < n).
Consider the coordinate system : W Km associated to the basis C, (y) =
m
P
(y1 , ..., ym ), for y =
yi wi . Then (T (vj )) =t Cj . Since is an isomorphism, it
i=1
preserves the linear dependence and the linear independence of vectors, and so does
1 . Thus T (v1 ), ..., T (vr ) are linearly independent and any p r + 1 elements of
{T (v1 ), ..., T (vn )} are linear dependent.
On the other hand, Im T = Span (T (v1 ), ..., T (vn )). It follows that {T (v1 ), ..., T (vr )}
is a basis of Im T , therefore dim (Im T ) = r, then we apply the dimension formula.
QED
We notice that the rank of the associated matrix T does not depend on the chosen
bases. Also, the notation rank T = dim (Im T ) makes now more sense, since the
previous result shos that dim (Im T ) is indeed the rank of a certain matrix.
Using the above result, it is not hard to deduce the next corollary.
COROLLARY 3.6 Let T L(V, W), V, W finite dimensional, T = the matrix
of T w.r.t. B and C.
(i) T is surjective if and only if dim W = rank T.
(ii) T is injective if and only if dim V = rank T.
(iii) T is bijective if and only if dim V = dim W = rank T.
In this case T is invertible, and the matrix of T 1 w.r.t. C, B is T 1 .
(iv) If dim V = dim W = n, then:
T is injective T is surjective T is bijective n = rank T.
Note that a particular case of (iv) is the case V = W, i.e. T End(V), V finite
dimensional.
PROPOSITION 3.7 Let T End(V), dim V = n, n N and B = {v1 , . . . , vn }
B0 = {v10 , . . . , vn0 } two bases of V. Denote by A and B the matrices associated to
T w.r.t. the basis B, and B0 respectively.
Then B = C 1 AC, where C is the matrix of change from B to B0 .
Proof. Let C = [cij ], A = [aij ], B = [bij ] and C 1 = [dij ]. Then
vj0 =
(3.5)
n
X
cij vi ,
vj =
i=1
(3.6)
T (vi ) =
n
X
k=1
aki vk ,
n
X
dij vi0 ,
j = 1, ..., n.
i=1
i = 1, ..., n;
T (vj0 ) =
n
X
l=1
blj vl0 ,
j = 1, ..., n.
34
Combining (3.5) and (3.6) we obtain:

T (vj0 ) =
n
X
cij T (vi ) =
i=1
n X
n
X
cij aki vk =
i=1 k=1
n X
n X
n
X
(
dlk aki cij )vl0 ,
n X
n
X
i=1 k=1
cij aki
n
X
dlk vl0 =
l=1
j = 1, ...n.
l=1 i=1 k=1
By the uniqueness of the representation of T (vj0 ) w.r.t. B0 and (3.6), it follows that
n P
n
P
blj =
dlk aki cij , l, j = 1, ..., n, thus B = C 1 AC.
QED
k=1 i=1
DEFINITION 3.8 The matrices A, B Mn,n (K) are called similar if there exists
a nonsingular matrix C Mn,n (K), such that B = C 1 AC.
By the previous proposition, two matrices are similar if and only if they represent
the same endomorphism (with respect to different bases).
REMARKS 3.9 Properties of similar matrices
(i) Similarity of matrices is an equivalence relation on Mn,n (K).
(ii) Similar matrices have the same rank.
(iii) Nonsingular similar matrices have the same determinant. As a consequence, it
makes sense to define the determinant of an endomorphism as the determinant
of the associated matrix with respect to an arbitrary basis.
Particular Endomorphisms
DEFINITION 4.1 Let V be a K-vector space and F L(V,V).

(i) F is an automorphism if it is bijective.
(ii) F is a projection if F 2 = F.
(iii) F is an involution if F 2 = idV .
(iv) F is a complex structure if F 2 = idV , for K=R.
(v) F is nilpotent of index p if F p1 6= 0, F k = 0, k p, where p {2, 3, . . .}.
Denote Gl(V) = {F L(V,V)| F is an automorphism }. Gl(V) is not a vector
space, but it is a group with respect to the product (composition) of automorphisms.
Gl(V) is called the general linear group of V.
THEOREM 4.2 If F : V V is a projection, then V = Ker F Im F.
4. PARTICULAR ENDOMORPHISMS
35
Proof. Let v V, F(v) Im F, w = v F (v) V. Then F(w) = F(v) F 2 (v) =

0, thus w Ker F, which shows that V = Ker F + Im F. By the dimension formula,
dim V = dim Ker F + dim Im F, therefore dim (KerF Im F) = 0, so the sum is
direct.
QED
Obviously if F is a projection, then idV F is a projection too. The theorem
could be reformulated:
If F1 and F2 are projections such that F1 +F2 = idV , then V = Im F1 Im F2 .
The latter form has the following generalization.
THEOREM 4.3 If Fi : V V, i = 1, ..., p are projections such that Fi Fj =
p
P
0, i 6= j and
Fi = idV , then V = Im F1 ... Im Fp .
i=1
Proof. The conclusion follows easily from the previous theorem, by induction.
Note that for the induction step we need F1 + ... + Fp1 to be a projection. To show
this, use Fi2 = Fi , and Fi Fj = 0, i 6= j.
The details of the proof are left to the reader.
QED
THEOREM 4.4 A finite dimensional real vector space V admits a complex structure if and only if dim V is even.
Proof. Suppose dim V = n = 2m. Let B = {e1 , ..., em , em+1 , ..., e2m } be a basis of
V. Define the endomorphism F : V V by
F(ei ) = em+i ,
F(em+i ) = ei ,
i = 1, ..., m.
Then F(ej ) = ej , j = 1, ..., n, so F 2 = idV . Conversely, let F be a complex

struture on V. Choose v1 V, v1 6= 0; then v1 , F(v1 ) are linearly independent.
For, suppose by contradiction that F(v1 ) = kv1 , for some k R. It follows that
v1 = F 2 (v1 ) = kF(v1 ) = k 2 v1 ; but this is impossible for k R.
If dim V > 2, then we can choose v2 V such that v1 , v2 , F(v1 ) are linearly
independent. We can show similarly that v1 , v2 , F(v1 ), F(v2 ), are linearly independent. Continuing the process we obtain a basis which has an even number of elements.
QED
Note that the matrix associated to the complex structure F with respect to the
0 Im
basis {v1 , ..., vm , F(v1 ), ...F(vm )} is
.
Im
0
THEOREM 4.5 If N L(V, V) is a nilpotent endomorphism of index p, and
x0 V such that N p1 6= 0, then the vectors x0 , , N (x0 ), ..., N p1 (x0 ) are linearly
independent.
Proof. Let k0 , ..., kp1 K such that
p1
X
ki N i (x0 ) = 0.
i=0
p1
Applying N
to this equality we obtain k0 = 0. Next, apply successively N p2 ,
2
. . . , N , N to obtain k1 = ... = kp1 = 0.
QED
p1
Note that Span {x0 , N (x0 ), ...N
(x0 )} is an invariant subspace of V.
36
THEOREM 4.6 If dim V = n 1, T L(V, V), then there exist two subspaces U
and W of V, invariant with respect to T such that:
(i) V = U W;
(ii) T |U is nilpotent;
(iii) T |W is invertible, when W 6= {0}.
Proof. Denote Nk = Ker (T k ) and Rk = Im (T k ), k N? . Obviously, these are
invariant subspaces with respect to T , and
Nk Nk+1 , Rk Rk+1 , k.
Moreover,
(4.7)
Nk = Nk+1 Nk+1 = Nk+2 and Rk = Rk+1
Let us prove (4.7). Assume Nk = Nk+1 and v Nk+2 .

T k+1 (T v) = 0 T v Nk+1 = Nk T k (T v) = 0 v Nk+1 .
To prove Rk = Rk+1 , use the dimension formula
dim V = dim Nk + dim Rk = dim Nk+1 + dim Rk+1 ,
which implies dim Rk = dim Rk+1 , thus Rk = Rk+1 , as Rk Rk+1 .
Now, since V is finite dimensional, there exists the smallest p N? such that
N1 N2 ... Np = Np+1 = Np+2 = ...
and R1 R2 ... Rp = Rp+1 = Rp+2 = ...
Take U = Np , W = Rp . Using the dimension formula it suffices to prove either
V = U + W or U W = {0} in order to obtain (i). Let v U W. Then T p v = 0
and v = T p w for some w; it follows that w N2p . But N2p = Np , so T p w = 0, which
proves U W = {0}, therefore (i).
Since U is T -invariant, we can consider T |U : U U; by the definition of p and Np
it follows that T p (Np ) = {0} and T p1 (Np ) 6= {0}, so T |U is nilpotent of index p.
For (iii), assume W 6= {0}. If w W and T w = 0, then w = T p u, some u; it
follows that u Np+1 . From Np = Np+1 follows w = 0. This shows that
T |W : W W is injective, hence invertible (as W is finite dimensional).
QED
Examples
1 1
(i) T : R3 R3 defined by the matrix T = 0 1
0 0
2
since T = 0.
1
1 is nilpotent of index 2
1
(ii) Let D : P3 P3 , where P3 is the space of polynomials of degree less or equal

to 3, with real coefficients, D(p) = p0 . Then D is nilpotent of index 4, because
p(4) = 0, p P3 , while D3 6= 0, since the third derivative of polynomials of
degree 3 is not 0.
5. ENDOMORPHISMS OF EUCLIDEAN VECTOR SPACES
37
Endomorphisms of Euclidean Vector Spaces
Let (V, h , i) be a Euclidean K-vector space, K = R or K = C, and T End(V).

When V is finite dimensional it is not hard to prove that there exists an endomorphism T End(V), uniquely determined by T which satisfies
hx, T yi = hT x, yi ,
(5.1)
x, y V .
We will accept without proof that all (possibly infinite dimensional) Euclidean
vector spaces we are working with in this course have the property mentioned above
for the finite dimensional case. Then the following definition makes sense.
DEFINITION 5.1 The endomorphism T defined by (5.1) is called the adjoint of
T . If K = R, then T is also called the transpose of T .
REMARKS 5.2
(i) idV = idV .
(ii) Note that (5.1) is eqiuvalent to
hT x, yi = hx, T yi ,
(5.2)
(iii) (T ) = T ,
x, y V .
T End(V).
DEFINITION 5.3 If T = T , then T is called Hermitian, when K = C,

and symmetric when K = R, or
self adjoint in both cases (either K = C or K = R).
If T = T , then T is called skew Hermitian when K = C,
and skew symmetric (or antisymmetric), when K = R.
If T satisfies
(5.3)
hT x, T yi = hx, yi ,
x, y V
(i.e. T preserves the scalar product),
then T is called unitary, when K = C,

and orthogonal, when K = R.
The notions of Hermitian, symmetric, unitary etc. are used for matrices too. A
matrix A Mn,n (C) is called
symmetric if A = tA
Hermitian if A = tA
skew symmetric if A = tA
skew Hermitian if A = tA
orthogonal if A tA = I and A Mn,n (R)
unitary if A tA = I and A Mn,n (C)
38
We will see later that the definitions given above for endomorphisms and matrices
are related in a natural way.
PROPOSITION 5.4 Let T , S End(V).
(i) (T S) = S T .
(ii) If T is invertible, then T is invertible too; moreover, (T )1 = (T 1 ) .
(iii) (T + S) = T + S .
(iv) If K = C, k C, then (kT ) = kT .
(v) If K = R, k R, then (kT ) = kT .
Proof.
(i)
hx, (T S)yi = hx, T (Sy)i = hT x, Syi = hS (T x), yi = h(S T )x, yi.
The adjoint of T S is unique, thus (T S) = S T .

(ii) We have already noticed that idV = idV . If we take the adjoint of both
sides in T T 1 = idV and T 1 T = idV , and apply (i), we get (T 1 ) T = idV =
T (T 1 ) .
(iii) is immediate by the additivity of the inner product with respect to each
argument.
Now, let us prove (iv).
hx, (kT )yi = hx, k(T y)i = khx, T yi
by conjugate linearity
= khT x, ui = h(kT )x, yi.

Again, apply the uniqueness of the adjoint.
To prove (v) we proceed like in (iv), but use linearity with respect to the second
argument instead of conjugate linearity.
QED
We deduce easily the next corollary for self-adjoint endomorphisms.
COROLLARY 5.5 Let T , S be two self-adjoint endomorphisms of V (K = R or
K = C) and k R. Then
(i) The endomorphism T S is self-adjoint if and only if T S = ST .
(ii) If T is invertible, then T 1 is self-adjoint too.
(iii) kT + S is self-adjoint.
THEOREM 5.6 Assume K = C. Then T is Hermitian if and only if
hx, T xi R ,
x V.
Proof. If T = T , then hT x, xi = hx, T xi, by (5.1). But hx, T xi = hT x, xi, by

Hermitian symmetry of the inner product, so the scalar product hx, T xi = hT x, xi is
real, x V.
Conversely, if hx, T xi R, x V, it follows that
hx, T xi = hx, T xi = hT x, xi = hx, T xi ,
x V.
5. ENDOMORPHISMS OF EUCLIDEAN VECTOR SPACES
39
Then hx, (T T )xi = 0, x V. If we replace x by x + y, C, we obtain

hy, (T T )xi + hx, (T T )yi = 0 ,
x, y V , C .
Now take = 1, then = i to get hx, (T T )yi = 0, x, y V. If in the last

equality x = (T T )y, it follows that T T = 0.
QED
THEOREM 5.7 The endomorphism T preserves the scalar product if and only if it
preserves the norm, i.e.
hT x, T yi = hx, yi ,
x, y V kT xk = kxk ,
x V.
Proof. For y = x we obtain kT xk2 = kxk2 , thus kT xk = kxk.

If K = C, then
hx, yi =
1
kx + yk2 kx yk2 + ikx + iyk2 ikx iyk2 ;
4
If K = R, then
1
kx + yk2 kx yk2 .
4
In the left hand side of each identity replace hx, yi by hT x, T yi, then use kT (x +
cy)k = kx + cyk, c {1, i}, to end up with hx, yi.
QED
hx, yi =
PROPOSITION 5.8 T preserves the inner product if and only if T T = idV .

Proof.
hT x, T yi = hx, yi, x, y V
hT T x, yi = hx, yi, x, y V
h(T T idV )x, yi = 0, x, y V
T T = idV .
QED
REMARKS 5.9
(i) If T preserves the inner product, then T is injective. (This may be deduced
either from Prop.5.8 or Thm.5.7. By Prop.5.8, T admits a left inverse, thus it is
injective. It is an easy exercise for the reader to prove that Ker T = {0} using Thm.
5.7.
(ii) If V is finite dimensional, then T preserves the inner product if and only if
T T = T T = idV .
In some books an unitary endomorphism T is defined as an endomorphism which
satisfies T T = T T = idV . This condition is stronger than the one we used in the
definition here, but they are equivalent in the finite dimensional case. The equivalence
comes up easily from Prop.5.8, if we recall that a linear transformation of finite
dimensional vector spaces is injective if and only if it is bijective.
THEOREM 5.10 Let V be finite dimensional, dim V = n, B = {e1 , . . . , en } an
orthonormal basis of V, and T = [tij ] the matrix of T w.r.t. B.
(I) Assume K = C. Then:
(I.i) T is Hermitian
(I.ii) T is skew Hermitian
(I.iii) T is unitary
T is Hermitian.
T is skew Hermitian.
T is unitary.
40

(II) Assume K = R. Then:
(II.i)
(II.ii)
(II.iii)
T is symmetric
T is skew symmetric
T is orthogonal
T is symmetric.
T is skew symmetric.
T is orthogonal.
Proof. The proofs of (II) are almost the same as the ones for (I). We will only
prove (I.i), leaving the rest to the reader.
n
P
Denote by [tij ] the matrix of T w.r.t. B. Multiplying T ej =
tkj ek by ei in
k=1
the sense of the inner product, we obtain

hT ej , ei i =
(5.4)
tkj ek , ei
= tij .
Similarly, hT ej , ei i = tij . But

hT ej , ei i = hej , T ei i = hT ei , ej i = tji .
(5.5)
If T is Hermitian, i.e. T = T , from (5.4), (5.5) follows tij = tji , i, j, hence

T = tT .
Conversely, assume tij = tji , i, j. then
hx, T xi
X
n
j=1
n
X
j,k=1
n
X
xj ej ,
n
X
x k T ek
k=1
xj xk hT ek , ej i =
n
X
xj xk hej , T ek i
j,k=1
n
X
j,k=1
xj xk tjk =
n
X
xj xk tkj
j,k=1
xk xj tkj = hx, T xi ,
k,j=1
thus hx, T xi R and T is Hermitian by Theorem 5.6.

QED
Note that the theorem is no longer true if we remove the condition on the basis
to be orthonormal, as we see in the next example.
Example
2
Let T :C2 C
such that the matrix of T with respect to the canonical basis
0 i
B is T =
. The canonical basis of C2 is orthonormal, and T = t T , thus
i 2
T is Hermitian by the theorem. Consider the basis B1 = f1 = (1, 0), f2 = (1, 1) ;

2i
i
hf1 , f2 i = 1 6= 0. Denote by T1 the matrix of T w.r.t. B1 . Then T1 =
,
2 + 2i i
which is not a Hermitian matrix.
6. ISOMETRIES
41
Isometries
Throughout this section V denotes a real Euclidean vector space, and d : VV R is

the Euclidean distance. Orthogonal endomorphisms of V preserve the inner product,
therefore they preserve the Euclidian norm and the Euclidean distance. The origin is
a fixed point for any endomorphism. Let us introduce other maps which are distance
preserving, but in general do not fix the origin.
DEFINITION 6.1 Let a V. The map T : V V defined by T (x) = x + a,
x V, is called the translation by the vector a. In order to avoid confusion we
denote sometimes this map by Ta .
Note that in general translations are not linear; the only linear translation is the
one corresponding to a = 0, namely the identity map.
The next properties of translations are straightforward.
THEOREM 6.2 If a, b V and Ta , Tb are the translations by a, and b respectively,
then
1) Ta Tb = Tb Ta = Ta+b ;
2) Ta is bijective, and its inverse is (Ta )1 = Ta .
It follows that composition of functions defines an abelian group structure on the
set of all translations of V, and this group is isomorphic to the additive group (V, +).
PROPOSITION 6.3 Translations are distance preserving, i.e.
d(T (x), T (y)) = d(x, y), x, y V.
Proof.
d(T (x), T (y)) = k(y + a) (x + a)k = ky xk = d(x, y), x, y V. QED
DEFINITION 6.4 An isometry of V is a surjective map F : V V which preserves the Euclidean distance, i.e.
d(F(x), F(y)) = d(x, y), x, y V.
REMARKS 6.5 1) By the properties of distance, it is obvious that any distance
preserving map is injective, therefore isometries are bijective maps.
2) Surjective orthogonal transformations are isometries.
3) Orthogonal transformations of finite dimensional Euclidean vector spaces are
isometries.
4) Translations are isometries.
5) The product (composite map) of two isometries is an isometry. Moreover, the
product of maps defines a group structure on the set of all isometries of the Euclidean
vector space V. The group of translations is a subgroup in this group.
The next theorem is the main step in proving that any isometry is the product of
a translation and an orthogonal transformation.
42
THEOREM 6.6 An isometry F : V V satisfying F(0) = 0 is an orthogonal

transformation.
Proof. It is easy to see that F preserves the norm:
kF(x)k = kF(x) 0k = d(0, F(x)) = d(F(0), F(x)) =
= d(0, x) = kx 0k = kxk, x V.
Using this property, we show next that F preserves the inner product.
d(F(x), F(y)) = d(x, y) kF(y) F(x)k = ky xk
hF(y) F(x), F(y) F(x)i = hy x, y xi
hF(y), F(x)i = hx, yi,
x, y V.
It remains to show that F is linear. We show first that F is homogeneous.

hF(kx), F(y)i = hkx, yi = khx, yi =
= khF(x), F(y)i = hkF(x), F(y)i, x, y V, k R.
Then
hF(kx), F(y)i = hkF(x), F(y)i hF (kx) kF(x), F(y)i = 0, x, y V, k R.
Fix now x V and k R. By the surjectivity of F, there exits y V such that
F(kx) kF(x) = F(y). For this y, using the positivity of the inner product we get
F(kx) kF(x) = 0, thus F is homogeneous.
We proceed similarly to prove the additivity of F. Let x, y V.
hF(x + y), F(z)i = hx + y, zi = hx, zi + hy, zi =
= hF(x), F(z)i + hF(y), F(z)i = hF(x) + F(y), F(z)i.
Then
hF(x + y) F(x) F(y), F(z)i = 0, z V.
Take z such that F(x + y) F(x) F(y) = F(z), then use the positivity of the inner
product to obtain F(x + y) F(x) F(y) = 0, i.e. F additive.
QED
THEOREM 6.7 (characterization of isometries) If F is an isometry, then there
exist a translation T and an orthogonal transformation R such that F = T R.
Proof. Let T be the translation by vector F(0). Then T 1 is the translation by
F(0). The map T 1 F is an isometry which fixes the origin. By the previuos
theorem, T 1 F is an orthogonal transformation. Denote R = T 1 F. Then
F = T R.
QED
Suppose now dim V = n N? and consider an orthonormal basis B = {e1 , . . . , en }.
If R is an orthogonal transformation, {R(e1 ), . . . , R(en )} is an orthonormal basis
too. The matrix R associated to R w.r.t. B is orthogonal, thus (det R)2 = 1. If
det R = +1, R is called a rotation.
If F = T R is an isometry such that R is a rotation, F is said to be a positive
(or orientation preserving) isometry. When det R = 1, F is said to be a negative
(or orientation reversing) isometry.
7. PROBLEMS
43
Problems
1. Let a
6= 0 be a fixed vector in the space of the free vectors V3 , and the map
T : V3 V3 , T (
x) = a
x
.
1) Show that T is a linear transformation.
2) Show that T is neither injective, nor surjective.
3) Find Ker(T ), Im(T ), and show that Ker(T ) Im(T ) = V3 .
2. Let Pn be the complex vector space of the polynomial functions of degree at
most n. Show that the map T : Pn Pn , defined by T p(x) = p(x + 3) p(x),
x C, is a linear transformation. Is T injective?
3. In each of the following cases determine the matrix associated to T with respect
to the canonical bases, the rank, and the nullity of T .
1) T : R3 C3 , T (x) = ix.
2) T : M22 (K) M22 (K), T (A) = t A.
1 i
3) T : C M22 (C), T (x) = x
.
i 1
4.
1)
2)
3)
Let a
6= 0 be a fixed vector in V3 , and the map T : V3 R, T (x) = h
a, x
i.
Show that T is a linear form.
Study the injectivity and the surjectivity T .
Determine Ker(T ), Im(T ) and their dimensions.
5. Which of the following maps is a linear transformation? Determine its kernel

and its image.
1) T : R3 R3 , T (x) = (x1 + x2 , x1 x2 , 2x1 + x2 x3 )
2) T : R3 R3 , T (x) = (x1 + x2 , x1 x2 + 5, 2x1 + x2 x3 )
6. Determine the matrices of the endomorphisms Tj : R3 R3 , j = 1, 2, with
respect to the basis {v1 = (1, 1, 1), v2 = (1, 0, 1), v3 = (0, 1, 1)}, knowing that
0 1 1
1 0 1
1) T1 = 1 1 2 , 2) T2 = 2 1 2
3 2 1
0 1 2
are the matrices of the transformations, with respect to the basis {w1 = (1, 2, 3),
w2 = (3, 1, 2), w3 = (2, 3, 1)}.
7. 1) Show that the endomorphism T : M33 (R) M33 (R), T (A) = t A is a
symmetric involution.
2) Let V be the real Euclidean vector space of C functions f defined on [a,b],
with f (a) = f (b). Show that the differentiation operator D : V V, D(f ) = f 0 is
antisymmetric.
3) Show that the endomorphism T : R2 R2 ,
T (x) = (x1 cos x2 sin , x1 sin + x2 cos ),
is orthogonal.
R,
44

8. The endomorphism T : R3 R3 is defined by
T (x, y, z) = (x y + z, y, y).
1) Show that T is a projection.
2) Determine the dimension and a basis for Ker(T ) and for Im(T ).
9. Let T : C2 C2 be the endomorphism defined by the matrix
6 2i 2i
T =
4 + 8i 0
with respect to the canonical basis of the complex vector space C2 .

Find the Hermitian endomorphisms T1 and T2 such that T = T1 + iT2 , and their
matrices w.r.t. the canonical basis.
Chapter 3
Eigenvectors and Eigenvalues

1
General Properties
Let V be a K-vector space, and T : V V an endomorphism.

The concept of eigenvector is closely related to that of invariant subspace.
DEFINITION 1.1 A vector x V\{0} is an eigenvector of T if there exists K
such that
(1.1)
T x = x.
The scalar is called the eigenvalue associated to the eigenvector x. The set of all
eigenvalues of the endomorphism T is called the spectrum of T and is denoted by
Spec T .
Obviously, an eigenvector can be described as a basis of a one-dimensional T invariant subspace.
Geometrically, an eigenvector is a nonzero vector v, such that v and T v are
collinear.
If dim V = n and v is an eigenvector of T , T v = v, we can extend {v} to a basis
{v1 = v, v2, . . . , vn }. Then the matrix of T with respect to this basis will be of the
A
, for some A M1,n1 (K), B Mn1,n1 (K).
form: T =
0 B
DEFINITION 1.2 Let T be an n n matrix with entries in K. An eigenvector of
T is a column vector X Kn \ {0}, which is an eigenvector for left multiplication by
T, i.e. K such that
(1.2)
T X = X,
As above, is called an eigenvalue (of the matrix T ).
REMARK 1.3 Let V be finite dimensional, B a fixed basis, T the matrix of T
w.r.t. B, x V, K, and X Kn the coordinate column of x with respect to B.
Then
(1.3)
T x = x T X = X.
45
46
CHAPTER 3. EIGENVECTORS AND EIGENVALUES
We saw in Chap.2, Section 3 that similar matrices represent the same endomorphism
(with respect to different bases). By (1.1), (1.2), (1.3) follows that similar matrices
have the same eigenvalues; if B = C 1 AC and X an eigenvector for A, then C 1 X
is an eigenvector for B corresponding to the same eigenvalue.
Sometimes eigenvectors and eigenvalues are called characteristic (or proper) vectors,
and characteristic (or proper) values, respectively.
DEFINITION 1.4 Let K and denote
S() = {x V | T x = x}.
If is an eigenvalue of T , S() is called the eigenspace of .
The following proposition is immediate; this type of result, that describes eigenvalues without involving eigenvectors will prove very helpful.
PROPOSITION 1.5 (i) S() = Ker (T idV ).
(ii) is an eigenvalue of T if and only if Ker (T idV ) 6= {0}.
From (i) follows that S() is a subspace of V, since it is the kernel of a linear
operator.
Examples
(i) Consider V = C (R) and the endomorphism D : V V, D(f ) = f 0 .
Each R is an eigenvalue of D; it is easy to check that the function f ,
f (x) = ex is an eigenvector for . Moreover, for a fixed we can solve the
differential equation f 0 = f whose solution is
S() = {f | f (x) = cex , x R, for some c R} = Span {f }.
In this example it happens that all eigenspaces are one-dimensional.
(ii) V = R3 , T : V V, T (x, y, z) = (4x + 6y, 3x 5y, 3x 6y + z).
We can check that 1 = 2 is an eigenvalue with a corresponding eigenvector
v1 = (1, 1, 3), and 2 = 1 is an eigenvalue with a corresponding eigenvector
v2 = (2, 1, 0), and another one v3 = (0, 0, 1).
At the end of the next section it will be striaghtforward that that -2 and 1 are the
only eigenvalues, and S(2) = {(x, y, z) | x+y = 0, x2y+z = 0} = Span {v1 },
S(1) = {(x, y, z) | x + 2y = 0} = Span {v2 , v3 }; dim S(2) = 1, S(2) is a
straight line, and dim S(1) = 2, S(1) is a plane.
THEOREM 1.6 (i) For an eigenvector of T corresponds a single eigenvalue.
(ii) Eigenvectors corresponding to distinct eigenvalues are linearly independent.
(iii) S() is an invariant subspace of T .
(iv) Eigenspaces corresponding to two distinct eigenvalues are independent.
Proof. (i) Let v V \ {0} such that T v = v and T v = 1 v. Then ( 1 )v = 0,
v 6= 0 imply = 1 .
2. THE CHARACTERISTIC POLYNOMIAL
47
(ii) Let v1 , . . . , vp be eigenvectors, corresponding to the distinct eigenvalues

1 , . . . , p . We proceed by induction on p. For p = 1, (ii) is obvious since v1 6= 0, as
v1 is an eigenvector. Assume (ii) is true for p 1, p 2. Suppose that
(1.4)
k1 v1 + . . . + kp vp = 0.
Then T (k1 v1 + . . . + kp vp ) = 0, and thus

(1.5)
k1 1 v1 + . . . + kp p vp = 0
If we multiply (1.4) by p and subtract from (1.5), we get

k1 (1 p )v1 + . . . + kp1 (p1 p )vp1 = 0
By the inductive hypothesis this yields k1 = . . . = kp1 = 0, since j are distinct.
Replacing in (1.4), follows kp vp = 0, therefore kp = 0.
(iii) We have already noticed that S() is a subspace. If x S(), then T (T x) =
T (x) = (T x). Thus T x S().
(iv) Let 1 , 2 be distinct eigenvalues. We will show that S(1 ) S(2 ) = {0},
i.e. S(1 ) and S(2 ) are independent. Let v S(1 ) S(2 ); T v = 1 v = 2 v implies
(1 2 )v = 0. Since 1 6= 2 , it follows that v = 0.
QED.
The Characteristic Polynomial
Throughout this section, V denotes a finite dimensional K-vector space, K = R or

K = C, dimK V = n, and T : V V is an endomorphism.
Let also A = [aij ] be a n n matrix with entries in K.
LEMMA 2.1 Let T be the matrix of T with respect to an arbitrary basis, and K.
Then is an eigenvalue of T if and only if det (T I) = 0.
Proof. We saw that is an eigenvalue of T if and only if it is an eigenvalue
for T . The existence of an eigenvector of T is equivalent to the fact that the linear
homogeneous system (T I)X = 0 has nontrivial solutions. A necessary and sufficient condition for the existence of nontrivial solutions is that the determinant of the
system is zero.
QED
Now, in order to determine the eigenvalues of T , it suffices to solve a polynomial
equation of degree n. Once the eigenvalues are found, for each of them it remains to
solve a linear system.
Next we will take a closer look to the case of matrices. For the matrix A = [aij ]
Mn,n (K), the matrix equation (A I)X = 0 can be written as a system
(a11 )x1 + a12 x2 + . . . + a1n xn = 0
a21 x1 + (a22 )x2 + . . . + a2n xn = 0

...........................................................
an1 x1 + an2 x2 + . . . + (ann )xn = 0
48
which has nontrivial solutions if and only if
a11
a12
...
a21
a22 . . .
det(A I) =
..
an1
an2
...
= 0.
ann
a1n
a2n
DEFINITION 2.2 The polynomial P () = det(A I) is called the characteristic

polynomial of the matrix A, and the equation P () = det(A I) = 0, K, is
called the characteristic equation of the matrix A. Sometimes we write PA () in order
to avoid confusion.
The eigenvalues of the matrix A are the solutions of its characteristic equation. This
equation has always n complex solutions, counted with their multiplicities (the fundamental theorem of algebra!), but it might have no real ones. Sometimes the eigenvalues
of A are understood to be complex, even if K = R; in this case the corresponding
eigenvectors are in the complexification of Rn , denoted by C Rn ' Cn .
THEOREM 2.3 (i) The characteristic polynomial of A has the form
P () = (1)n [n (tr A)n1 + . . . + (1)n det A]
(ii) Similar matrices have the same characteristic polynomial.
(iii) The matrices A and tA have the same characteristic polynomial.
Proof. (i) We omit the general proof; it can be shown that the coefficient of
k is the sum of principal minors of order k of A, multiplied by (1)k . A direct
computation sorts out the problem for low dimensional cases.
For instance, suppose A is of order 3. Then
a11
a12
a1n
a22
a2n = 3 + 2 (tr A) J + det A,
P () = a21
a31
a32
a33
a11 a12 a11 a13 a22 a23
.
where tr A = a11 + a22 + a33 , J =
+
+
a21 a22 a31 a33 a32 a33
(ii) Assume B = C 1 AC. Then
det (B I) =
=
det (C 1 AC I)
det C 1 det (A I)det C
=
=
det (C 1 (A I)C)
det (A I).
(iii) PA () = det (A I) = det t (A I) = det (t A I) = PtA ().

THEOREM 2.4 The eigenvalues of a complex Hermitian matrix are real. In particular, all eigenvalues of a symmetric real matrix are real.
Proof. Recall that a Hermitian matrix A is equal to the complex conjugate of its
transpose: A = tA, or equvalently A = tA.
Let be an eigenvalue of A, and X an associated eigenvector. By conjugating the
relation
= X.
(?) AX = X,
we find
(??) AX
2. THE CHARACTERISTIC POLYNOMIAL
49
In (?) we left multiply by tX, then take the transpose of both sides to get tX tAX =
tXX. Then ( )
tXX = 0
=
tXX. Left multiplicatin by tX in (??) implies tX AX
t
t
2
since A = A. But X 6= 0 since it is an eigenvector, so XX = kXk 6= 0. Therefore

i.e. R.
= ,
QED.
DEFINITION 2.5 Let T be the matrix of T associated to an arbitrary basis. The
polynomial P () = det(T I) is called the characteristic polynomial of the endomorphism T . When we want to point out the endomorphism in order to avoid
confusion, we will use the notation PT (). The equation P () = 0, K, is called
the characteristic equation of the endomorphism T
Note that for each fixed K, PT () does not depend on the basis we choose; for
any other basis the matrix of (T idV ) is similar to T I, thus it has the same
determinant. This shows that the characteristic polynomial of an endomorphism of a
finite dimensional vector space depends on the endomorphism only, so the definition
makes sense.
We may also define the terms trace and determinant of T to be those obtained
using the matrix of T with respect to an arbitrary basis: tr T = tr T, det T = det T ,
since these are coefficients of PT (), thus independent of the choice of basis (see also
Chap2, Remarks 3.9).
The degree of P () is n, since dim V = n = the order of the matrix (T I). It
follows that T has at most n distinct eigenvalues in K. If this is this is the case, and
we pick one eigenvector for each eigenvalue, we obtain a basis (see Thm 1.6 (ii)). The
associated matrix of T is a diagonal matrix whose diagonal entries are the eigenvalues
of T . In the next section we will see what happens when P () has multiple roots.
Examples
(i) Consider again the endomorphism T : R3 R3 , T (x, y, z) = (4x + 6y, 3x
5y, 3x 6y + z), from the previous section. Our purpose is to determine all eigenvalues and the eigenvectors of T .
4
6 0
The associated matrix with respect to the canonical basis is T = 3 5 0 ,
3 6 1
and the characteristic polynomial
4
6
0
5
0 = ( + 2)( 1)2 .
P () = 3
3
6
1
Then the eigenvalues are the roots of P (), namely -2 and 1. In order to find the
eigenvectors, for each eigenvalue we solve the system (T I)X = 0. Here, for
= 2, we have to solve:
6x + 6y = 0
x+y =0
3x 3y = 0
(T + 2I)X = 0
x 2y + z = 0
3x 6y + 3z = 0
1
x=
y = X = = 1 , R.
z =
50
The eigenspace of = 2 is the solution space of the above system, namely S(2) =
Span {(1, 1, 1)}
For = 1, solve
3x + 6y = 0
3x 6y = 0 x + 2y = 0
(T I)X = 0
3x 6y = 0
0
2
2
+ 0 , , R.
=
X=
1
0
and S(1) = Span {(2, 1, 0), (0, 0, 1)}.

The eigenvectors associated to an eigenvalue are all the nonzero elements of
S (). We accept, however the answer of this problem given in the form:
the eigenvalues are: 1 = 2, with eigenvector v1 = (1, 1, 3); 2 = 1 with
eigenvectors v2 = (2, 1, 0) and v3 = (0, 0, 1). We will understand that for each
eigenspace the answer points out the vectors of a basis.
(ii) Let P3 be the vector space of polynomials with real coefficients, of degree at
most 3, and T : P3 P3 , T (p) = q, where q(X) = p(X + 1). We want to determine
the eigenvalues and the eigenvectors of T .
2
3
First we need to fix a basis of P3 . Let us choose
the canonical basis {1, X, X , X }.
1 1 1 1
0 1 2 3
4
The associated matrix is T =

0 0 1 3 , and P () = ( 1) with the only
0 0 0 1
root = 1, of multiplicity 4. The coordinates of the eigenvectors are the nontrivial
solutions of the system

1
x2 + x3 + x4 = 0
0
2x3 + 3x4 = 0
(T I)X = 0
X =
0 , R.
3x4 = 0
0
The column t [1, 0, 0, 0] represents the constant polynomial 1; S(1) = Span {1} =
{the constant polynomials} ' R.
(iii) Let us determine also the eigenvalues
0 1
A= 1 1
0 1
and the eigenvectors of the matrix
0
1 .
0
From
= (2 2)
P () = 1 1
0
1
0
1
we obtain: 1 = 1, with eigenvector v1 = t [1, 0, 1]; 2 = 0, with eigenvector

v2 = t [1, 1, 1]; 3 = 2, with eigenvector v3 = t [1, 2, 1].
3. THE DIAGONAL FORM
51
The Diagonal Form
V, K and T are as in the previous section; n = dimK V.

Working with endomorphisms of finite dimensinal spaces, usually requires a choice
of basis. It is desirable to choose a basis such that the associated matrix is as simple
as possible. We noticed that when the characteristic polynomial has n distinct roots
in K, there exists a basis made of eigenvectors; the matrix of T with respect to that
basis is diagonal.
DEFINITION 3.1 The endomorphism T is called diagonalizable if there exists a
basis of V such that the matrix of T with respect to this basis is diagonal.
A matrix A Mn,n (K) is diagonalizable if it is similar to a diagonal matrix.
REMARK 3.2 Let T be the matrix of T with respect to an arbitrary basis. Then T
is diagonalizable if and only if T is diagonalizable.
In this section we will give necessary and sufficient conditions for an endomorphism
of a finite dimensional vector space (and for a square matrix) to be diagonalizable.
We call diagonalization of a (diagonalizable) endomorphism T the process of
finding a basis with the property in Definition 3.1; we say that T has a diagonal form
with respect to that basis.
For a matrix A of order n we call diagonalization of A the process of finding
(if possible) an invertible matrix C and a diagonal matrix D of order n such that
D = C 1 AC; the matrix D is called a diagonal form of A and the matrix C is called
the diagonalizing matrix.
PROPOSITION 3.3 The endomorphism T is digonalizable if and only if there
exists a basis made up of eigenvectors of T .
Proof. If T is diagonalizable, then there exists a basis B = {e1 , . . . , en } such that
d1 0 . . . 0
0 d2 . . . 0
T
MB
= D is diagonal. Let D =
.
..
.
0 0 . . . dn
This means that T ej = dj ej j = 1, ..., n i.e. the vectors in B are eigenvectors of
T , and d1 , ..., dn are the associated (not necessarily distinct) eigenvalues.
Conversely, If B1 = {v1 , ..., vn } is a basis of V such that each vj is an eigenvector,
then T vj = j vj , for some j K. It follows that the matrix of T with respect to
B1 is
1 0 . . . 0
0 2 . . . 0
QED
(where the eigenvalues j need not be distinct).

..
.
0
0 . . . n
THEOREM 3.4 The dimension of an eigenspace of the endomorphism T is less
or equal to the multiplicity order of the corresponding eigenvalue, as a root of the
characteristic polynomial.
52
Proof. Let 0 be an eigenvalue of multiplicity order m. Denote dim S(0 ) = p.

If p = n, then S(0 ) = V and T = 0 idV . It follows that P () = (0 )n , thus
m = n > p.
Assume now p < n. Let {v1 , ..., vp } be a basis of S(0 ). We extend this basis
to a basis B of V, B = {v1 , ..., vp , vp+1 , ..., vn }. Since the first p vectors of B are
eigenvectors associated to 0 , we have:
T (vj ) = 0 vj , j = 1, ..., p;
T (vj ) =
n
X
akj ej , j = p + 1, ..., n.
k=1
The matrix of T with respect to
T = 0
...
0
B is
...
..
.
...
...
...
a1p+1
0 app+1
...
...
0 anp+1
...
a1n
,
. . . apn
... ...
. . . ann
thus P () = (0 )p D(), where D() is a determinant of order n p.

On the other hand, the multiplicity of 0 as a root of P () is m, so P () =
( 0 )m Q(), for some polynomial Q() such that 0 is not a root of Q(). From
these two factorizations of P () follows that p m.
QED
THEOREM 3.5 The endomorphism T is diagonalizable if and only if its characteristic polynomial has all its n roots (counted with their multiplicities) in K, and the
dimension of each eigenspace is equal to the multiplicity order of the corresponding
eigenvalue.
Proof. Let 1 , 2 , . . . , p K be the distinct eigenvalues of T , and denote by
m(j ) = mj the multiplicity of j , j = 1, ..., p. Then P () = ( 1 )m1 (
p
P
2 )m2 . . . ( p )mp Q(), and
mj n. Obviously, the condition that P () has
j=1
all roots in K is equivalent to
p
P
j=1
mj = n; if this is the case, Q() 1.
Suppose first that T is diagonalizable, so there is a basis B = {e1 , . . . , en } of V,

made up of eigenvectors.
Denote sj = the number of vectors in B which are eigenvectors for j , and pj =
dim S(j ). The elements of B are linearly independent, thus sj pj . Since card B =
p
P
n, and each eigenvector corresponds to a single eigenvalue, it follows that
sj = n.
j=1
By the previous theorem, pj mj . Now

p
p
P
P
sj pj mj , j = 1, ..., p and
mj n =
sj
force sj = pj = mj , j, and
Conversely, assume that
p
P
j=1
j=1
p
P
j=1
j=1
mj = n.
mj = n and dim S(j ) = mj , j. Consider the
ordered set B1 = {v1 , ..., vn } whose n elements are chosen such that the first m1 form
3. THE DIAGONAL FORM
53
a basis in S(1 ), the next m2 form a basis in S(2 ), and so on, up to the last mp
elements which form a basis in S(p ). Then the elements of B are distinct eigenvectors
(see Thm. 1.6 (iv)). Moreover, using induction on p, one shows that B is a basis of
V. By Prop. 3.3, this means that T is diagonalizable.
QED
COROLLARY 3.6 If T is diagonalizable and 1 , 2 , . . . , p are its distinct eigenvalues, then
V = S(1 ) . . . S(p ).
It is clear now that not all endomorphisms of finite dimensional spaces (and not
all matrices) are diagonalizable.
From the proof of Theorem 3.5 follows that for diagonalizable endomorphisms
(matrices) the diagonal form is unique up to the order of the diagonal entries. The
diagonalizing matrix is not unique either. Moreover, there are infinitely many diagonalizing matrices corresponding to the same diagonal form.
Diagonalization Algorithm
1) Fix a basis of V and determine the matrix T of T with respect to that basis.
2) Determine the eigenvalues of T by solving the characteristic equation, P () = 0.
3) If K = R and there are non-real roots of P (), then we stop with the conclusion
that T is not diagonalizable.
Otherwise, move to step 4).
4) For each eigenvalue j check whether the multiplicity mj is equal to dim S(j ).
For, it suffices to verify if
mj = n rank (T j I), j.
If there exists at least one j such that mj > n rank (T j I), then we stop; T
is not diagonalizable, by Thm. 3.5.
If all equalities hold, point out that T is diagonalizable and go to step 5).
5) Solve the p systems (T j I)X = 0, where p is the number of distinct eigenvalues. For each system chose mj independent solutions, that represent the coordinates
of vectors of a basis in S(j ). Form a basis of V such that the first m1 vectors form
a basis of S(1 ), the next m2 form a basis of S(2 ), and so on.
6) The matrix of T associated to the basis formed in 4) is diagonal; its diagonal
entries are:
1 , . . . 1 ; 2 , . . . , 2 ; p , . . . , p
where each j appears mj times. Let us denote this diagonal matrix by D.
7) The diagonalizing matrix is C whose columns are the solutions of the systems
in 5), i.e. the coordinate-change matrix from the initial basis to the basis formed by
eigenvectors.
Examples
(i) The endomorphism T : R3 R3 , T (x, y, z) = (4x+6y, 3x5y, 3x6y+z).
studied in the previous sections is diagonalizable. Its matrix with respect to the basis
54
2
0 0
{v1 = (1, 1, 1), v2 = (2, 1, 0), v3 = (0, 0, 1)} is D = 0 2 0 . The
0
0 1
1 2 0
1 0 , which satisfies D = C 1 T C.
diagonalizing matrix is C = 1
1
0 1
(ii) The endomorphism T : P3 P3 , T (p) = q, q(X) = p(X + 1) whose eigehenvalues and eigenspaces were determined in Section
nor is its
2 is not diagonalizable,
1 1 1 1
0 1 2 3
matrix with respect to the canonical basis, T =

0 0 1 3 .
0 0 0 1
We saw that = 1 is the only eigenvalue and m(1) = 4, while dim S(1) =
4 rank (T I) = 1.
The Canonical Jordan Form
Let V be a finite dimensional Kvector space, K = R or K = C as before; n =

dimK V, T : V V an endomorphism, T = the matrix of T with respect to an
arbitrary basis, A Mn,n (K).
In this section we will see that even when an endomorphism (or matrix) is not
diagonalizable, but all its eigenvalues are in K, it is still possible to find a canonical
form that is convenient to work with.
DEFINITION 4.1 Let K, k N .
1 0 ...
0 1 ...
..
..
.
.
.
..
0 0 ... ...
The matrix
0 0
0 0
Mk,k (K)
..
.
1
is said to be the Jordan cell (or Jordan block) of order k associated to the scalar .
The Jordan cells of order 1, 2 and 3 respectively
1
1
[],
, 0
0
0 0
are:
0
1 .
DEFINITION 4.2 We say that the endomorphism T is Jordanizable (or admits a

canonical Jordan form) if there exists a basis in V such the associated matrix of T
has the form
J1 0 . . . 0
0 J2 . . . 0
J =
(4.1)
(the canonical Jordan form,)
..
.
0
. . . Js
4. THE CANONICAL JORDAN FORM
55
where each Ji is a Jordan cell associated to some scalar i .; 1 , ..., s are not necessarily distinct.
A matrix J of type (4.1) is said to be in Jordan form or to be a Jordan matrix.
We call the matrix A Jordanizable if it is similar to a Jordan matrix; if J =
C 1 AC, then C is called the Jordanizing matrix of A.
If T is the matrix of T with respect to the basis B, and the matrix of T with
respect to some other basis B1 is the Jordan matrix J, let us denote by C the
coordinate-change matrix from B to B1 .Then J = C 1 T C, so the endomorphism
T is Jordanizable if and only if its matrix T is Jordanizable.
The basis B1 is called a Jordan basis.
We saw that if an endomorphism admits a diagonal form, the corresponding basis
is made up of eigenvectors. Let us take a closer look at the Jordan basis.
Let J be a Jordan matrix, which is the matrix of T , {e1 , ..., en } the correspoding
Jordan basis, and J1 the first cell of J. Assume J1 is of dimension k1 2 and its
diagonal entries are equal to 1 K. Then we observe that
(4.2)
T e1 = 1 e1 , T e2 = e1 + 1 e2 , ..., T ek1 = ek1 1 + 1 ek1 ,
Therefore 1 is an eigenvalue of T with eigenvector e1 . The vectors e2 , ..., ek1 are

called principal vectors associated to the eigenvector e1 .
Similarly, the scalar on the diagonal of each Jordan block is an eigenvalue, and for
each Ji of dimension ki corresponds a sequence of ki basis vectors, such that the first
one is an eigenvector; if ki 2, the other ki 1 are principal vectors.
REMARKS 4.3 (i) The diagonal form is a particular case of the Jordan form,
where the dimension of each cell is 1.
(ii) The Jordan form is unique up to the permutation of blocks.
(iii) The order of the diagonal blocks depends on the order of basis vectors.
(iv) Assume e1 , . . . , ek1 are nonzero vectors satisfying (4.2). Then
(T 1 idV )e1 = 0, (T 1 idV )2 e2 = 0, (T 1 idV )k1 ek1 = 0.
Moreover, ei Ker (T 1 idV )i \ Ker (T 1 idV )i1 , i = 1, . . . , k1 .
Now it takes two more lines to show that e1 , . . . , ek1 are linearly independent.
(v) Let L be an ordered set of distinct t vectors formed by l 2 sequences, each of
length si N? , i = 1, . . . , l such that the first vector in each sequence is an
eigenvector, and the next ones are associated principal vectors; assume also that
the subset of L formed with all eigenvectors in L (which need not correspond
to distinct eigenvalues) is linearly independent.
Then it can be shown by induction on t that L is linearly independent.
Note that if we are able to produce such a set L with t = n, then we deduce
that T is Jordanizable and L is a Jordan basis.
56
Based on the fact that if is an eigenvalue for the matrix T , then m is an

eigenvalue for T m , m N , we deduce the next lemma; the details of its proof are
left as an execise.
LEMMA 4.4 The only complex eigenvalue of a nilpotent matrix (nilpotent endomorphism of a finite dimensional complex vector space) is zero.
THEOREM 4.5 If 1 , ..., p are the distinct eigenvalues of T , with multiplicities
p
P
m(j ) = mj , j = 1, ..., p and
mj = n, then there exist p T -invariant subspaces
j=1
Vj V, j = 1, . . . , p such that:
(i) dim Vj = mj , j = 1, . . . , p
(ii) V = V1 V2 . . . Vp
(iii) T |Vj = N| + | idV , | = , . . . , where N , . . . , N are nilpotent endomorphisms of various orders.
Proof. for each fixed j {1, . . . , p}, consider the endomorphisms Tj = T j idV
and apply Thm. 4.6, Chap.2 to obtain the subspaces Vj and Wj such that V =
Vj Wj , and Tj |Vj is nilpotent and Tj |Wj is invertible. Since Vj is Tj -invariant,
it follows that it is also T = Tj + j idV -invariant.
Let T |Vj End (Vj ) and T |Wj End (Wj ) be the restrictions of calT to V j
and to Wj respectively. From V = Vj Wj follows
det(T idV ) = det(T |Vj idVj )det(T |Wj idWj ), ; as polynomials in .
Then j is an eigenvalue for T |Vj , of multiplicity mj , since T |Wj j idWj is
invertible. On the other hand, j is the only eigenvalue of T |Vj since 0 is the only
eigenvalue of the nilpotent endomorphism Tj |Vj , by the lemma.
It is clear now that the degree of the polynomial det(T |Vj idVj ) is mj , thus
dim Vj = mj . Therefore (i) and (iii) are proved.
p
P
(ii) is immediate by induction on p, using
mj = n.
j=1
We will accept the next theorem without proof, but we note that the missing proof
relies on Thm. 4.5.
THEOREM 4.6 (Jordan) The endomorphism T admits a Jordan form if and only
if its characteristic polynomial has all its n roots (counted with their multiplicities) in
K.
COROLLARY 4.7 Any endomorphism of a finite dimensional complex vector space
(and any complex matrix) admits a Jordan form.
REMARKS 4.8 We would like to point out in more detail the relationship of the
Jordan form J and the decomposition in Theorem 4.5. In the next set of remarks we
keep the notation used in Thm. 4.5. Denote also the Jordan basis corresponding to
J by B1 , and dj = dim S(j ).
57
(i) The number of Jordan cells having j on the diagonal is dj , i.e. the maximal
number of linearly independent eigenvectors.
(ii) mj = dim Vj = the sum of dimensions of Jordan cells corresponding to j .
(iii) Assuming that the Jordan blocks of J are ordered such that the first d1 correspond to 1 , the next d2 correspond to 2 , and so on, it follows that the first
d1 vectors of B1 form a basis of V1 , the next d2 form a basis of V2 , and so on,
up to the last dp vectors of B1 which form a basis of Vp .
In practice, if all multiplicities are small enough we may use the following algorithm.
Algorithm for Finding the Jordan Form and the Jordan Basis
1) Find the matrix T of the endomorphism with respect to an arbitrary (fixed)
basis.
2) Solve the characteristic equation. If this equation has all n roots in K, then
the endomorphism is Jordanizable, otherwise it is not.
3) Compute dim S(j ) = n rank (T j I) mj . We have already noticed that
the number of Jordan blocks corresponding to the eigenvalue j is equal to dim S(j ).
Sometimes, for small values of mj this fact allows us to figure out the Jordan blocks
corresponding to j (see the Remark below).
If dim S(j ) = mj , then there are mj Jordan cells of dimension 1, corresponding
to j .
4) For each eigenvalue j determine the eigenspace S(j ), by solving the linear
homogeneous system
(Syst.1)
(T j I)X = 0.
Denote the general solution of (Syst.1) by X1 .

5) If dim S(j ) < mj , some of the eigenvectors admit principal vectors. We impose
the compatibility conditions on X1 and require X1 6= 0, to solve the system
(Syst.2)
(T j I)X = X1 .
Denote general solution of (Syst.2) by X2 . We continue to determine the sequence of

principal vectors (actually we determine their coordinates with respect to the basis
fixed in 1)) by trying to solve
(Syst.3)
(T j I)X = X2 .
This process stops when the compatibility of

(Syst.k)
(T j I)X = Xk
leads to a contradiction. The number k = k(j ) found in this way represents the
dimension of the largest Jordan cell having j on the diagonal.
6) Pick particular values for the parameters which appear in X1 , . . . , Xk to obtain
the basis vectors corresponding to the part of the Jordan matrix that has j on
the diagonal. These particular values must be chosen such that the compatibility
conditions are all satisfied, and the eigenvectors are linearly independent.
58
REMARK 4.9 Suppose j = is an eigenvalue with m() = m. Sometimes it

is possible to find out the part of the Jordan matrix with diagonal entries using
dim S() only, without determining the corresponding part of the basis. For m small,
we list below all possible combinations of Jordan blocks with on the diagonal.
m=1
[]
0
0
0
1
if dim S() = 1;
0
0
if dim S() = 3.
0
0
0
1
0
0
0
0
0
1
0
0
0
0
0
0
0
0
if dim S() = 3;
0
0
0
m=4
m=2
m=3
( the only possibility) .
0
0
0
0
0
0
if dim S() = 1;
if dim S() = 2.
0
0
0
0
if dim S() = 2;
if dim S() = 1;
or
0
0
0
0
0
0
0
0
0
0
0
0
if dim S() = 2;
0
0
0
0
if dim S() = 4.
0
Note that dim S() distinguishes all above cases except for m = 4, dim S() = 2.
Examples (i) Find the Jordan form and the Jordan basis for the endomorphism
T : R4 R4 ,
T (x) = (3x1 x2 x3 2x4 , x1 + x2 x3 x4 , x1 x4 , x2 + x3 + x4 ).
We apply the algorithm.
1) The matrix of T with respect to the canonical basis is
3 1 1 2
1
1 1 1
.
T =
1
0
0 1
0 1
1
1
59
2) P () = ( 1)3 ( 2) with roots: 1 = 1, 2 = 2; m(1) = 3, m(2) = 1. All

roots are in R (m(1) + m(2) = dim R4 ), thus T admits a Jordan form.
3) dim S(1) = 4 rank (T I) = 2 < m(1), so there are 2 Jordan cells with 1
on the diagonal (see the remark above). For the eigenvalue 2 = 2, the multiplicity
m(2) = 1 forces dim S(2) = 1 (Why?). It follows that T admits a canonical Jordan
form
1 1
0 1
.
J =
1
2
This is the matrix of T with respect to a Jordan basis B = {e1 , e2 , e3 , e4 }, where
e1 , e3 are linearly independent eigenvectors in S(1), e2 is a principal vector for e1 ,
and e4 is an eigenvector in S(2).
We will apply the rest of the algorithm for 1 , then for 2 .
For 1 = 1:
4) Solve
(Syst.1)
(T I)X = 0;
X1 =t [ ],
, R.
e1 and e3 are of the form X1 with and chosen for each of them, such that e1 , e3
are linearly independent, and e1 admits a principal vector.
5) The system
(Syst.2)
(T I)X =t [ ]
requires the compatibility condition = 0, so e1 = (, 0, 0, ), 6= 0.

For = , the general solution of (Syst.2) is
X2 =t [ + + + ],
, R.
We know from 3) that there is no need to look for more principal vectors.
6) For e1 , choose = 1 = in X1 ; e1 = (1, 0, 0, 1). For e2 , = 1 (like in e1 ),
and pick = 1, = 0, which give e2 = (0, 1, 0, 0).
If in X1 we take = 1, = 0 for e3 , we obtain e3 = (1, 1, 1, 0).
For 2 = 2:
4) The system (T 2I)X = 0 has the general solution t [2a a a 0] a R.
Any nontrivial solution does the job for e4 ; let us take a = 1. Then e4 = (2, 1, 1, 0).
The problem is now completely solved. The Jordanizing matrix is
1
0 1 2
0 1 1 1
.
C=
0
0 1 1
1
0 0 0
We may check that J = C 1 T C, or equivalently, CJ = T C.
Note that we determined the Jordan form J immediately after we found out
dim S(1); then it was clear that an eighenvector associated to the eigenvalue 1 = 1
60
admits either one principal vector, or none, so we stoped looking for principal vectors
after solving (Syst.2). Let us see what would happen if we tried to determine one
more principal vector, namely to solve
(T I) =t [ + + + ].
The compatibility condition for this system is = 0, that contradicts e1 6= 0.
(ii) Determine the Jordan form and the Jordanizing matrix of
4 1 1 1
1
2
0 1
M4,4 (R).
A=
0
0
3 0
0
0
1 3
We find P () = ( 3)4 , 1 = 3, m(3) = 4 = dim R4 , thus A is Jordanizable.
dim S(3) = 4rank (A3I) = 2, thus the Jordan form of A is one of the following:
3 1 0 0
3 1 0 0
0 3 0 0
0 3 1 0
J =
0 0 3 1 or J = 0 0 3 0 .
0 0 0 3
0 0 0 3
We will find out which of J or J is the correct one, while looking for the Jordan
basis. The system
(Syst.1)
(T I)X = 0
has the general solution X1 =t [

The next system,
(Syst.2)
+ ],
(T 3I)X =t [
, R.
+ ]
is compatible for , R; its general solution is X2 =t [a b

Following the algorithm, we try to solve the system:
(Syst.3)
(T 3I)X =t [a b
a + b].
a + b],
which is compatible if and only if = = 0. But this contradicts e1 6= 0.

Therefore no eigenvector admits sequences of 2 principal vectors.
Each eigenvector e1 , e3 of the Jordan
Consequently, the Jordan form is J = J.
basis {e1 , e2 , e3 , e4 } is followed by one principal vector.
We may pick any linearly independent e1 , e3 S(3), and for each of them we may
take any values of a and b to obtain e2 and e4 . For = 1, = 0, a = b = 0, we
get e1 = (1, 0, 0, 1), e2 = (0, 0, 1, 0), and for = 0, = 1, a = b = 0, we get
e3 = (0, 1, 0, 1), e4 = (0, 0, 1, 1).
1
0 0 0
0
0 1 0
The Jordanizing matrix C =

0 1 0 1 and the Jordan form
1
0 1 1
3 1 0 0
0 3 0 0
1
J =
0 0 3 1 will certainly satisfy J = C AC.
0 0 0 3
5. THE SPECTRUM OF ENDOMORPHISMS ON EUCLIDEAN SPACES
61
The Spectrum of Endomorphisms on Euclidean

Spaces
V is an Euclidean K-vector space, K = R or K = C, T End(V). Recall that the

spectrum of T is
Spec(T ) = { K | v 6= 0, T v = v} = the set of all eigenvalues .
REMARK 5.1 If is an eigenvalue of T and v a corresponding eigenvector, then
=
hT v, vi
.
hv, vi
THEOREM 5.2 If K = C and T is Hermitian, then

(i) Spec(T ) R
(ii) Eigenvectors corresponding to distinct eigenvalues are orthogonal.
(iii) If dimV = n N , then there exists an orthonormal basis of V, made up of
eigenvectors. Thus a Hermitian endomorphism is always diagonalizable.
Proof. hx, T yi = hT x, yi, x, y V, by hypothesis.
(i) Assume is an eigenvalue and v a corresponding eigenvector. Then
=
hT v, vi
hT v, vi
hv, T vi
=
=
= .
hv, vi
hv, vi
hv, vi
Thus R.
(ii) Let 1 6= 2 be eigenvalues with corresponding eigenvectors v1 and v2 respectively. From
1 hv1 , v2 i = hT v1 , v2 i = hv1 , T v2 i = 2 hv1 , v2 i = 2 hv1 , v2 i,
follows (1 2 )hv1 , v2 i = 0, i.e. hv1 , v2 i = 0, as 1 6= 2.
(iii) will be proved by induction on n. For n = 1 it is trivially true. Let n 2
and assume (iii) is true for vector spaces of dimension n 1. From (i) follows that
T has at least one real eigenvalue 1 . Let v1 be an eigenvector associated to 1 and
denote U = {v1 } . We know from Chapter 1 that U is a subspace of V; moreover,
V = Span {v1 } U , thus dim U = n 1.
In order to apply the inductive hypothesis for U we need to show that U is T invariant. For, let x U ; then hT x, v1 i = hx, T v1 i = 1 hx, v1 i, thus T x U.
The restriction T |U : U U is Hermitian too, so by the inductive hypothesis
there exists an orthonormal basis {u2 , . . . , un } of U , made up of eigenvectors. Then
1
v1 , u2 , . . . , un } is the required basis of V.
QED.
{
||v1 ||
Translating the theorem in matrix form we obtain
COROLLARY 5.3 Let A Mn,n (C), A Hermitian. Then
62
(i) A is diagonalizable.
(ii) If D is a diagonal form of A, then D Mn,n (R).
(iii) There exists a diagonalizing unitary matrix (in Mn,n (C).)
REMARK 5.4 Let A Mn (R), A symmetric. Since real symmetric matrices are
particular cases of Hermitian matrices, the previous corollary applies, thus all eigenvalues of A are real and A is diagonalizable; there exists a diagonalizing orthogonal
matrix C Mn,n (R). In particular. Theorem 2.4 comes out as corollary of (i).
COROLLARY 5.5 If K = R and T is symmetric, then
(i) Assuming V finite dimensional, all roots of the characteristic polynomial of T
are real.
(iii) If dim(V) = n N , then there exists an orthonormal basis of V, made up of
eigenvectors. Thus a symmetric endomorphism is always diagonalizable.
Similarly we can deduce some properties of eigenvalues and eigenvectors of skew
Hermitian and skew symmetric endomorphisms.
REMARK 5.6 (i) The eigenvalues of a skew Hermitian endomorphism are purely
imaginary or zero. Assuming V finite dimensional over R, it follows that all
roots of the characteristic polynomial are purely imaginary or zero.
(ii) Eigenvectors corresponding to distinct eigenvalues of a skew Hermitian endomorphism are orthogonal.
(iii) Note that skew symmetric endomorphisms are not necessarily diagonalizable.
THEOREM 5.7 Let V be a complex (real) Euclidian space and T End(V) an
unitary (orthogonal) endomorphism.
(i) If there exist any eigenvalues, their absolute value is 1.
(iii) If K = C, V is finite dimensional, and T is unitary, then T is diagonalizable
and V admits an orthonormal basis made up of eigenvalues.
(iv) If K = R, V is finite dimensional, T is orthogonal and the characteristic
polynomial has no other roots in C besides 1 or -1, then T is diagonalizable and
V admits an orthonormal basis made up of eigenvectors.
6. POLYNOMIALS AND SERIES OF MATRICES
63
Polynomials and Series of Matrices
V will denote a Kvector space, dim V = n N? , and A : V V an endomorphism.

The matrix of A w.r.t. a fixed basis of V is A = [aij ] Mn,n (K).
DEFINITION 6.1 Consider the polynomial Q(t) K[t], Q(t) = am tm +am1 tm1 +
. . . + a1 t + a0 . The matrix
Q(A) = am Am + am1 Am1 + . . . + a1 A + a0 I
is called the value of Q for the matrix A. The endomorphism
Q(A) = am Am + am1 Am1 + . . . + a1 A + a0 idV
is called the value of Q for A.
Q(A) is also called polynomial of matrix or matrix polynomial; Q(A) is called
polynomial of endomorphism.
Since endomorphisms (of finitely dimensional vector spaces) are completely determined by their associated matrices, it suffices to study matrix polynomials. All
results about matrices may be reformulated in terms of endomorphisms.
REMARK 6.2 If the matrix A admits the canonical Jordan form J, then the following observations make the computation of Q(A) easier.
A = CJC 1 ,
A2 = CJ 2 C 1 ,
. . . , Am = CJ m C 1 .
If A admits the diagonal form, i.e. J = D is a diagonal matrix, then the diagonal
form D of A is even more convenient to use for computing powers of A.
A = CDC1,
A2 = CD2 C 1 ,
. . . , Am = CDm C 1 ,
since D = diag(d1 , . . . , dn ) Dk = diag(dk1 , . . . , dkn ), k N.

THEOREM 6.3 (CayleyHamilton) If P () is the characteristic polynomial of
the matrix A, then P (A) = 0
Proof. Recall that for any matrix M = mij Mn,n (K), there exists a matrix
M + , called the reciprocal of M such that
M M + = (det M )I.
(Some books use the notation M instead of M + and call it the adjoint of M instead
of the reciprocal; here reciprocal and adjoint have different meanings.)
The (i, j) entry of M + is the algebraic complement of mji in M . For M = A I,
we obtain
(6.1)
(A I)(A I)+ = P ()I,
since P () = det(A I). Moreover, by the construction of the reciprocal matrix
follows that each entry of (A I)+ is a polynomial in of degree at most n 1;
consequently, we can express
(6.2)
(A I)+ = Bn1 n1 + Bn2 n2 + . . . + B0 ,
64
where Bi Mn,n (K), i = 0, . . . , n 1. Let

(6.3)
P () =
n
X
ak k , ak K.
k=0
Replacing (6.2), (6.3) in (6.1) we obtain

(AI)(Bn1 n1 +B n2 n2 +. . .+B1 +B0 ) = (an1 n1 +an2 n2 +. . .+a1 +a0 )I
or
(Bn1 )n +(ABn1 Bn2 )n1 +. . .+(AB1 B0 )+AB0 = (an I)n +. . .+(a1 I)+a0 I.
Now we identify the coefficients
Bn1 = an I, ABn1 Bn2 = an1 I, . . . , AB1 B0 = a1 I, AB0 = a0 I.
Multiplying on the left by An , An1 , . . ., A, I and adding up each side we get
P (A) =
=
an An + an1 An1 + . . . + a1 A + a0 I
An Bn1 +
n
+A Bn1 An1 Bn2 +
+An1 Bn2 An2 Bn3 +
.
+..................+
2
+A B1 AB0 +
+AB0
=
0
QED.
COROLLARY 6.4 If dim V = n N? , A : V V is an endomorphism, and

P () the characteristic polynomial of A, then P (A) = 0.
THEOREM 6.5 If A Mn,n (K), then the value at A of any polynomial of degree
at least n can be expressed as a matrix polynomial in A, of degree at most n 1.
Proof. Let P () be the characteristic polynomial of A. By the Cayley-Hamilton
Theorem, P (A) = 0. Since the degree of P is n, the coefficient of An in P (A) is not
zero, thus we can express
An = bn1 An1 + . . . + b0 I
This proves the theorem in the case of polynomials of degree n. If we multiply the
equality by A and replace An , we get An+1 as a linear combination of An1 , . . . , A, I.
Inductively, the same holds for An+p , p N.
QED.
P
Assume now K = C or R and consider the power series f (t) =
am tm with
m
coefficients in K. Such a series makes sense for t V, when V is endowed with

multiplication, such that the powers tm are well defined. If V is a Euclidean vector
space, we consider the Euclidean metric on V, and we can discuss about the conver2
gence of the series. In particular, we are interested in V = Mn,n (K) ' Kn with the
Euclidean norm
s X
kAk =
|aij |.
1i,jn
6. POLYNOMIALS AND SERIES OF MATRICES
65
DEFINITION 6.6 Let A : V V be an endomorphism, dim V = n N? and A

the associated basis w.r.t. a fixed basis. We define a series of matrices and a series
of endomorphisms by
X
X
am Am , and
am Am respectively.
m
If the series are convergent, their sums are denoted by f (A), and fP
(A) respectively,
where f is the function defined by the convergent series f (t) =
am tm , t K.
m
f (A) is called function of matrix or matrix function, and f (A) is called function of
endomorphism.
On finite dimensional vector spaces, the study of series of endomorphisms reduces
to the studyPof matrix series. On the other hand, as a consequence of theorem Cayley
Hamilton,
am Am can be expressed as a matrix polynomial Q(A), of degree n 1
m
P
whose coefficients are numerical series. If am Am is convergent, then the coefficients
of Q(A) are convergent numerical series.
If A admits
eigenvalues 1 , . . . , n , then the polynomial of degree n 1
Pdistinct
am Am can be written in the Lagrange form
associated to
m
f (A) =
n
X
(A 1 I) . . . (A j1 I)(A j+1 I) . . . (A n I)
j=1
(j 1 ) . . . (j j1 )(j j+1 ) . . . (j n )
or
f (A) =
n
X
f (j ),
Zj f (j ),
j=1
where Zj Mn,n (K) are matrices independent of the function f .

If A has multiple eigenvalues, a more general formula works
f (A) =
p mX
k 1
X
Zkj f (j) (k ),
k=1 j=0
where f (j) denotes the derivative of order j of the funtion f , and Zkj are matrices
independent of f .
The following series are convergent for any matrix A.
eA =
X
1 m
A ,
m!
m=0
sin A =
(1)m
A2m+1 ,
(2m
+
1)!
m=0
cos A =
X
(1)m 2m
A .
(2m)!
m=0
We will also consider the series

etA =
X
tm m
A ,
m!
m=0
where t K.
This series is very useful in the theory of linear differential with constant coefficients.
66
Problems
1. Determine the eigenvalues and the eigenvectors of the endomorphisms in problems 4 - 7, Chapter 2.
2. Let V be the real Euclidean vector space of the continuous real functions on
[0, 2]. Let T : V V be the endomorphism defined by
g = T (f ), g(x) =
Z2
[1 + cos(x t)] f (t)dt, x [0, 2].
0
1) Show that the subspace Im(T ) is finite dimensional and find an orthogonal
basis of this subspace of V.
2) Determine Ker(T ).
3) Show that T is symmetric; find its eigenvalues and eigenvectors.
3. Determine the diagonal form and the
7 4 1
1) A : R3 R3 , A = 4 7 1
4 4 4
1 0 0
1
0
1
0
0
3) A : R4 R4 , A =
0 0 1 2
1 0 2 5
corresponding basis, then compute
4
6
2)A : R3 R3 , A = 3 5
3 6
eA .
0
0
1
4. Determine the canonical forms (diagonal or Jordan) of the following endomorphisms, and the corresponding bases.
1 0 3
6 6 15
3
3
3
3
1) A : R R , A = 3 2 3 2) A : R R , A = 1 5 5
3 0 1
1 2 2
0 1 0
2 1
3) A : R2 R2 , A =
4) A : R3 R3 , A = 0 0 1
1 4
2 5 4
2
0 0 0
1
3 1 1
4
4
5) A : R R , A =
0
0 0 1
1 1 0 2
5. For each matrix A in the previous problem compute An , eA , sin A, cos A using
the Cayley-Hamilton theorem.
6. Use the Cayley - Hamilton theorem to determine A1 and the value of the
matrix polynomial Q(A) = A4 + 3A3 9A2 28A, where
1 2
2
A = 2 1 2 .
2
2 1
Chapter 4
Bilinear Forms. Quadratic

Forms
1
Bilinear Forms
Let V be K-vector space.

The basic example for bilinear forms is the canonical inner product on Rn . On the
other hand, the notion of bilinear form extends the one of linear form. Recall that a
linear form on V is a linear map : V K.
DEFINITION 1.1 A map A : V V K is called a bilinear form on V if it is
linear in each argument i.e.
A(kx + ly, z) = kA(x, z) + lA(y, z),
A(x, ky + lz) = kA(x, y) + lA(x, z),
x, y, z V, k, l K.
Example Any inner product on a real vector space is a bilinear form.

Counterexample An inner product on a complex vector space is not a bilinear form,
since it is not C-linear in the second argument.
Denote by B(V, K) the set of all bilinear forms on V. Addition and multiplication
of bilinear forms are defined in the usual way for K-valued functions. These operations
define a K-vector space structure on B(V, K).
REMARK 1.2 Assume that V is finite dimensional, dim V = n N? , and let
n
n
P
P
B = {e1 , . . . , en } be a basis of V. Let x, y V, x =
xi ei , y =
yj ej . If
i=1
A B(V, K), then

n
n
n X
n
X
X
X
A(x, y) = A
xi ei ,
yj ej =
xi yj A(ei , ej ),
i=1
j=1
i=1 j=1
which shows that A is uniquely determined by its values on B B.

67
j=1
68
CHAPTER 4. BILINEAR FORMS. QUADRATIC FORMS
Using the notation aij = A(ei , ej ), A = [aij ] Mn,n (K), we may describe the
bilinear form A by the one of the following equalities
A(x, y) =
n X
n
X
aij xi yj
(the analytic expression of A)
i=1 j=1
or
A(x, y) = t XAY
(the matrix expression of A),
where X, Y represent as usually the coordinate columns of x and y respectively.

A is called the matrix associated to A w.r.t. the basis B. It is straightforward
that the correspondence between a bilinear form and its associated matrix w.r.t.
a fixed basis defines an isomophism from B(V, K) onto Mn,n (K). It follows that
dim B(V, K) = n2 .
Example Consider the cannonical inner product on Rn ,
hx, yi = x1 y1 + . . . + xn yn .
If B is the canonical basis, then the associated matrix is the n n identity matrix In ,
since hei , ej i = ij .
DEFINITION 1.3 The bilinear form A is said to be symmetric if
A(x, y) = A(y, x), x, y V.
The bilinear form A is said to be skew symmetric if
A(x, y) = A(y, x), x, y V.
From the definition of the inner product on a real vector space follows that inner
products on real vector spaces are symmetric bilinear forms.
THEOREM 1.4 Let A : V V K be a bilinear form and dim V = n N? .
Then A is symmetric (skew symmetric) if and only if the associated matrix w.r.t. a
fixed basis is symmetric (skew symmetric).
Proof. Fix a basis B = {e1 , . . . , en } in V. Denote A = [aij ] the associated matrix.
Assume A symmetric. Then aij = A(ei , ej ) = A(ej , ei ) = aji , i, j = 1, . . . , n,
thus A = t A.
Conversely, assume A = t A. Then
A(y, x) = t Y AX = t ( t Y AX) = t X t AY = t XAY = A(x, y), x, y V.
The skew symmetric case is similar.
QED.
THEOREM 1.5 Let A be a bilinear form on V, dim V = n N? , B and B0 two

bases in V with C Mn,n (K)-the matrix of change from B to B0 .
If A and B Mn,n (K) are the matrices associated to A w.r.t. B and B0 respectively, then
B = t CAC.
1. BILINEAR FORMS
69
Proof. Let x, y V arbitrary, and X, Y the coordinate columns of x, respectively y

w.r.t. B. Denote by X 0 , Y 0 the coordinate columns of the same x, y w.r.t. B0 . The
coordinate columns are related by X = CX 0 , Y = CY 0 . Starting with the matrix
form of A w.r.t. B we obtain
A(x, y) = t XAY = t (CX 0 )A(CY 0 ) = t (X 0 )( t CAC)(Y 0 ).
On the other hand, the matrix form of A w.r.t. B0 is A(x, y) = t (X 0 )B(Y 0 ), and
B is uniquely determined by this equality; therefore B = t CAC.
QED.
DEFINITION 1.6 If A is nonsingular (singular), then the bilinear form A is called
nondegenerate (degenerate). The rank of A is also called the rank of the bilinear form
A.
The rank of the associated matrix is the same regardless the choice of basis (see
Corollary 1.8), so the above definition makes sense.
DEFINITION 1.7 Let A : V V K be a symmetric bilinear form. The set
Ker A = {x V| A(x, y) = 0, y V}
is called the kernel of the form A.
PROPOSITION 1.8 Ker A is a vector subspace of V.
Proof. Let u, v Ker A i.e. A(u, y) = 0, A(v, y) = 0 y V. If k, l K, then
A(ku + lv, y) = kA(u, y) + lA(v, y) = 0,
thus ku + lv Ker A.
QED.
THEOREM 1.9 If A : V V K is a symmetric bilinear form, and dim V =

n N? , then
rank A = n dim KerA.
Proof. Let {e1 , . . . , en } be a basis of V and A = [aij ], aij K the matrix associn
n
P
P
ated to A w.r.t. this basis. If x, y V are arbitrary, x =
xi ei , y =
yj ej ,
i=1
A(x, y) =
n X
n
X
i=1 j=1
aij xi yj =
n X
n
X
j=1
j=1
aij xi yj
i=1
which shows that x Ker A if and only if x is a solution of the linear homogeneous
system
n
X
aij xi = 0 j = 1, . . . , n.
i=1
So Ker A is isomorphic to the solution space of this system, whose dimension is

n rank A = n rank A.
QED.
COROLLARY 1.10 (i) rank A is independent of the choice of basis; it depends
only on the symmetric bilinear form A.
(ii) A is nondegenerate Ker A = {0}.
70
Quadratic Forms
Throughout the section V denotes a K-vector space and A a symmetric bilinear form
on V.
DEFINITION 2.1 A map Q : V K is called a quadratic form on V if there
exists a symmetric bilinear form A such that Q(x) = A(x, x), x V.
REMARK 2.2 In the definition of the quadratic form, Q and A are uniquely determined by each other. Obviously Q is determined by A. For the converse, the
following computation gives a formula for A in terms of Q.
Q(x + y) = A(x + y, x + y) = A(x, x) + A(y, y) + A(x, y) + A(y, x).
From the symmetry of A,
Q(x + y) = A(x, x) + A(y, y) + 2A(x, y), therefore
1
(Q(x + y) Q(x) Q(y)).
2
The symmetric bilinear form A associated to Q is called the polar of Q.
A(x, y) =
Example The quadratic form corresponding to the inner product on a real vector
space is the square of the Euclidean norm
Q(x) = hx, xi = kxk2 , x V.
Assume dim V = n N? ; let {e1 , . . . , en } be a basis of V and A = [aij ] the matrix

n
P
of A w.r.t. this basis. For any x V, x =
xi ei ,
i=1
Q(x) = A(x, x) =
n X
n
X
aij xi xj = t XAX,
i=1 j=1
where X = t [x1 , . . . , xn ]. It follows that the matrix A characterizes Q as well as

A. The symmetric matrix A is also called the matrix of the quadratic form Q; by
definition, the rank of Q is the rank of the matrix A, so rank Q = rank A = rank A.
DEFINITION 2.3 Let A : V V K be a symmetric bilinear form and Q the
associated quadratic form. The vectors x, y V are said to be orthogonal w.r.t. A
(or Q) if A(x, y) = 0.
DEFINITION 2.4 Let U V, a vector subspace of V. The vector space
U = {y V | A(x, y) = 0, x V}
is called the orthogonal complement of U in V w.r.t. A.
2. QUADRATIC FORMS
71
THEOREM 2.5 Let A : V V K be a symmetric bilinear form. Denote by A|U

the restriction of A to U U.
(i) U is a vector subspace of V.
(ii) If {u1 , u2 , . . . , up } is a basis in U, then
(2.1)
y U A(u1 , y) = A(u2 , y) = . . . A(up , y) = 0.
(iii) If dim V = n N? , then dim U + dim U dim V; equality holds if and only
if A|U is nondegenerate.
(iv) If dim V = n N? , then
U U = V A|U is nondegenerate.
Proof. (i) Let y1 , y2 U , i.e. A(x, y1 ) = 0, A(x, y2 ) = 0, x U. For any
k, l K
A(x, ky1 + ly2 ) = kA(x, y1 ) + lA(x, y2 ) = 0,
thus ky1 + ly2 U .
(ii) y U implies A(x, y) = 0, x U; in particular A(ui ) = 0, i = 1, . . . , p,
n
P
since ui U. Conversely, if x U, then x =
xi ui , using the given basis of U.
i=1
Then
A(x, y) =
n
X
xi A(ui , y) = 0, i.e. y U .
i=1
(iii) If V is finite dimensional, so is U. Fix a basis of U using the notation of (ii),

and complete it to a basis of V; in (2.1) write y as a linear combination of this basis.
Then the rank of the system in (2.1) is at most p. Consequently, dim U n p, so
dim U + dim U = n. Equality holds when the rank of the system is exactly p, i.e.
dim U = rank A|U ; but this means that A|U is nondegenerate (see the last part of
Section 1).
(iv) As a consequence of (iii), it suffices to show that
U U = {0} A|U is nondegenerate.
But this is obvious by Corollary 1.10.
QED.
Example Note that if a symmetric bilinear form A (or the quadratic form Q) is
nondegenerate on the space V , it might happen that its restriction to some subspace of V is degenerate. For example, Q(x) = x21 x22 + x23 is nondegenerate on
R3 ; however, the restriction of Q to U = {x R3 | x1 + x2 = 0} is degenerate,
as 0 6= x = (1, 1, 0) Ker(A|U ). Indeed, for A(x, y) = x1 y1 x2 y2 + x3 y3 ,
A((1, 1, 0), (a, a, b)) = 0, a, b R.
72
DEFINITION 2.6 Let A : V V K be a symmetric bilinear form. A basis

{e1 , . . . , en } is called an orthogonal basis w.r.t. the form A (or w.r.t. the quadratic
form Q) if A(ei , ej ) = 0 for i =
6 j, i, j = 1, . . . , n, i.e. the vectors of this basis are
pairwise orthogonal. The associated matrix is a diagonal matrix. If we denote its
diagonal entries by ai = aii = A(ei , ei ), i = 1, . . . , n, then the corresponding analytic
expressions of A and Q
A(x, y) =
n
X
ai x i yi ,
Q(x) =
i=1
n
X
ai x2i
i=1
are called canonical expressions of A and Q respectively.
Reduction of Quadratic Forms to

Canonical Expression
Let V be a K-vector space, K = R or K = C, dim V = n N? and Q(x) = t XAX,

a quadratic form on V, expressed by the associated matrix A w.r.t. a fixed basis of
V. A change of basis corresponds to X = CX 0 , where C is the matrix of change. The
matrix associated w.r.t. the new basis is B = t CAC. Note that both A and B are
symmetric matrices.
THEOREM 3.1 (Gauss method) If Q : V R is a quadratic form, then there
exists a basis of V which is orthogonal w.r.t. Q (the associated analytic expression is
canonical).
Proof. We will describe an inductive algorithm which reduces the problem to the
case of a quadratic form defined on a vector space of smaller dimension.
Denote by {e1 , . . . , en } the initial basis and by A = [aij ] the associated matrix.
n P
n
P
Suppose that the analytic expression Q(x) =
aij xi xj is not canonical and Q is
i=1 j=1
not identically zero (otherwise the problem is solved and nothing needs to be done).
Case 1. There exists i such that aii 6= 0.
We may assume without loss of generality that i = 1, so we can write
Q(x) = a11 x21 + 2
n
X
j=2
By factoring out
we obtain
a1j x1 xj +
n X
n
X
aij xi xj .
i=2 j=2
1
, from the terms which contain x1 , then completing the square
a11
n X
n
X
1
2
Q(x) =
(a11 x1 + a12 x2 + . . . + a1n xn ) +
a0ij xi xj .
a11
i=2 j=2
The change of coordinates

x01 = a11 x1 + a12 x2 + . . . + a1n xn ; x0j = xj , j = 2, . . . n,
3. REDUCTION OF QUADRATIC FORMS TO CANONICAL EXPRESSION 73

which correspond to a new basis {e01 , e02 = e2 , . . . , e0n = en } with matrix of change
1
a12
a1n
...
a11
a11
a11
0
0
1
...
0
C =
,
...
...
...
...
0
gives
Now
...
n
n
1 02 X X 0 0 0
Q(x) =
x +
aij xi xj .
a11 1
i=2 j=2
n P
n
P
i=2 j=2
a0ij xi xj is the analytic expression of a quadratic form on Span {e02 , . . . , e0n };
if this new quadratic form is identically zero, or in canonical form, the reduction is
done, otherwise we continue by applying the algorithm for this new quadratic form,
which is defined on a vector space of dimension n 1.
Case 2. aii = 0, i = 1, . . . , n.
As Q is not identically zero, there is at least one element aij 6= 0, i 6= j. After the
change of coordinates
xi = x0i + x0j , xj = x0i x0j , xk = x0k , k 6= i, j
the expression of the quadratic form becomes Q(x) =
n P
n
P
i=1 j=1
a00ij x0i x0j where a00ii 6= 0,
02
since xi xj = x02
i xj . The matrix of change of basis which corresponds to this change
of coordinates is the nonsingular matrix
1
0 ... 0 ... 0 ... 0
... ... ... ... ... ... ... ...
0
0 ... 1 ...
1 ... 0
C 00 =
... ... ... ... ... ... ... ... .
0
0
.
.
.
1
.
.
.
1
.
.
.
0
... ... ... ... ... ... ... ...

0
0 ... 0 ... 0 ... 1
Now Case 1. applies for the new analytic expression of Q.

After at most n 1 steps we obtain an orthogonal basis w.r.t. Q.
The corresponding canonical expression represents a linear combination of squares
of linearly independent forms. The number of these squares in the canonical expression
is minimal, and it is equal to rank Q.
QED.
Example 1. Determine the canonical form of Q : R3 R, Q(x) = 2x1 x2 + 2x1 x3 ,
using Gauss method. Point out the change of coordinates and the matrix of change.
Solution. Note that aii = 0, i = 1, 2, 3, so we need to start with a change of
coordinates described in Case 2. Let
x1 = x01 + x02 ,
x02 = x01 x02 ,
x3 = x03 ,
74
02
0 0
0 0
as a12 = 1 6= 0. Then Q(x) = 2x02
1 2x2 + 2x1 x3 + 2x2 x3 .
1
1
0 0
Next, we follow Case 1 and obtain Q(x) = (2x01 + x03 )2 x02
2x02
2 + 2x2 x3 .
2
2 3
The change of coordinates
x001 = 2x01 + x03 ,
x002 = x02 ,
x003 = x03 ,
1 002 1 002
x x2 .
2 1
2
Combining the two changes of coordinates we get
leads to Q(x) =
x001 = x1 + x2 + x3 ,
x02 =
1 0
1
x1 x02 ,
2
2
The corresponding matrix of change is
1/2 0
1 1 0
C = 1 1 0 0 1
0 0 1
0 0
C 1

1/2
1/2
0 = 1/2
0
1
1
1
1
= 1/2 1/2 0 .
0
0
1
x3 = x03 .
1 1/2
1 1/2 ;
0
1
The columns of C represent the coordinates of the vectors of the new basis w.r.t.
1
1
the initial basis. Thus the new basis is {f1 , f2 , f3 }, f1 = e1 + e2 , f2 = e1 e2 ,
2
2
1
1
f3 = e1 e2 + e3 .
2
2
THEOREM 3.2 (Jacobis method) Let Q : V R be a quadratic form, and
A = [aij ] the associated matrix w.r.t. an arbitrary fixed basis B = {e1 , . . . , en }.
Denote 0 = 1. If all the determinants
a
1 = a11 , 2 = 11
a21
a12
a22
a11
, 3 = a21
a31
a12
a22
a32
a13
a23
a33
, . . . , n = det A
are not zero, then there exists a basis B1 = {f1 , . . . , fn } such that the expression of
Q w.r.t. this new basis is
(3.1)
Q(x) =
n
X
i1
i=1
x02
i ,
where x0i , i = 1, . . . , n are the coordinates of x w.r.t. the basis B1 .

Proof. We are looking for the vectors f1 , . . . , fn of the form
(3.2)
f1 = c11 e1 , f2 = c12 e1 + c22 e2 , . . . fn = c1n e1 + c2n e2 + . . . cnn en
satisfying
(3.3)
A(ei , fj ) = 0, if 1 i < j n,
A(ei , fi ) = 1, i = 1, . . . n
3. REDUCTION OF QUADRATIC FORMS TO CANONICAL EXPRESSION 75

where A is the polar of the quadratic form Q. We will prove that this is the desired
basis. Using the bilinearity of A in (3.3), we obtain the system
A(e1 , fi )
=
A(e2 , fi )
=
...
...
A(ei1 , fi ) =
A(ei , fi )
=
c1i a11
c1i a21
...
c1i ai1,1
c1i ai1
+
+
...
+
+
c2i a12
c2i a22
...
c2i ai1,2
c2i ai2
+
+
...
+
+
...
...
...
...
...
+
cii a1i
+
cii a2i
...
...
+ cii ai1,i
+
cii aii
= 0
= 0
= 0
= 1
of i equations, with i unknowns: c1i , c2i , . . . , cii .

The determinant of this system is i 6= 0, thus the solution is unique. By Cramers
rule we obtain
a11
. . . a1,i1
0
...
...
...
...
i1
cii =
ai1,1 . . . ai1,i1 0 = i .
ai1
...
ai,i1
1
So the basis B1 = {f1 , . . . , fn } is uniquely determined by (3.2), (3.3). It remains to
show that the matrix B = [bij ] associated to Q w.r.t. this basis is diagonal, with the
required diagonal entries. For, compute
bij = A(fi , fj ) = . . . = c1i A(e1 , fj ) + . . . + cii A(ei , fj ), i j.
i1
. The matrix B is
i
symmetric, since A is symmetric (as the polar of a quadratic form), thus bij = 0, for
i > j too.
QED.
Using (3.2), it follows that bij = 0, if i < j, and bii = cii =
Note that Jacobis method does not apply for Example 1.

THEOREM 3.3 (eigenvalues method) Let V be a real n-dimensional vector
space as before, Q : V R a quadratic form, and A = [aij ] the matrix of Q w.r.t.
an arbitrary fixed basis B. Then there exists a basis B1 = {f1 , . . . , fn } such that the
associated expression of Q is
n
X
Q(x) =
i x02
i ,
i=1
where i , i = 1, . . . , n are the eigenvalues of A (each eigenvalue appears as many

times as its algebraic multiplicity), and x0i , i = 1, . . . , n are the coordinates of x w.r.t.
the basis B1 .
Proof. The matrix associated to a quadratic form is always symmetric, thus A
has n real eigenvalues (counted with their multiplicities) and admits a diagonal form
over R. Moreover, there is a basis B1 = {f1 , . . . , fn } made of eigenvectors whose
coordinate columns w.r.t. the initial basis B are orthonormal in Rn , or equivalently,
the matrix of change C is orthogonal (tC = C 1 ). Denote D = diag(1 , . . . , n ) the
diagonal matrix whose diagonal entries are the eigenvalues of A. Then
D = C 1 AC = t CAC
implies that D is the matrix associated to Q w.r.t. the basis B1 . It follows that the
expression of Q w.r.t. B1 is the one in the conclusion of the theorem.
QED.
76
REMARKS 3.4 Comparison of the three methods

1) Gauss method is the most elementary and it provides a sequence of coordinate
changes, not the new basis (which can be deduced by multiplying the corresponding
matrices of change obtained at each step). The basis corresponding to the canonical
expression has no special properties.
2) Jacobis method is the most recommended (if it applies) when we need a canonical expression, and we are not interested in the corresponding basis, since it takes
longer to deduce it.
3) The eigenvalues method is the most convenient for further purposes. If on
V we consider the scalar product which makes the initial basis into an orthonormal
basis, then the basis corresponding to the canonical expression is orthonormal too.
For example, if V = Rn and the initial basis is the canonical basis in Rn , then the
new basis is orthonormal (w.r.t. the canonical inner product).
The method provides the canonical expression and the corresponding basis. The
basis has additional useful properties.
The Signature of a Real Quadratic Form
Real quadratic forms of constant sign are useful in many applications, for example
in extemum problems for functions of several real variables. This why it is helpful
to study these forms, and state some methods which indicate whether the sign is
constant or not.
V will denote a real vector space.
DEFINITION 4.1 A quadratic form Q : V R is said to be:
(i) positive definite if Q(x) > 0, x V \ {0};
(ii) negative definite if Q(x) < 0, x V \ {0};
(iii) positive semidefinite if Q(x) 0, x V;
(iv) negative semidefinite if Q(x) > 0, x V;
(v) indefinite if x1 , x2 V such that Q(x1 ) > 0 and Q(x2 ) < 0.
The notions defined above are used for symmetric bilinear forms too, with the
obvious meaning. A real symmetric bilinear form A : V V R is positive definite
if the associated quadratic form Q(x) = A(x, x) is positive definite, etc.
Example The inner product on a real vector space is a positive definite symmetric
bilinear form.
From now on, throughout the section we assume that dim V = n N? .
The following remarks are easy observations relating some of the resuls in this
chapter.
4. THE SIGNATURE OF A REAL QUADRATIC FORM
77
REMARKS 4.2 1) The number of nonzero terms in a canonical expression of Q is

the same, regardless the method used, and it is equal to rank Q.
2) If Q is positive or negative definite, then Q is nondegenrate, i.e., in the canonical
expression appear n nonzero terms.
3) Q is positive definite in a fixed canonical expression all coefficients are strictly
positive.
In particular, from the eigenvalues method follows that: Q is positive definite
all eigenvalues of Q are strictly positive.
4) Q is negative definite in a fixed canonical expression all coefficients are
strictly negative.
From the eigenvalues method follows that: Q is negative definite all eigenvalues
of Q are strictly negative.
If we combine Remarks 3), 4) above and Jacobis method we obtain the following
result.
THEOREM 4.3 (Sylvesters criterion) If the hypotheses of Jacobis method are
fulfilled, then
(i) Q is positive definite all determinants i , i = 1, . . . , n are strictly positive.
(ii) Q is negative definite the determinants i , i = 1, . . . , n have alternating signs:
1 < 0, 2 > 0, 3 < 0, . . ., (1)k k > 0, . . ..
DEFINITION 4.4 Let Q(x) =
n
P
i=1
ai x2i be a canonical expression of Q : V R,
dim V = n N? , where p of the coefficients a1 , . . . , an are strictly positive, q are

strictly negative, and d = n (p + q) coefficients are equal to zero.
The triplet (p, q, d) N3 is called the signature of the form Q and is denoted by
sign Q.
As we have seen, the canonical expression of a quadratic form is not unique. The
next result states that the signature depends only on the quadratic form.
THEOREM 4.5 (Inertia law) The signature of a quadratic form Q is the same,
for any canonical expression of Q.
Proof. Consider two bases B = {e1 , . . . , en } and B0 = {e01 , . . . , e0n } in V such that
the expression of Q w.r.t. each of them is canonical
Q(x) =
n
X
ai x2i and Q(x) =
i=1
n
X
a0i x02
i respectively.
i=1
We may assume that in the expression of Q w.r.t. B (B0 ) the first p (p0 ) coefficients
are strictly positive, the next q (q 0 ) are strictly negative, and the last d (d0 ) are zero.
Morever, we may assume that ai , a0i {1, 0, 1}. Then
(4.1)
Q(x) =
p
X
i=1
x2i
p+q
X
i=p+1
x2i
p
X
i=1
x02
i
0
pX
+q 0
i=p0 +1
x02
i .
78
We need to show that p = p0 , q = q 0 . Suppose p 6= p0 and assume p > p0 . Let U =

Span(e1 , . . . , ep ) and U0 = Span(e0p0 +1 , . . . , e0n ), thus dim U = p, dim U0 = n p0 .
This yields
dim U + dim U0 = p + n p0 > n,
which shows that the subspaces U, U0 are not independent. Therefore there exists
x U U0 , x 6= 0, and x can be written as
x = x1 e1 + . . . + xp = x0p0 +1 e0p0 +1 + . . . + x0n e0n .
Replacing this particular x in (4.1) we obtain
02
0 x21 + . . . + x2p = Q(x) = x02
p0 +1 . . . xn 0.
But this implies

x1 = . . . = xp = 0;
x0p0 +1 = . . . = x0n = 0,
thus x = 0 which contradicts the choice of x. It follows that p = p0 . Similarly, q = q 0 ,

hence d = d0 and (p, q, d) = (p0 , q 0 , d0 ).
QED.
COROLLARY 4.6 1)The quadratic form Q : V R, dim V = n N? is positive
definite if and only if one of the following conditions is satisfied:
(i) sign Q = (n, 0, 0);
(ii) the determinants i , i = 1, . . . , n are strictly positive;
(iii) the eigenvalues of the matrix of Q are all strictly positive.
2) Q is negative definite if and only if one of the following conditions is satisfied:
(i) sign Q = (0, 0, n);
(ii) (1)i i > 0, i = 1, . . . , n ;
(iii) the eigenvalues of the matrix of Q are all strictly negative.
Example 2. Determine the canonical form of Q : R3 R, Q(x) = x21 + 7x22 + x23
8x1 x2 16x1 x3 8x2 x3 , and the corresponding basis. Use Jacobis method, then the
eigenvalues method. Check the inertia law.
Solution.
1 4 8
A = 4 7 4 ;
8 4 1
1 4
= 9, 3 = det A = 729.
0 = 1, 1 = a11 = 1, 2 =
4 7
1 02
1 02
Using formula (3.1) we obtain Q(x) = x02
x .
1 x2 +
9
81 3
We look for the corresponding basis B1 = {f1 , f2 , f3 } of the form (3.2). The coefficients cij , 1 i < j n = 3 are given by the solutions of the linear equations (3.3)
as follows.
A(e1 , f1 ) = 1 c11 = 1.
5. PROBLEMS
79
4
1
A(e1 , f2 ) = 0
c12 4c22 = 0
c12 =
; c22 =
.
A(e2 , f2 ) = 1
4c12 + 7c22 = 1
9
9
A(e1 , f3 ) = 0
c13 4c23 8c33 = 0
4
1
8
A(e2 , f3 ) = 0
4c13 + 7c23 4c33 = 0 c13 =
; c23 =
; c23 =
.
81
81
81
A(e3 , f3 ) = 1
8c13 4c23 + c33 = 0
4
1
8
4
1
It follows that f1 = e1 , f2 = e1 e2 , f3 = e1 e2 + e3 , and the matrix
9
9
81
81
81
of change is
1 4/9 8/81
C = 0 1/9 4/81 .
0
0
1/81
Next we will use the eigenvalues method.
The eigenvalues of A are 1 = 9, 2 = 9, 3 = 9. Therefore Q(x) = 9y12 +
2
9y2 9y32 , where y1 , y2 , y3 are the coordinates of x w.r.t. a basis made up of eigenvectors, such that the matrix of change is orthogonal. To determine such a basis
{u1 , u2 , u3 }, pick first some linearly independent eigenvectors. Let us take v1 =
(1, 0, 1), v2 = (1, 2, 0), v3 = (2, 1, 2). Using the Gram-Schmidt procedure for
the first two (which happen not to be orthogonal, as they correspond
tothe same
eigenvalue),
and
normalizing
the
third
we
obtain
the
u
=
(1/
2,
0,
1/
2), u2 =
1
( 2/6, 2 2/3, 2/6), u3 = (2/3, 1/3, 2/3).

As the inertia law states, from either one of the canonical expressions obtained
using different methods, the signature is (2, 1, 0).
Problems
1. Let V be a real vector space and : V R a linear form. Show that

Q(x, y) = (x)(y), x, y V is a positive semidefinite symmetric bilinear form on V .
2. Let P3 be the real vector spaceR ofRthe polynomial functions of degree at most
1 1
two and Q : P3 P3 R, Q(x, y) = 0 0 x(t)y(s)dtds.
1) Show that Q is a positive semidefinite symmetric bilinear form.
2) Determine the matrices of Q with respect to the bases {1, t, t2 } and {1t, 2t, 1+
t + t2 }. What is the relationship between these matrices ?
3) What is the canonical expression of g ?
3. Consider the bilinear form
Q(x, y) = x1 y2 x2 y1 + x1 y3 x3 y1 + x1 y4 x4 y1 + x4 y4 .
1) Write Q(x, y) in matrix form.
2) Find the matrix associated to Q with respect to the basis f1 = (1, 1, 0, 1),
f2 = (0, 1, 1, 0), f3 = (0, 1, 0, 1), f4 = (1, 0, 0, 1).
4. Reduce the quadratic forms to the canonical expression using orthogonal transformations, determine the signatures and the sets of constant level.
80

1) 5x21 + 4x22 + 6x23 + 4x1 x2 + 4x1 x3 ,
2) 3x21 5x22 7x23 8x1 x2 + 8x2 x3 ,
3) 2x1 x2 + 2x3 x4
5. Show that the quadratic forms

1) Q(x) = 3x21 + 4x22 + 5x23 + 4x1 x2 4x2 x3 ,
2) Q(x) = 2x21 + 5x22 + 5x23 + 4x1 x2 4x1 x3 8x2 x3 ,
are positive definite.
Chapter 5
Free Vectors
1
Free Vectors
E3 will denote the 3-dimensional space of elementary geometry.
For any two fixed points A, B E3 consider the oriented segment AB.
A is called the origin of AB and B the vertex of AB (Fig. 1).
Fig. 1
If A 6= B, the straight line AB is called the supporting line of the oriented segment
AB. The length (module or norm) of AB is the length of the segment AB and it is
denoted by k AB k, AB or d(A, B).
If A = B we say that AB=AA is the zero oriented segment with the origin at A.
The zero segment has length zero.
Note that AB and BA are distinct oriented segments if A 6= B.

We say that 2 oriented segments have the same direction if their supporting lines
are parallel or identical.
Since each straight line admits two different orientations, a nonzero oriented segment is characterized by its direction, sense, length and origin.
DEFINITION 1.1 Two nonzero oriented segments are said to be equipolent if they
have the same direction, sense and length (but possibly different origins).
By definition, all zero oriented segments are equipolent.
If AB and CD are equipolent, we write ABCD (Fig. 2).

81
82
CHAPTER 5. FREE VECTORS
Fig. 2
PROPOSITION 1.2 The equipolence relation is an equivalence relation on the set

of all oriented segments in E3 , i.e., it satisfies the following properties
(i) ABAB, for any A, B E3 (reflexivity)
(ii) ABCDCDAB (symmetry)
(iii) ABCD and CDEF ABEF (transitivity).

DEFINITION 1.3 A free vector is an equivalence class of the equipolence relation.
The set of all free vectors is denoted by V3 .
The equivalence class of the oriented segment AB is denoted by AB and it is
called the free vector AB. Thus AB = {CD | CDAB}, and AB AB. Any element
of a free vector is called a representative of that free vector. AB is a reprezentative
of AB; if ABM N (or equivalently, M N AB), then M N is another representative

of AB, and AB = M N .
The equivalence class of all zero oriented segments is called zero free vector and
is denoted by 0. Arbitrary free vectors will be denoted by overlined letters a
, b, c . . ..
According to the previous notation that involves a representative of the free vector, we
may write either AB a

or AB = a
, if a
is the set of all oriented segments equipolent
to AB.
DEFINITION 1.4 The length (norm) of a free vector a
is the length of a represen
tative AB, AB a
.
We denote the length of a

by k
ak. Thus k
ak = kABk = k AB k = d(A, B).
It follows that k
ak is a positive real number and k
ak = 0 a
= 0. Similarly,
the direction and sense of a nonzero free vector are the direction and respectively the
sense of a representative.
The above definitions are correct since all representatives of AB have the same
length, same direction and same sense (for A 6= B).
A free vector of length one is called a unit vector or a versor and it is usually
denoted by e or u
.
Nonzero free vectors are characterized by direction, sense and length. The length
zero determines the zero free vector.
PROPOSITION 1.5 For any P in E3 and any free vector a
V3 , there exists a
unique oriented segment P A (i.e. a unique vertex A) such that P A a

.
2. ADDITION OF FREE VECTORS
83
This proposition allows us to use representatives which have a convenient origin.

In E3 we fix a point O, called the origin of the space E3 . By the previous proposition there is a bijective correspondence between V3 and the set of all oriented segments
which have the origin at O.
E3 can be viewed as the set of all vertices of the oriented segments which have the
origin at O. Therefore there is also a bijective correspondence between E3 and V3 .
DEFINITION 1.6 Two nonzero free vectors of same direction are called collinear
vectors (Fig. 3).
By definition, 0 is collinear with any other vector.
Three nonzero vectors are called coplanar if there is a plane containing their directions (Fig. 4).
By definition, 0 and any other two vectors are coplanar.
REMARKS 1.7 (i) Let a
= P A, b = P B V3 \ {0}. Then a
, b, are collinear
the points P, A, B are collinear.
(ii) Let a
= P A, b = P B, c = P C V3 \ {0}. Then a
, b, c are coplanar the
points P, A, B, C are coplanar.
The collinear vectors AB and BA are called opposite vectors. It follows that the
opposite of 0 is 0. If a
6= 0, then the opposite of a
is the unique free vector which has
the same direction and the same length as a
, but it has opposite sense. In the picture
below
a is the opposite of a
(Fig. 3).
Fig. 3
Fig. 4
Addition of Free Vectors
DEFINITION 2.1 Let a

, b V3 such that P A = a
and AB = b, where P is an
arbitrary point in E3 .
The sum of a
, b is the free vector denoted by c = a
+ b and defined by (Fig. 5)
a
+ b = P B
Fig. 5
84
This addition rule is called the triangle law.

If a
, b are not collinear, we may use the equivalent parallelogram law and define
a
+ b as the class of the oriented diagonal P B of the parallelogram P ABC, where
PA = a
, P C = b.
Addition of free vectors + : V3 V3 V3 is a well defined binary operation
since it does not depend on the choice of the point P .
THEOREM 2.2 Addition of free vectors has the following properties:
(i)
a, b, c V3 , a
+ (b + c) = (
a + b) + c (associativity)
(ii)
a, b V3 , a
+b=b+a
(commutativity)
(iii)
a V3 , a
+ 0 = 0 + a
=a
(0 is the identity element)
(iv)
a V3 a
V3 such that a
+ (
a) = (
a) + a
= 0 (each vector has an
00 00
inverse with respect to + )
Proof. (ii) and (iii) are immediate from the definition.
For (iv) we check easily that
a is the opposite of a
, i.e. if a
= P A, then
a = AP .
If a
, b, c are pairwise not collinear, then associativity follows from Fig. 6.
Fig. 6
Fig. 7
COROLLARY 2.3 (V3 , +) is an abelian group.

Like in any abelian group with additive notation, we may define the subtraction
(difference) of vectors. The difference a
b is the unique solution of the equation
b + x
=a
; this is x
=a
+ (b).
If a
= AB and b = AD, then x
=a
b = DB (Fig. 7).
By the observation which proves (iv), it makes sense to write AB = BA, for any
A, B E3 .
Multiplication by Scalars
In this section we will define on V3 a natural structure of a real vector space. Addition
of free vectors was already defined. We still need an external operation R V3 V3
which satisfies the definition of the vector space.
DEFINITION 3.1 (multiplication of free vectors by scalars, Fig. 8).
Let k R and a
V3 . Then k
a V3 is defined as follows
(i) if a
= 0 or k = 0, then k
a = 0;
(ii) if a
6= 0 and k 6= 0, then k
a is the vector which has the direction of a
, length
|k| k
ak, the sense of a
for k > 0, and the sense of
a for k < 0.
Note that a
and k
a are collinear, and kk
ak = |k| k
ak k R, a
V3 .
4. COLLINEARITY AND COPLANARITY
85
Fig. 8
THEOREM 3.2 Multiplication of free vectors by real scalars has the following properties:
(i) k, l R,
a V3 , k(l
a) = (kl)
a
(ii)
a V3 , 1
a=a
(iii) k, l R,
a V3 , (k + l)
a = k
a + l
a
(iv) k R,
a, b V3 , k(
a + b) = k
a + kb.
Proof. (i) (iii) are left to the reader.
We will prove (iv) for a
, b not collinear. The collinear case is left to the reader as
well.
Let OA = a
and AB = b. Then a
+ b = OA + AB = OB.
Suppose k > 0. Then there is A0 such that OA0 = k
a and B 0 such that OB 0 =
k(
a + b).
Fig. 9
The triangles 4OAB and 4OA0 B 0 are similar since they have a common angle
whose sides are proportional. This similarity implies ABkA0 B 0 and kA0 B 0 k = kkABk.
The same orientation of OA, OA0 and OB, OB 0 gives the same orientation of AB and
A0 B 0 , thus A0 B 0 = kAB (Fig. 9).
The case k < 0 is similar to k > 0, and the case k = 0 is trivial.
QED.
COROLLARY 3.3 V3 is a real vector space.
Collinearity and Coplanarity
The collinearity and coplanarity of free vectors are notions we defined geometrically.
In this section we will see they are related to the algebraic structure of a real vector
space that was defined on V3 .
and b are collinear and a
6= 0, then there
PROPOSITION 4.1 If the free vectors a
exists a unique real number k such that b = k

a.
Proof. If b = 0, pick k = 0.
86

If b 6= 0 and b has the sense of a
, then
b = kbk a
,
k
ak
kbk
a
has the direction, sense and length of b.
k
ak
Similarly, if b 6= 0 and the sense of b is opposite to sense of a
(i.e., b has the sense
kbk
of
a), b =
a
, and the existence of k is proved.
k
ak
For the uniqueness, let k, l R such that b = k
a = l
a. Then k = l since a
6= 0,
by the properties of vector space operations.
QED
since
COROLLARY 4.2 (i) The vectors a

, b are collinear if and only if they are linearly
dependent.
(ii) If a
6= 0, then the set V1 = {b| a
, b collinear} is a 1-dimensional vector space.
More precisely, V1 = {b| k R, b = k

a} = Span {
a}.
(iii) Two noncollinear vectors are linearly independent.
, b, c are coplanar and a
, b are not collinear,
PROPOSITION 4.3 If the vectors a
then there exist the unique numbers s, t R such that c = s
a + tb.
Proof. If c = 0, take s = t = 0. If c and a
are collinear, take s such that c = s
a
and t = 0.
If c and a
are not collinear and c, b are not collinear either, let A, B, C such that
OA = a
, OB = b, OC = c. Then O, A, B, C are coplanar points.
Fig. 10
The parallel lines through C to the supporting lines of OA and OB respectively,

determine the vectors OE and OF such that (Fig. 10):
OC = OE +OF , OA and OE collinear, and OB, OF collinear. Then we can write
OE = sOA, OF = sOB for some s, t R, and the existence of the decomposition is
proved.
For the uniqueness assume
c = s
a + tb = s1 a
+ t1b
Then (ss1 )
a+(tt1 )b = 0. Since a
, b are not collinear, they are linearly independent,
so s s1 = 0 and t t1 = 0.
QED
COROLLARY 4.4 (i) The vectors a
, b, c are coplanar if and only if they are linearly dependent.
(ii) If a
, b are not collinear, then the set V2 = {
c| a
, b, c coplanar} is a 2
dimensional vector space. More precisely, V2 = {
c = s
a + tb| s, t R} = Span {
a, b}.
(iii) Three noncoplanar vectors are linearly independent.
5. INNER PRODUCT IN V3
87
THEOREM 4.5 V3 is a 3-dimensional real vector space.

Proof. We saw that any three vectors which are not coplanar, are linearly independent. Let a
= OA, b = OB, c = OC such that O, A, B, C are not coplanar. We
will show that a
, b, c span V3 .
Let d = OD V3 . If d is coplanar with 2 of the vectors a

, b, c, we are done.
Assume it is not. Then the planes through D which are parallel to the planes
(OAB), (OAC), (OBC) determine a parallelipiped.
Fig. 11
OD is an oriented diagonal of this parallelipiped and it decomposes as (Fig. 11)

OD = OD1 + OD2 + OD3 = rOA + sOB + tOC.
Therefore d = OD Span {
a, b, c}.
QED
Inner Product in V3

, b V3 \ {0} and a
= OA, b = OB.
The angle 6 (
a, b) between a
and b is the angle 6 AOB [0, ] of their representa
tives OA and OB.
Fig. 12
The definition of the angle does not depend on the representatives chosen.
DEFINITION 5.2 Let u
V3 be a versor, b 6= 0 and = 6 (
u, b).
The vector u b = kbk cos u
is called the orthogonal projection of b onto u
.
Fig. 13
88
The number pru b = kbk cos is called the algebraic measure of the orthogonal
projection of b onto u
(Fig. 13).
By definition u 0 = 0, pru 0 = 0.
Note that
i
pru b < 0
, ; pru b = 0 = or b = 0.
2
2
LEMMA 5.3 For any versor u
, the functions u : V3 V3 and pru : V3 R are
linear transformations.
We leave the proof as an exercise. The picture below (Fig. 14) describes one of
the cases that arise in order to prove additivity, where OA is collinear to u
, OB = b,
OC = c, OABC is a parallelogram, BB 0 OA, CC 0 OA, DD0 OA.
Then AD = b + c, u b = OB 0 = C 0 D0 , u c = OC 0 , u (b + c) = u (OD) = OD0 .
The additivity of u follows from OD0 = OC 0 + C 0 D0 .
Fig. 14
THEOREM 5.4 The function h, i : V3 V3 R defined by

(
k
ak kbk cos , if a
6= 0, b 6= 0 and = 6 (
a, b)
h
a, bi =
0,
if a
= 0 or b = 0
is a scalar (dot, inner) product in V3 .
Proof. We need to check the symmetry (commutativity), homogeneity, additivity
and positivity of the function defined above (see the definition of the scalar product
on a real vector space). All are straightforward except for additivity. Let us prove
the additivity of h, i with respect to the second argument. For, assume a
6= 0 (the
a
case a
= 0 is obvious), and let u
=
be the versor of a
. Then
k
ak
h
a, b + ci = hk
ak
u, b + ci = k
akh
u, b + ci, by homogeneity.
From the definitions
Then
pru b = h
u, bi,
b V3 .
h
u, b + ci = h
u, bi + h
u, ci by the previous lemma.
5. INNER PRODUCT IN V3
It follows that
h
a, b + ci
89
= k
ak(h
u, bi + h
u, ci)
= hk
ak
u, bi + hk
ak
u, ci
= h
a, bi + h
a, ci.
The scalar product defined above is called the canonical scalar product on V3 .
COROLLARY 5.5 V3 is a Euclidean vector space.
p
REMARKS 5.6 1) k
ak = h
a, a
i,
a V3 . Therefore the length of a free vector
is the Euclidean norm of that vector with respect to the canonical scalar product.
2) The Cauchy-Schwarz inequality follows directly from the definition using:
| cos | 1 |h
a, bi| k
ak kbk.
3) h
a, bi = 0 a
= 0 or b =
0 or 6 (
a, b) = .
2
As in a general Euclidean vector space, the free vectors a
, b are called orthogonal
if h
a, bi = 0.
It is convenient to work with orthonormal bases.
The coordinates of a vector with respect to an orthonormal basis are called Euclidean coordinates.
Fig. 15
be an orthonormal basis (Fig. 15), i.e. the values of the scalar products
Let {i, j, k}
are given by the table
h, i i j k
i 1 0 0
j 0 1 0
k 0 0 1
which leads to the canonical expression of the scalar product of a
= a1i + a2 j + a3 k
and b = b1 i + b2 j + b3 k, namely
h
a, bi = a1 b1 + a2 b2 + a3 b3 .
p
In particular the norm of a
is k
ak = a21 + a22 + a23 and
cos =
a1 b1 + a2 b2 + a3 b3
h
a, bi
p
=p 2
,
k
ak kbk
a1 + a22 + a23 b21 + b22 + b23
for a
6= 0, b 6= 0, [0, ].
90
Vector (cross) Product in V3
Let a
, b V3 . If a
6= 0 and b 6= 0, then denotes the angle of a
and b.
DEFINITION 6.1 The vector denoted a
b and defined by (Fig. 16)
k
ak kbk sin
e, if a
, b noncollinear
a
b=
0,
if a
, b collinear (in particular a
= 0 or b = 0),
where e is a versor whose direction is given by e a
, e b and the sense of e is given
by the right-hand rule for (

a, b, e) is called the vector (or cross) product of a
and b.
Fig. 16
The function: V3 V3 V3 ,
proposition below.
(
a, b) a
b is bilinear. This follows from the
PROPOSITION 6.2 The vector product has the following properties:

(1) a
b = 0 a
, b collinear; in particular a
0, a
a
= 0
(2) b a
=
a b (anticommutativity)
(3) t(
a b) = (t
a) b = a
(tb), t R (homogeneity)
(4) a
(b + c) = a
b+a
c (additivity or distributivity)
(5) k
a bk2 = k
ak2 kbk2 h
a, bi2 (the Lagrange identity)
(6) If a
, b are not collinear, then k
a bk represents the area of the parallelogram
determined by a
and b.
Proof. (1), (2), (3) are immediate from the definition. For (5) multiply
sin2 = 1 cos2
ak2 kbk2 . The area of the parallelogram in Fig.16 is
by k
k
ak kbk sin = k
a bk,
so 6) is an obvious geometric interpretation of the vector product. For (4) assume
that a
is a versor and let be a plane perpendicular on a
(Fig. 17). Let also b0 and
0
c be the projections of b respectively c onto the plane P (i.e. b0 is determined by the
intersections of the lines passing through O and B and perpendicular to P , and P ,
where b = OB)
Then
a
b = a
b0 , a
c = a
c0
and
a
b0 , a
c0 , a
(b0 + c0 )
6. VECTOR (CROSS) PRODUCT IN V3
91
are obtained from b0 , c0 , b0 + c0 respectively, by a rotation of angle about the axis a

.
2
Since the rotation of the sum is the sum of rotations, i.e.
a
(b0 + c0 ) = a
b0 + a
c0 ,
it follows that
a
(b + c) = a
b + a
c.
Fig. 17
of V3 is an orthonormal basis satisfying

From now on the canonical basis {i, j, k}
also k = i j (Fig. 15).

i.e. i, j, k are the versors of a Cartesian system of coordinate axes in E3 .
Then the cross products of these basis vectors are in the following table
i
j
k
i
0
k
j
j
k
k j
i
0
i 0
and the expression of a

b can be written as a symbolic determinant,
i
j
k
a
b = a1 a2 a3 = (a2 b3 a3 b2 )i + (a3 b1 a1 b3 )j + (a1 b2 a2 b1 )k.
b1 b2 b3
(b
c) is called the double vector product of a
, b, c.
DEFINITION 6.3 The vector a
Fig. 18
It can be shown that
a
(b c) =
b
c
= h
a, cib h
a, bi
c.
h
a, bi h
a, ci
Note that w
=a
(b c), b, c are coplanar (Fig. 18). In general a
(b c) 6= (
a b) c
thus the vector product is not associative.
92
Mixed Product

, b, c V3 . The number h
a, b ci is called the mixed
product of the vectors a

, b, c.
PROPOSITION 7.2 (1) h
a, b ci = 0 a
, b, c are coplanar.
(2) If a
, b, c are not coplanar, then the absolute value |h
a, b ci| represents the
volume of the parallelepiped determined by the representatives with common origin of
a
, b, c (Fig. 19).
Fig. 19
Proof. (1) Assume b, c noncollinear (otherwise b c = 0 and the equivalence is

obvious). Then
h
a, b ci = 0
a
b c
a
= 0 or the direction of a
is parallel to a plane containing the directions of b and c

a
, b, c coplanar.
Let V be the volume

V = kb ck h = kb ck k
ak | cos |,
where = 6 (
a, b c). Since kb ck is the area of the parallelipiped basis. Then
V = |h
a, b, ci|.
PROPOSITION 7.3 (Properties of the mixed product)
(1)
If
a
= a1i + a2 j + a3 k
b = b1i + b2 j + b3 k ,
c = c1i + c2 j + c3 k
h
a, b ci
(2)
(3)
a1
then h
a, b ci = b1
c1
= h
c, a
bi = hb, c a
i
= h
a, c bi
a b, ci.
= h
ht
a, b ci = h
a, (tb) ci = h
a, b t
ci = th
a, b ci.
(The mixed product is linear in each of its three arguments).

(4)
a2
b2
c2
h
u + v, b ci = h
u, b ci + h
v , b ci.
a3
b3
c3
7. MIXED PRODUCT
(5)
93
h
a, ci h
a, di
h
a b, c di =
hb, ci hb, di
(the Lagrange identity.)
Proof. We will prove only (5) leaving the other proofs to the reader. Note that (1)
is a straightforward computation using the formula for b c and the known products
(2), (3), (4) are easy consequences of (1). For (5) denote w
of i, j, k;
=a
b. Then
hw,
c di
But
w
= hd,
ci =
c wi
= hd,
c w
= c (
a b) =
by (2)
h
c, a
i
h
c, bi
by the formula for the double vector product. We replace
= hd,
h
hw,
c di
c, bi
a h
c, a
ibi
= h
c, bihd, a
i + h
c, a
ihd, bi
and get

h
a, ci h
a, di
.
hb, ci hb, di
A basis {
a, b, c} of V3 is said to have positive orientation if h
a, b ci is strictly
positive, and negative orientation if h
a, b ci is strictly negative.
has positive orientation since hi, j ki
= 1.
The canonical basis {i, j, k}
94
Problems
1. Let A(1, 2, 0), B(1, 0, 3), C(2, 1, 1) be points in E3 . Are these points
collinear? Find the area of the triangle ABC, and the altitude from the base BC to
the vertex A.
2.
1)
2)
3)
Given the points A(1, 1, 2), B(2, 3, 0), C(0, 1, 1), D(1, 2, 3), compute:
the mixed product hAB, AC ADi; are the points coplanar?
the volume of the tetrahedron ABCD;
the altitude of the tetrahedron, from the base ACD to the vertex B.
3. Show that
a
(b c) + b (
ca
) + c (
a b) = 0
h
a (b c), (b (
ca
)) (
c (
a b))i > 0.
4. Solve the equations: 1) h

a, x
i = ,
2) a
x
= b.
5. Find the volume of the parallelipiped constructed on some representatives with

common origin of the vectors a
= 2
u v + w,
b = u
w,
c = u
+ w,
where
k
uk = 1, k
v k = 2, kwk
= 3,
6
(
u, v) =
,
2
6
(
u, w)
=
,
3
6
(
v , w)
=
.
4
b = i + j + 3k,
c = i + 4j. If the
6. Consider the vectors a
= i + 4j + 6k,
vectors are coplanar, determine the possible values of . In this case, decompose a
along b and c.
7. Show that if a
b + b c + c a
= 0, then the vectors a
8. Let a
, b, c V3 be noncoplanar vectors such that h
a, bi 6= 0. Compute
a
b a
(b c)
E=
,
+ 2.
h
a, bi h
a, b ci
9. Given the points A(4, 2, 2), B(3, 1, 1), C(4, 2, 0), determine the vertex D of
the tetrahedron ABCD such that D Oz and the volume of ABCD is 4. Determine
also the altitude from the base ABC to D.
Chapter 6
Straight Lines and Planes in

Space
1
Cartesian Frames
We mentioned that a fixed origin O E3 provides a natural one-to-one correspondence between E3 and V3 , assigning to each point M E3 a unique vector r = OM
V3 , called the position vector of M . Each fixed basis of V3 determines a bijective
correspondence between V3 and R3 .
a fixed orThroughout this chapter, O E3 will be a fixed origin and {i, j, k}
thonormal basis of V3 . The set {O; i, j, k} is called a Cartesian frame in E3 . The

is the basis of the frame.
point O is called the origin of the frame, and {i, j, k}
The Euclidean coordinates of the position vector OM are also called the Cartesian
If
coordinates of the point M with respect to the orthonormal frame {O; i, j, k}.
OM = xi + y j + z k, we write M (x, y, z).

The bijection between E3 and R3 determined by a fixed Cartesian frame is called
a Cartesian coordinate system in E3 .
The bijections mentioned above allow the identification of E3 , V3 and R3 .
will also be assumed to have positive (or right-handed) orienThe basis {i, j, k}
= 1, or equivalently i j = k.
tation in V3 , i.e. hi, j ki

The oriented lines Ox, Oy, Oz passing through O, of direction and sense given
by i, j, k respectively are called the Cartesian axes of the frame {O; i, j, k}.
The
Cartesian coordinates of the point M are the algebraic measures of the orthogonal
projections of OM onto the coordinate axis (Fig. 20).
Fig. 20
95
96
CHAPTER 6. STRAIGHT LINES AND PLANES IN SPACE

The axes are characterized by the equations
y=0
z=0
Ox :
,
Oy :
,
z=0
x=0
Oz :
x=0
.
y=0
Any two axes determine a plane called a coordinate plane. The coordinate planes
xOy, yOz, zOx are characterized by the equations
xOy : z = 0,
yOz : x = 0,
zOx : y = 0.
Equations of Straight Lines in Space
A straight line in E3 may be determined by either:

a point and a nonzero vector,
two distinct points,
the intersection of two planes, etc.
2.1. The Line Determined by a Point and a Nonzero Vector

The point M0 (x0 , y0 , z0 ) and a
= li + mj + nk V3 \ {0} determine a straight
line D passing through M0 , and having the direction of a
(Fig. 21).
Let M (x, y, z) E3 , r = OM , r0 = OM0 . Then M D M0 M and a
are
collinear i.e.
(2.1)
(
r r0 ) a
= 0 (the vector equation of D)
The vector a
is called a director vector of D. Any vector of the form k
a, k R\{0}
is another director vector of D and may be used to obtain the equations of D, instead
of using a
.
Fig. 21
The position vector r = OM of an arbitrary point M D is of the form

(2.2)
r = r0 + t
a,
t R,
which is another equivalent condition to the collinearity of r r0 and a

. The vector
equation of D is equivalent to
x = x0 + lt
y = y0 + mt ,
(2.3)
t R (the parametric equations of D in R3 )
z = z0 + nt
or to
(2.4)
y y0
z z0
x x0
=
=
l
m
n
(the canonical Cartesian equations of D

in R3 ),
3. EQUATIONS OF PLANES IN SPACE
97
with the convention that if a denominator is zero , then the corresponding numerator
is zero too. More precisely,
(1) if l = 0, mn 6= 0, then
x = x0 ,
y y0
z z0
=
m
n
and D k yOz;
(2) if l = m = 0, then
x = x0 ,
y = y0
and D k Oz.
Note that a
6= 0, thus at most two of the coordinates of a
could possibly be zero.
2.2 The Line Determined by Two Distinct Points

Let M1 (x1 , y1 , z1 ) 6= M2 (x2 , y2 , z2 ) E3 , and D =the straight line M1 M2 . Then
a
= M1 M2 = (x2 x1 )i + (y2 y1 )j + (z2 z1 )k is a director vector of D. In the
previous equations we may replace a
by M1 M2 and M0 by M1 (Fig. 22). For example
the canonical Cartesian equations (2.4) become
(2.5)
x x1
y y1
z z1
=
=
.
x2 x1
y2 y1
z2 z1
Fig. 22
Equations of Planes in Space
A plane in E3 may be determined by either:

- a point and a normal direction,
- three noncollinear points,
- a straight line and an exterior point,
- two concurrent straight lines,
- two parallel straight lines, etc.
3.1. The Plane Determined by a Point and a Nonzero Vector

The plane determined by M0 (x0 , y0 , z0 ) and a nonzero vector n
= ai + bj + ck is
the unique plane P passing through M0 such that the line D through M0 of direction
n
is perpendicular onto P (Fig. 23).
n
is called a normal vector to the plane P .
D is normal line of P .
98

Fig. 23
Let M (x, y, z). Then

M P M0 M n
hM0 , M , n
i = 0.
The last equality is called the vector equation of the plane. Replacing M0 M and n
by
their analytic expressions it follows that
(3.1)
a(x x0 ) + b(y y0 ) + c(z z0 ) = 0
(the Cartesian equation in R3 of the plane passing through M0 , perpendicular to n

).
If we denote d = (ax0 + by0 + cz0 ), the previous equation becomes
(3.2)
ax + by + cz + d = 0
(the general Cartesian equation of a plane).
Conversely, any equation of the form (3.2) with a2 + b2 + c2 6= 0 represents a plane of
normal vector n
= ai + bj + ck.
3.2. Particular Planes

1) The coordinate planes and planes parallel to these planes are described by the
following particular forms of (3.2). We start with the coordinate planes,
xOy : z = 0;
yOz : x = 0;
zOx : y = 0.
k is a normal vector for xOy and for any plan P parallel to xOy. Then (Fig. 24)
P : z = a,
a R \ {0},
for P kxOy.
Fig. 24
Similarly, since i is normal to yOz and is normal to zOx we obtain

P : x = a,
a R \ {0},
for P kyOz
P : y = a,
a R \ {0},
for P kzOx.
2) The equations of planes perpendicular to the coordinate planes are:

P : ax + by + d = 0
a2 + b2 6= 0,
if P xOy
P : by + cz + d = 0
b2 + c2 6= 0,
if P yOz
P : ax + cz + d = 0
a2 + c2 6= 0,
if P zOx.
3. EQUATIONS OF PLANES IN SPACE
99
3) The equation of a plane passing through O(0, 0, 0) is

ax + by + cz = 0.
4) The equations of planes containing one of the coordinate axes Oz, Ox, Oy are
ax + by = 0, by + cz = 0,
and
ax + cz = 0 respectively.
3.3. The Plane Determined by Three Noncollinear Points

Given the noncollinear points Mi (xi , yi , zi ), i = 1, 2, 3 there exists a unique plane P
containing M1 , M2 , M3 (Fig. 25). If M (x, y, z) is another point, then
M P
M, M1 , M2 , M3
are coplanar in E3
M1 M , M1 M2 , M1 M3
hM1 M , M1 M2 M1 M3 i = 0 the vector equation of P.
are coplanar in V3
The above vector equation is equivalent to the equation
x x1 y y1 z z1
x
y
y
z
z
(3.3)
1
2
1
2
1 =0
2
x3 x1 y3 y1 z3 z1
in R3 or to
(3.4)
x1
y1
z1
x2
y2
z2
x3
y3
z3
1
=0
1
(the equation of the plane determined

by three points; coplanarity of four points).
Fig. 25
A particular case of a plane determined by three points is M1 (a, 0, 0) Ox,

M2 (0, b, 0) Oy, M3 (0, 0, c) Oz, a, b, c R (i.e. 0
/ P and M1 , M2 , M3 are the
intersections of P with the coordinate axes). Replacing in (3.3) or (3.4) we obtain
x y z
+ + 1 = 0 (the intercept form of the equation of a plane, Fig. 26).
a
b
c
Fig. 26
100
3.4. The Plane Determined by a Point and

Two Noncollinear Vectors
There exists a unique plane P passing through a given point M0 (x0 , y0 , z0 ) and
containing lines of directions v1 = l1i+m1 j+n1 k and v2 = l2i+m2 j+n2 k respectively,
where v1 , v2 are noncollinear (Fig. 27).
Let M1 , M2 such that v1 = M0 M1 , v2 = M0 M2 . Then P is determined by
M0 , M1 , M2 . It follows that M (x, y, z) P if and only if M0 M , M0 M1 , M0 M2
are coplanar, this is equivalent to M0 M = r
v1 + s
v2 for some r, s R. Identifying
the coordinates we obtain
x = x0 + rl1 + sl2
y = y0 + rm1 + sm2
(3.6)
r, s R
z = z0 + rn1 + sn2
(the parametric equations of the plane in R3 ).
The coplanarity of M0 M , M0 M1 , M0 M2 is also equivalent to hM0 M , v1 v2 i = 0,
i.e.,
x x0 y y0 z z0
= 0.
l
m
n
(3.7)
1
1
1
l2
m2
n2
Fig. 27
The Intersection of Two Planes
It is known that any two distinct planes are either parallel or they have a common
line.
i = aii + bi j + ci k,
Let Pi : ai x + bi y + ci z + di = 0. a2i + b2i + c2i 6= 0, i = 1, 2, n
and consider the system
(
a1 x + b1 y + c1 z + d1 = 0
(4.1)
a2 x + b2 y + c2 z + d2 = 0 .
The planes P1 , P2 are identical their equations are equivalent, i.e.,
b1
a1
=
=
a2
b2
c1
d1
= . This means that the system is compatible of rank 1. The planes are parallel
c2
d2
n
1, n
2 are collinear but the system is not compatible, i.e.,
a1
b1
c1
d1
=
=
6= .
a2
b2
c2
d2
5. ORIENTATION OF STRAIGHT LINES AND PLANES
101
The planes intersect along a straight line D the system (4.1) is compatible of
rank 2. In this case the equations (4.1) represent the line D = P1 P2 .
The set of all planes passing through the straight line D is called the pencil of
planes determined by P1 and P2 . The line D is called the axes of the pencil (Fig. 28).
The equation of an arbitrary plane of the pencil is
(4.2) r(a1 x + b1 y + c1 z + d1 ) + s(a2 x + b2 y + c2 z + d2 ) = 0, s, t R, s2 + t2 6= 0,
which also said to be the equation of the pencil.
The set of all planes parallel or coincident to a given plane P1 is called a pencil of
parallel planes, whose equations are of the form:
a1 x + b1 y + c1 z + = 0,
R.
Note that from (4.1) one can deduce easily the equations of the line D = P1 P2
in any of the forms studied previously, using the director vector a
= n
1 n
2 and
M0 (x0 , y0 , z0 ), where (x0 , y0 , z0 ) is a particular solution of the system (4.1)
Fig. 28
Orientation of Straight Lines and Planes
5.1. Oriented Straight Lines

An oriented straight line D is a straight line together with a chosen sense of way
along D. Choosing a sense on D is equivalent to choosing a director vector a
of D.
We say that the pair (D, a
) represents an oriented line.
Any line admits two orientations: the one given by a
(or by any vector k
a, k > 0)
and the orientation given by
a (or by any vector k
a, k < 0). If a
is fixed, we agree
to call the orientation given by a
the positive sense on D, and the other the negative
sense on D. For example the coordinate axes Ox, Oy, Oz are oriented by i, j and k
respectively.
a
The orientation given by a

is uniquelly determined by the versor e =
. Therek
ak
fore an oriented straight line may be seen as a pair (D, e), where D is a line and e
one of the two versors of direction D (Fig. 29).
are called the director angles of
The angles = 6 (
e, i), = 6 (
e, j), = 6 (
e, k)
(D, e), where
e = cos i + cos j + cos k.
102
Fig. 29
The numbers cos , cos , cos are called the director cosines of (D, e). From
k
ek = 1 follows
cos2 + cos2 + cos2 = 1.
If e =
then
,a
= li + mj + nk,
k
ak
cos =
l2
l
,
+ m2 + n2
cos =
l2
m
,
+ m2 + n2
cos =
l2
n
.
+ m 2 + n2
5.2. Orientation of a Plane. Semispaces

An oriented plane is a plane together with a choice of a normal vector i.e. a pair
(P, n
). Intuitively, a plane has two faces; the face which corresponds to the sense of
a fixed normal vector n
is denoted by +, and the opposite one denoted by -. Of
course, a plane may be oriented according to the right hand rule.
Any plane admits two orientations.
Let n
= ai + bj + ck be a fixed normal vector of P : ax + by + cz + d = 0. Denote
f (x, y, z) = ax + by + cz + d. Then E3 = S + P S where (Fig. 30)
S + = {(x, y, z)|f (x, y, z) > 0},
S = {(x, y, z)|f (x, y, z) < 0}
Fig. 30
Angles in Space
6.1. The Angle between Two Oriented Lines

If D1 , D2 are two lines oriented by a
= l1i + m1 j + n1 k and b = l2i + m2 j + n2 k
a, b), given by (Fig.
respectively, the angle of (D1 , a
) and (D2 , b) is the angle = 6 (
31)
h
a, bi
l1 l2 + m1 m2 + n1 n2
p
cos =
=p2
, [0, ].
k
ak kbk
l1 + m21 + n21 l22 + m22 + n22
6. ANGLES IN SPACE
103
Fig. 31
Note that
(1) D1 D2 h
a, bi = 0 l1 l2 + m1 m2 + n1 n2 = 0
(2) D1 kD2 a
b = 0, D1 6= D2
l1
m1
n1
=
=
, D1 6= D2 .
l2
m2
n2
6.2. The Angle between Two Oriented Planes

The dihedral angle of (P1 , n
1 ) and (P2 , n
2 ) is the angle of their normals n
1, n
2
(Fig. 32)
Fig. 32
Denote this angle by . Then

cos =
h
n1 , n
2i
a1 a2 + b1 b2 + c1 c2
p
, [0, ],
=p 2
kn1 k kn2 k
a1 + b21 + c21 a22 + b22 + c22
i = 1, 2.
where n
i = aii + bi j + ci k,
6.3 The Angle between an Oriented Straight Line and

an Oriented Plane
Let (D, a
) be an oriented line and (P, n
) an oriented plane (Fig. 33).
Fig. 33
The angle between (D, a

) and (P, n
) is the angle between a
and the projection
of a
onto P . It turns out that + = 90 where = 6 (
a, n
). Then
sin = cos =
al + bm + cn
h
a, n
i
=
.
2
2
k
ak k
nk
a + b + c2 l2 + m2 + n2
DkP or D P h
a, n
i = 0,
DP
a
n
= 0
b
c
a
=
= .
l
m
n
104
Distances in Space
7.1. The Distance from a Point to a Straight Line

The distance from a point A to the line D passing through M0 (x0 , y0 , z0 ), and of
director vector a
is
k
a M0 Ak
d(A; D) =
k
ak
since d(A; D) = AA0 is the height of a parallelogram constructed on a
and M0 A (if
A
/ D) corresponding to the basis of length k
ak. If A D, the formula is still true
since d(A; D) = 0.
7.2. The Distance from a Point to a Plane

Let M0 (x0 , y0 , z0 ) and P : ax + by + cz + d = 0, M0
/ P (Fig. 34).
Fig. 34
If M1 is the projection of M0 onto P , then d(M0 , P ) = kM1 M0 k.

The collinearity of n
= ai + bj + ck and M1 M0 implies cos(
n, M1 M0 ) =+ 1. This
means
|h
n, M1 M0 i|
= 1.
k
nk kM1 M0 k
Then
d(M0 , P ) = kM1 M0 k =
|h
n, M1 M0 i|
|ax0 + by0 + cz0 + d|
=
.
k
nk
a2 + b2 + c2
The formula still works when M0 P , i.e., d(M0 , P ) = 0.
7.3. The Common Perpendicular of two Noncoplanar Lines

Consider the straight lines D1 and D2 of director vectors a
1 and a
2 respectively.
D1 and D2 are noncoplanar if and only if D1 D2 = and a
1 , a
2 are noncollinear.
In this case there exists a unique straight line D such that D D1 , D D2 and
D D1 6= , D D2 6= . The line D is called the common perpendicular of D1 and
D2 (Fig. 35).
Obviously, a director vector of D is n
=a
1 a
2 .
Fig. 35
7. DISTANCES IN SPACE
105
In order to find out the equations of D we may consider the plane P1 determined
by D and D1 , and the plane P2 determined by D and D2 ; since D = P1 P2 , we are
going to use the equations of P1 and P2 (see Section 4, (4.1)).
For i = 1, 2 let Mi be an arbitrary fixed point on Di ; it turns out that Pi is the
plane determined by Mi and the noncollinear vectors n
and a
i . As we have seen at
the end of Section 3, it follows that
Pi : hMi M , a
i n
i = 0,
i = 1, 2,
where M is the current point of Pi .

Consequently, the equations of the common perpendicular are
hM1 M , a
1 n
i = 0
D:
,
hM2 M , a
2 n
i = 0
where M is the current point of D.
7.4. The Distance between two Straight Lines

Let D1 , D2 be two straight lines. It is known that the number inf {d(M, N )| M
D1 , N D2 } is called the the distance between D1 and D2 and it is denoted by
d(D1 , D2 ). It is also known that d(D1 , D2 ) may be computed as follows (Fig. 36):
1) if D1 D2 6= , then d(D1 , D2 ) = 0;
2) if D1 ||D2 , pick a point M1 D1 ; then d(D1 , D2 ) = d(M1 , D2 );
3) if D1 and D2 are noncoplanar, then d(D1 , D2 ) = d(A, B) = kABk, where
{A} = D D1 , {B} = D D2 , and D is the common perpendicular of D1 and D2 .
Using the equations of D found previously we may compute the coordinates of A
and B for the given lines D1 , D2 . However it is more convenient to regard kABk as
the height of a certain parallelepiped in the following way.
Let M1 , M2 be arbitrary fixed points on D1 and D2 respectively. Consider the
straight line D20 passing through M1 , parallel to D2 . Denote by Q1 the plane determined by D1 and D20 .
Fig. 36
Then Q1 is the plane parallel to D2 , passing through D1 , and kABk is the distance
from B to Q1 . But d(B, Q1 ) = d(M2 , Q1 ) = d(D2 , Q1 ), since D2 kQ1 . On the other
hand d(M2 , Q1 ) is the height of the parallelepiped constructed on the supports of the
vectors M1 M2 , a
1 , a
2 , that corresponds to the basis determined by a
1 , a
2 . Then
d(D1 , D2 ) =
1 a
2 i|
|hM1 M2 , a
.
k
a1 a
2 k
8. PROBLEMS
107
Problems
1. Write the equations of a straight line passing through the point A(1, 1, 2) and
parallel to the straight line D.
y+2
z+1
x4
=
=
2
3
4
x y 3z + 2 = 0
2) D :
2x y + 2z 3 = 0
1) D :
2. Compute the distance from the point A(1, 1, 1) to the straight line D.
x1
y+1
z1
=
=
2
1
3
xy+z =0
2) D :
x+yz =0
1) D :
3. Write the equation of the plane passing through A(1, 1, 1) and perpendicular
to D.
x
y1
z+1
1) D : =
=
1
2
1
xy =0
2) D :
x + 2y z + 1 = 0
4. Write the equation of the plane which passes through the point A(0, 1, 1) and
through the straight line D given as follows:
1) D : x = 4 + 2t, y = 2 + 3t, z = 1 + 4t;
2x y + z + 1 = 0
2) D :
x + y + z = 0.
5. Consider the planes
P : x 2y + 2z 7 = 0, Q : 2x y 2z + 1 = 0, R : 2x + 2y + z 2 = 0.
1) Show that the planes are pairwise perpendicular.
2) Find the common point of these three planes.
3) Find the distance from A(2, 4, 7) to the plane P .
6. Determine the projection of the straight line D onto the plane
P : 2x + 3y + 4z 10 = 0, where D is given by:
x y 3z + 2 = 0
1) D :
2x y + 2z 3 = 0;
x4
y+2
z+1
=
=
;
2
3
4
y+2
z+1
x1
=
=
.
3) D :
3
2
0
2) D :
7. Consider the straight lines D1 , D2 of equations
108
x
y
z
x1
y1
z
= = , D2 :
=
= .
1
2
3
2
1
1
Prove that D1 , D2 are noncoplanar, determine the equations of their common
perpendicular, and the distance between D1 and D2 .
D1 :
8. Write down the equation of the plane P which contains the point A(3, 1, 2)
and the line
2x y 3z 2 = 0
D:
x + 3y z + 4 = 0.
Determine also the distance from A to D.
9. Consider the point A(1, 2, 5) and the plane Q : x + 2y + 2z + 1 = 0. Find the
projection of A onto Q, the symmetric of A w.r.t. Q, and the distance from A to Q.
10. Given the points A(3, 1, 3), B(5, 1, 1), C(0, 4, 3), determine:
1) the parametric equations of the straight lines AB and AC
2) the angle between AB and AC
3) the distance from A to the straight line BC.
11. If the point M (3, 4, 2) is the projection of the origin onto the plane P , determine the equation of P .
12. Determine the parameters , such that the planes
P : 2x y + 3z 1 = 0, Q : x + 2y z + = 0, R : x + y 6z + 10 = 0
1) have exactly one common point;
2) have a common straight line;
3) intersect about three parallel and distinct straight lines.
13. Consider the points A(1, 3, 2), B(1, 2, 1), C(0, 1, 1), D(2, 0, 1), and the
plane P : 2x + y z 1 = 0. Which of the given points is on the same side as the
origin, with respect to the plane P ?
Chapter 7
Transformations of
Coordinate Systems
Changes of Cartesian frames are related to isometries of R3 . Using the group of
isometries of R3 we may define in a natural way the congruence of space figures in
E3 . On the other hand, isometries may be described geometrically. It turns out
that the fundamental isometries are: rotation, translation, symmetry w.r.t. a plane,
symmetry w.r.t. a point, and translation. Any isometry is a composite of isometries
listed above.
Rotations and symmetries are called orthogonal transformations and they actually
correspond to linear orthogonal transformations on V3 ' R3 .
As we have seen, any isometry is the product of a translation and an orthogonal
transformation.
and
Let I = T R be an isometry determined by the frames F = {O, i, j, k}
F0 = {O0 ; i0 , j0 , k0 }. The isometry I is said to be positive (displacement) if the basis
has positive orientation, and negative (antidisplacement) otherwise.
{i, j, k}
Translations and rotations are positive isometries; symmetries are negative isometries.
Translations of Cartesian Frames
We say that the Cartesian frame O0 x0 y 0 z 0 is obtained by the translation of the Cartesian frame Oxyz, if the axes of the new frame O0 x0 y 0 z 0 are parallel and of the same
sense as the axes of the initial frame (Fig. 37).
The translation T is described by:
= k.
O0 = T (O), i0 = T (i) = i, j0 = T (j) = j, k0 = T (k)

109
110
CHAPTER 7. TRANSFORMATIONS OF COORDINATE SYSTEMS
Fig. 37
If the coordinates of the new origin w.r.t. the initial frame are O0 (a, b, c), let us
determine now the relationship between the coordinates x, y, z and x0 , y 0 , z 0 of the
same point M w.r.t each coordinate system. For, note that OM = OO0 + O0 M . In
this relation becomes
terms of the basis {i, j, k},
xi + yj + z k = ai + bj + ck + x0i + y 0 j + z 0 k,
thus
x = x0 + a, y = y 0 + b, z = z 0 + c.
In matrix form we obtain

0

x
a
x
a
1 0
y = b + y0 = b + 0 1
z
c
z0
c
0 0
0
0
x
0 y0 .
1
z0
Particular case. A translation in the plane xOy is described by

x = x0 + a, y = y 0 + b.
Rotations of Cartesian Frames
means passing from this frame to a new one

Rotating the Cartesian frame {O; i, j, k}
{O; i0 , j0 , k0 }, keeping the same origin O and changing the orthonormal basis {i, j, k}
of V3 into another orthonormal basis {i0 , j0 , k0 }, such that {i0 , j0 , k0 } has positive
orientation. If the coordinates of the new basis vectors w.r.t. the old basis are known
(i.e. the matrix of change is known), then we can express a relationship between the
coordinates x, y, z and x0 , y 0 , z 0 of the same point M w.r.t each coordinate system.
This is immediate since
OM = xi + yj + z k = x0 i0 + y 0 j0 + z 0 k0 .
Denote the matrix of change by R = [aij ] M3,3 (R). Then
0
x
x
y = R y0 .
(2.1)
z0
z
2. ROTATIONS OF CARTESIAN FRAMES
111
to {i0 , j0 , k0 }
Since both bases are orthonormal, the linear transformation sending {i, j, k}
is
is an orthogonal transformation R of V3 whose associated matrix w.r.t. {i, j, k}
R. The entries of R can be expressed in terms of inner products of the bases vectors
(see Chapter 1, section 7) as follows
k
R(i) = i0 = hi0 , iii + hi0 , jij + hi0 , ki
k
R(j) = j0 = hj0 , iii + hj0 , jij + hj0 , ki
= k0 = hk0 , iii + hk0 , jij + hk0 , ki
k.
R(k)
Then
a11 = hi0 , ii, a21 = hi0 , ji, a31 = hi0 , ki,
a12 = hj0 , ii, a22 = hj0 , ji, a32 = hj0 , ki,
a13 = hk0 , ii, a23 = hk0 , ji, a33 = hk0 , ki.
Since i0 , j0 , k0 are pairwise orthonormal unit vectors, their coordinates represent

director cosines, which implies that R is an orthogonal matrix, i.e. R t R = I =
t
RR, or R1 = t R. (As we mentioned above, it is a general fact that a linear
transformation of a Euclidean space, which sends an orthonormal basis to another
orthonormal basis, is an orthogonal transformation, therefore its associated matrix
w.r.t. any orthonormal basis is an orthogonal matrix.) An equivalent form of (2.1) is
0
x
x
y0 = t R y .
z0
z
means hi0 , j0 , k0 i = 1. But hi0 , j0 , k0 i = det R,
The positive orientation of {i, j, k}
thus R is a rotation.
If we do not impose the positive orientation of the new frame, then the determinant
of the matrix of change can be -1. This is the case of a symmetry, or rotation followed
by a symmetry.
Particular cases.
1) Rotation about the Oz axis. Denote by the rotation angle (Fig. 38).
112

Fig. 38
From the figure above it follows that

R(i) = i0 = i cos + j sin
R(j) = j0 = i sin + j cos
= k0 = k.
R(k)
Therefore R is described by
x = x0 cos y 0 sin
y = x0 sin + y 0 cos
R:
z = z0.
Obviously, the determinant of the associated matrix is +1, thus R is a positive isometry. In particular, a rotation in the xOy plane, of angle is described by
R:
x = x0 cos y 0 sin
y = x0 sin + y 0 cos .
The composite of a translation and a rotation is called a roto-translation (Fig.

39). Any roto-translation in the xOy plane is characterized by
R:
x = x0 cos y 0 sin + a
y = x0 sin + y 0 cos + b.
Fig. 39
2) Symmetry with respect to a plane. Consider the Cartesian frame {O; i, j, k}

and S the symmetry w.r.t. the plane (O; i, j). Then
= k0 = k.
S(i) = i0 = i; S(j) = j0 = j; S(k)
3. CYLINDRICAL COORDINATES
113
Fig. 40
it follows
By xi + yj + z k = x0i + y 0 j + z 0 k,
S : x = x0 , y = y 0 , z = z 0
or in matrix form

0
x
1 0 0
x
y = 0 1 0 y0 .
z
0 0 1
z0
The determinant of S is -1, thus S is a negative isometry.
Cylindrical Coordinates
Let Oxyz be a fixed Cartesian system of coordinates in E3 . Any point M is uniquelly

determined by its Cartesian coordinates (x, y, z). Denote E3 = E3 \ Oz. The point
M is also characterized by the ordered triplet (, , z), where p
is the distance from
0
the origin to the projection M of M onto the xOy plane, = x2 + y 2 , and is the
angle of the semilines Ox and OM (Fig. 41).
Fig. 41
114
The numbers , , z are called the cylindrical coordinates of the point M . The
cylindrical coordinates and the Cartesian coordinates of M are related by
x = cos
y = sin
z = z.
If we impose > 0, [0, 2), then the above relations give a one-to-one correspondence between E3 \ Oz and (0, ) [0, 2) R.
Coordinate sufaces
= 0 : circular cylinder with generator lines parallel to Oz.
= 0 : semiplane bounded by Oz.
z = z0 : plane parallel to xOy, without the point (0, 0, z0 ).
Coordinate curves
= 0 , z = z0 : semiline parallel to xOy, with the origin on Oz.
= 0 , z = z0 : circle whose center is on Oz, contained in a plane parallel to xOy.
= 0 , = 0 : line parallel to Oz.
The coordinate curves of different types are orthogonal, so the coordinate surfaces
of different types are orthogonal too.
Consider the point M (, , z). The unit vectors e , e , ez tangent to the coordinate
curves passing through M are pairwise orthogonal. The moving orthonormal frame
{M (, , z); e , e , ez } is called cylindrical frame (Fig. 42)
Fig. 42
to the cylindrical frame

The change from the Cartesian frame {O; i, j, k}
{M (, , z); e , e , ez } is described by
e = cos i + sin j
e = sin i + cos j
ez = k.
These formulas are based on the rule which gives the components of a vector w.r.t. an
orthonormal basis as projections of that vector onto the basis vectors. For example
k = cos i + sin j.
e = h
e , iii + h
e , jij + h
e , ki
4. SPHERICAL COORDINATES
115
Spherical Coordinates
Sometimes it is convenient to characterize the position of a point M E3 = E3 \ Oz

using the ordered triplet (r, , ), where r represents the distance d(O, M ), is the
angle of the semilines Ox and OM 0 , and is the angle between Oz and OM . M 0
denotes the projection of M onto xOy (Fig. 43).
Fig. 43
The numbers r, , are called the spherical coordinates of the point M . The
Cartesian and the spherical coordinates of M are related by
x = r sin cos
y = r sin sin
z = z cos .
These formulas provide a one-to-one correspondence between the sets E3 \ Oz and
(0, ) (0, ) [0, 2).
Coordinate sufaces
r = r0 : sphere of radius r, center at O, without the north and the south poles.
= 0 : semiplane bounded by Oz.
= 0 : semicone without vertex (the origin).
Coordinate curves
= 0 , = 0 : semiline with origin at O.
r = r0 , = 0 : circle whose center is on Oz, contained in a plane parallel to xOy.
= 0 , r = r0 : open semicircle.
The coordinate curves of different types are orthogonal, so the coordinate surfaces
of different types are orthogonal too.
Consider the point M (r, , ). The unit vectors er , e , e tangent to the coordinate
curves passing through M are pairwise orthogonal. The moving orthonormal frame
{M (r, , ); er , e , e } is called spherical frame (Fig.44).
116
Fig. 44
to the spherical frame

The change from the Cartesian frame {O; i, j, k}
{M (r, , ); er , e , e } is described by
er = sin cos i + sin sin j + cos k

e = cos cos i + cos sin j sin k
e = sin i + cos j.
Problems
1. If the spherical coordinates of a point are r = 5, = 600 , = 450 , determine

its cylindrical coordinates and its Cartesian coordinates.
2. Express the cylindrical coordinates of an arbitrary point in terms of its spherical
coordinates.
3. Express the spherical coordinates of an arbitrary point in terms of its cylindrical
coordinates.
4. Determine the change of coordinates formulas for the following transformation:
O = O0 ; the Ox0 axis is contained in the xOy plane, and the angle (as oriented straight
lines) between Ox0 and Ox is acute; the Oz 0 axis has the same direction and the same
the Oy 0 axis is such that the new Cartesian system
sense as the vector v = 2i + j + 2k;
0 0 0
Ox y z has positive orientation.
5. Consider the Cartesian frame Oxyz, and the the points A(3, 0, 0), B(0, 2, 0),
C(0, 0, 6). After a rotation we obtain the Cartesian frame Ox0 y 0 z 0 such that: Oz 0 has
the direction and sense of the height OO0 of the tetrahedron OABC; Oy 0 is parallel
to O0 A0 , where A0 is the orthogonal projection of A onto (ABC), and the Ox0 axis is
such that the the system O0 x0 y 0 z 0 has positive orientation.
Write the matrix of the rotation and determine its invariant direction (its real
1-dimensional eigenspace).
6. Let D be a straight line of equations
D:
x2
y+1
z3
=
=
.
2
1
2
Let i0 be the unit director vector of D (choose a sense on D). Let j 0 be a versor
contained in yOz and perpendicular on D, and the versor k0 such that {i0 , j0 , k0 } is
an orthonormal basis.
5. PROBLEMS
117
Find out the change of basis formulas and compare the orientations of the two
frames.
7. Consider three points given by their cylindrical coordinates:
A(5,
4
5
, 4), B(7,
, 2), C(2,
, 1).
3
3
6
Show that A and B belong to a plane passing through Oz; determine the Cartesian
coordinates of A and C, and the distance d(A, C).
8. Write the following equations in spherical coordinates.
(x2 + y 2 + z 2 )2 = 3(x2 + y 2 );
(x2 + y 2 + z 2 )2 (x2 + y 2 ) = 4x2 y 2 .
118
Exam Samples
I.
1. Eigenvalues and eigenvectors; characteristic polynomial.
2. Determine the equation of the plane equidistant from
D1 :
x1
y+1
z
=
= ,
1
1
1
D2 :
x
y
z1
=
=
.
2
1
1
3. Find the canonical form of x1 x2 + x2 x3 + x3 x1 .
II.
1. Bases and dimension.
2. Prove the identity
h
a, ci
h
a b, c di =
hb, ci

h
a, di
.

hb, di
3. Determine the canonical form of the matrix
5 6 6
1
4
2 .
3 6 4
III.
1. Scalar product and vector product of free vectors.
2. Choose the basis {1, x, x2 , x3 } in the vector space Vof all real polynomials of degree 3. Let D denote the differentation operator and T : V V, T (p)(x) = xp0 (x).
Determine the matrix of each of the following linear transformations: D, T , DT , T D.
Find the eigenvalues and eigenvectors of T .
3. Let V be the real vector space of all polynomials of order 2 and
Z 1Z 1
A(x, y) =
x(t)y(s)dtds.
0
119
120
EXAM SAMPLES
Show that A is a bilinear form which is symmetric and positive semidefinite. Find
the matrix of A with respect to the basis {1, t, t2 }.
IV.
1. Euclidean vector spaces.
2. Show that if the vector b is perpendicular to c, then (
a b) c = bh
a, ci.
Give sufficient conditions for the orthogonality of (

a b) c and c.
3. Let V be the vector space of all real polynomials p(x) of degree n. Define
T : V V, T (p) = q, q(t) = p(t+1). Determine the eigenvalues and the eigenvectors
of T .
V.
1. Linear transformations: general properties, kernel and image.
2. Show that the area of a figure F contained in the plane P : z = px + qy + l and
the area of its projection F onto the xOy plane are related by
p
S(F ) = 1 + p2 + q 2 S(F ).
Z b
3. Show that hf, gi =
f (t)g(t)dt is a scalar product on the real vector of all
a
continuous functions on [a, b].
VI.
1. Scalar product and mixed product of free vectors.
2. Let V and W be vector spaces, each of dimension 2, and {e1 , e2 } a basis in
Vand {f1 , f2 } a basis in W. Let T : V W, T (e1 + e2 ) = 3f1 + 9f2 , T (3e1 + 2e2 ) =
7f1 + 23f2 , be a linear transformation. Compute T (e2 e1 ) and determine the nullity
and rank of T . Determine T 1 .
3. Determine an orthogonal matrix C which reduces the quadratic form Q(x) =
2x21 + 4x1 x2 + 5x22 to a diagonal form.
VII.
1. The matrix of a linear transformation. Particular endomorphisms.
2. Find the equation of the plane P passing through the point (1,1,1), which is
perpendicular to the planes Q1 : x + y + z + 1 = 0, Q2 : x y + z = 0. Using the
normal vectors of P, Q1 , Q2 and the Gram-Schmidt procedure, find an orthonormal
basis of R3 .
3. Determine the canonical form of
Q(x) = 3x21 5x22 7x23 8x1 x2 + 8x2 x3 .
EXAM SAMPLES
121
VIII.
1. Free vectors: collinearity, coplanarity.
2. Let V be the vector space of all continuous functions on (, ) and such that
Zx
Zx
the integral
f (t)dt exists for all x. Define T : V V, g = T (f ), g(x) =
f (t)dt.
Prove that every positive real number is an eigenvalue for T and determine the
eigenfunctions corresponding to .
3. Let Q : R3 R, Q(x) = x1 x2 + x2 x3 + x3 x1 . Determine the canonical form
of Q and the image of
x1 1
x2
x3
=
=
D:
1
1
2
by Q.
IX.
1. Vectors spaces. Vector subspaces.
2. Let T : C3 C3 be represented by the matrix
1 + 2i
i
4
12
1+i
1i
b
.
4
12
1
2i
c
4
12
Determine a, b and c so that T is unitary (for C3 with the usual inner product).
3. Find the projection of
D:
x1
y
z2
= =
2
1
2
onto
P :x+y+z =0
and the symmetric of P with respect to D.
X.
1. The spectrum of endomorphisms on Euclidean spaces.
2. Determine the relation between a, b, c, d, , , , , , such that the three planes
ax + by + cz + d = 0, x + y + z + = 0, (ax + by + cz) + (x + y + z) = 0
have no points in common.
3. Let {x, y} be a linearly independent set in a Euclidean space V.
Define f : R R by f () = kx yk.
122
EXAM SAMPLES
(a) Where does f take on its minimum value ?

(b) Give a geometric interpretation of (a). What happens if f : C R ?
XI.
1. Diagonal form of an endomorphism.
2. Let P : x + y 2z = 0. Find the equation of the plane Q symmetric to P about
xOy (about the origin O).
3. Let A : R4 R4 R,
A(x, y) = x1 y2 x2 y1 + x1 y3 x3 y1 + x1 y4 x4 y1 + x4 y4 .
Find the matrix of A with respect to the basis f1 = (1, 1, 0, 0), f2 = (0, 1, 1, 0),
f3 = (0, 1, 0, 1), f4 = (1, 0, 0, 1).
XII.
1. Basic facts on straight lines and planes.
2. Let P4 denote the space of all polynomials in t of degree
at most
3. Find the
d
dx
2
matrix representation of T : P4 P4 , y = T x, y(t) =
(t 1)
with respect
dt
dt
3
1 5
3
to the basis B = 1, t, t2 , t3 t .
2
2 2
2
3. LetP
V consist of all infinite sequences
P x = {xn } of real numbers for which
the series
x2n converges. Define hx, yi =
xn yn . Prove that this series converges
1
1
absolutely and hx, yi is a scalar product. Compute hx, yi if xn = and yn =
.
n
n+1
XIII.
1. Quadratic forms.
2. Let a
, b, c V3 such that b perpendicular to c. Show that (
a b) c = bh
a, ci.
3. Let V = C[0, T ] and define (Px)(t) = x(0)(1 t), for 0 t T . Show that P
is a projection and determine the range of P.
XIV.
1. Polar, cylindrical and spherical coordinates.
2. ZConsider the linear transformation T : V V given by y = T x, where
2
y(t) =
x(0)4 cos 2(t s)ds, and V = L{1, cos s, cos 2s, sin s, sin 2s}.
0
(a) Express T as a matrix.

(b) Is T one-to-one ?
EXAM SAMPLES
123
(c) Does it map V onto itself ?

3. Let V be the vector space of all real-valued functions defined on the real line.
Which of the subsets {1, eax , xeax }, {1, cos 2x, sin2 x} is linearly independent in V?
Compute the dimension of the subspace spanned by each subset.
XV.
1. Equations of a straight line.
2. Let V = L2 [, ], and V1 = L(A1 ), V2 = L(A2 ), where A1 = {1, cos t, cos 2t, . . .},
A2 = {sin t, sin 2t, . . .}. Show that A1 , A2 are linearly independent, and the sum of
V1 and V2 is direct.
Z
(1 + cos(x t))f (t)dt. Find a basis for
3. Let T : V V, g = T (f ), g(x) =
T (V).
(a) Determine the kernel of T .
(b) Find the eigenvalues of T .
XVI.
1. Equation of a plane in space.
2. Let V be the real vector space of all functions of the form x(t) = a cos(t + ),
where is fixed. Show that B = {cos t, sin t} is a basis of V. Give examples of
linear transformations on V.
3. Let V be the real Euclidean space of real polynomial functions on [-1,1]. Determine which of the following linear transformations is symmetric or skew symmetric:
T (f )(x) = f (x), T (f )(x) = f (x) + f (x), T (f )(x) = f (x) f (x).
XVII.
1. Transformations of Cartesian frames.
2. Let V denote the real vector space of all polynomials in t, of degree at most
dx
four, and define T : V V by T = D2 + 2D + I, where Dx =
(differential
dt
operator).
(a) Represent T by a matrix T w.r.t. the basis {1, t, t2 , t3 , t4 }.
(b) Represent T 2 by a matrix.
n
X
i=1
3. Let x = (x1 , .. . , xn ), y = (y1 , . . . , yn ) Rn . Find out whether hx, yi =

n
n
X
xi
yj , hx, yi =
xi yi are scalar products or not. When hx, yi is not a scalar
y=1
i=1
product, point out the axioms which are not fulfilled.
XVIII.
1. Jordan form of an endomorphism.
124
EXAM SAMPLES
2. What is the locus of points with the property that the ratio of distances to two
given planes is constant?
Z
3.
In the linear space of all real polynomials, with scalar product hx, yi =
x(t)y(t)dt, let xn (t) = tn for n = 0, 1, 2, . . .. Prove that the functions y0 (t) = 1,
y1 (t) = 3(2t 1), y2 (t) = 5(6t2 6t + 1) form an orthonormal set spanning the
same subspace as {x0 , x1 , x2 }.
1
XIX.
1. Free vectors: addition, multiplication of a vector by a scalar.
2. Let V be the vector
Z x space of all continuous functions defined on (, ), and
such that the integral
tf (t)dt exists for all real numbers x. Define T : V V,
Z x
g = T (f ), g(x) =
tf (t)dt. Prove that every negative is a proper value for T
and determine the eigenvectors corresponding to .

3. Determine the signature of the quadratic form
Q(x) = 3x21 5x22 + x23 2x1 x2 + x2 x3 .
XX.
1. Linear transformations on Euclidean spaces.
2. Show that the locus of points equidistant from three pairwise non-parallel
planes is a straight line.
3. Show that the quadratic forms
Q(x) = 3x21 + 4x22 + 5x23 + 4x1 x2 4x2 x3 ,
Q(x) = 2x21 + 5x22 + 5x23 + 4x1 x2 4x1 x3 8x2 x3
are positive definite.
XXI.
1. Isometries.
2. Write the equations of the straight line D passing through the point (1,1,1)
and parallel to the planes P : x y + z = 0, Q : x + 2y z = 0. Find the points of P
which are at the distance 2 from D.
3. Define T : C2 C2 by
y1
y2
a b
c d
x1
x2
EXAM SAMPLES
125
where a, b, c, d are complex numbers. Determine necessary and sufficient conditions

on a, b, c, d such that T is self-adjoint, or unitary.
XXII.
1. Polynomials of matrices. Functions of matrices.
2. Find out the angles between the coordinate axes and the plane P : x+y+2z = 0.
x
y1
z
Determine the symmetric of P with respect to the line D : =
= .
1
1
1
3. Let x, y be vectors in a Euclidean space Vand assume that
kx + (1 )yk = kxk,
[0, 1].
Show that x = y.
XXIII.
1. Orthogonality. The Gram-Schmidt orthogonalization procedure.
2. Prove that: if the vectors a
and b are perpendicular to the vector c, then
(
a b) c = 0. Give sufficient conditions for the collinearity of (
a b) c and c.
3. Find the canonical form of T : R3 R3 ,
7
4 1
7 1 .
T = 4
4 4
4
XXIV.
1. Linear dependence and independence.
2. Find the angles between the coordinate planes and the straight line
D:
x1
y
z+2
=
=
.
1
2
1
Determine the equations of the symmetric of D with respect to the plane

P : x + y + z = 0.
3. Determine the canonical form of T : R3 R3 given by
0 1
0
0
1 .
T = 0
1 3
3
126
EXAM SAMPLES
XXV.
1. Bilinear forms.
2. Prove that the straight lines joining the mid-points of the opposite edges of a
tetrahedron intersect at one point. Express the coordinates of this point in terms of
the coordinates of the vertices of the tetrahedron.
3. Show that the set of all functions xn (t) = eint , n = 0, 1, 2, . . . is linearly
independent in L2 [0, 2].
XXVI.
1. Eigenvalues and eigenvectors of an endomorphism.
2. Given the point A(0, 1, 2) and the line
x+y =0
D:
x + z + 1 = 0,
detemine the symmetric of A with respect to D and the symmetric of D w.r.t. A.
3. Let Q : R3 R,
Q(x) = 4x21 + x22 + 9x23 + 4x1 x2 12x1 x3 6x2 x3 .
Reduce Q to the canonical expression using the eigenvalues method; determine
the corresponding basis.
Use another method of reduction to the canonical form and check the inertia law.
XXVII.
1. The common perpendicular of two noncoplanar straight lines. The distance
between two noncoplanar straight lines.
2. Consider the real vector space V = C (R) and D : V V, D(f ) = f 0 .
Find the eigenvalues and the eigenvectors of D. Is = 2 an eigenvalue? What is the
dimension of S(2)?
3. Let T : R2 [X] R2 [X] be a linear map such that
T (1 + X) = X 2 , T (X + X 2 ) = 1 X, T (1 + X + X 2 ) = X + X 2 .
Find the matrix associated to T w.r.t. canonical basis of R2 [X]. Determine the
dimension and a basis for each of Ker(T ) and Im(T ).
XXVIII.
1. Endomorphisms of Euclidean vector spaces.
2. Consider v = (14, 3, 6) R3 and S = {v1 , v2 , v3 }, where v1 = (3, 0, 7),
v2 = (1, 4, 3), v3 = (2, 2, 2). Determine the orthogonal projection w of v on Span S
and the vector w .
EXAM SAMPLES
127
3. Let T : R3 R3 be an endomorphism, and
4
6 0
T = 3 5 0
3 6 1
the matrix of T w.r.t. the canonical basis. Find the diagonal form of T and the
corresponding basis. Is T injective? Is T surjective?
XXIX.
1. The vector space of free vectors: mixed product - properties, geometric interpretation.
2. Let T : R3 R3 be an endomorphism such that T (1, 0, 0) = (3, 1, 0),
T (0, 0, 1) = (1, 2, 0), Ker T = {(, 2, 3) | R}. Determine a basis for Ker T
and Im T respectively; determine also the matrix of T w.r.t. the canonical basis.
3. Given
1
A = 1
2
1
3
1
2
1 ,
1
find the canonical expression of the quadratic form Q : R3 R, Q(x) = t XAX and
the correspoding change of coordinates.
XXX.
1. Matrix polynomials.
2. Consider the vectors v1 = (3, 2, 1), v2 = (1, 2, 1), v3 = (1, 0, 2). Prove that
there exists a unique linear form f : R3 R such that f (v1 ) = 5, f (v2 ) = 3,
f (v3 ) = 6. Write f (x) in terms of the components of x, for x arbitrary in R3 , and
determine an orthonormal basis of Ker f . Is Im f a proper subspace of R?
3. Let M (1, 2, 5) and Q : x + 2y + 2z + 1 = 0. Determine the projection of the
point M onto the plane Q, the symmetric of M w.r.t. Q, the distance from M to Q,
and the symmetric of Q w.r.t. M .
XXXI.
1. Bilinear forms.
2. Use the Cayley-Hamilton theorem to determine the inverse of the matrix
1 2
2
A = 2 1 2
2
2 1
and the value of the matrix polynomial R(A) = A3 + 3A2 8A 27I.
128
EXAM SAMPLES
3. Let V = C 0 [0, 2], S = {f0 , f1 , . . . , fk . . .} V , where f0 (x) = 1, f2n1 (x) =

cos nx, f2n (x) = sin nx, n N , x [0, ].
Prove that S is linearly independent and determine an orthonormal basis of
Span S.
XXXII.
1. Quadratic forms.
2. If a
, b, c V3 are noncoplanar vectors such that h
a, bi 6= 0, show that
(b c)
a
b a
E=
,
h
a, bi h
a, b ci
does not depend on a
, b, c.
3. Determine the eigenvalues and the eigenvectors of the linear transformation
T : R3 R3 , T (x) = (x1 2x2 x3 , x1 + x2 + x3 , x1 x3 ).
(a) Is T diagonalizable? (b) Find orthonormal bases in KerT and ImT .
XXXIII.
1. Kernel and image of a linear transformation.
2. Given the vectors
b = i j + k,
c = 3i + j k,
a
= i j + 3k,
find the value of R such that a
For = 2, determine the altitude of the parallelipiped constructed on some representatives with common origin of the vectors a
, b, c, corresponding to the base whose
sides are the representatives of a
and b.
3. Let
4
T = 2
6
2
1
3
6
3
9
be the matrix associated to the endomorphism T : R3 R3 w.r.t. the canonical

basis of R3 .
(a) Find out whether T admits a Jordan form or not.
(b) Compute T n , n N .
(c) Is T an injective endomorphism?
XXXIV.
1. Collinearity and coplanarity in V3 .
EXAM SAMPLES
129
2. (a) Write the analytic expression of the bilinear form A : R3 R3 R, if the

associated matrix w.r.t. the canonical basis is
1 2 1
A = 1 2 3 .
3 0 1
(b) Is A a symmetric bilinear form?
(c) Let Q : R3 R be a quadratic form,
Q(x) = x21 + 8x22 + x23 + 16x1 x2 + 4x1 x3 + 4x2 x3 .
Determine the canonical form of Q and the corresponding basis, using Jacobis
method.
3. Prove that the straight lines
D1 :
y
z
x
= = ,
1
2
3
x1
y1
z
=
=
2
1
1
are noncoplanar, determine the equations of their common perpendicular and the
distance between D1 and D2 .
D2 :
XXXV.
1. Linear dependence. Linear independence.
2. Consider the straight line D and the plane P of equations
D:
y
z
x
= =
1
2
1
P : x 2y + z = 0.
(a) Find the projection of D onto P .
(b) If D and P are viewed as vector subspaces of R3 , determine D + P , D P
and an orthonormal basis of P .
3. Given the matrix
1
A= 0
0
0
2
0
1
0 ,
3
use the Cayley - Hamilton theorem in order to:

(a) write A1 as a polynomial of degree 2 in A;
(b) show that
1 0 (3n 1)/2
.
0
An = 0 2n
0 0
3n
130
Bibliography
[1] M.Artin, Algebra, Prentice-Hall, 1991.
[2] Gh.Atanasiu, Gh.Munteanu, M.Postolache, Algebr
a liniar
a, geometrie analitic
a si diferential
a (in Romanian), Editura All, Bucuresti, 1994.
a liniar
a si diferential
a (in Romanian),
[3] V. Balan, Algebr
Universitatea Politehnica Bucuresti, 1998.
[4] R.M.Bowen, C.C.Wang, Introduction to Vectors and Tensors, vol. 1-2,
Plenum Press, New York, 1976.
[5] V.Brnzanescu, O.Stanasila, Matematici speciale (in Romanian), Editura
All, Bucuresti, 1994.
[6] J.Dieudonne, Linear Algebra and Geometry, Paris, Hermann, 1969.
a liniar
a (in Romanian), Geometry Balkan
[7] O.Dogaru, M.Doroftei, Algebr
Press, Bucuresti, 1998.
[8] Gh.Th.Gheorghiu, Algebr
a liniar
a si diferential
a si programare (in Romanian), Editura Didactica si Pedagogica, Bucuresti, 1977.
[9] S.Ianus, Curs de geometrie diferential
a (in Romanian), Universitatea Bucuresti, 1981.
[10] I.D.Ion, R.Nicolae, Algebra (in Romanian), Editura Didactica si Pedagogica,
Bucuresti, 1981.
[11] N.Jacobson, Lectures in Abstract Algebra, II - Linear algebra, SpringerVerlag, 1975.
[12] W.Klingenberg, Lineare Algebra und Geometrie, Springer-Verlag, Berlin,
1990.
[13] S. Lang, Algebra, Addison-Wesley, 1984.
[14] A.V.Pogorelov, Analytic Geometry, Mir Publishers, Moscow, 1961.
[15] C.Radu, Algebr
a liniar
a si diferential
a (in Romanian),
Editura All, Bucuresti, 1996.
131
132
BIBLIOGRAPHY
[16] C.Radu, C.Dragusin, L.Dragusin, Aplicatii de algebr
a, geometrie si matematici speciale (in Romanian), Editura Didactica si Pedagogica, Bucuresti,
1991.
[17] L.Smith, Linear Algebra, Springer-Verlag, 1978.
[18] N.Soare, Curs de geometrie (in Romanian), Universitatea Bucuresti, 1996.
[19] C.Udriste, Linear Algebra, University Politehnica of Bucharest, 1991-1992.
[20] C.Udriste, Problems in Algebra, Geometry and Differential Equations I, II,
University Politehnica of Bucharest, 1992.
[21] C.Udriste, Aplicatii de algebr
a, geometrie si ecuatii diferetiale (in Romanian), Editura Didactica si Pedagogica, Bucuresti, 1993.
[22] C.Udriste, Algebr
a liniar
a, Geometrie analitic
a (in Romanian), Geometry
Balkan Press, Bucuresti, 1996.
[23] C.Udriste, O.Dogaru, Algebr
a liniar
a, Geometrie analitic
a (in Romanian),
Universitatea Politehnica Bucuresti, 1991.
[24] C.Udriste, C.Radu, C.Dicu, O.Malancioiu, Probleme de algebr
a, geometrie
si ecuatii diferentiale (in Romanian), Editura Didactica si Pedagogica, Bucuresti, 1981.
[25] C.Udriste, C.Radu, C.Dicu, O.Malancioiu, Algebr
a, geometrie si ecuatii
diferentiale (in Romanian), Editura Didactica si Pedagogica, Bucuresti,
1982.

Linear Algebra

Uploaded by

Copyright:

Available Formats

Linear Algebra

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Linear Algebra

Uploaded by

Copyright:

Available Formats

Differential Geometry - Dynamical Systems

Geometry Balkan Press

Differential Geometry - Dynamical Systems * Monographs

Analytic and Differential Geometry

Includes bibliographical references.

c Balkan Society of Geometers, Differential Geometry - Dynamical Systems

3 Eigenvectors and Eigenvalues

4 Bilinear Forms. Quadratic Forms

6 Straight Lines and Planes in Space

7 Transformations of Coordinate Systems

M1,n (R) = v = [a1 , . . . , an ] ; aj R, j = 1, n

CHAPTER 1. VECTOR SPACES

DEFINITION 1.1 A field K is a set endowed with two laws of composition:

called addition and respectively multiplication, satisfying the following axioms :

(iii) the element 1 K acts as identity:

(iv) the distributive laws hold:

Examples of vector spaces

CHAPTER 1. VECTOR SPACES

COROLLARY 1.4 Let V be a Kvector space. Then:

Throughout this paragraph V denotes a Kvector space and S a nonempty subset of

(i) If W is a vector subspace, then 0V W.

(ii) Let S be a nonempty subset (possibly infinite) of V. A vector of the form

is called a finite linear combination of vectors of S.

(llj ) wj is also a finite linear combination with

elements of S. It follows that Span S is a vector subspace, by Remark 2.2(iii). QED

CHAPTER 1. VECTOR SPACES

(i) Span (W1 W2 ) = W1 + W2 .

(ii) Span (S1 S2 ) = Span S1 + Span S2 .

3. LINEAR DEPENDENCE. LINEAR INDEPENDENCE

Linear Dependence. Linear Independence

Let V be a Kvector space and S V, S 6= .

(ii) An infinite subset S of V is called linearly dependent if there exists a finite

CHAPTER 1. VECTOR SPACES

(iv) If f1 (x) = ex ; f2 (x) = ex ; f3 (x) = sh x, then {f1 , f2 , f3 } is linearly dependent

We may assume without loss of generality that k1 6= 0. Then

Conversely, if w S and w Span (S \ {w}), then it can be written as w =

for some q 1, cj K, wj S \ {w}. It follows that

therefore the set {w, w1 , . . . , wq } is linearly dependent. Then S is linearly dependent

aij vi Span S, j = 1, . . . , p + 1 and k1 , . . . , kp+1 K such

We replace the vectors wj to obtain

4. BASES AND DIMENSION

since v1 , . . . , vp are linearly independent.

Bases and Dimension

V is a K vector space. Let B V be a linearly independent set. Then Span B

If S is linearly independent, then S is a basis. If S is not linearly independent,

CHAPTER 1. VECTOR SPACES

en = (0, 0, . . . , 1) form a basis of Kn ,

(ii) B = {Eij |1 i m, 1 j n} is a basis of Mm,n (K), where Eij is the matrix

ISOMORPHISMS. CHANGE OF COORDINATES

CHAPTER 1. VECTOR SPACES

PROPOSITION 5.1 The set B = {v1 , . . . , vn } V is a basis if and only if every

is called a linear transformation (or a vector space morphism).

ISOMORPHISMS. CHANGE OF COORDINATES

(i.e. T is compatible with linear combinations)

CHAPTER 1. VECTOR SPACES

maximal number of linearly independent columns.

Then B0 is another basis of V if and only if det [cij ] 6= 0.

DEFINITION 5.10 If B0 in the previous theorem is a basis, then the matrix C is

By the uniqueness of the representation of x with respect to the basis B, we get

6. EUCLIDEAN VECTOR SPACES