Lecture-Note

Presented to students of the second year Maths.

By

D JAMEL BELLAOUAR
University 8 Mai 1945 Guelma

Theme

Algebra 4, summary of lessons, examples,

solved and suggested exercises

University Year : 2023/2024

 Thanks

I wish to express my sincere thanks to my colleague Rafael Jakimczuk (Universidad

Nacional de Luján Buenos Aires, Argentina) for valuable comments on refinements of an
earlier version. Special thanks are given to Özen Özer (University of Kirklareli, Turkey)
for the careful proof reading of this lecture-note and for many helpful suggestions.

i
 Dedications

I d e
di c
at
e t hi s L e
ctur e-Not
e of Al
g
ebr a 4 t
o :

M y Stu d ents

M at h em ati ci an s

R e
a d er s

L es p ar ol
es s' en-

vol
ent m ai s l
es é
crits r est
ent... á c
et eff
et c
e tr avail.

ii
 Table of contents

Introduction 1

1 Linear forms, Duality (vector space and its dual space 2

1.1 Vector space (a summary of lessons) . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Linear mappings and linear forms . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Proposed Problems on linear forms . . . . . . . . . . . . . . . . . . . . . . . . 12

2 Bilinear forms over a vector space 14

2.1 Bilinear forms (Definitions) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 Orthogonal matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.1 Matrix norms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.2 Scalar Product (Inner product) over a real vector space . . . . . . . . 19
2.3 Gram-Schmidt Orthonormalization Theorem . . . . . . . . . . . . . . . . . . 21
2.4 Diagonalization of real symmetric matrices . . . . . . . . . . . . . . . . . . . 22
2.5 Proposed Problems on bilinear forms . . . . . . . . . . . . . . . . . . . . . . . 24

3 Symmetric bilinear forms and quadratic forms 27

3.1 Relation between a quadratic form and its polar form . . . . . . . . . . . . . 27
3.2 Quadratic forms over Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3 Quadratic forms over an arbitrary vector space . . . . . . . . . . . . . . . . . 30
3.4 Orthogonalization method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.5 Definitions and results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.5.1 Is the isotropic cone a vector space ? . . . . . . . . . . . . . . . . . . . 33
3.5.2 When is a quadratic form surjective ? . . . . . . . . . . . . . . . . . . 37
3.6 Gauss Decomposition (Silvester’s Theorem) . . . . . . . . . . . . . . . . . . . 38
3.7 Proposed problems (quadratic forms) . . . . . . . . . . . . . . . . . . . . . . 44

4 Introduction to hermitian space 48

4.1 Sesquilinear forms and hermitian quadratic forms . . . . . . . . . . . . . . . 48
4.1.1 Definitions and examples . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.1.2 Hermitian matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

iii
 4.2 Hermitian quadratic forms over Cn . . . . . . . . . . . . . . . . . . . . . . . . 52
4.3 Gauss decomposition for hermitian forms . . . . . . . . . . . . . . . . . . . 53

5 Spectral decomposition of self-adjoint linear mappings 57

5.1 Scalar Product over a complex vector space . . . . . . . . . . . . . . . . . . . 57
5.2 Spectral decomposition of self-adjoint linear mapping . . . . . . . . . . . . . 58
5.3 Proposed problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

6 Solutions to some exercises and problems 64

Conclusion 96

Bibliography 97

iv
 Notations

The following notations allow the reader to clearly understand the content of this ma-
nuscript.
• α, β, λ, µ are scalars and x, y, u, v, w, u0 , v 0 , w0 are vectors.
• ⊕ This is an internal composition law.
• ⊗ This is an external composition law.
• E a vector space over K.
• Kn The field of n-tuples of real or complex numbers.
• (x1 , x2 , ..., xn ) An element of Kn (vector).
• Kn [x] The vector space of all polynomial of degree not exceeding n with real or
complex coefficients.
• C ([a, b] , R) The vector space of all continuous functions on [a, b] .
• C ∞ ([a, b] , R) The v. space of all infinitely differentiable functions on [a, b] .
• Mn (K) The vector space of all n by n real (or complex) matrices.
• Sn (K) The vector space of all n by n real (or complex) symmetric matrices.
• An (K) The vector space of all n by n real (or complex) skew-symmetric matrices.
• GLn (K) The vector space of all n by n invertible matrices.
• Mf (B) The matrix of the mapping f with respect to the basis B.
• P The passage matrix.
• {e1 , e2 , ..., en } In general denotes for the canonical basis.
• V ect {u1 , u2 , ..., un } The vector space of all linear combinations of the vectors ui (1 ≤
1 ≤ n).
• ker f The kernel of the linear mapping f or the kernel of the bilinear symmetric form
f.
• Im f The vector subspace {f (v) : v ∈ E} .
• F ⊕ G Direct sum between F and G.
• L2 (E) The v. space 1 of all bilinear forms on E.
• L (E, F ) The v. space of all linear mappings from E to F .
• L (E, K) The v. space of all linear mappings from E to K.
• q or Q Quadratic forms.
• C The isotropic cone ; C = {v ∈ E : f (v, v) = 0}.
1. v. space means vector space.

v
• E ∗ dual v. space of a vector space E.
• Φ∗ The dual mapping of Φ.
• S2 (E) The v. subspace of all symmetric bilinear forms on E.
• A2 (E) The v. subspace of all skew-symmetric bilinear forms on E.
• Q2 (E) The set of all quadratic forms on E.
• diag {a1 , a2 , ..., an } Diagonal matrix whose diagonal entries are a1 , a2 , ..., an .
• tr (A) The trace of an n by n matrix A.
• Sp (A) The spectral set of A = The set of eigenvalues of A.
• z The conjugate of the vector z ∈ Cn .
• i The imaginary pure number (i2 = −1).
• I The identity matrix.
• Re (z) The real part of a complex number z.
• At The transpose of a matrix A.
• det (A) Determinant of a square matrix A.
• A∗ The transpose conjugate of a complex n by n matrix A.
• kvk the norm of the vector v.
• hu, vi The scalar product (or inner product) between the vectors u and v.

vi
 General Introduction

T his work is the fruit of teaching of this subject at the University of 8 Mai 45
Guelma. It is intended for students of the 2nd year mathematics. This volume
is devoted to a part of the program of Algebra 4 (bilinear forms, quadratic
forms, sesquilinear forms and hermitian forms. One can see [1], [2], [3], [4], [5], [6]).
Each chapter begins with a clear presentation of definitions, lemmas and theorems,
illustrated with numerous examples. This is followed by a graduated number of a set of
solved exercises.
The course summary is sufficiently developed so that everyone will find the results
they need to solve proposed problems. Although the large number of additional pro-
blems makes their solution difficult, special importance should nevertheless be given to
those presented in the first two chapters. After engaging in it, the student will feel more
confident.
I had been teaching this material by French from 2012 to 2016. Then I have taught it to
students a second time, but by English, starting 2020 to now. Being the first subject presen-
ted to students at the beginning of their education, they gladly accepted presenting it in
English language. Indeed, this course, which is based on bilinear forms (linearity from the
right and those from the left), is a continuation study of Algebra II taught in the first year
M.I. It is composed of five chapters. In Chapter 1, we recall some definitions and give wi-
thout proof some classical results on vector spaces and linear mappings, that is, we list in
the this chapter the basic notions on a vector space and its dual space. In chapter 2 we deal
with bilinear forms over a real vector space. It is not possible to understand such proper-
ties without examining the related concepts of linear forms. More precisely, this chapter
describes the most properties of bilinear forms on a vector space and gives examples of
the three most common types of such forms as well as symmetric, skew-symmetric and
alternating bilinear forms. Chapter 3 deals with the spectral decomposition of self-adjoint
linear mappings. The important condition of nondegeneracy for a bilinear form, Gauss
decomposition theorem and the orthogonal basis for a symmetric bilinear form are the
subject of Chapter 4. An introduction to Hermitian space is given in Chapter 5. At the end
of this lecture-note, the reader will find a conclusion and a bibliography.

1
 C HAPTER 1

L INEAR FORMS , D UALITY ( VECTOR

SPACE AND ITS DUAL SPACE

A s a continuation of Algebra 2, we present in this chapter many relationships bet-

ween scalars, vectors and linear mappings having many variables defined on a
finite-dimensional v. space. Recall that the v. space is a basic object in the study of linear
algebra. It is a set of several vectors which are objects that can be added together and
multiplied by numbers, which are called scalars in this context. This chapter deals with
mappings defined on some special v. spaces that display one or two variables.

1.1 Vector space (a summary of lessons)

Let K = R or C and let E be a non-empty set equipped with two operations ⊕ and ⊗,
where

1. ⊕ is an internal composition law ; i.e., ∀ u, v ∈ E : u ⊕ v ∈ E.

2. ⊗ is an external composition law ; i.e., ∀ λ ∈ K, ∀ v ∈ E : λ ⊗ v ∈ E.

We say that (E, ⊕, ⊗) is a v. space on the field K if the following conditions hold :
1. (E, ⊕) is a commutative (Abelian) group.
2. ∀ λ ∈ K, ∀ u, v ∈ E : λ ⊗ (u ⊕ v) = (λ ⊗ u) ⊕ (λ ⊗ v) ,
3. ∀ λ, µ ∈ K, ∀ v ∈ E : (λ + µ) ⊗ v = (λ ⊗ v) ⊕ (µ ⊗ v) ,
4. ∀ λ, µ ∈ K, ∀ v ∈ E : λ ⊗ (µ ⊗ v) = (λ.µ) ⊗ v,
5. ∀ v ∈ E : 1K ⊗ v = v. (if K = R or C ⇒ 1K = 1).
To make statements (things) easier ; in a v. space (E, ⊕, ⊗) over K, the internal law ⊕
we designate it + and the external law ⊗ we designate it · or nothing. The definition of a
v. space becomes :
We say that (E, +, .) is a vector space (or just v.s.) over the field K if :
i. + is an internal composition law on E ; i.e., ∀ u, v ∈ E : u + v ∈ E.
ii. · is an external composition law on E ; i.e., ∀ λ ∈ K, ∀ v ∈ E : λv ∈ E.
1. (E, +) is a commutative (abelian) group.
2. ∀ λ ∈ K, ∀ u, v ∈ E : λ (u + v) = λu + λv,

2
1.1. VECTOR SPACE (A SUMMARY OF LESSONS) 3

3. ∀ λ, µ ∈ K, ∀ v ∈ E : (λ + µ) v = λv + µv,
4. ∀ λ, µ ∈ K, ∀ v ∈ E : λ (µv) = (λµ) v,
5. ∀ v ∈ E : 1K v = v.
We must know the following facts :
• The elements of the vector space E are called vectors and the elements of the field K
are called scalars (⇒ the sum of two vectors is a vector and the multiplication of a
vector by a scalar is a vector).
• The neutral element with respect to + in the vector space E we designate it 0E ; and
we call it the zero vector.
• In the v. space E over K ; we have ∀ v ∈ E : −v = (−1) v ; where −v is the symmetric
element of v with respect to +, and (−1) v is the multiplication of the vector v by the
scalar −1.
• For two vectors u and v of the vector space E, we write by convention u − v instead
of u + (−v) and u + (−1) v :
Let (E, +, .) a vector space over the field K and let F be a subset of E.

Definition 1.1. We say that F is a vector subspace (or subspave) of E if (F, +, .) is a vector
space over K, where 0F = 0E .

Remark 1.1. From the above definition we deduce that every vector space is a vector
subspace of itself.

Let E be a v. space, and let F and G be two subspaces of E.

• We have F ∩G = {v ∈ E / v ∈ F and v ∈ G} and F +G = {u + v ∈ E / u ∈ F and v ∈ G}
are vector subspaces of E.
• Note that F + G = G + F, F + F = F, F ∩ G ⊂ F ⊂ F + G and F ∩ G ⊂ G ⊂ F + G.
• Note that if v ∈ F + G, then ∃ a ∈ F, ∃ b ∈ G : v = a + b ; where a and b are not
unique.
• The v. space E is a direct sum of G and F ; and write E = G ⊕ F , if E = G + F and
G ∩ F = {0E } .
• We say that G is supplementary of F in E (or the opposite) if E = G ⊕ F.
• We have E = G ⊕ F ⇔ Every vector v of E is written in a unique way a + b, where
a ∈ G and b ∈ F.

Proposition 1.1. Let F be a subset of E. We have


iff
 ∀ u, v ∈ F : u + v ∈ F

F is a v. subspace of E ⇔ ∀ λ ∈ K, ∀ v ∈ F : λ · v ∈ F.

F 6= ∅ (0E ∈ F )


c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
1.1. VECTOR SPACE (A SUMMARY OF LESSONS) 4

Or equivalently,
(
iff F 6= ∅ (0E ∈ F )
F is a v. subspace of E ⇔
∀ λ, µ ∈ K, ∀ u, v ∈ F : λ · u + µ · v ∈ F.

Example 1.1. Suppose that K = R or C. Then

Kn = {(x1 , x2 , ..., xn ) / xi ∈ K}

is a vector space on K with the laws + and · defined by

1. ∀ (x1 , x2 , ..., xn ) , (y1 , y2 , ..., yn ) ∈ Kn :

(x1 , x2 , ..., xn ) + (y1 , y2 , ..., yn ) = (x1 + y1 , x2 + y2 , ..., xn + yn ) ,

2. ∀ λ ∈ K, ∀ (x1 , x2 , ..., xn ) ∈ Kn :

λ (x1 , x2 , ..., xn ) = (λx1 , λx2 , ..., λxn ) ,

where 0Kn = (0, 0, ..., 0) is the zero vector of this space.

| {z }
n-times
For these laws, we have

• Rn is a v. space over R,
• Rn is not a v. space over C,
• Cn is a v. space over C,
• Cn is a v. space over R.
Let E be a v. space on K, and let v, v1 , v2 , ..., vn ∈ E.
def
– We have : v is a linear combination of v1 , v2 , ..., vn ⇔ ∃ λ1 , λ2 , ..., λn ∈ K :

v = λ1 v1 + λ2 v2 + ... + λn vn .

– We always have 0E = 0.v1 + 0.v2 + ... + 0.vn (where 0E is the zero vector of space E).
– The sum of two linear combinations is a linear combination.
– Multiplying a linear combination by a scalar is a linear combination.
Let E be a v. space on K and let v1 , v2 , ..., vn ∈ E. The set of all linear combinations of
vectors v1 , v2 , ..., vn we note it V ect (v1 , v2 , ..., vn ) or hv1 , v2 , ..., vn i and we call it the subspace
generated by the vectors v1 , v2 , ..., vn . We have then

V ect (v1 , v2 , ..., vn ) = {λ1 v1 + λ2 v2 + ... + λn vn : λ1 , λ2 , ..., λn ∈ K} .

Moreover, we have

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
1.1. VECTOR SPACE (A SUMMARY OF LESSONS) 5

– V ect (0E ) = {0E } .

– V ect (v1 , v2 , ..., vn ) is a vector subspace of E (with v1 , v2 , ..., vn ∈ V ect (v1 , v2 , ..., vn )).
Therefore, the subspace generated by vectors of a space is a vector subspace. of this
space.
– If F is a vector subspace of E, then we have v1 , v2 , ..., vn ∈ F ⇔ V ect (v1 , v2 , ..., vn ) ⊂
F. Therefore, the subspace generated by vectors is the smallest v. subspace contains
these vectors.
– If v = λ1 v1 + λ2 v2 + ... + λn vn , then V ect (v1 , v2 , ..., vn , v) = V ect (v1 , v2 , ..., vn ) .
– V ect (v1 , v2 , ..., vn , 0E ) = V ect (v1 , v2 , ..., vn ) .
– If F = V ect (v1 , v2 , ..., vn ) and G = V ect (u1 , u2 , ..., um ), then

F + G = V ect (v1 , v2 , ..., vn , u1 , u2 , ..., um ) .

Let E be a v. space over K, and let v1 , v2 , ..., vn ∈ E.

We call a linear relationship between the vectors v1 , v2 , ..., vn ; any relation of the form

λ1 v1 + λ2 v2 + ... + λn vn = 0E , where λ1 , λ2 , ..., λn ∈ K.

1. If λ1 , λ2 , ..., λn are all zero, we say that this linear relation is trivial.

2. If λ1 , λ2 , ..., λn are not all zero, we say that this linear relation is non-trivial.

We say that the vectors v1 , v2 , ..., vn are linearly independent (or free) if there is no non-
trivial linear relationship between the vectors v1 , v2 , ..., vn , in other words ; any linear rela-
tionship between vectors v1 , v2 , ..., vn is trivial ; i.e.,
def
v1 , v2 , ..., vn are free ⇔

∀ λ1 , λ2 , ..., λn ∈ K : λ1 v1 + λ2 v2 + ... + λn vn = 0E ⇒ λ1 = λ2 = ... = λn = 0.

– We say that the vectors v1 , v2 , ..., vn are linearly dependent (or linked) if they are not
free, in other words ; if there is at least one non-trivial linear relationship between
the vectors v1 , v2 , ..., vn ; i.e.,
def
v1 , v2 , ..., vn are linked ⇔

∃ λ1 , λ2 , ..., λn ∈ K (are not all null) : λ1 v1 + λ2 v2 + ... + λn vn = 0E .

– The family of vectors {v1 , v2 , ..., vn } are said to be free if the vectors v1 , v2 , ..., vn are
free.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
1.1. VECTOR SPACE (A SUMMARY OF LESSONS) 6

– The family of vectors {v1 , v2 , ..., vn } are said to be linked if the vectors v1 , v2 , ..., vn are
linked.
– Note that if a family contains a linked part, then this family is linked.
– If v ∈ E, then v 6= 0E ⇔ v is free ; since we have

v 6= 0E ⇔ (∀λ ∈ K : λv = 0E ⇒ λ = 0) .

– The null vector or the zero vector 0E is linked ; since we have 1.0E = 0E , which is a
non-trivial linear relationship.
– If a family of vectors contains the zero vector, then that family is related ; i.e., family
{v1 , v2 , ..., vn , 0E } is linked, since {0E } is linked ; or because

0.v1 + 0.v2 + ... + 0.vn + 1.0E = 0E .

Let E be a vector space over K, and let v1 , v2 , ..., vn ∈ (

E.
1) E = V ect (v1 , v2 , ..., vn )
– The family {v1 , v2 , ..., vn } is a base(or basis ) of E ⇔
2) v1 , v2 , ..., vn are free.
Note that E = V ect (v1 , v2 , ..., vn ) i.e., E is spanned by v1 , v2 , ..., vn ; or, we say that
{v1 , v2 , ..., vn } a generated part of E.
– Please note, a basis of E is not always exists or unique.
– If E = V ect (v1 , v2 , ..., vn ), then E admits at least one basis {u1 , u2 , ..., um } ; with m ≤
n, and all the bases of E have the same number of vectors m. This unique number
m ; denoted by dim E, is called the dimension of E.
– Ifv1 , v2 , ..., vn are free, then by definition {v1 , v2 , ..., vn } is a basis of E, and so dim E =
n. Notice, in this case, that every other basis E contains exactly n vectors.
– If v1 , v2 , ..., vn are linked, then a vector of them is a linear combination of the other
vectors. For example, v1 = λ2 v2 + ... + λn vn . Therefore,

V ect (v1 , v2 , ..., vn ) = V ect (v2 , v3 , ..., vn ) .

Now, if v2 , v3 , ..., vn are free, then by definition {v2 , v3 , ..., vn } is a basis of E. Hence,
dim E = n − 1. But, if v2 , v3 , ..., vn are linked, then, a vector of them is a linear combination
of the others ; For example vn = α2 v2 + α3 v3 ... + αn−1 vn−1 . Hence,

E = V ect (v2 , v3 , ..., vn ) = V ect (v2 , v3 , ..., vn−1 ) ...and so on.

– Note that the vector subspace {0E } has no basis ; but by convention we put dim {0E } =
0 ({0E } = V ect ({0E }), where {0E } is linked).

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
1.1. VECTOR SPACE (A SUMMARY OF LESSONS) 7

– Note that if E = V ect (v), where v 6= 0E (i.e., v is free), then {v} is a base of E. In this
case, dim E = 1.
– For the vector space Kn over K, we have dim Kn = n ; since the family of vectors
{e1 , e2 , ..., en } form a basis of Kn ; which is called the canonical basis of Kn , where

e1 = (1, 0, ..., 0) , e2 = (0, 1, ..., 0) , ..., en = (0, 0, ..., 1) .

– For the vector space Cn on the field R, we have dim Cn = 2n ; since the family of
vectors
{e1 , ie1 , e2 , ie2 , ..., en , ien } , where i2 = −1

form a basis of Cn over R ; which is called the canonical basis of Cn over R.

– For the vector space Rn [x], we have dim Rn [x] = n + 1 ; because the family of vectors
{1, x, x2 , ..., xn } form a basis of Rn [x] ; which is called the canonical basis of Rn [x].
– If dim E = n, then

{v2 , v3 , ..., vn } is a basis of E ⇔ E = V ect (v2 , v3 , ..., vn ) ⇔ v2 , v3 , ..., vn are free.

– If F is a vector subspace of E, then we have dim F ≤ dim E.

– If F is a vector subspace of E, then we have dim F = dim E ⇔ F = E.
– Dimension theorem. If F and G are two vector subspaces E, then we have

dim (F + G) + dim (F ∩ G) = dim F + dim G. (1.1)

– Assume that B = {v1 , v2 , ..., vn } is a basis of E and let v ∈ E. Then

∃ λ1 , λ2 , ..., λn ∈ K : v = λ1 v1 + λ2 v2 + ... + λn vn ,

since E = V ect (v1 , v2 , ..., vn ). But ; since the vectors v1 , v2 , ..., vn are free, then the
scalars λ1 , λ2 , ..., λn are unique. In this case the scalars ( λ1 , λ2 , ..., λn ) we call them
the coordinates of v in the basis B.
– In the vector space Kn over K, we have ∀ (x1 , x2 , ..., xn ) ∈ Kn :

(x1 , x2 , ..., xn ) = x1 e1 + x2 x2 + ... + xn en ,

where {e1 , e2 , ..., en } is the canonical basis of Kn . Therefore, (x1 , x2 , ..., xn ) are the co-
ordinates of the vector (x1 , x2 , ..., xn ) in the canonical basis {e1 , e2 , ..., en }.
– In the vector space Cn on R, we have ∀ (z1 , z2 , ..., zn ) ∈ Cn :

(z1 , z2 , ..., zn ) = x1 e1 + y1 (ie1 ) + x2 x2 + y2 (ie2 ) + ... + xn en + yn (ien ) ,

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
1.2. LINEAR MAPPINGS AND LINEAR FORMS 8

where zk = xk + iyk (1 ≤ k ≤ n) and {e1 , ie1 , e2 , ie2 ..., en , ien } is the canonical ba-
sis of Cn over R. Hence, (x1 , y1 , x2 , x2 , ..., xn , yn ) are the coordinates of the vector
(z1 , z2 , ..., zn ) in the canonical basis.
– In the v. space Rn [X], we have

∀P ∈ Rn [x] : P = a0 + a1 · x + a2 · x2 + ... + an · xn ,

where {1, x, x2 , ..., xn } is the canonical basis of Rn [x] . Hence, (a0 , a1 , a2 , ..., an ) are the
coordinates of P = a0 + a1 · x + a2 · x2 + ... + an · xn in the canonical basis.
– In the vector space M2 (R), we have : ∀A = (aij ) ∈ M2 (R) :
! ! ! !
1 0 0 1 0 0 0 0
A = a11 · + a12 · + a21 · + a22 · ,
0 0 0 0 1 0 0 1

where
( ! ! ! !)
1 0 0 1 0 0 0 0
e1 = , e2 = , e3 = , e4 =
0 0 0 0 1 0 0 1

is the canonical basis of M2 (R) . Hence, dim M2 (R) = 4. More generally, dim Mn (R) =
n2 .

1.2 Linear mappings and linear forms

Let E and F be two vector spaces over the same field K, and let f : E → F be a
mapping 1 from E to F .
def
• f is a linear mapping 2 ⇔ ∀ u, v ∈ E, ∀ α ∈ K :
(
f (u + v) = f (u) + f (v)
(1.2)
f (α · v) = α · f (v) .

prop
• f is a linear mapping ⇔ ∀ α, β ∈ K, ∀ u, v ∈ E :

f (α · u + β · v) = α · f (u) + β · f (v) . (1.3)

• We denote by L (E, F ) the set of all linear mappings from E to F.

• If E = F , then we denote by L (E) instead of L (E, E) .

1. Sometimes we say a map instead of mapping.

2. – – we say a linear functional instead of a linear mapping.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
1.2. LINEAR MAPPINGS AND LINEAR FORMS 9

Definition 1.2. Let E be a vector space over K. A linear form over K is a linear mapping
from E to K. The vector space of all linear forms on E, denoted by E ∗ , is called the dual
vector space of E.

Example 1.2. Using (1.2) or (1.3), we can easily prove that the following mappings are
linear forms on E

1. The mapping f : R2 → R such that f (x, y) = 2x − y is a linear form on R2 .

2. The mapping f : Kn → R such that

f (x1 , x2 , ..., xn ) = a1 x1 + a2 x2 + ... + an xn

is a linear form on Rn , where ai ∈ R, for i = 1, 2, ..., n.

3. The mapping f : Rn [x] → R such that
Z b
f (p) = p (t) dt
a

is a linear form on Rn [x].

4. The mapping f : Mn (K) → R such that f (A) = tr (A) is a linear form on Mn (K).
5. Let E be vector space of finite dimension (or a finite-dimensional vector space), say
dim E = n and let B = {u1 , u2 , ..., un } be a basis of E. Note that every vector v ∈ E
can be written (uniquely) as u = α1 u1 + ... + αn un . For each i ∈ 1, n, the mapping

u∗i : E→K (1.4)

u 7→ u∗i (u) = αi

is a linear form on E.

The dual space of E, denoted E ∗ , is the v. space of all linear mappings on E. In other
words, E ∗ = L (E, K) .
We have the following facts :
• If E has finite dimension, then dim E = dim E ∗ .
• If u1 , u2 , ..., un is a basis of E, then the dual basis of u1 , u2 , ..., un is the list Φ1 , Φ2 , ..., Φn
of elements of E ∗ , where each Φi : E → K is a linear mapping such that
(
1, if i = j
Φi (uj ) = (1.5)
0, otherwise.

In the case when E = Kn , we can easily find the corresponding dual basis of the

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
1.2. LINEAR MAPPINGS AND LINEAR FORMS 10

canonical basis of Kn , namely (e) = {e1 , e2 , ..., en }. Define the mappings :

Φi : Kn → K
(x1 , x2 , ..., xn ) 7→ xi

We see that Φi (ej ) satisfies (1.5). Hence, {Φ1 , Φ2 , ..., Φn } is the corresponding dual
basis of the canonical basis (e) of Kn .
– Every basis of E ∗ is the dual basis of a unique basis of E, it is called the predual basis.
– Let f be a nonzero linear form over E. Then there exists a nonzero vector v such that
f (v) = 1. In fact, since f 6= 0, there exists a nonzero vector x such that f (x0 ) 6= 0.
x0
The results holds for v = .
f (x0 )
– Let E be a finite-dimensional v. space, namely dim E = n. If v ∈ E is a nonzero
vector, then there exits a linear form f ∈ E ∗ such that f (v) = 1. Indeed, let v =
α1 u1 + ... + αn un be a a nonzero vector. Then there exists i0 ∈ 1, n such that αi0 6= 0.
Define u∗i0 as in (1.4). That is, u∗i0 (v) = αi0 6= 0. Hence, the result holds if we put
u∗
f = ∗ i0 .
ui0 (v)
Proposition 1.2 (Changing dual basis). Let B1 and B2 be two basses of E and let P be the
t
passage matrix from B1 to B2 . Then (P −1 ) is passage matrix from B1∗ to B2∗ .

Definition 1.3 (dual mapping). If Φ ∈ L (E, F ), then the dual mapping of f is the linear
mapping Φ∗ ∈ L (E ∗ , F ∗ ) defined by Φ∗ (f ) = f ◦ Φ, for f ∈ E ∗ .

Example 1.3. Define the mapping

Φ : Rn [x] → R
p 7→ Φ (p) = p0 ,

where p0 denotes the derivative of p. Let us take, for example f : Rn [x] → R such that
f (p) = p (n) (here n is a positive integer). Then Φ∗ (f ) is the linear mapping on Rn [x]
given by
(Φ∗ (f )) (p) = (f ◦ Φ) (p) = f [Φ (p)] = f (p0 ) = p0 (n) .

Hence, Φ∗ (f ) is the linear map on Rn [x] that takes p to p0 (n) .

Rb
Suppose further that f : Rn [x] → R such that f (p) = a p (t) dt. Then Φ∗ (f ) is the
linear mapping on Rn [x] given by
Z b
∗ 0
(Φ (f )) (p) = (f ◦ Φ) (p) = f [Φ (p)] = f (p ) = p0 (t) dt = p (b) − p (a) .
a

Hence, Φ∗ (f ) is the linear map on Rn [x] that takes p to p (b) − p (a) .

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
1.2. LINEAR MAPPINGS AND LINEAR FORMS 11

Let us state some algebraic properties of dual maps :

1. (Φ1 + Φ2 )∗ = Φ∗1 + Φ∗2 for every Φ1 + Φ2 ∈ L (E, F ) .

2. (α · Φ2 )∗ = α · Φ∗ for all Φ ∈ L (E, F ) and α ∈ K.
3. (Φ1 ◦ Φ2 )∗ = Φ∗2 ◦ Φ∗1 for all Φ1 ∈ L (E, F ) and Φ2 ∈ L (F, G) .

• Any linear mapping is a homomorphism (we can talk about the kernel, the image and
so on).
• If f : E → F is linear mapping, then f (0E ) = 0F (the converse is false).
• If f (0E ) 6= 0F , then f : E → F is not a linear mapping.
• Be careful, if f : E → F is linear and v ∈ E, then : f (v) = 0F ; v = 0E (in general).
But, if f is injective, then f (v) = 0F ⇒ v = 0E (since f (v) = 0F ⇔ f (v) = f (0E )).
• Every linear mapping f : E → E is called Endomorphism of E.
• Every linear mapping f : E → F bijective is called Isomorphism.
• Every bijective Endomorphism of E is called Automorphism of E.
• Every linear mapping f : Rn −→ Rm is uniquely defined as follows :

f (x1 , x2 , ..., xn ) = (a11 x1 + a12 x2 + ... + a1n xn , ..., am1 x1 + am2 x2 + ... + amn xn ) ,

where (aij ) ∈ R for all i, j (i = 1, 2, ..., m and j = 1, 2, ..., n).

• The kernel of a linear mapping f : E → F is the set defined by

ker f = {v ∈ E : f (v) = 0F } . (1.6)

We can easily prove that ker f is a vector subspace of E.

• The image of a linear mapping f : E → F is the set defined by

Im f = {f (v) : v ∈ E} .

We can easily prove that Im f is a vector subspace of F.

• If f : E → F is the zero linear mapping (i.e., f (v) = 0, ∀ v ∈ E), then ker f = E and
Im f = {0F } .
• The identical mapping of E, i.e., idE is linear, where ker (idE ) = {0E } and Im (idE ) =
E.
• If f : E → F is linear, then

E = V ect (v1 , v2 , ..., vn ) ⇒ Im f = V ect (f (v1 ) , f (v2 ) , ..., f (vn )) .

In practice, we use the canonical basis of E. So, we have in particular :

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
1.3. PROPOSED PROBLEMS ON LINEAR FORMS 12

• If f : Rn −→ F is linear and {e1 , e2 , ..., en } is the canonical basis of Rn , then we have

Im f = V ect (f (e1 ) , f (e2 ) , ..., f (en )) ,

where e1 = (1, 0, ..., 0) , e2 = (0, 1, ..., 0) , en = (0, 0, ..., 1) .

• If f : Rn [x] −→ F is linear and {1, x, x2 , ..., xn } is the canonical basis of Rn [x], then
we have
Im f = V ect (f (1) , f (x) , ..., f (xn )) ,

where e1 = (1, 0, ..., 0) , e2 = (0, 1, ..., 0) , en = (0, 0, ..., 1) .

• If f is linear, then the number dim (Im f ) is called the rank of f and we note it by
rank (f ), i.e.,
rank (f ) = dim (Im f ) .

• If f : E → F is linear, then

f is injective ⇔ ker f = {0E } ⇔ dim (ker f ) = 0,

and also, we have

f is surjective ⇔ Im f = F ⇔ dim (Im f ) = dim F.

• If f : E → F is linear, then

dim E = dim ker f + dim Im f. (Rank Theorem)

• If f : E → F is linear with dim E = dim F, then

f is surjective ⇔ f is bijective ⇔ f is injective.

In practice, we use this result if E = F , i.e., if f is an Endomorphism on E.

1.3 Proposed Problems on linear forms

Exercise 1. Determine the linear form f defined by

f (1, 1, 1) = 0, f (2, 0, 1) = 1 and f (1, 2, 3) = 4,

then, determine ker f . The same question for g, where g (1, 0, 1) = −1, g (0, 1, 1) = 0 and
g (−1, 1, 1) = 2.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
1.3. PROPOSED PROBLEMS ON LINEAR FORMS 13

Exercise 2. Let E = R2 and f1 , f2 ∈ E ∗ such that

f1 (x, y) = x + y, f2 (x, y) = x − y.

1. Show that {f1 , f2 } is a base of E ∗ .

2. Express g and h, in this base, where g (x, y) = x and h (x, y) = 2x − 6y.
3. Determine the predual base of {f1 , f2 }.
4. Note that {(1, 2) , (−1, 1)} is a base of E, find its dual base.

Exercise 3. Let {e1 , e2 , e3 } be the canonical basis of E = R3 and let f1 , f2 , f3 ∈ E ∗ defined

by

∗ ∗ ∗
 f1 = 2e1 + e2 + e3

f2 = −e∗1 + 2e∗3

f3 = e∗1 + 3e∗2 .


1. Prove that {f1 , f2 , f3 } is a basis of E ∗ .

2. Determine the predual basis of {f1 , f2 , f3 }.
3. Prove that A = {(1, 1, 1) , (−1, 2, 1) , (0, 1, 3)} is a basis of E, and find its dual basis,
say A∗ .
4. Calculate ϕ the passage matrix from {f1 , f2 , f3 } to A∗ .

Exercise 4. Consider the vector space of real polynomials of degree not exceeding 2,
i.e., E = R2 [x]. Define the mappings ϕ0 , ϕ1 , ϕ2 from E to R by
Z 1
∀ p ∈ E, ϕ0 (p) = p (0) , ϕ1 (p) = p (1) et ϕ2 (p) = p (t) dt.
0

1. Prove that ϕi ∈ E ∗ for i = 0, 1, 2.

2. Show that {ϕ0 , ϕ1 , ϕ2 } is a basis of E ∗ .
3. Determine the predual basis of {ϕ0 , ϕ1 , ϕ2 }.
4. Prove that {1, 1 + x, 1 + x + x2 } is a basis of E, and find its dual.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 C HAPTER 2

B ILINEAR FORMS OVER A VECTOR SPACE

I n this chapter we present a basic introduction on Bilinear forms over a vector space
including rank, kernel, Orthogonalization of Gram-Schmidt, Orthogonal matrices and
diagonalization of real symmetric matrices.

2.1 Bilinear forms (Definitions)

In this section, R is the field of real numbers and E is a vector space over R. For example,
E = Rn , Rn [x] or Pn [x], C ([a, b] , R), C ∞ ([a, b] , R) and Mn (R) with n ≥ 1, and so on.
Let E be a vector space on R. As above, a linear form 1 is a mapping f from E to R such
that for every (x, y) ∈ E 2 and λ ∈ R, we have
(i) f (x + y) = f (x) + f (y) ,
(ii) f (λx) = λf (x).
Similarly, we have the following definition :

Definition 2.1. Let E be a vector space on R. A bilinear form is a mapping f from E 2 to R

such that for every (x, x0 , y, y 0 ) ∈ E 4 and λ ∈ R, one has
(i) f (λx + x0 , y) = λf (x, y) + f (x0 , y) ,
(ii) f (x, λy + y 0 ) = λf (x, y) + f (x, y 0 ) .

As in (1.2) and (1.3), note that a bilinear form is a mapping f from E 2 to R such that f is
linear from the left and linear from the right. For details, we present the following remark.

Remark 2.1. Let f : E × E → R be a bilinear form on E. This means that for all x, x0 , y, y 0 ∈
E and λ ∈ R we have
• f (x + x0 , y) = f (x, y) + f (x0 , y),
• f (x, y + y 0 ) = f (x, y) + f (x, y 0 ),
• f (λx, y) = λf (x, y),
• f (x, λy) = λf (x, y) .

Definition 2.2. Let f : E 2 → R be a bilinear form.

1. If f : E × E → F is bilinear, then f is called a bilinear mapping. However, if f : E × E → K is bilinear,
then f is called a bilinear form.

14
2.1. BILINEAR FORMS (DEFINITIONS) 15

1. f is said to be symmetric if for each (x, y) ∈ E 2 , f (x, y) = f (y, x) .

2. f is said to be skew-symmetric if for each (x, y) ∈ E 2 , f (x, y) = −f (y, x) .
3. f is said to be alternating if for each x ∈ E, f (x, x) = 0.

Example 2.1. We can easily check that the following mappings are symmetric bilinear
forms.
1. f : R × R → R, f (x, y) = xy.

2. f : R2 ×R2 → R, ((x, y) , (x0 , y 0 )) 7→ xx0 + yy 0 .

Rb
3. ϕ : P [x] ×P [x] → R with ϕ (p, q) = a p (t) q (t) dt.
4. Let x = (x1 , x2 , x3 ), y = (y1 , y2 , y3 ) ∈ R3 with f (x, y) = x1 y1 + x2 y2 − x3 y3 .

Example 2.2. Let u = (x, y) , v = (x0 , y 0 ) ∈ R2 with f (u, v) = xy 0 − x0 y. Then f is a skew-

symmetric alternating bilinear form.

Notation 2.1. Let L2 (E) denote the vector space of all bilinear forms over E, S2 (E) denote
the vector space of all symmetric bilinear forms over E and A2 (E) denote the vector space
of all skew-symmetric bilinear forms over E.

We can prove the following fact : L2 (E) = S2 (E) ⊕ A2 (E) . Indeed, we have f0 = 0 ∈
S2 (E) ∩ A2 (E) . Also, if f ∈ S2 (E) ∩ A2 (E), then by Definition 2.2 f (x, y) = f (y, x) =
−f (y, x) for every x, y ∈ E. Hence, f (x, y) = 0 for every x, y ∈ E. So, f = f0 . Thus, we
have proved that S2 (E) ∩ A2 (E) = {f0 } . Now, let f ∈ L2 (E). For any x, y ∈ E, we see
that
f (x, y) − f (y, x) f (x, y) + f (y, x)
f (x, y) = + = h1 (x, y) + h2 (x, y) ,
2 2
where h1 is skew-symmetric and h2 is symmetric.

Theorem 2.1. Let    

x1 y1
x2 y2
   
   
x= , y =   ∈ Rn .
.. ..
. .
   
   
xn yn
Let A ∈ Mn (R) be a square matrix. Define f : Rn × Rn → R, (x, y) 7→ xt · A · y. Then f is a
bilinear form over Rn . Moreover, if A is symmetric, then f is also symmetric.

Proof. For all x, x0 , y ∈ Rn and for all λ ∈ R we have

t
f (λx + x0 , y) = (λx + x0 ) · A · y
t
= λxt · A · y + (x0 ) · A · y
= λf (x, y) + f (x0 , y) .

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
2.1. BILINEAR FORMS (DEFINITIONS) 16

Thus, f is linear from the left. We use the same manner to show that f is linear from the
right. For every x, y, y 0 ∈ Rn and λ ∈ R we have

f (x, λy + y 0 ) = xt A (λy + y 0 )
= λxt Ay + xt Ay 0
= λf (x, y) + f (x, y 0 ) .

Next, assume that A is symmetric. We show that f is also symmetric. In fact, we have

f (x, y) = xt Ay

t
= xt Ay (since xt Ay ∈ R)

t
= y t At xt (well-known result)
= y t Ax (since A is symmetric)
= f (y, x) .

Hence, f (x, y) = f (y, x) .

The proof is finished.

We conclude from Theorem 2.1 the following corollary.

Corollary 2.1. Every matrix A ∈ Mn (R) produces a bilinear form over Rn and every
symmetric matrix A ∈ Mn (R) produces a symmetric bilinear form over Rn .

Theorem 2.2. Let B and B 0 be two bases of E. Let P be the passage matrix from B to B 0 and let
f : E × E → R be a bilinear form over E. If A = Mf (B) and A0 = Mf (B 0 ), then A0 = P t · A · P.

Proof. Assume that B = {e1 , e2 , ..., en } and B 0 = {e01 , e02 , ..., e0n }. For every x, y ∈ E, we see
that  Pn  Pn
 x = i=1 xi · ei
  y = i=1 yi · ei

and
Pn 0 0
y = ni=1 yi0 · e0i
  P
x = i=1 xi · ei
 

That is, x = P · x0 and y = P · y 0 . Therefore,

t t
f (x, y) = xt · A · y = (P · x0 ) · A · (P · y 0 ) = (x0 ) · P t
| {zAP} ·y 0 .

Thus, the matrix of f with respect to the basis B 0 is given by : A0 = P t · A · P, where A is

the matrix of f with respect to the basis B. The proof is finished.

Theorem 2.3. If dim E = n, then dim L2 (E) = n2 .

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
2.2. ORTHOGONAL MATRICES 17

Proof. Let {u1 , u2 , ..., un } be a basis of E. Define the bilinear forms fi,j by
(
1, for (i, j) = (r, s)
fi,j (er , es ) =
0, for (i, j) 6= (r, s)
Pn Pn
Let x = i=1 xi ui and y = j=1 yj uj be two vectors of E. It is clear that

fi,j (x, y) = xi yj , for i = 1, 2, ..., n.

Now, let f ∈ L2 (E) and put f (er , es ) = ars . It follows that

n n
! n X
n
X X X
f (x, y) = f xi ui , yj uj = xi yj f (ei , ej )
i=1 j=1 i=1 j=1
n X
X n n X
X n
= xi yj aij = xi yj fi,j (x, y) .
i=1 j=1 i=1 j=1

Then these n2 bilinear forms fi,j generated the vector space fi,j . Since (fi,j )1≤i,j≤n form a
free family, we conclude that (fi,j )1≤i,j≤n is a basis of L2 (E) . The proof is finished.

2.2 Orthogonal matrices

Definition 2.3. An invertible square matrix A is said to be orthogonal if At = A−1 .

Clearly, a sufficient and necessary condition for A to be orthogonal is that AAt = At A =

I, , where I is the identity matrix.

Example 2.3. By the above definition, the following matrices :

!
cos δ sin δ
with δ ∈ R (2.1)
− sin δ cos δ

and  
! √1 √1 √1
√1 √1  2 6 3 
2 2 −1 √1 √1 .
−1
,  √
√ √1  2 6 3 
2 2 −2 √1
0 √
6 3

are orthogonal.

Proposition 2.1. Let A ∈ Mn (R) be an orthogonal matrix. Then det (A) = ±1.

Proof. Since At = A−1 , we conclude that At A = In . This gives

det At A = det At det (A) = (det (A))2 = det (In ) = 1.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
2.2. ORTHOGONAL MATRICES 18

Hence, det (A) = ±1.

We need to define matrix norms and scalar product over a vector space E.

2.2.1 Matrix norms

Definition 2.4. Let E be a vector space over K ( R or C). The norm over E, denoted by k.k,
is a mapping

k.k : E → R+
x 7→ kxk (we say : the norm of x)

which satisfy the following properties :

1. For every x ∈ E : kxk ≥ 0 and kxk = 0 ⇔ x = 0E ;

2. For every x ∈ E and scalar α ∈ K : kαxk = |α| . kxk ;
3. For every x, y ∈ E : kx + yk ≤ kxk + kyk .

In this case, the couple (E, k.k) is called normed vector space or normed space. So, a nor-
med space E is a v. space endowed by a norm.

Example 2.4. Here, we only use the two vector spaces, Kn and Mn (K) with K = R or C.

1. Define over Kn the following norms :

n n
! 12
X X
kxk1 = |xi | , kxk2 = |xi |2 ,
i=1 i=1
kxk∞ = max (|xi |) .
1≤i≤n

2. Define over Mn (K) the following norms :

n
X n
X
kAk1 = max |aij | and kAk∞ = max |aij |
j i
i=1 j=1

n
! 21
X
kAk2 = |aij |2 .
i,j

 t
As an application, for x = −1 1 −2 , we have

√
kxk1 = 4, kxk2 = 6 and kxk∞ = 2.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
2.2. ORTHOGONAL MATRICES 19

!
−1 −2
and for A = ∈ Mn (R), we also have
7 3
√
kAk1 = max (8, 5) = 8, kAk2 = 3 7 and kAk∞ = max (3, 10) = 10.

Lemma 2.1. For each matrix A ∈ Mn (K) and for each x ∈ Kn , we have the following inequality :

kAxk ≤ kAk kxk .

The above lemma is not used in this lecture-note ; but it remains interesting for future
study.

2.2.2 Scalar Product (Inner product) over a real vector space

Definition 2.5. Let E be real vector space. The inner product of E (over E) is a function
h., .i defined by

h., .i : E×E →R
(x, y) 7→ hx, yi

which satisfy the following properties :

1. For all x ∈ E, hx, xi ≥ 0 and hx, xi = 0 ⇔ x = 0.
2. For all x, y ∈ E, we have hx, yi = hy, xi .
3. For all x ∈ E and scalar α ∈ R, we have hλx, yi = λ hx, yi
4. For all x, y, z ∈ E, we have hx + y, zi = hx, zi + hy, zi .
We say, the scalar product between x and y, or the inner product between x and y.
Example 2.5. Define over Rn the scalar product h., .i by
   
x1 y1
n
x2 y2
    X
   
∀x= .. , y =  ..  ∈ Rn : hx, yi = xi y i . (2.2)
. .
   
    i=1
xn yn
 t  t
We can write (2.2) as 2 : hx, yi = xt ·y. In particular, for x = x1 x2 and y = y1 y2 ,
we have
hx, yi = h(x1 , x2 ) , (y1 , y2 )i = x1 y1 + x2 y2 .
2. Sometimes we use the notation t x · y instead of xt · y. We also write t A · A instead of At · A when A is
a square matrix.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
2.2. ORTHOGONAL MATRICES 20

Example 2.6. Define over the vector space C([a, b]) the inner product :
Z b
∀ f, g ∈ C([a, b]) : hf, gi = f (x) · g (x) dx.
a

Theorem 2.4. Let A ∈ Mn (R). The following properties are equivalent :

(i) A is orthogonal.
(ii) For every x ∈ Rn , kAxk = kxk .
(iii) For every x, y ∈ Rn , hAx, Ayi = hx, yi .

Proof. 1)⇒2). Assume that A is orthogonal. Let x ∈ Rn we have

kAxk2 = hAx, Axi = x, At Ax = hx, In xi = hx, xi = kxk2 . (2.3)

Therefore, kAxk = kxk .

2)⇒3). Suppose that ∀ x ∈ Rn : kAxk = kxk . Let x, y ∈ Rn we see that

kA (x + y)k2 = kx + yk2 .

This means that hAx + Ay, Ax + Ayi = hx + y, x + yi, or, equivalently,

hAx, Axi + hAy, Ayi + 2 hAx, Ayi = hx, xi + hy, yi + 2 hx, yi .

Thus, hAx, Ayi = hx, yi .

3)⇒1). Assume that ∀ x, y ∈ Rn : hAx, Ayi = hx, yi . Then

x, At Ay = hx, yi ,

i.e., hx, At Ay − yi = 0. In particular, for x = xt Ay − y, we obtain

2
At Ay − y = 0.

Hence, At Ay = y. Consequently, At A = In .

Example 2.7 (Homework). Consider the matrix

!
0 −1
A= .
1 0

For any θ real, show that eθA is orthogonal 3 .

3. We can prove that A is diagonalizable, where eθA is given by (2.1).

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
2.3. GRAM-SCHMIDT ORTHONORMALIZATION THEOREM 21

2.3 Gram-Schmidt Orthonormalization Theorem

Let E be a Euclidean space and let B = {u1 , u2 , ..., un } be a basis of E. There exists a
unique orthonormal basis {e1 , e2 , ..., en } of E satisfying the following conditions :

1. V ect {e1 , e2 , ..., en } = V ect {u1 , u2 , ..., un } .

2. hei , ui i = 0 for i = 1, 2, ..., n.

The following formulas permit us to find such orthonormal basis recursively as fol-
lows :  u1
 e1 =
 ,

 ku1 k
vk = uk − k−1
P
i=1 hei , uk i · ei ,
(2.4)

 vk

 ek = .
kvk k
Example 2.8. Let us take E = R2 and B = {(1, −1) , (1, 1)} = {u1 , u2 }. Clearly, B is a
basis of R2 . Now, we construct the corresponding orthonormal basis using Gram-Schmith
method. First, we have  
(1, −1) 1 −1
e1 = √ = √ ,√
2 2 2
and

v2 = u2 − he1 , u2 i · e1
    
1 −1 1 −1
= (1, 1) − √ , √ , (1, 1) · √ , √ = (1, 1) .
2 2 2 2

Hence,  
v2 1 1
e2 = = √ ,√ .
kv2 k 2 2
We deduce that B = {e1 , e2 } is an orthonormal basis of R2 . Similarly, let

B = {(1, 1, 1, ) , (0, 1, 1, ) , (0, 0, 1)} = {u1 , u2 , u3 } .

u1  
We have e1 = = √13 , √13 , √13 . Next, by (2.4) we have
ku1 k

v2 = u2 − he1 , u2 i · e1
    
1 1 1 1 1 1
= (0, 1, 1) − √ , √ , √ , (0, 1, 1) · √ ,√ ,√
3 3 3 3 3 3
   
2 1 1 1 −2 1 1
= (0, 1, 1) − √ · √ , √ , √ = , , .
3 3 3 3 3 3 3

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
2.4. DIAGONALIZATION OF REAL SYMMETRIC MATRICES 22

v2 
−2 √1 √1

Hence, e2 = = √ , , . Also, by by (2.4),
kv2 k 6 6 6

v3 = u3 − he1 , u3 i e1 − he2 , u3 i e2
    
1 1 1 1 1 1
= (0, 0, 1) − √ , √ , √ , (0, 0, 1) · √ ,√ ,√ −
3 3 3 3 3 3
    
−2 1 1 −2 1 1
√ , √ , √ , (0, 1, 1) · √ ,√ ,√
6 6 6 6 6 6
 
−1 1
= 0, , .
2 2
 
−1 √1
Thus, e3 = 0, √
2
, 2 . We deduce that B = {e1 , e2 , e3 } is an orthonormal basis of R3 .

Example 2.9 (Homework). Let B = {(1,   Prove o

n2) , (3, 4)}. that B is a basis of R2 . Transform
√1 , √2 −1
B to an orthonormal basis. Ans. B = 5 5
, √25 , √5
.

2.4 Diagonalization of real symmetric matrices

Recall that a matrix A ∈ Mn (C) is said to be symmetric if At = A. It is well known
that for any matrix A ∈ Mn (R), the matrices At A and AAt are symmetric. If A is skew-

symmetric, then A2 is symmetric. Moreover, A + At is always symmetric and A − At is

always skew-symmetric. So we can easily prove that

Mn (R) = Sn (R) ⊕ An (R) .

We can easily show that if A is symmetric, then eA is also symmetric. Indeed, by definition,
if A is symmetric then we have

+∞
!t +∞  i t +∞ +∞
i

t X Ai X A X (At ) X Ai
eA = = = = = eA .
i=0
i! i=0
i! i=0
i! i=0
i!

The result holds. Another important result is given by :

Lemma 2.2. Every symmetric matrix A ∈ Mn (R) is diagonalizable. Moreover, every symmetric
matrix A ∈ Mn (R) can be represented in the form :

A = P · D · P t, (2.5)

where P is orthogonal and D is diagonal whose diagonal entries are the eigenvalues of A.

Proof. The proof is found in the course of Algebra 3.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
2.4. DIAGONALIZATION OF REAL SYMMETRIC MATRICES 23

Definition 2.6. Let A ∈ Mn (R) be a symmetric matrix. We have

• A is said to be positive if xt Ax ≥ 0 for every x ∈ Rn .
• A is said to be definite positive if xt Ax > 0 for every x ∈ Rn − {0Rn }.

Next, we present the following theorem.

Theorem 2.5. Let A ∈ Mn (R). Then A is symmetric definite positive if and only if there exists
an invertible matrix M such that
A = M tM . (2.6)

Proof. Assume that A = M t M with M is invertible. Then A is symmetric ; since

t
t
At = M t M = Mt Mt = M t M = A.

On the other hand, let x 6= 0 be a column vector. We put

 
t
(M x) = y1 y2 . . . yn .

Since M ∈ GLn (R), then M x = y 6= 0. Therefore,

n
X
xt Ax = xt M t M x = (M x)t (M x) = y t y = yi2 > 0.


i=1

Thus, A is definite positive.

Conversely, assume that A is symmetric definite positive. By Lemma 2.5, we write

A is symmetric ⇒ ∃ P ∈ GLn (R) such that A = P DP t ,

where D = (λii ) is diagonal whose diagonal elements are the eigenvalues of A. However,
since A est definite positive, the matrix D is also definite positive, that is, its diagonal
entries are strictly positive. Thus, we can define the diagonal matrix :
√ np p p o
D = diag λ1 , λ2 , ..., λn ,

and rewriting, we get

√ √ √ √ t t  √   √ t
A = P DP t = P D DP t = P D D P = P D P D = M t M,

 √ t √
where M = P D ∈ GLn (R) ; since P, D ∈ GLn (R) . The proof of Theorem 2.2 is
finished.

Corollary 2.2. Let A be a symmetric definite positive matrix. Then det (A) > 0.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
2.5. PROPOSED PROBLEMS ON BILINEAR FORMS 24

Proof. First method. Since A is a symmetric definite positive then by Theorem 2.5, A =
M t M , where M is invertible. Therefore,

det (A) = det M t M = det M t det (M ) = (det (M ))2 > 0.

Second method. Since A is a symmetric definite positive then Sp (A) ⊂ R∗+ . On the other
hand, it is well-known that
Y
det (A) = λi ,

from which it follows that det (A) > 0.

Another interesting property of symmetric matrices is the following result :

Proposition 2.2. Let A be a symmetric matrix and let (λ1 , x), (λ2 , y) be two eigenpairs of A with
α 6= β. Then hx, yi = 0.

Proof. Indeed, we see that

λ1 hx, yi = hλ1 x, yi = hAx, yi = x, At y = hx, Ayi = hx, λ2 yi = λ2 hx, yi ,

and since α 6= β, we deduce that hx, yi = 0.

2.5 Proposed Problems on bilinear forms

Exercise 1. Let v1 = (2, 1) and v2 = (1, −2) two elements of th real vector space R2
reported by its canonical basis. Show that {v1 , v2 } is a basis of R2 . Consider the linear form
ϕ over R2 defined by ϕ (v1 ) = 15 and ϕ (v2 ) = −10. Find ϕ (x) for anyx = (x1 , x2 ) in R2 .
Give the dual basis {v1 , v2 }.
Exercise 2. Let E be real vector space R2 related to its canonical basis {e1 , e2 } and let f
be the bilinear form defined on E setting for every x = (x1 , x2 ) and (y1 , y2 ) in E,

f (x, y) = 33x1 y1 − 14 (x1 y2 + x2 y1 ) + 6x2 y2 .

1. Find the matrix of f relative to the basis {e1 , e2 }.

2. Prove that the vectors v1 = e1 + 2e2 , v2 = 2e1 + 5e2 form a basis of E.
3. Write the matrix of f with respect to the basis {v1 , v2 } .
4. What is the rank of f ?
Exercise 3. Let f be the bilinear form defined on the vector space R2 by
(
f (e1 , e1 ) = 1, f (e1 , e2 ) = 1
f (e2 , e1 ) = −1, f (e1 , e2 ) = 3,

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
2.5. PROPOSED PROBLEMS ON BILINEAR FORMS 25

where {e1 , e2 } is the canonical basis of R2 . Specify the value f (x, y) for every x, y in R2 .
Exercise 4. Let f be the bilinear form on R2 setting x = (x1 , x2 ) and (y1 , y2 ) in R2 ,

f (x, y) = 2x1 y1 − 3x1 y2 + x2 y2 .

1. Find the matrix A0 of f related to the basis {u1 = (1, 0) , u2 = (1, 1)} .
2. Find the matrix B of f related to the basis {v1 = (2, 1) , u2 = (1, −1)} .
3. Find the passage matrix P from the basis {u1 , u2 } to the basis {v1 , v2 } and verify that
B = P t A0 P.
4. What is the rank of f ?
Exercise 5. Let E be a vector space over a commutative field K (R or C). We denote
by S the set of symmetric bilinear forms on E and by A the set of antisymmetric bilinear
forms on E.
1. Show that S and A are two vector subspaces of E.
2. Show that the vector space of bilinear forms B(E) over E is the direct sum of S and
A.
3. We assume that E is of finite dimension n. What are the dimensions of S and A.
Exercise 6. Are the following functions E × E → R bilinear forms over the vector
space E? If yes, write their matrix in the canonical basis. Are they symmetric ? When
E = R3 , give their matrix in the basis B = {v1 , v2 , v3 } , where v1 = (1, 0, 1), v2 = (1, 1, 0)
and v3 = (−1, 0, 1) .
• f (x, y) = x1 y1 + x2 y2 + x3 y3 , E = R3
• f (x, y) = y1 y2 + x1 y1 + x3 y3 , E = R2
• f (x, y) = x22 y1 + 3x2 y2 , E = R3
• f (x, y) = x1 y2 − 2x3 (y2 + 2y1 ) + 4x3 y2 − y1 x2 , E = R3
Exercise 7. Let f1 , f2 be bilinear forms on R3 whose matrices in the canonical basis are
   1

1 −1 0 0 1 2
−1
A1 =  −1 −3 2  and A2 =  1 −2 .
   
2
1 −1
0 −2 −1 2 2
0

Write the matrices B1 and B2 of f1 and f2 with respect to the basis {v1 , v2 , v3 }, where
v1 = (1, 0, 0) , v2 = 21 , 12 , 0 , v3 = −1 −1



2
, 2
, 1 . Deduce the rank of each of the linear forms
f1 and f2 .
Exercise 8. Prove that the vectors e1 = (1, 0, 2) , e2 = (0, 1, 1) and e3 = (−1, 0, 1) form a
basis of R3 .
Determine the matrix with respect to this basis of the bilinear form R3 × R3 → R

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
2.5. PROPOSED PROBLEMS ON BILINEAR FORMS 26

defined, for every x = (x1 , x2 , x3 ) andy = (y1 , y2 , y3 ) in R3 by

f (x, y) = 2x1 y2 + x2 y2 − x2 y3 − 2x3 y1 + x3 y2 − x3 y3 .

Exercise 9. Let f be a bilinear form on E and A its matrix representation in a given

basis. f is said to be symmetrical or symmetric (resp., skew-symmetric, alternating) if
f (x, y) = f (y, x) (resp., f (x, y) = −f (y, x) , f (x, x) = 0) for every x, y belong to E.
1. Prove that f is symmetric (resp., skew-symmetric) if and only if At = A (resp.,
At = −A).
2. Prove that if f is alternating, then f is skew-symmetric.
3. Recall that the basic field K of E is infinite. Show that if f is antisymmetric, then f
is alternating.
4. Define f : R × R → R and g : R2 × R2 → R such that f (x, y) = xy and

g ((x1 , x2 ) , (y1 , y2 )) = x1 y2 − x2 y1 ,

where x, y ∈ R and (x1 , x2 ) , (y1 , y2 ) ∈ R2 . Study whether f and g are symmetric,

skew-symmetric or alternating.
Exercise 10. Let (e) = {e1 , e2 , e3 } be the standard basis of R3 and let f the bilinear
symmetric form over R3 given by

f (x, y) = x1 y1 + 6x2 y2 + 56x3 y3 − 2 (x1 y2 + x2 y1 ) + 7 (x1 y3 + x3 y1 ) − 18 (x2 y3 + x3 y2 ) ,

for x = (x1 , x2 , x3 ) and y = (y1 , y2 , y3 ) in R3 .

1. Find the matrix of f with respect to the basis (e).
2. Prove that the vectors e01 = e1 , e02 = 2e1 + e2 and e03 = −3e1 + 2e2 + e3 form a basis of
R3 .
3. Write the matrix of f with respect to the basis (e0 ) = {e01 , e02 , e03 }.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 C HAPTER 3

S YMMETRIC BILINEAR FORMS AND

QUADRATIC FORMS

I n this chapter we focus on the goal of symmetric bilinear forms which define quadratic
forms, where every bilinear form is uniquely represented as the sum of a symmetric
bilinear form and a skew-symmetric bilinear form. Let us start by the following definition :

Definition 3.1. Let E be a vector space over R. A mapping q : E → R is said to be quadra-

tic form if there exists a symmetric bilinear form f : E × E → R such that f (x, x) = q (x)
for any x ∈ E. In this case, f is said to be the polar form of q. Thus, f is the polar for of q
if and only if f is bilinear, symmetric and f (x, x) = q (x) for any x ∈ E.

Example 3.1. Using the above definition, we can easily show that the following mappings
are quadratic forms over E.

1. E = R and q : E→ R, x 7→ x2 .
2. E = R2 and q : E→ R, (x, y) 7→ x2 + y 2 .
Rb
3. E = P [x] and q : E→ R, p 7→= a p2 (t) dt.
4. E = R3 and q : E→ R, (x1 , x2 , x3 ) 7→ x21 + x22 − x23 .
The corresponding polar forms are given in Example 2.1.

Notation 3.1. Let q : E → R be a quadratic form over E. We denote by Q2 (E) the set of
all quadratic forms over E.

3.1 Relation between a quadratic form and its polar form

Let q : E → R be a quadratic form and let u, v ∈ E. Then the polar form of q, namely f
satisfies :

1 1
f (u, v) = [q (u + v) − q (u − v)] = [q (u + v) − q (u) − q (v)] . (3.1)
4 2

27
3.2. QUADRATIC FORMS OVER RN 28

In general, if we would like to prove that a mapping q : E → R is a quadratic form, we

first define the mapping f from E 2 to R by f : E × E→ R,

1
(u, v) 7→ [q (u + v) − q (u) − q (v)] .
2

and by Definition 3.1 we must prove the following facts :

1. f is bilinear,
2. f is symmetric,
3. f (x, x) = q (x) for any x ∈ E.

Example 3.2. Define the mapping Q : P2 [x] → R, p 7→ p (0) p (1). Show that Q is a quadratic
form over P2 [x]. In deed, by (3.1) we obtain ϕ : P2 [x] × P2 [x] → R, where

1 1
(p, q) 7→ ϕ (p, q) = p (0) q (1) + q (0) p (1) .
2 2

We can easily check that ϕ is bilinear, symmetric and ϕ (p, p) = Q (p) .

3.2 Quadratic forms over Rn

First, the analytic expression of q is given by :
n
X n X
X n
q= aij · xi xj = aij · xi xj , (3.2)
i,j i=1 j=1

where (aij ) are real numbers. There are two cases to consider :
Case 1. aij = aji for 1 ≤ i, j ≤ n. The quadratic form q is given by the following matrix
form :
n
X n
X
q (x1 , x2 , ..., xn ) = aii x2i + aij · xi xj
i=1 i6=j
Xn X n
= aii x2i + 2 aij · xi xj
i=1 i<j
  
a11 a12 ... a1n x1
   a12 a22 ... a2n
 
 x2


= x1 x2 ... xn  ..
 .. ,

 . 
 .


a1n a2n ... ann xn
= xt · A · x,

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
3.2. QUADRATIC FORMS OVER RN 29

Case 2. aij 6= aji for some 1 ≤ i, j ≤ n. Here, we see that

n
X n
X n
X n
X
q = aii · x2i + aij · xi xj = aii · x2i + (aij + aji ) · xi xj
i=1 i6=j i=1 i<j
n n
X X aij + aji
= aii · x2i + 2 · xi x j
i=1 i<j
2
Xn Xn
= aii · x2i + 2 bij · xi xj ,
i=1 i<j

where bij = bji (1 ≤ i, j ≤ n). It follows that

 a12 +a21 a1n +an1
 
a11 2
... 2
x1
a12 +a21 a2n +an2
 a22 ... x2
  
2 2
  
q = x1 x2 ... xn  ..
 .. 

 . 
 .


a1n +an1 a2n +an2
2 2
... ann xn
t
= x · A · x.

In both cases, q can be written in the form q = xt · A · x with A symmetric.

Corollary 3.1. Every symmetric matrix A ∈ Mn (R) produces a quadratic form over Rn .

Example 3.3. For q = x21 + 5x1 x2 + 7x22 , we have

! !
5
  1 2
x1
q= x1 x2 5
.
2
7 x2

Here E = R2 . But, if E = R3 we also have

  5

  1 0 x12
 5
q= x1 x2 x3  2 7 0   x2  .
 

0 0 0 x3

Similarly, for q = −x21 + 5x1 x2 + x1 x3 + 2x22 + 2x2 x3 − x23 , we also have

 5 1

  −1 2
x1
2
5
q= x1 x2 x3 2 1   x2  .
  
 2
1
2
1 −1 x3

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
3.3. QUADRATIC FORMS OVER AN ARBITRARY VECTOR SPACE 30

3.3 Quadratic forms over an arbitrary vector space

Let E be a vector space of dimension n (finite-dimensional space) and let (e) = {e1 , e2 , ..., en }
be a basis of E. Let f ∈ S2 (E) and let q ∈ Q2 (E) be the corresponding quadratic form. For
each u, v ∈ E we have
n
X n
X
u= xi ei and v = yi ei , where xi , yi ∈ R for 1 ≤ i, j ≤ n.
i=1 i=1

Then
n n
! n n
X X X X
f (u, v) = f xi ei , yi ei = xi yi f (ei , ei ) + (xi xj + xj xi ) f (ei , ej ) .
i=1 i=1 i=1 i<j

In the matrix form (for E = Rn ), we obtain

  
f (e1 , e1 ) f (e1 , e2 ) ... f (e1 , en ) y1
   f (e2 , e1 ) f (e2 , e2 )

... f (e2 , en )

 y2


f (x, y) = x1 x2 ... xn  ..
 .. 

 . 
 .


f (en , e1 ) f (en , e2 ) ... f (en , en ) yn
= xt · A · y.

By definition, the following matrix

 
f (e1 , e1 ) f (e1 , e2 ) ... f (e1 , en )
 f (e2 , e1 ) f (e2 , e2 ) ... f (e2 , en )
 

Mf ((e)) =  ..
.

 . 

f (en , e1 ) f (en , e2 ) ... f (en , en )

is called the matrix of f in the basis (e) .

Example 3.4 (Homework). Show that the mappings :

q1 : A 7→ q1 (A) = tr At A

q2 : A 7→ q2 (A) = tr A2

are quadratic forms, where tr (M ) denotes the trace 1 of M.

1. Recall that the trace of an n by n matrix M = (aij ) is defined by tr (M ) = a11 + a22 + ... + ann .

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
3.4. ORTHOGONALIZATION METHOD 31

3.4 Orthogonalization method

Using some properties of symmetric matrices, we prove the following theorem :

Theorem 3.1. Every quadratic form over Rn is diagonalizable. That is, if q = xt Ax with A
symmetric, then q = v t Dv with D diagonal.

Theorem 3.2. Every quadratic form over Rn is of the form :

q = λ1 · v12 + λ2 · v22 + ... + λn · vn2 , (3.3)

where the scalars λ1 , ..., λn and the vectors (v1 , v2 , ..., vn ) ∈ Rn satisfy Avi = λi vi . That is, by
(3.3) we get
  
λ1 v1
 λ2   v2 
  

q = v1 v2 ... vn  ..   ..  .
   
 .  . 
λn vn

Proof. We know that

q = xt · A · x,

where A is symmetric. By Lemma 2.2, we have

t
q = xt · P DP t · x = xt P · D · P t x = P t x · D · P t x.

Now, if we put v = P t x, then we obtain q = v t · D · v. Since

 
λ1
λ2
 
 
D=
 ...


 
λn
 
t
is diagonal and v = v1 v2 ... vn , the proof is finished.

Example 3.5. Let q = 2x21 − 4x1 x2 + 5x22 .

1. Write q in the form xt Ax, where A ∈ M2 (R).

2. Using the Orthogonalization method, write q in the form λ1 v12 + λ2 v22 , where λ1 , λ2
are the eigenvalues of A. Solution. 1) In fact, we have
! !
  2 −2 x1
q = 2x21 − 4x1 x2 + 5x22 = x1 x2 .
−2 5 x2

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
3.4. ORTHOGONALIZATION METHOD 32

!
2 −2
2) We put A = . After few computation, the eigenpairs of A are :
−2 5

λ1 = 1 → u1 = (2, 1) ,
λ2 = 6 → u2 = (1, −2) .
√
We see that ku1 k2 = ku2 k2 = 5. Setting
! !
√2 √1 1 0
5 5
P = −2
and D = .
√1 √ 0 6
5 5

Clearly, P is orthogonal (P P t = I2 ). Moreover, we have

! ! ! !
√2 √1 1 0 √2 √1 2 −2
P DP t = 5
−2
5 5
−2
5
= = A.
√1 √ 0 6 √1 √ −2 5
5 5 5 5

It follows that

t
q = xt Ax = xt P DP t x = xt P · D · P t x = P t x · D · P t x.

Now, we put v = P t x. That is,

! ! √ √ !
√2 √1 x1 2
5x1 + 15 5x2
5 5 5√ √
v= −2
= .
√1 √ x2 1
5x1 − 25 5x2
5 5 5

Therefore,

q = v t · D · v = λ1 v12 + λ2 v22
2 2
2√ 1√ 1√ 2√
 
= 5x1 + 5x2 + 6 5x1 − 5x2
5 5 5 5
(2x1 + x2 )2 (x1 − 2x2 )2
= 1· +6· .
5 5

Thus, we have written q as in (3.3).

Example 3.6 (Homework). Let q = 2x1 x2 .

1. Write q in the form xt Ax, where A ∈ M2 (R).

2. Using the Orthogonalization method, write q in the form λ1 v12 + λ2 v22 , where λ1 , λ2
are the eigenvalues of A.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
3.5. DEFINITIONS AND RESULTS 33

3.5 Definitions and results

Let f ∈ S2 (E) (that is, f is a bilinear symmetric form over E).

Definition 3.2. A bilinear form f is called nondegenerate 2 if it satisfies the condition :

f (x, y) = 0 for all x ∈ E implies that y = 0.

Thus, f is called nondegenerate if ker f = {0E }. In the case when ker f 6= {0E }, f is said
to be degenerate.

Definition 3.3. A vector v ∈ E is said to be isotropic if f (v, v) = 0. A subset A ⊂ E is

called isotropic if f (v, v) = 0 for any v ∈ E.

We denote by C the set of all isotropic vectors. That is ,

C = {v ∈ E : f (v, v) = 0} . (3.4)

The set C is called the isotropic cone of E. Note that if v ∈ C, then α · v ∈ C. In fact, for
every v ∈ C and α ∈ R we have

q (αv) = f (αv, αv) = α2 f (v, v) = 0. (3.5)

Proposition 3.1. If f is skew-symmetric, then every vector is isotropic.

Proof. If f is skew-symmetric, i.e., f (x, y) = −f (y, x) for any x, y ∈ E, then f (x, x) =

−f (x, x) for any x ∈ E, so f (x, x) = 0 for any x ∈ E.

Proposition 3.2. Let f ∈ L2 (E). Then f is alternating if and only if C = E.

3.5.1 Is the isotropic cone a vector space ?

In general, the isotropic cone of E is not a vector subspace of E. Thus, we have the
following theorem.

Theorem 3.3. Let f be a nonzero bilinear symmetric form defined on a vector space E. Then

iff
C = ker f ⇔ C is a vector subspace of E.

For the proof we need to the following lemma.

2. In some references we find the word ”nonsingular” instead of nondegenerate.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
3.5. DEFINITIONS AND RESULTS 34

Lemma 3.1. Let q be a quadratic form defined over E and let f be its polar form, where E is vector
space over R. Assume that C is a vector subspace of E and there exists an element x0 ∈ C/ ker f .
Then
∀ y ∈ E ; if f (x0 , y) 6= 0, then y ∈ C,

where C denotes the isotropic cone.

Proof. Let y ∈ E such that f (x0 , y) 6= 0. We have

∀ λ ∈ R : q (λx0 + y) = f (λx0 + y, λx0 + y)

= 2λf (x0 , y) + q (y) (since q (x0 ) = 0).

−q (y)
If we let λ0 = ∈ R, then clearly q (λ0 x0 + y) = 0, from which we deduce that
2f (x0 , y)
λ0 x0 + y ∈ C. But, C is given as a subspace containing x0 . Thus, we deduce that y belongs
to C, as claimed.

Proof of Theorem 3.3. (⇒) Let q be the quadratic form of f . For every (x, y) ∈ C 2 and λ ∈ R,
we get that
• q (x + y) = q (x) + q (y) + 2f (x, y) = 0. Implies x + y ∈ C.
• q (λx) = λ2 q (x) = 0; i.e., λx ∈ C.
Thus, C is a subspace of E.
(⇐) Suppose that C is a subspace of E. Note that the inclusion ker f ⊂ C is always
true ; since

f (x, y) = 0, ∀ y ∈ E ⇒ f (x, x) = 0 (by taking the case y = x).

We would like to prove that if C is a subspace of E, then C ⊂ ker f. Assume by the way of
contradiction that C * ker f . There exists a nonzero vector x0 with x0 ∈ C/ ker f. Define

H = {y ∈ E ; f (x0 , y) = 0} .

It suffices to verify that E ⊂ C. In fact, let z ∈ E and y ∈

/ H. We have

z = y + z − y.

From Lemma 3.1, we have y ∈ C. We distinguish two cases :

Case 1. If z is in H, then y + z ∈
/ H, since

f (x0 , y + z) = f (x0 , y) + f (x0 , z) 6= 0

| {z } | {z }
6=0 =0

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
3.5. DEFINITIONS AND RESULTS 35

Likewise, from Lemma 3.1, we have y + z ∈ C. In this case, we have

z = y + z − y ∈ C (since C is a subspace of E).

| {z } |{z}
∈C ∈C

Case 2. If z ∈
/ H, by Lemma 3.1 once again, we have z ∈ C. Thus, E = C, and so f = 0 ;
since q = 0. But this is a contradiction with our hypothesis that f is a nonzero bilinear
form. Our proof of Theorem 3.3 is finished.

Definition 3.4. Two vectors x and y are said to be orthogonal by f if f (x, y) = 0. We

denote by x ⊥ y.

We deduce from the above definition that C consists all vectors x such that x ⊥ x. Also,
ker f consists vectors which are orthogonal with all the vectors of E.

Definition 3.5 (Orthogonal set). Let A ⊂ E. The orthogonal 3 of A with respect to f is

usually denoted by A⊥ and defined by

A⊥ = {x ∈ E, f (x, y) = 0 for every y ∈ A} . (3.6)

Example 3.7 (Homework). When does A ⊂ A⊥ ?

Remark 3.1. In the case when a nondegenerate bilinear form on E is not symmetric, there
are two different orthogonals of A :

1. A⊥,R = {x ∈ E, f (x, y) = 0 for every y ∈ A} .

2. A⊥,L = {x ∈ E, f (y, x) = 0 for every y ∈ A} .

Here, we can prove that

⊥,R
⊥,L
A⊥,L = A⊥,R = A.

Definition 3.6 (Kernel of a bilinear symmetric form). Let f ∈ S2 (E). The kernel of f is
defined by
ker f = {x ∈ E, f (x, y) = 0 for every y ∈ E} . (3.7)

From Definition 3.5, we deduce that ker f = E ⊥ . Note that x0 ∈ ker f iff f (x, y) = 0 for
every y ∈ E. Similarly, x0 ∈ / ker f if and only if there exists y ∈ E such that f (x0 , y) 6= 0, or
equivalently f (x0 , y) 6= 0 for some y ∈ E.

Theorem 3.4. Let f ∈ S2 (E) and let A, B ⊂ E. Then

3. In some references we say ”perp space to A” instead of the orthogonal of A.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
3.5. DEFINITIONS AND RESULTS 36

⊥
1. A⊥ ⊃ A.
2. (A ∩ B)⊥ ⊃ A⊥ + B ⊥ .
3. (A ∪ B)⊥ ⊃ A⊥ ∩ B ⊥

Proof. 1. Let v ∈ A. For any u ∈ A⊥ , we have f (u, x) = 0 for any x ∈ A. In particular, for

⊥
x = v we have f (u, v) = 0. This means that A⊥ contains v, are required.
2. Let v = a + b ∈ A⊥ + B ⊥ , where A⊥ contains a and B ⊥ contains b. We will prove
that (A ∩ B)⊥ contains v. For every x ∈ A ∩ B we have f (a, x) = f (b, x) = 0 and so
f (a + b, x) = 0 since f ∈ S2 (E). Thus, f (v, x) = 0.
3. Let v ∈ A⊥ ∩ B ⊥ . For every x ∈ A ∪ B we have
• If x ∈ A, then f (v, x) = 0 since v ∈ A⊥ .
• If x ∈ B, then f (v, x) = 0 since v ∈ B ⊥ .
In both cases we have f (v, x) = 0 for any x ∈ A ∪ B. Thus, v ∈ (A ∪ B)⊥ , as asked.

Proposition 3.3. Let f be a bilinear form over E. Two subsets A and B of E are called orthogonal
with respect to f if f (x, y) = 0 for any x in A and y in B. The following conditions are equivalent :

1. A and B are orthogonal,

2. A ⊂ B ⊥ ,
3. B ⊂ A⊥ .

Proof. We prove (a) ⇒ (b). Let a0 ∈ A. For each vector v ∈ B, f (a0 , v) = 0 since A and
B are orthogonal. Hence, a0 ∈ B ⊥ . Next, (b) ⇒ (c). Let b0 ∈ B. For each vector v ∈ A, we
have v ∈ B ⊥ , and so f (b0 , v) = 0 since b0 ∈ B. Hence, b0 ∈ A⊥ . Finally, (c) ⇒ (a). Let u ∈ A
and v ∈ B. Since v ∈ A⊥ , then f (u, v) = 0.

Definition 3.7. Let E be a v. space on R and let {e1 , e2 , ..., en } be a family of n vectors of E.
We have

• {e1 , e2 , ..., en } is orthogonal by f if f (ei , ej ) = 0 for i 6= j.

• {e1 , e2 , ..., en } is orthonormal by f if f (ei , ej ) = 0 for i 6= j and f (ei , ei ) = 1 for
i = 1, 2, ..., n.

Definition 3.8. Let q ∈ Q2 (E) and f its polar form. Then

• q or f is said to be positive if q (x) ≥ 0 for every x ∈ E.
• q or f is said to be definite positive if q (x) > 0 for every x ∈ E − {0E }.

Example 3.8. q = x21 − 4x1 x2 + 4x22 is positive. In fact, we see that

q = (x1 − 2x2 )2 ≥ 0 for every (x1 , x2 ) ∈ R2 .

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
3.5. DEFINITIONS AND RESULTS 37

But, q = x21 − 2x1 x2 + 2x22 is definite positive. In fact, we have

q = (x1 − x2 )2 + x22 > 0 for every (x1 , x2 ) ∈ R2 − {(0, 0)} .

Theorem 3.5. Let f ∈ S2 (E). If f is definite positive, then f is nondegenerate.

Proof. Let x ∈ ker f . Then by (3.7), f (x, y) = 0 for every y ∈ E. In the case when y = x,
we get f (x, x) = 0. But, since f is definite positive, f (x, x) = 0 implies x = 0.

Theorem 3.6 (Cauch-Schwarz inequality). Let q ∈ Q2 (E) and f ∈ S2 (E). If q is positive,

then
(f (x, y))2 ≤ q (x) q (y) for every x, y ∈ E.

3.5.2 When is a quadratic form surjective ?

Let q ∈ Q2 (E) . Here we ask if any real number is represented by this quadratic form.
We present the following result :

Theorem 3.7. Let q be a quadratic form on a real vector space E. The following three properties
are statements :

1. q is surjective.
2. q is neither positive nor negative.
3. There exists an isotropic vector which is not in the kernel.

?
Proof. (1) ⇒ (2). Since q is surjective, then there exists x0 ∈ E (resp. x1 ∈ E) such that
(
q (x0 ) > 0,
q (x1 ) < 0.

Then q is not negative (resp. positive).

?
(2) ⇒ (3). Let x0 and x1 in E such that q (x0 ) > 0 and q (x1 ) < 0. Consider then a vector
of the form λx0 + x1 , where λ ∈ R. Let f be the polar form of q, we have

q (λx0 + x1 ) = f (λx0 + x1 , λx0 + x1 )

= λ2 q (x0 ) + 2λf (x0 , x1 ) + q (x1 ) = p (λ) .

Assume that p (λ) = 0. By computation, we find

∆ = f 2 (x0 , x1 ) − q (x0 ) q (x1 ) > 0.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
3.6. GAUSS DECOMPOSITION (SILVESTER’S THEOREM) 38

Then the equation p (λ) = 0 has two roots. Let λ0 be one of them. Then the vector y0 =
λ0 x0 + x1 is by construction, isotropic. We prove by the way of contradiction that y0 is not
in the kernel of f , that is, assume that y0 ∈ ker f. Hence, f (y0 , x) = 0 for each x ∈ E . In
particular, for x = x0 and for x = x1 we have
(
0 = f (y0 , x0 ) = λ0 q (x0 ) + f (x0 , x1 ) ,
0 = f (y0 , x1 ) = λ0 f (x0 , x1 ) + q (x1 ) .

That is, (
λ20 q (x0 ) + λ0 f (x0 , x1 ) = 0,
λ0 f (x0 , x1 ) + q (x1 ) = 0.
We deduce that λ20 q (x0 ) = q (x1 ) < 0. A contradiction.
?
(3) ⇒ (1). Let y0 be an isotropic vector which is not in the kernel of f. There exists
y1 ∈ E such that f (y0 , y1 ) 6= 0. Then for each γ ∈ R, we put

γ − q (y1 )
λ= ∈ R,
2f (y0 , y1 )

from which it follows that

 
γ − q (y1 )
q (λy0 + y1 ) = q y0 + y1
2f (y0 , y1 )
 
γ − q (y1 ) γ − q (y1 )
= f y0 + y1 , y0 + y1
2f (y0 , y1 ) 2f (y0 , y1 )
= γ.

Thus, q is surjective. The proof of Theorem 3.7 is finished.

3.6 Gauss Decomposition (Silvester’s Theorem)

First, we need to the following definition :

Definition 3.9 (signature of a quadratic form). Assume that

q = +f12 + f22 + ... + fr2 − fr+1

2 2
− .... − fr+s , (3.8)

where f1 , f2 , ..., fr+s are linearly independent forms over Rn . The couple (r, s) is called the
signature of q.
(i) (i) (i) (i)
Recall that if fi (x1 , x2 , ..., xn ) = a1 x1 + a2 x2 + ... + an xn with aj ∈ R for 1 ≤ i ≤ r + s

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
3.6. GAUSS DECOMPOSITION (SILVESTER’S THEOREM) 39

and 1 ≤ j ≤ n, then f1 , f2 , ..., fr+s are linearly independent if and only if

(1) (2) (r+s)

a1 a1 . . . a1
(1) (2) (r+s)
a2 a1 . . . a1
.. .. .. 6= 0.
. . ... .
(1) (2) (r+s)
an a1 . . . a1

In the rest of this section we show how to write any quadratic form over Rn as in (3.8).
To make this, we use the well-known identity :

(a + b + c + ...)2 = a2 + b2 + c2 + 2ab + 2ac + ... + 2bc + 2bd + .... (3.9)

Let q (x1 , x2 , ..., xn ) = ni,j aij · xi xj be a quadratic form over Rn , where aij = aji for
P

1 ≤ i, j ≤ n. We distinguish two cases :

Case 1. When a11 6= 0, we put
 1

 x1 = y1 − (a12 y2 + ... + a1n yn )


 a11

x2 = y 2 (3.10)
..
.





xn = y n


It follows that
q (x1 , x2 , ..., xn ) = a11 y12 + q 0 (y2 , y3 , ..., yn ) ,

where q 0 is also a quadratic form ; but over Rn−1 . Then we repeat the same argument with
q0.
Case 2. When a11 = 0, but a12 6= 0. Here, we put



 x1 = y 1 + y 2

 x2 = y 1 − y 2



x3 = y 3 (3.11)

 ..



 .

 x =y .
n n

It follows that n
X
q (x1 , x2 , ..., xn ) = bij · yi yj ,
i,j

where b11 6= 0. This is the first case (we have transformed q so that we can apply the first

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
3.6. GAUSS DECOMPOSITION (SILVESTER’S THEOREM) 40

case). By this method we can write q in the following form :

q = ±f12 ± f22 ± ... ± fm

2
,

where m ≤ n and f1 , f2 , ..., fm are linearly independent forms over Rn .

Example 3.9. Using Gauss’ Method, diagonalize the following two quadratic forms and
deduce their signatures :
• q1 = x21 + x22 + 2x23 − 4x1 x2 + 6x2 x3
• q2 = 2x1 x2 + 2x2 x3 + 2x1 x3 .
Solution : For q1 , since a11 = 1 it follows from (3.10) that

 x1 = y1 + 2y2

x2 = y 2

x3 = y 3 .


This implies

q1 = x21 + x22 + 2x23 − 4x1 x2 + 6x2 x3

= (y1 + 2y2 )2 + y22 + 2y32 − 4 (y1 + 2y2 ) y2 + 6y2 y3
= y12 − 3y22 + 6y2 y3 + 2y32
= y12 + q10 (y2 , y3 ) .

Likewise by (3.10), let us take (

y2 = z2 + z3
y3 = z3 .
It follows that

q10 = −3y22 + 6y2 y3 + 2y32

= −3 (z2 + z3 )2 + 6 (z2 + z3 ) z3 + 2z32
= −3z22 + 5z32 .

Finally, we obtain

q1 = (x1 − 2x2 )2 + 5x23 − 3 (x2 − x3 )2 = f12 + f22 − f32 ,

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
3.6. GAUSS DECOMPOSITION (SILVESTER’S THEOREM) 41

where f1 , f2 and f3 are linearly independent forms over R3 since

1 −2 0
0 0 5 = −5 6= 0.
0 1 −1

Thus, the signature of q1 is (2, 1). For the quadratic form q2 , by (3.11) we put

 x1 = y 1 + y 2

x2 = y 1 − y 2

x 3 = y3 .


We obtain

q2 = 2 (y1 + y2 ) (y1 − y2 ) + 2 (y1 − y2 ) y3 + 2x1 y3

= 2y12 + 4y3 y1 − 2y22
= q20 .

Setting once again


 y1 = z1 − z3

y2 = z2

y3 = z3 .


It follows that

q20 = 2y12 + 4y3 y1 − 2y22

= 2 (z1 − z3 )2 + 4z3 (z1 − z3 ) − 2z22
= 2z12 − 2z22 − 2z32 .

Hence,

q2 = 2 (y1 + y3 )2 − 2y22 − 2y32

 2  2
x1 + x2 x1 − x2
= 2 + x3 − 2 − 2x23
2 2
= f12 − f22 − f32 ,

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
3.6. GAUSS DECOMPOSITION (SILVESTER’S THEOREM) 42

where f1 , f2 and f3 are linearly independent forms over R3 ; since

1 1
2
1 2
1 −1 0 = 2 6= 0
0 0 −2

The signature of q2 is (1, 2) and the rank is 3.

Example 3.10. Let q = x1 x3 + x2 x3 , E = R3 . Find the signature of q.

Solution : We put

 x1 = y 1 + y 2

x2 = y 1 − y 2

x3 = y 3


We obtain
q = (y1 + y2 ) y3 + (y1 − y2 ) y3 = 2y1 y3

We put once again (

y1 = z1 + z2
y3 = z1 − z2
Then

q = 2 (z1 + z2 ) (z1 − z2 ) = 2z12 − 2z22 (the signature is (1, 1) )

 2  2
y1 + y3 y1 − y3
= 2 −2
2 2
 x1 +x2 2  x1 +x2 2
2
+ x3 2
− x3
= 2 −2 .
2 2

Remark 3.2. The inner product is a bilinear form, symmetric and definite positive. For each
(x, y) ∈ Rn × Rn , we have
 
y1
 y2
 
 
hx, yi = x 1 x2 . . . xn  ..  = xt y.
.
 
 
yn

Corollary 3.2. Let A ∈ Mn (R). Then there exists a symmetric matrix B ∈ Sn (R) such that

xt Ax = xt Bx for every x ∈ Rn .

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
3.6. GAUSS DECOMPOSITION (SILVESTER’S THEOREM) 43

Proof. Since xt Ax = a ∈ R, for every x ∈ Rn we have

t
xt Ax = xt Ax = xt At x.

It follows that
A + At
 
1 1
x Ax = xt Ax + xt At x = xt
t
x.
2 2 2
A + At
Note that the matrix B = is always symmetric.
2

Proposition 3.4. Let A ∈ Mn (R) be a symmetric and let (α, x), (β, y) be two eigenpairs of A
with α 6= β. Then x and y are orthogonal, i.e., x ⊥ y. Or, equivalently, hx, yi = 0.

Proof. Indeed, we have

α hx, yi = hαx, yi = hAx, yi = x, At y = hx, Ayi = hx, βyi = β hx, yi ,

and since α 6= β, it follows that hx, yi = 0.

Example 3.11 (Homework). 1. Consider the equation

ax2 + 2hxy + by 2 = 0. (3.12)

 t
Write (3.12) in the form v t Av = 0, where A ∈ M2 (R) and v = x y .
2. Write the equation λ1 x21 + λ2 x22 = 0 in the matrix form.
3. Let A ∈ Mn (R). We ask if v t Av = 0 ∀ v ∈ Rn , implies A = 0 ?
!
0 −1
Ans. No, take the matrix A = .
1 0

Definition 3.10. Let E be a real vector space equipped with an inner product h., .i. The
couple (E, h., .i) is said to be a real pre-Hilbert space. A real pre-Hilbert space of finite
dimension is said to be Euclidian space.

Let (E, h., .i) be a pre-Hilbert space. The related norm is defined by
p
∀ x ∈ E : kxk = hx, xi. (3.13)

Note that (3.13) a well-known identity which is used in (2.3).

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
3.7. PROPOSED PROBLEMS (QUADRATIC FORMS) 44

3.7 Proposed problems (quadratic forms)

Exercise 1. Determine in the canonical basis of R3 the matrix of the symmetric bilinear
form f such that for v1 = (1, 2, 1) , v2 = (−1, 2, 0), v3 = (1, 0, 1), one has f (v1 , v2 ) = 0,
f (v2 , v3 ) = 4, f (v1 , v3 ) = −1, f (v1 , v1 ) = 5, f (v2 , v2 ) = 1, f (v3 , v3 ) = 0. Find the quadratic
form associated with f .
Exercise 2. Let f be a symmetric bilinear form on E and q the quadratic form associated
with f . Show that for all x, y in E, one has

1
f (x, y) = (q (x + y) − q (x − y)) . (3.14)
4

Consider the mapping q : E → K such that for all x ∈ E, and λ ∈ K we have q (λx) =
λ2 q (x) . The map f : E × E → K given by (3.14) is bilinear. Show that q is the quadratic
form associated with f .
Exercise 3. In the vector space R2 define the quadratic form :

q (x) = 33x21 − 28x1 x2 + 6x22 ,

 t
where x1 x2 are the coordinates of x in the canonical basis {e1 , e2 } of R2 . Determine
the expression of q when we take as basis {e01 , e02 } = {e1 + 2e2 , 2e1 + 5e2 } . Write the polar
form of q in both bases.
Exercise 4. Let f be a symmetric bilinear form on E and let A and B be two parts of E.

⊥
Prove that A ⊂ A⊥ and if A ⊂ B, then B ⊥ ⊂ A⊥ .
In the vector space R3 related to its canonical base the symmetrical bilinear form defi-
ned by
f (x, y) = x1 y1 + x2 y2 ,
 t  t
where x1 x2 x3 and y1 y2 y3 are the coordinates of x and y. Find e⊥ 1 and
⊥
e⊥


1 .
What can we deduce from this ?
Exercise 5. Let f be a symmetric bilinear form on E and let F and G be two subspaces
of E. Show that

(F + G)⊥ = (F ∪ G)⊥ = F ⊥ ∩ G⊥ and F ⊥ + G⊥ ⊂ (F ∩ G)⊥ .

In the vector space R2 related to its canonical base the symmetric bilinear form defined
by
f (x, y) = x1 y1 ,

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
3.7. PROPOSED PROBLEMS (QUADRATIC FORMS) 45

 t  t
where x1 x2 and y1 y2 are the coordinates of x and y. Calculate (V ect {e1 })⊥ ,
(V ect {e1 + e2 })⊥ , (V ect {e1 })⊥ + (V ect {e1 + e2 })⊥ and (V ect {e1 } ∩ V ect {e1 + e2 })⊥ . What
can we deduce from this ?
Exercise 6. In the vector space R3 related to its canonical base the quadratic form defi-
ned by
q (x) = x21 + x22 + x23 − 4 (x1 x2 + x1 x3 + x2 x3 ) ,
 t
where x1 x2 x3 are the coordinates of x. Without using the Gauss method, find a
basis of R3 which is orthogonal by f , where f is the polar form of q.
Exercise 7. In the vector space E = R3 define to its canonical basis the quadratic form

q (x) = x21 + x22 + x23 − (x1 x2 + x1 x3 + x2 x3 ) ,

 t
where x1 x2 x3 are the coordinates of x,and let f be the polar form of q.
1. Decompose q into sum of squares using the Gaussian method.
2. Find a base of E which is orthogonal to f .
3. Find the matrix A and B of f respectively in the canonical and orthogonal basis of
E.
4. Verify by calculation that P t AP = B, where P is the passage matrix from the cano-
nical basis to the orthogonal basis.
Exercise 8. Let A ∈ Mn (C) . Show that if A is symmetric then there exists B ∈ Mn (C)
such that A = B t · B.
Exercise 9. Let (e) = {e1 , e2 , e3 } a basis of a real vector space E of dimension 3 and let

q (x) = x21 + 4x22 + x23 + 4x1 x2 − 2x1 x3 − 12x2 x3 ,

be a quadratic form over E, where x = x1 e1 + x2 e2 + x3 e is any vector of E. Reduce q (x)

in sum of squares using the Gaussian method and deduce its rank and signature.
Exercise 10. Let (e) = {e1 , e2 , e3 } a basis of a real vector space E of dimension 3. Define

q (x) = x21 + 4x22 + x23 + 4x1 x2 − 2x1 x3 − 12x2 x3 ,

which is a quadratic form on E, where x = x1 e1 + x2 e2 + x3 e is any vector of E. Construct,

without using the Gauss method, a basis (e0 ) of E which is orthogonal with respect to q.
Exercise 11. Let (e) = {e1 , e2 , e3 } a basis of a real vector space E of dimension 3. Define
the quadratic form on E,

q (x) = x21 + ax22 + 5x23 + 2x1 x2 − 6x1 x3 + 2x2 x3 ,

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
3.7. PROPOSED PROBLEMS (QUADRATIC FORMS) 46

where a ∈ R and x = x1 e1 + x2 e2 + x3 e3 is any vector of E.

1. Give the polar form f of q as well as the matrix A associated with q relative to the
base (e).
2. Reduce q to sum of squares using the Gaussian method.
3. Construct, without using the Gauss method, a basis (e0 ) of E which is orthogonal
for f .
4. Give the matrix B associated with f in the basis (e0 ) .
5. Deduce the rank and signature of q.
Exercise 12. Let (e) = {e1 , e2 , e3 } be a basis of a real vector space E of dimension 3.
Define
q (x) = 4x21 + 25x22 + ax23 − 12x1 x2 + 4x1 x3 + 2x2 x3

a quadratic form on E, where a is a real number and x = x1 e1 + x2 e2 + x3 e3 is any vector

of E, and let f be the polar form of q.
1. Give the matrix A associated with f relative to the base (e).
2. Reduce q to sum of squares using the Gaussian method.
3. Deduce the rank and signature of q.
4. Study if f is degenerate, positive, definite.
5. Construct, without using the Gauss method, a basis (e0 ) of E which is orthogonal
by f .
6. Give the quadratic form associated with f in the basis (e0 ).
Exercise 13. Let q be a real quadratic form of signature (s, t). Show that
1. q is non-degenerate if and only if s + t = n;
2. q is positive if and only if t = 0;
3. q is negative if and only if s = 0;
4. q is definite positive if and only if s = n;
5. q is definite negative if and only if t = n.
Exercise 14.

1. Compute the signature of each of the following quadratic forms :

• q = 2x1 x3 + 2x2 x3 .
Pk
• q= rs · xr xs , k = 1, 2, ...
r,s

2. Show that the eigenvalues of a real skew-symmetric matrix are imaginary pure (use
two methods).

Exercise 15. Let E be a real vector space and let q be a nondegenerate quadratic form

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
3.7. PROPOSED PROBLEMS (QUADRATIC FORMS) 47

over E of the polar form f . Let a ∈ E be a nonisotropic vector. Define the mapping :

Sa E→E
:
f (x, a)
x →
7 x−2 a.
q (a)

1. Verify the equality

f (Sa (x) , Sa (y)) = f (x, y) for any (x, y) ∈ E 2 .

2. Let x1 and x2 be two vectors of E such that q (x1 ) = q (x2 ) 6= 0. Prove that at least one
of the vectors x1 + x2 and x1 − x2 is nonisotropic (use the way of contradiction).
3. Deduce that there is a nonisotropic vector a0 ∈ E such that

Sa0 (x1 ) = x2 or Sa0 (x1 ) = −x2 .

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 C HAPTER 4

I NTRODUCTION TO HERMITIAN SPACE

T hroughout this chapter, the field used here is the field of complex numbers and E
is a vector space over C. For example, E = Cn with n ≥ 2, Cn [x], Mn (C), and so
on. The basic goal of this chapter is to define quadratic forms over a complex pre-Hilbert
space of finite dimension, namely, hermitian space.

4.1 Sesquilinear forms and hermitian quadratic forms

In this section, we deal with a sesquilinear form defined over a complex vector space
E, which is a mapping from E × E to C, linear according to one of the variables and
semi-linear with respect to the other variable.

4.1.1 Definitions and examples

Definition 4.1. Let E be a vector space on C. A semi-linear form is a mapping f from E
to C such that for every (u, v) ∈ E 2 and α ∈ C, one has
1. f (u + v) = f (u) + f (v) ,
2. f (αv) = αf (v).

Example 4.1. The mapping

f : C→C
z 7→ f (z) = z

is a semi-linear form over C. In fact, we see that

– For every z1 , z2 ∈ C,

f (z1 + z2 ) = z1 + z2 = z1 + z2 = f (z1 ) + f (z2 ) .

– For every z ∈ C and α ∈ K = C we have

f (αv) = αz = α · z = αf (z) .

48
4.1. SESQUILINEAR FORMS AND HERMITIAN QUADRATIC FORMS 49

Definition 4.2. A sesquilinear form is a mapping f from E 2 to C such that f is linear from
the left and semi-linear from the right. Thai is, for every (x, x0 , y, y 0 ) ∈ E 4 and λ ∈ C, one
has
1. f (λx + x0 , y) = λf (x, y) + f (x0 , y) ,
2. f (x, λy + y 0 ) = λf (x, y) + f (x, y 0 )

Example 4.2. Let f : C × C → C, (z1 , z2 ) 7→ z1 · z2 and we prove that f is a sesquilinear

form. In fact, for every (z1 , z2 , z10 , z20 ) ∈ C4 and λ ∈ C, we have

f (λz1 + z2 , z10 ) = (λz1 + z2 ) · z10 = λz1 z10 + z2 z10

= λf (z1 , z10 ) + f (z2 , z10 ) .

That is, f is linear from the left.

f (z1 , λz10 + z20 ) = z1 · (λz10 + z)02 = z1 · λz10 + z20

= λ · z1 z10 + z1 z20 = λf (z1 , z10 ) + f (z1 , z20 ) .

That is, f is semi-linear from the right.

As we have done above, we deduce :

Theorem 4.1. Let B and B 0 be two bases of E. Let P be the passage matrix from B to B 0 and
let f : E × E → R be a sesquilinear form over E. If A = Mf (B) and A0 = Mf (B 0 ), then
A0 = P t · A · P .

Definition 4.3. A hermitian sesquilinear form is a sesquilinear form f over E satisfying

f (x, y) = f (y, x), for each (x, y) ∈ E 2 .

Example 4.3. The sesquilinear form defined over C by

f : C×C→C
(z1 , z2 ) 7→ z1 z2

is hermitian. In fact, for each (z1 , z2 ) ∈ C2 , one has

f (z1 , z2 ) = z1 z2 = z1 z2 = z1 z2 = z2 z1 = f (z2 , z1 ).

That is, f is hermitian.

Theorem 4.2. Let A be a hermitian matrix, and let f : Cn × Cn → C, f (x, y) 7→ xt Ay. Then f
is a hermitian sesquilinear form over Cn .

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
4.1. SESQUILINEAR FORMS AND HERMITIAN QUADRATIC FORMS 50

Proof. We use the same argument as in the proof of Theorem 2.1.

Theorem 4.3. Let f be a sesquilinear form over E. Then f is hermitian if and only if f (x, x) ∈ R
for every x ∈ E.

Proof. Assume that f is Hermitian. Then by definition, f (x, y) = f (y, x) for every x, y ∈
E. In particular, when x = y we have f (x, x) = f (x, x) for every x ∈ E. Thus, f (x, x) ∈ R
for every x ∈ E.
Conversely. Assume that f (x, x) ∈ R for every x ∈ E. Then for every x, y ∈ E we also
have (
f (x + y, x + y) ∈ R,
f (ix + y, ix + y) ∈ R.
It follows that 


 f (x, x) + f (y, y) +f (x, y) + f (y, x) ∈ R,
 | {z } | {z }
∈R ∈R


 f (x, x) + f (y, y) +i [f (x, y) − f (y, x)] ∈ R.
 | {z } | {z }
∈R ∈R

We put (
α = f (x, y) + f (y, x) ∈ R,
β = i [f (x, y) − f (y, x)] ∈ R.
It is clear that
α + iβ α − iβ
= f (y, x) and = f (x, y) ,
2 2
and so f (y, x) = f (x, y). This completes the proof.

4.1.2 Hermitian matrices

At first, define hermitian matrices :

Definition 4.4. Let A = (aij )1≤i,j≤n ∈ Mn (C). The matrix (aij )1≤i,j≤n is called conjugate
of A, denoted by A. The transpose conjugate matrix of A is called the adjoint of A, and
denoted by A∗ .

t
Note that for any matrix A ∈ Mn (C), we have A∗ = At = A . That is, the conjugate
transpose is the same with the transpose conjugate.

Definition 4.5. A matrix A ∈ Mn (C) is said to be hermitian if A∗ = A. That is, if At = A.

Thus,
def
A is hermitian ⇔ aij = aji for 1 ≤ i, j ≤ n.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
4.1. SESQUILINEAR FORMS AND HERMITIAN QUADRATIC FORMS 51

Example 4.4. The matrices

 
! 1 1 + i 2 + 3i
3 2+i
A= , B =  1 − i −2 −i 
 
2−i 7
2 − 3i i 0

are hermitian.

We also state the following elementary properties :

1. I ∗ = I,
2. (A∗ )∗ = A,
3. (A + B)∗ = A∗ + B ∗ ,
4. (αA)∗ = α · A∗ ,
5. (AB)∗ = B ∗ A∗ .

Remark 4.1. Let A ∈ Mn (C). We can easily prove that the matrices A + A∗ , AA∗ and A∗ A
are hermitian.

Proposition 4.1. The diagonal entries of a hermitian matrix A are real numbers.

Proof. Let A = (aij )1≤i,j≤n ∈ Mn (C) be a hermitian matrix. Since aij = aji for each 1 ≤
i, j ≤ n, then
aii = aii , ∀ i = 1, 2, ..., n.

It follows that aii ∈ R for i = 1, 2, ..., n.

Proposition 4.2. Let A and B be two hermitian matrices. Then AB is hermitian if and only if
AB = BA.

Proof. We see that (AB)∗ = AB iff B ∗ A∗ = AB iff BA = AB, as desired.

Definition 4.6. Let A ∈ Mn (C).

1. A is said to be skew-hermitian if A∗ = −A. That is, if At = −A.

Proposition 4.3. Let A ∈ Mn (C). The diagonal entries of a skew-hermitian matrix A are zero or
imaginary pure.

Proof. Let A = (aij )1≤i,j≤n ∈ Mn (C) be a skew-hermitian matrix. Since −aij = −aji for
each 1 ≤ i, j ≤ n, then
−aii = aii , ∀ i = 1, 2, ..., n.

It follows that Re (aii ) = 0, so aii = 0 or αi · i with αi ∈ R∗ for i = 1, 2, ..., n.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
4.2. HERMITIAN QUADRATIC FORMS OVER CN 52

Proposition 4.4. Let A ∈ Mn (C). Then A is skew-hermitian if and only if iA is hermitian.

Proof. We have
(iA)∗ = iA ⇔ −iA∗ = iA ⇔ A∗ = −A.

The proof is finished.

Example 4.5 (Homework). 1. Find the complex number b for which the matrix
 
0 b 0
A= b 0 1 − b , b ∈ C
 

0 b−1 0

is hermitian.
2. Let  
0 x y
A =  −x 0 z  , x, y, z ∈ C
 

−y −z 0
Find the complex numbers x, y, z such that (i) A∗ = A, (ii) A∗ = −A, (i) A is unitary.

4.2 Hermitian quadratic forms over Cn

Let E be a v. space over C. Recall that a map q : E → C is said to be hermitian quadratic
form if there exists a hermitian sesquilinear form f : E × E → R such that f (x, x) = q (x)
for any x ∈ E.

Remark 4.2. Every hermitian matrix A ∈ Mn (C) produces a hermitian quadratic form
over Cn .

Next, the analytic expression of q is given by :

n
X n X
X n
q= aij · xi xj = aij · xi xj .
i,j i=1 j=1

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
4.3. GAUSS DECOMPOSITION FOR HERMITIAN FORMS 53

Thus, every hermitian quadratic form over Cn is given by the following matrix form 1 .
  
a11 a12 ... a1n x1
 a12 a22 ... a2n
 x2
  
 
q (x1 , x2 , ..., xn ) = x1 x2 ... xn  ..
 .. 

 . 
 .


a1n a2n ... ann xn
t
= x · A · x,

where aij = aji for 1 ≤ i, j ≤ n.

Definition 4.7. 1. A hermitian sesquilinear form f : E × E → C is said to be positive if for

any v ∈ E,
f (v, v) ∈ R+ .

2. A hermitian sesquilinear form f : E × E → C is said to be definite positive if for any

v ∈ E,
f (v, v) ∈ R∗+ .

Theorem 4.4. Let A ∈ Mn (C). Then A is hermitian definite positive iff there exists an invertible
matrix M such that
A = Mt · M. (4.1)

Definition 4.8. Let E be a vector space over C. An inner product over E is a sesquilinear
form, hermitian and definite positive.

Thus, a vector space E over C equipped with a sesquilinaer form which is hermitian
and definite positive is called pre-Hilbert space. If a pre-Hilbert space E has finite dimen-
sion, it is called Euclidean space.

4.3 Gauss decomposition for hermitian forms

Here, we have not a direct method as in Section 3.6 ; but we usually use the following
well-known facts :
– For every z ∈ C : z · z = |z|2 .
– For every z ∈ C : z + z = 2Re (z).
– For every z1 , z2 ∈ C :

1 1
z1 · z2 + z1 · z2 = |z1 + z2 |2 − |z1 − z2 |2 .
2 2
1. In some references xt · A · x is the matrix representation of a quadratic hermitian form over Cn , where
t
x · A · x = xt · A · x.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
4.3. GAUSS DECOMPOSITION FOR HERMITIAN FORMS 54

Example 4.6. Diagonalize the following hermitian quadratic forms :

1. q1 = ix1 x2 − ix2 x1 , E = C2 .
2. q2 = x1 x1 + ix1 x2 − ix2 x1 + x2 x2 , E = C2 .
3. q3 = x1 x1 + a12 x1 x2 + a21 x2 x1 + a22 x2 x2 .
4. Deduce the signature of the quadratic form given by :

q20 = αx1 x1 + ix1 x2 − ix2 x1 + x2 x2 , α ∈ R.

Solution. We can write

q1 = ix1 x2 − ix2 x1
= x1 (ix2 ) + x1 (−ix2 )


= x1 −ix2 + x1 (−ix2 ) (which is of the form z1 z2 + z1 z2 )
1 1 1 1
= |x1 − ix2 |2 − |x1 + ix2 |2 (since z1 z2 + z1 z2 = |z1 + z2 |2 − |z1 − z2 |2 )
2 2 2 2
2 2
= |f1 | − |f2 | ,

where f1 et f2 are linearly independent forms over C2 , since

1 −i
6= 0.
1 i

The signature of q1 is (1, 1) . Likewise, we have

q2 = x1 x1 + ix1 x2 − ix2 x1 + x2 x2
= (x1 − ix2 ) (x1 + ix2 )
= (x1 − ix2 ) (x1 − ix2 )
= |x1 − ix2 |2
= |f1 |2 .

The signature of q2 is (1, 0) .

For the quadratic form q20 = αx1 x1 + ix1 x2 − ix2 x1 + x2 x2 , α ∈ R. We see that

q20 = (α − 1) x1 x1 + q2 = (α − 1) |x1 |2 + |x1 − ix2 |2 .

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
4.3. GAUSS DECOMPOSITION FOR HERMITIAN FORMS 55

We deduce that 
 α = 1, the signature is (1, 0) .

α > 1, the signature is (2, 0) .

α < 1, the signature is (1, 1) .


Finally, we have

q3 = x1 x1 + a12 x1 x2 + a21 x2 x1 + a22 x2 x2

= (x1 + a21 x2 ) (x1 + a12 x2 ) + (a22 − a12 a21 ) x2 x2
= (x1 + a21 x2 ) (x1 + a21 x2 ) + (a22 − a12 a21 ) x2 x2
= |x1 + a21 x2 |2 + (a22 − a12 a21 ) |x2 |2 .
| {z }
∈R

Example 4.7. Let E = C, and let q be the Hermitian quadratic form over E given by

q = a12 x1 x2 + a13 x1 x3 + a12 x2 x1 + a23 x2 x3 + a13 x3 x1 + a23 x3 x2 .

Give the diagonal form of Gauss.

Solution. We have

q = a12 x1 x2 + a13 x1 x3 + a12 x2 x1 + a23 x2 x3 + a13 x3 x1 + a23 x3 x2

= x1 (a12 x2 + a13 x3 ) + x1 (a12 x2 + a13 x3 ) + a23 x2 x3 + a23 x3 x2
= x1 (a12 x2 + a13 x3 ) + a23 x2 x3 + x1 (a12 x2 + a13 x3 ) + a23 x3 x2
a23 a23 a12
= x1 (a12 x2 + a13 x3 ) + x2 (a12 x2 + a13 x3 ) − x2 x2 +
a13 a13
a23 a23 a12
x1 (a12 x2 + a13 x3 ) + x2 (a12 x2 + a13 x3 ) − x2 x2
a13 a13
| {z }
     
a23 a23 a23 a12 a23 a12
= x1 + x2 (a12 x2 + a13 x3 ) + x1 + x2 (a12 x2 + a13 x3 ) − + x2 x2
a13 a13 a13 a13
  2   2  
1 a23 1 a23 a23 a12
= x1 + + a12 x2 + a13 x3 − x1 + − a12 x2 − a13 x3 − 2Re |x2 |2 .
2 a13 2 a13 a13

Example 4.8. Diagonalize the Hermitian quadratic form given by its matrix :
 
0 1−i 0
Mq =  1 + i 0 i .
 

0 −i 0

Here, Mq is the matrix of the hermitian quadratic form q with respect to the standard basis
of C3 .

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
4.3. GAUSS DECOMPOSITION FOR HERMITIAN FORMS 56

Solution. We have

q = (1 − i) x1 x2 + (1 + i) x2 x1 + ix2 x3 − ix3 x2
= x2 [(1 + i) x1 + ix3 ] + x2 [(1 − i) x1 − ix3 ]
= x2 [(1 − i) x1 − ix3 ] + x2 [(1 − i) x1 − ix3 ] (which is of the form z1 z2 + z1 z2 )
1 1
= |x2 + (1 − i) x1 − ix3 |2 − |x2 − (1 − i) x1 + ix3 |2
2 2
2 2
= |f1 | − |f2 | .

The signature is (1, 1) .

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 C HAPTER 5

S PECTRAL DECOMPOSITION OF
SELF - ADJOINT LINEAR MAPPINGS

I n this chapter we present a sufficiently and necessary condition for a linear form to be
normal in a complex pre-Hilbert space of finite dimension. But first, define the inner
product on a complex vector space and then we state, without proof, the spectral decom-
position theorem of self-adjoint linear mappings.

5.1 Scalar Product over a complex vector space

Definition 5.1. Let E be complex v. space. The inner product of E (over E) is a function
h., .i defined by

h., .i : E×E →C
(x, y) 7→ hx, yi

satisfying the following properties :

1. For all x ∈ E, hx, xi ∈ R+ and hx, xi = 0 ⇔ x = 0.
2. For all x, y ∈ E, we have hx, yi = hy, xi.
3. For all x ∈ E and scalar α ∈ R, we have hλx, yi = λ hx, yi
4. For all x, y, z ∈ E, we have hx + y, zi = hx, zi + hy, zi .

We say, the scalar product between x and y, or the inner product between x and y.

Definition 5.2. Let E be a complex vector space equipped with an inner product h., .i. The
couple (E, h., .i) is said to be a complex pre-Hilbert space. A complex pre-Hilbert space
of finite dimension is said to be hermitian space.

Example 5.1. Define over Cn the scalar product h., .i by

n
X
n
∀ x, y ∈ C : hx, yi = xi y i . (5.1)
i=1

57
5.2. SPECTRAL DECOMPOSITION OF SELF-ADJOINT LINEAR MAPPING 58

 t  t
1 t
We can write (5.1) as : hx, yi = x ·y. In particular, for x = x1 x2 and y = y1 y2 ,
we have
hx, yi = h(x1 , x2 ) , (y1 , y2 )i = x1 y1 + x2 y2 .

Lemma 5.1. For every x, y ∈ Cn :

|hx, yi| ≤ kxk · kyk . (5.2)

5.2 Spectral decomposition of self-adjoint linear mapping

At first, define unitary and normal matrices or linear mapping.

Definition 5.3. Let A ∈ Mn (C).

1. A is said to be unitary if A∗ = A−1 .

2. A is said to be normal if A∗ A = AA∗ . This means that A commutes with its transpose
conjugate.

Example 5.2. We can check that the matrix

" #
0 −i
U=
i 0

is unitary ; however for the matrix :

" #
−i −i
N=
−i i

we can easily check that N ∗ N = N N ∗ , so N is normal.

Proposition 5.1. Every n by n complex invertible matrix A can be represented as A = U · T ,

where U is unitary and T = (tij ) is upper triangular with tij ≥ 0.

Proof. The proof is similar to the real case.

Lemma 5.2. Every hermitian matrix A ∈ Mn (R) can be represented in the form :

A = Pt · D · P, (5.3)

where P is orthogonal and D is diagonal whose diagonal entries (∈ R) are the eigenvalues of A.
1. Sometimes we use the notation t x · y instead of xt · y.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
5.2. SPECTRAL DECOMPOSITION OF SELF-ADJOINT LINEAR MAPPING 59

From the above lemma, we deduce that every hermitian definite positive matrix A can
√
written as A = M t · M , where M = DP is invertible.

Definition 5.4. Let f ∈ L (E), the adjoint (or the hermitian conjugate) of f is the mapping
f ∗ ∈ E ∗ satisfying
hf (u) , vi = hu, f ∗ (v)i ,

for any u, v ∈ E. Further, f is said to be self-adjoint or hermitian if f = f ∗ .

Theorem 5.1. Let A ∈ Mn (C) be a hermitian matrix (resp. self-adjoint mapping). Then xt Ax ∈
R for each x ∈ Cn .

Proof. We have

t
xt Ax = xt Ax (since xt Ax = a ∈ C)
= (x)t At x (known result)
= (x)t A∗ x
= xt A∗ x
= xt Ax (since A∗ = A).

This implies that xt Ax = xt Ax . Hence, xt Ax ∈ R.

Second proof. We know that

  
a11 a12 ... a1n x1
   a12 a22 ... a2n
 
 x2


t ,
x Ax = x 1 x2 ... xn  ...
 ..
.
  
  
a1n a2n ... ann xn

where aii ∈ R for 1 ≤ i ≤ n and aij = aji for i 6= j because the matrix A is Hermitian.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
5.2. SPECTRAL DECOMPOSITION OF SELF-ADJOINT LINEAR MAPPING 60

Therefore,
X
xt Ax = aij xi xj
i,j
Xn X
= aii xi xi + aij xi xj
i=1 i6=j
Xn X
= aii |xi |2 + (aij xi xj + aji xj xi )
i=1 i<j
| {z }
∈R
n
X X
aii |xi |2 +


= aij xi xj + aij xi xj
|i=1 {z } i<j

∈R
n
X X
= aii |xi |2 + 2Re aij xi xj
i=1 i<j
| {z } | {z }
∈R ∈R

The proof is finished.

Remark 5.1. By a second method we prove that the eigenvalues of a hermitian matrix
A (resp. self-adjoint mapping) are real numbers. Let fA be the corresponding hermitian
sesquilinear form of A and let (λ, x) be an eigenpair of A. Applying Theorem 5.1, we
obtain

f (x, x) = (x)t Ax = (x)t Ax = (x)t λx = λ (x)t x

| A {z }
∈R
n
X
= λ· |xi |2 ∈ R.
|i=1{z }
∈R

Hence, λ ∈ R.
Theorem 5.2. The eigenvalues of a hermitian matrix (resp. self-adjoint mapping) are real numbers.
Proof. Let (λ, x) be an eigenpair of a hermitian matrix A (note that x 6= 0) 2 . We can write

λ hx, xi = hλx, xi = hAx, xi = (Ax)t x = xt At x


t t
t
t
= x A x (since A = A) = xt Ax
= xt Ax = hx, Axi = hx, λxi = λ hx, xi .

Thus, λ = λ and so λ ∈ R.
2. The eigenvectors are always nonzero.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
5.3. PROPOSED PROBLEMS 61

Theorem 5.3 (Spectral decomposition of self-adjoint linear mapping). Let E be a pre-

Hilbert space over C with dim E = n and let f ∈ L (E). Then f is normal iff there exists an
orthonormal basis for E formed by the eigenvectors of f .

Proof. For the proof, one can see [1].

We finish this subsection by a simple comparison between linear algebra and sesquili-
near algebra.

Linear Algebra Sesquilinear Algebra

Linear Semi-linear
f is bilinear f is sesquilinear
f is bilinear symmetric f is sesquilinear hermitian
q is a quadratic form q is a hermitian quadratic form
Eulidian space Hermitian space
symmetric matrix Hermitian matrix
Anti-symmetric (skew-symmetric) matrix Anti-hermitian (skew-hermitian) matrix
Orthogonal matrix Unitary matrix
... ...

5.3 Proposed problems

Exercise 1. Let f be a sesquilinear form on E. Show that f is hermitian sesquilinear
form on E if, and only if, for every x in E, f (x, x) is real.
Exercise 2. Show that
i. The set of all sesquilinear forms over E, equipped with the usual sum of functions
and multiplication by a scalar, is a vector space over C.
ii. The set of all hermitian forms over E, equipped with the usual sum of functions
and multiplication by a scalar, is a vector space over C.
Exercise 3. Let f be a hermitian sesquilinear form on E. Two parts A and B of E are
said to be orthogonal with respect to f if f (x, y) = 0 for any x in A and y in B. Prove that
the following conditions are equivalent :
(i) A and B are orthogonal.
(ii) A ⊂ B ⊥
(iii) B ⊂ A⊥ .
Exercise 4. Let f be a hermitian sesquilinear form on E and q its associated Hermitian
quadratic form. Prove that for all x, y in E and α, β in C we have

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
5.3. PROPOSED PROBLEMS 62

• q (x + y) + q (x − y) = 2q (x) + 2q (y) ,
• q (αx + βy) = |α|2 q (x) + 2Re αβf (x, y) + |β|2 q (y) .

Exercise 5. Let (e) = {e1 , e2 , e3 } a basis for a vector space E of dimension 3.

1. Let f be the sesquilinear form defined by

f (x, y) = 3x1 y1 +2ix1 y2 −5ix1 y3 +(2 + i) x2 y1 −7x2 y2 +x2 y3 +ix3 y1 −x3 y2 +(1 − i) x3 y3 .

1.1. Determine the matrix of f with respect to the basis (e) .

1.2. Is f hermitian ?
2. Explain the hermitian form g whose matrix in the base (e) is given by
 
−2 i 5
 −i −1 3 − 2i  .
 

5 3 + 2i 4

Give the hermitian quadratic form associated with g.

3. Determine, in the basis (e), the matrix of the hermitian form h whose associated
hermitian quadratic form is

q (x) = 3x1 x1 − 5ix1 x3 + (2 − i) x2 x3 − 7x2 x2 + 5ix3 x1 + (2 + i) x3 x2 + x3 x3 .

Exercise 6. Show that the product of two hermitian matrices A and B is a Hermitian
matrix if and only if AB = BA.
Exercise 7. Let A be the hermitian matrix :
 
1 1+i 2i
A= 1−i 4 2 − 3i  .
 

−2i 2 + 3i 7

Find an invertible matrix P such that P t · A · P is diagonal. Deduce the rank and signature
of A.

t
Exercise 8. Let A be an invertible complex matrix. Show that the matrix A A is her-
mitian definite positive.
Exercise 9. Let q be a hermitian quadratic form on E with polar form f and let x be an
isotropic vector for q.
1. Show that if q is defined then f is non-degenerate.
2. Show that for all y ∈ E and λ ∈ C, we have

q (y + λx) = q (y) + 2Re (λf (x, y)) .

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
5.3. PROPOSED PROBLEMS 63

3. Deduce that if q is positive then for all y ∈ E and µ ∈ R, we get

0 ≤ q (y) + 2µ |f (x, y)|2 .

4. Using the previous questions, show that if q is positive and f is nondegenerate, then
q is definite.
Exercise 10. Let A be an invertible complex matrix. Show that if A is hermitian then
−1
A is also hermitian.

t
Exercise 11. A complex matrix A is said to be anti-hermitian if A = −A. Show that
the matrix A is anti-hermitian if and only if iA is hermitian.
Exercise 12. Give a Gaussian decomposition of the Hermitian quadratic forms of C3
whose matrices in the canonical basis are
   
1 1−i 0 0 −i i
A= 1+i 3 i  and B =  i 0 −i  .
   

0 −i 1 −i i 0

Deduce their core, rank and signature.

Exercise 13. Show that a hermitian quadratic form on a vector space E is non-degenerate
if, and only if, the matrix A which represents it in a basis of E is invertible.
Exercise 14. We consider the hermitian quadratic form on C3 given by :

q (x) = x1 x1 + (1 + a) x2 x2 + 1 + a + a2 x3 x3 + ix1 x2 − ix2 x1 − iax2 x3 + iax3 x2 ,

 t
where a is a real number and x1 x2 x3 are the coordinates of x in the canonical
3
basis of C .
1. Give the matrix of q in the canonical basis as well as its polar form f .
2. Using the Gauss method, decompose q into the sum of squares of modules of inde-
pendent linear forms.
3. Deduce an orthogonal basis of C3 relative to f and give the matrix of q in the new
basis.
4. Discuss according to the values of a the rank, signature and kernel of q.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 C HAPTER 6

S OLUTIONS TO SOME EXERCISES AND

PROBLEMS

T he present chapter consists a detailed solution to some exercises and problems re-
lated to symmetric bilinear forms and quadratic forms. These problems were the
subject of some previous TD’s at department of mathematics.
Exercise 01. Find the corresponding symmetric matrix of each of the following qua-
dratic forms :

1. q (x, y) = 4x2 − 6xy − 7y 2 , where E = R2 .

2. q (x, y) = xy + y 2 , where E = R2 .
3. q (x, y, z) = x2 + y 2 − 2z 2 + xy + yz, where E = R3 .
4. q (x, y, z) = 2x2 + 2y 2 + 2z 2 + 2xy + 2yz + 2xz, where E = R3 .
5. q (x, y, z, t) = 2x2 + 2y 2 + 2z 2 + 2xy + 2yz + 2xz, where E = R4 .

Solution. We can easily

! write ! !
4 −3   4 −3 x
1) A = ; since q (x, y) = x y . In fact, we have
−3 −7 −3 −7 y
! ! !
  4 −3 x   x
x y = 4x − 3y −3x − 7y
−3 −7 y y
= 4x2 − 6xy − 7y 2 = q (x, y) .

64
 65

Using the same manner, we obtain

1
 

1
 1 0
2
 0
 
 1 1
2

2) A =  1  , 3) A = 
 2 1 2
 

1  1 
2 0 −2
 2 
  2 1 1 0
2 1 1  
 1 2 1 0 
4) A =  1 2 1  , 5) A = 
  
 1 1 2 0 
1 1 2
 
0 0 0 0

Exercise 02. Consider the quadratic form

q : R2 → R
(x, y) 7→ x2 − y 2 .

1. Calculate the polar form of q, say f .

2. Write f in the matrix form.
3. Calculate the isotropic cone C.
4. Verify that q is nondegenerate.

Solution.
1. We know that
1
f (u, v) = (q (u + v) − q (u − v)) ,
4
where u = (x, y) and v = (x0 , y 0 ) ∈ R2 ; i.e.,

f : R2 × R2 → R
(u, v) 7→ f (u, v) .

Then

f (u, v) = f ((x, y) , (x0 , y 0 ))

1
= (q (x + x0 , y + y 0 ) − q (x − x0 , y − y 0 ))
4
1 2 2 2 2

= (x + x0 ) − (y + y 0 ) − (x − x0 ) + (y − y 0 )
4
= xx0 − yy 0 .

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 66

2. We see that the matrix form of f is given by

! !
  1 0 x0
f (u, v) = x y
0 −1 y0
= ut Av, where u = (x, y) and v = (x0 , y 0 ) .

3. We calculate the isotropic cone C. By (3.4), we have

(x, y) ∈ R2 ; q (x, y) = 0

C =
(x, y) ∈ R2 ; x2 − y 2 = 0

=
(x, y) ∈ R2 ; (x − y) (x + y) = 0

=
(x, y) ∈ R2 ; y = x or y = −x .

=

4. We verify that q is nondegenerate. Indeed, we have

(x, y) ∈ R2 ; f ((x, y) , (x0 , y 0 )) = 0, ∀ (x0 , y 0 ) ∈ R2


ker f =
(x, y) ∈ R2 ; xx0 − yy 0 = 0, ∀ (x0 , y 0 ) ∈ R2

=
= {(0, 0)} .

Thus, f or q is nondegenerate.
Exercise 03. Let f ∈ S2 (E), and let q be the associated quadratic form. Let x0 ∈ E with
q (x0 ) 6= 0. Setting
(
F : is the subspace generated (spanned) by x0 ,
G = {y ∈ E ; f (x0 , y) = 0} .

Prove that E = F ⊕ G.
Solution. At first, we can check that F ∩ G = {0E } .
Let u ∈ F ∩ G. Since u ∈ F , u = kx0 for some scalar k ∈ K. Since u ∈ G, then
f (x0 , kx0 ) = kf (x0 , x0 ) = 0. But, f (x0 , x0 ) 6= 0, then k = 0. This gives u = 0. Thus,
F ∩ G ⊂ {0E } .
Second, we prove that E = F + G. Let x ∈ E and let

f (x0 , x) f (x0 , x)
x= · x0 + x − · x0 ,
f (x0 , x0 ) f (x0 , x0 )
| {z } | {z }
u v

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 67

f (x0 , x)
where u ∈ F (since u is of the form λx0 with λ = ∈ R). Likewise, since
f (x0 , x0 )
 
f (x0 , x)
f (x0 , v) = f x0 , x − · x0 = f (x0 , x) − f (x0 , x) = 0,
f (x0 , x0 )

then v ∈ G. Thus, we have shown that F ∩G = {0E } and F +G = E, and hence E = F ⊕G.

Exercise 04. Let A ∈ Mn (R) and x ∈ Rn . Prove that

A + At
 
t t
x Ax = x x.
2

Solution. For each A ∈ Mn (R) and x ∈ Rn , we have

t
xt Ax = xt Ax = xt Ax (since xt Ax ∈ R)

t
= xt At xt (well-known result)
= xt At x.

Then we can write

A + At
 
1 1 1 1
x Ax = xt Ax + xt Ax = xt Ax + xt At x = xt
t
x.
2 2 2 2 2

This completes the proof.

Exercise 05. Define the quadratic form

q = x21 + 4x1 x2 + 3x22 .

Calculate the polar form associated with q, denoted by f.

Solution. The polar form f of q is given by

f : R2 × R2 → R
(u, v) 7→ f (u, v) = ut Av,
!
1 2
where u = (x1 , x2 ), v = (y1 , y2 ) ∈ R2 and A = . Hence,
2 3
! !
 1 2  y1
f = x1 x2
2 3 y2
= x1 y1 + 2x1 y2 + 2x2 y1 + 3x2 y2 .

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 68

 t
Exercise 06. Let x = x1 x2 ∈ R2 . Show that there are infinitely many matrices
A ∈ M2 (R) such that !
1 4
xt Ax = xt x, (6.1)
0 0
where x ∈ R2 .
Solution. Let n ∈ N. From Exercise 06, we have
! !
1 4 1 0
! +
t 1 4 t
0 0 4 0
X X = X X
0 0 2
!
t 1 2
= X X (in this case, the matrix is symmetric)
2 0
!
1 n
= Xt X.
4−n 0

Then (6.1) is true for infinitely many matrices A.

Exercise 7. Let f ∈ S2 (E) and let F be a subspace of E. Prove that

F ⊂ F ⊥ ⇔ f (x, x) = 0, for every x ∈ F. (6.2)

Assume that E = R3 , and let

f : R3 × R3 → R
((x1 , x2 , x3 ) , (y1 , y2 , y3 )) 7→ x1 y1 + x2 y2 − x3 y3 .

Define F = {(x1 , x2 , x3 ) ∈ R3 ; x1 = x3 and x2 = 0}. Prove by two methods that F ⊂ F ⊥ .

Solution.
1. Suppose that f (x, x) = 0 for all x ∈ F , and we prove that F ⊂ F ⊥ . Let y ∈ F, we
have

f (x + y, x + y) = 0 for each x ∈ F
= f (x, x) + f (y, y) + 2f (x, y) .

Then for each x ∈ F , f (x, y) = 0. Hence, y ∈ F ⊥ .

Assume that F ⊂ F ⊥ and we show that f (x, x) = 0 for each x ∈ F. In fact, let x ∈ F .
For all y ∈ F , we have
f (y, x) = 0. (since x ∈ F ⊥ ).

In particular, f (x, x) = 0 for each x ∈ F.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 69

2. 1st Method. For every x = (λ, 0, λ) ∈ F , we have by (6.2) that

f (x, x) = f ((λ, 0, λ) , (λ, 0, λ)) = λ2 + 02 − λ2 = 0.

Hence, F ⊂ F ⊥ .
2nd Method. By Definition 3.5, we can compute F ⊥ as follows :

F⊥ = y ∈ R3 ; f (x, y) = 0; ∀ x ∈ F


(y1 , y2 , y3 ) ∈ R3 ; f ((λ, 0, λ) , (y1 , y2 , y3 )) = 0; ∀ x = (λ, 0, λ) ∈ F


=
(y1 , y2 , y3 ) ∈ R3 ; λy1 − λy3 = 0; ∀ x = (λ, 0, λ) ∈ F

=
(y1 , y2 , y3 ) ∈ R3 ; λ (y1 − y3 ) = 0; ∀ x = (λ, 0, λ) ∈ F

=
(y1 , y2 , y3 ) ∈ R3 ; y1 = y3

=
= V ect {(1, 0, 1) , (0, 1, 0)} .

Since

(x1 , x2 , x3 ) ∈ R3 ; x1 = x3 and x2 = 0

F =
= V ect {(1, 0, 1)} .

Then clearly, F ⊂ F ⊥ .
Exercise 8.
1. Using Gauss’ Method, diagonalize the following two quadratic forms :
a. q1 = x21 + x22 + 2x23 − 4x1 x2 + 6x2 x3
b. q2 = 2x1 x2 + 2x2 x3 + 2x1 x3 .
Then, determine their associated signatures.
2. Diagonalize the following quadratic form (use two methods).

q (x1 , x2 ) = −12x1 x2 + 5x22 .

Solution.
a. Using Gauss’s method, we put

 1

 x1 = y 1 − (a12 y2 + ... + a1n yn )


 a11

x2 = y2
..
.





xn = yn


c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 70

That is, 
 x1 = y1 + 2y2

x2 = y2

x3 = y 3 .


This implies

q1 = x21 + x22 + 2x23 − 4x1 x2 + 6x2 x3

= (y1 + 2y2 )2 + y22 + 2y32 − 4 (y1 + 2y2 ) y2 + 6y2 y3
= y12 − 3y22 + 6y2 y3 + 2y32
= y12 + q10 (y2 , y3 ) .

Likewise, let us take (

y2 = z2 + z3
y3 = z3 .
It follows that

q10 = −3y22 + 6y2 y3 + 2y32

= −3 (z2 + z3 )2 + 6 (z2 + z3 ) z3 + 2z32
= −3z22 + 5z32 .

Finally, we obtain

q1 = (x1 − 2x2 )2 + 5x23 − 3 (x2 − x3 )2 = f12 + f22 − f32 ,

where f1 , f2 and f3 are linearly independent forms over R3 since

1 −2 0
0 0 5 = −5 6= 0.
0 1 −1

The signature of q1 is (2, 1).

b. Consider the quadratic form

q2 = 2x1 x2 + 2x2 x3 + 2x1 x3 .

In this case, we put


 x1 = y1 + y2

x2 = y1 − y2

x3 = y3 .


c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 71

We obtain

q2 = 2 (y1 + y2 ) (y1 − y2 ) + 2 (y1 − y2 ) y3 + 2x1 y3

= 2y12 + 4y3 y1 − 2y22
= q20 .

Setting once again


 y1 = z1 − z3

y2 = z2

y3 = z3 .


It follows that

q20 = 2y12 + 4y3 y1 − 2y22

= 2 (z1 − z3 )2 + 4z3 (z1 − z3 ) − 2z22
= 2z12 − 2z22 − 2z32 .

Hence,

q2 = 2 (y1 + y3 )2 − 2y22 − 2y32

 2  2
x1 + x2 x1 − x2
= 2 + x3 − 2 − 2x23
2 2
= f12 − f22 − f32 ,

where f1 , f2 and f3 are linearly independent forms over R3 ; since

1 1
2
12
1 −1 0 = 2 6= 0
0 0 −2

The signature of q2 is (1, 2) . The rank is 3.

2. Using two methods, we diagonalize the following quadratic form

q (x1 , x2 ) = −12x1 x2 + 5x22 .

1st method. Setting

 
−6 6
x2 = y2 − y1 = y2 + y1 , x1 = y1
5 5

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 72

We obtain

q (x1 , x2 ) = −12x1 x2 + 5x22

   2
6 6
= −12y1 y2 + y1 + 5 y2 + y1
5 5
36
= 5y22 − y12
5
 2
6 36
= 5 x2 − x1 − x21
5 5
= |f1 |2 − |f2 |2 .

where f1 , f2 and f3 are linearly independent forms over R2 , since

6
1 −
5 6= 0.
1 0

The signature of q is (1, 1) .

2nd method. We have
! !
  0 −6 x1
q (x1 , x2 ) = x 1 x2
−6 5 x2
!
0 −6
= xt Ax, where A = ∈ S2 (R) .
−6 5

Then we can write A in the form P DP t , where P is orthogonal and D is diagonal whose
diagonal entries are the eigenvalues of A. The eigenpairs of A are
(
λ1 = −4, v1 = (3, 2)
λ2 = 9, v2 = (−2, 3) .

Therefore, !

v1 v2

√3 −2
√
13 13
P = = .
kv1 k2 kv2 k2 √2 √3
13 13

which gives

q (x1 , x2 ) = xt Ax
= xt P DP t x (since A = P DP t )

t
= P tx D P tx

= v t Dv, where v = P t x.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 73

It follows that
! ! √ √ ! !
√3 √2 x1 3
13x1 + 2
13x2 v1
13 13 13 √ 13 √
v= −2
= = .
√ √3 x2 3
13x2 − 2
13x1 v2
13 13 13 13

That is,
! !
  −4 0 v1
q = v1 v2
0 9 v2
= 9v22 − 4v12
2 2
3√ 2√ 3√ 2√
 
= 9 13x2 − 13x1 − 4 13x1 + 13x2
13 13 13 13
= |f1 |2 − |f2 |2 .

where f1 are f2 linearly independent form over R2 , since

√ √
3 13 −2 13
√ √ 6= 0.
3 13 2 13

The signature is (1, 1).

Exercise 9. Let E = M2 (R) be the vector space of 2 × 2 square matrices on R. Let
!
1 2
M= ,
3 5

and let f (A, B) = tr (At M B), where A, B ∈ E.

1. Prove that f is a bilinear form on the space E.

2. Find the matrix of f with respect to the canonical (standard) basis of E :
( ! ! ! !)
1 0 0 1 0 0 0 0
B= , , , .
0 0 0 0 1 0 0 1

Solution.
1. We first prove that f is a bilinear form on the space E. Indeed, ∀ A, A0 , B ∈ M2 (R),

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 74

∀ λ ∈ R we have
 
t
f (λA + A0 , B) = tr (λA + A0 ) M B
 
t
= tr λAt M B + (A0 ) M B
 
t 0 t


= λtr A M B + tr (A ) M B
= λf (A, B) + f (A0 , B) .

2. We compute Mf (B), where

 

 

 ! ! ! !


1 0 0 1 0 0 0 0 
B= , , ,


 0 0 0 0 1 0 0 1 

 | {z } | {z } | {z } | {z }


e1 e2 e3 e4

From a simple calculation, we obtain

f (e1 , e1 ) = tr et1 M e1


( ! ! !)
1 0 1 2 1 0
= tr
0 0 3 5 0 0
!
1 0
= tr
0 0
= 1.

Similarly, we have f (e1 , e2 ) = 0, f (e1 , e3 ) = 2,...

It follows that
   
f (e1 , e1 ) f (e1 , e2 ) f (e1 , e3 ) f (e1 , e4 ) 1 0 2 0
   
 f (e2 , e1 ) f (e2 , e2 ) f (e2 , e3 ) f (e2 , e4 )   0 1 0 2 
Mf (B) =   f (e , e ) f (e , e ) f (e , e ) f (e , e )
= .
 3 1 3 2 3 3 3 4
 
  3 0 4 0 

f (e4 , e1 ) f (e4 , e2 ) f (e4 , e3 ) f (e4 , e4 ) 0 3 0 4

Exercise 10. Recall that a bilinear form on a vector space E is called alternating form if
and only if
∀ x ∈ E, f (x, x) = 0.

1. Let f be an alternating bilinear form on a vector space E. Prove that f is skew-

symmetric.
2. Assume that f 6= 0 and 2 ≤ dim E < ∞. Prove that there exist two vectors u1 , u2 ∈ E

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 75

such that
f (u1 , u2 ) = 1.

Calculate f (u2 , u1 ).
3. Let U be the v. subspace spanned by u1 and u2 . Verify that {u1 , u2 } is a base of U .
Write the associated matrix of f in this basis.
4. Setting
W = {w ∈ E; f (w, u) = 0, ∀ u ∈ U } = U ⊥ .

Prove that E = U ⊕ W and deduce that there exists a basis B of the vector space E
for which  
0 1 × ··· × ×
 −1 0 × · · · × × 
 
 × × ...
 .. 
 . 

 . . 
Mf (B) =  . ..  ∈ Mn (R).
 . 0 1 
 
 −1 0 × 
..
 
 × × ··· . × 
 

× × ··· × × 0
Solution.
1. For each (x, y) ∈ E 2 we have

f (x + y, x + y) = 0 (since f is alternating)
= f (x, x) + f (y, y) +f (x, y) + f (y, x) .
| {z } | {z }
=0 =0

Hence, f (x, y) = −f (y, x). Then f is skew-symmetric.

2. Since f 6= 0, there exist two vectors x, y ∈ E such that f (x, y) = α 6= 0, and so

x 
f , y = f (u1 , u2 ) = 1.
α

Since f (u1 , u2 ) = −f (u2 , u1 ), then f (u2 , u1 ) = −1.

3. Let U be the vector subspace generated by u1 and u2 . We prove that u1 and u2 are
linearly independent. By the way of contradiction, if we put u2 = ku1 , then

f (u1 , u2 ) = f (u1 , ku1 ) = kf (u1 , u1 ) = 0.

A contradiction. Then {u1 , u2 } is a base of U.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 76

The matrix of f associated of {u1 , u2 } is

!
0 1
Mf ({u1 , u2 }) = .
−1 0

4. Setting W = {w ∈ E; f (w, v) = 0, ∀ v ∈ U }. We prove that E = U ⊕ W.

It is clear that {0E } ⊂ U ∩ W. Further, if x ∈ U ∩ W implies

 x = αu1 + βu2 ,

f (x, u1 ) = 0,

f (x, u2 ) = 0.


where α, β ∈ R. Hence, (
αf (u1 , u2 ) = 0 ⇒ α = 0
βf (u2 , u1 ) = 0 ⇒ β = 0.
Then x = 0. Therefore, U ∩ W = {0E }.
It remains to be shown that E = U + W. For each x ∈ E, setting

u = f (x, u2 ) u1 − f (x, u1 ) u2

We see that x = u + x − u. Let u is a linear combination of u1 and u2 , then u ∈ U .

It suffices to prove that x − u ∈ W. In fact, we see

f (x − u, u1 ) = f (x − f (x, u2 ) u1 + f (x, u1 ) u2 , u1 )
= f (x, u1 ) − f (x, u1 )
= 0.

Similarly, we also see

f (x − u, u2 ) = f (x − f (x, u2 ) u1 + f (x, u1 ) u2 , u2 )
= f (x, u2 ) − f (x, u2 )
= 0.

Hence f (x − u, v) = 0, ∀ v ∈ U. Then x − u ∈ W , which gives the result.

Now, the restriction of f on the set W is an alternating bilinear form. By induction,

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 77

there exists a basis B = {u3 , u4 , ..., un } of W with

 
0 1 × ··· × ×
 −1 0 × · · · × × 
 
 × × ...
 .. 
 . 

 . .. 
Mf (B) =  .  ∈ Mn−2 (R).
 . . 0 1 
 
 −1 0 × 
..
 
 × × ··· . × 
 

× × ··· × × 0

Thus, u1 , u2 , ..., un is a basis of E for which the matrix representing f has the desired form.
Exercise 11. Let E be a vector space over R with dimension 2. Let f ∈ S2 (E), and let q
be the associated quadratic form. Prove that the following three statements are equivalent :
a. f is nondegenerate and there is a nonzero vector e1 such that q (e1 ) = 0.
b. There exists a basis of E for which the matrix of f is given by
!
0 1
A= .
1 0

c. There exists a basis of E for which the matrix of f is given by

!
1 0
D= .
0 −1

?
Solution. (a) ⇒ (b). Since f is nondegenerate and e1 6= 0, there exists a vector y ∈ E
such that
f (e1 , y) 6= 0.

We put
1
z= y,
f (e1 , y)
so we get  
1
f (e1 , z) = f e1 , y = 1.
f (e1 , y)
For the vector e2 = z − 21 q (z) e1 , we find
(
f (e1 , e2 ) = f e1 , z − 12 q (z) e1 = 1 = f (e2 , e1 )

f (e2 , e2 ) = 0.

The family {e1 , e2 } is a basis of E. Otherwise, e2 = ke1 and f (e1 , e2 ) = 0. Here, the matrix

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 78

of f is given by !
0 1
A= .
1 0
?
(b) ⇒ (c). Conserving the previous notations. The vectors

 e01 = 1 e1 + e2 ,

2
0 1
 e2 = e1 − e2

2

satisfy the following equations

  
1 1
q (e01 ) =f e1 + e2 , e1 + e2 = 1


 2 2



 

0 1 1
q (e2 ) = f e1 − e2 , e1 − e2 = −1

 2 2 

 0 0 1 1
f (e2 , e1 ) = f e1 + e2 , e1 − e2 = 0.


2 2


The family {e01 , e02 } is a basis of E. Otherwise, we get e02 = αe01 , where α ∈ R. Then

−1 = q (e02 ) = q (αe01 ) = α2 q (e01 ) = α2 .

A contradiction. In this basis the matrix of f is given by

!
1 0
D= .
0 −1

?
(c) ⇒ (a). Conserving the previous notations. The quadratic form is nondegenerate
since the matrix D is invertible. For the nonzero vector v 0 = e01 + e02 , we have

q (v 0 ) = f (v 0 , v 0 ) = f (e01 + e02 , e01 + e02 ) = 0.

Exercise 12. Let E be a real vector space and let a ∈ E. Let q be a quadratic form over
E with the polar form f . Define the mapping q 0 from E to R, by setting :

∀ x ∈ E, q 0 (x) = q (a) q (x) − (f (a, x))2 .

1. Prove that q 0 is a quadratic form whose polar form f will be specified.

2. Verify that a ∈ ker f 0 and that ker f ⊂ ker f 0 . Deduce the following inclusion set :
R.a ⊂ ker f 0 .
3. If a is nonisotropic, i.e., q (a) 6= 0, then prove that ker f ⊕ R · a = ker f 0 .

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 79

Solution.
1. We see that q 0 is a quadratic form because the mapping

f0 : E×E →R
(x, y) 7→ q (a) f (x, y) − f (a, x) f (a, y)

is a symmetric bilinear form (since f is also a symmetric bilinear form). Further,

f 0 (x, x) = q 0 (x) for every x ∈ E.

2. We Verify that a ∈ ker f 0 and ker f ⊂ ker f 0 .

For each y ∈ E, we have

f 0 (a, y) = q (a) f (a, y) − f (a, a) f (a, y) = 0.

Hence, a ∈ ker f 0 .
We show that ker f ⊂ ker f 0 . In fact, if x ∈ ker f , then for each y ∈ E we have

f 0 (x, y) = q (a) f (x, y) − f (a, x) f (a, y) = 0.

| {z } | {z }
=0 =0

Thus, x ∈ ker f 0 .

We show that Ra ⊂ ker f 0 . Let λ ∈ R. For each y ∈ E, we have

f 0 (λa, y) = f (λa, y) q (a) − f (a, λa) f (a, y)

= λq (a) f (a, y) − λq (a) f (a, y)
= 0.

Therefore, Ra ⊂ ker f 0 .

If a is nonisotropic, we prove that ker f ⊕ Ra = ker f 0 .

Since ker f and R.a are two subspace of E, then {0E } ⊂ ker f ∩ Ra. If x ∈ ker f ∩ Ra,
then x = λa and f (λa, y) = 0 for each y ∈ E. That is,

f (λa, a) = λq (a) = 0.

Hence λ = 0 (since q (a) 6= 0). Which implies x = 0. Consequently, ker f ∩ Ra ⊂ {0E }.

Finaly, we obtain ker f ∩ Ra = {0E } .

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 80

For each x ∈ ker f 0 and y ∈ E, we write

q (a) x = f (a, x) a + q (a) x − f (a, x) a, (6.3)

| {z }

where f (a, x) a ∈ Ra. It suffices to prove that q (a) x − f (a, x) a ∈ ker f. In fact, for each
y ∈ E, we have

f (q (a) x − f (a, x) a, y) = q (a) f (x, y) − f (a, x) f (a, y)

= 0 (since x ∈ ker f 0 ).

From (6.3), we have

f (a, x) q (a) x − f (a, x) a
x= a+ ,
q (a) q (a)
| {z } | {z }
∈Ra ∈ker f

since u ∈ ker f ⇔ αu ∈ ker f.

Exercise 13.
1. Let q be a quadratic form on a vector space E and let {e1 , e2 , ..., en } be a finite ortho-
gonal set for q. Prove the following equality :

q (e1 + e2 + ... + er ) = q (e1 ) + q (e2 ) + ... + q (er ) .

2. Let (E, h., .i) be a Hilbert space and let {e1 , e2 , ..., en } be an orthonormal basis of E.
Prove that n
X
∀x∈E :x= hx, ei i ei .
i=1

3. Let A = {u1 , u2 , ..., un } be a finite orthonormal set. Show that A is free. Further, for
each x ∈ E prove that the vector

y = x − hx, u1 i u1 − hx, u2 i u2 − ... − hx, un i un

is orthogonal with ui , for i = 1, 2, ..., n.

Solution.
1. Let q be a quadratic form on a vector space E and let {e1 , e2 , ..., en } be a finite ortho-
gonal set for q. We have

q (e1 + e2 + ... + er ) = f (e1 + e2 + ... + er , e1 + e2 + ... + er )

= q (e1 ) + q (e2 ) + ... + q (er ) (since f (ei , ej ) = 0) for i 6= j)

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 81

2. Let {e1 , e2 , ..., en } be an arbitrarily orthonormal basis of E. We prove that

n
X
∀x∈E :x= hx, ei i ei .
i=1

For each x ∈ E, we have x = α1 e1 + α2 e2 + ... + αn en . Further, we have

hx, ei i = hα1 e1 + α2 e2 + ... + αn en , ei i = αi hei , ei i = αi , (6.4)

for i = 1, 2, ..., n. We replace αi by hx, ei i in the equation (6.4), we obtain for the
desired result.

3. Let A = {u1 , u2 , ..., un } be a finite orthonormal set. We show that A is free.

For each α1 , α2 , ..., αn ∈ R, we have

α1 u1 + α2 u2 + ... + αn un = 0,

implies
0 = h0, ui i = hα1 u1 + α2 u2 + ... + αn un , ui i = αi , ∀ i = 1, 2, ..., n.

Further, for each x ∈ E, the vector

y = x − hx, u1 i u1 − hx, u2 i u2 − ... − hx, un i un

is orthogonal with ui , i = 1, 2, ..., n; since

hy, ui i = hx − hx, u1 i u1 − hx, u2 i u2 − ... − hx, un i un , ui i

= hx, ui i − hx, ui i hui , ui i
| {z }
=1
= 0.

Exercise 14. Let q be a quadratic form over Rn which has the matrix A in the standard
basis, and let λmax be the greatest eigenvalue of A. Prove the following inequality :

q (x1 , x2 , ..., xn ) ≤ λmax x21 + x22 + ... + x2n .

Solution. Let x ∈ Rn with

q
kxk2 = x21 + x22 + ... + x2n = 1,

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 82

and let {u1 , u2 , ..., un } an orthonormal basis formed by the eigenvectors of A. We have

x = α1 u1 + α2 u2 + ... + αn un

with α12 + α22 + ... + αn2 = 1, since kxk22 = hx, xi = 1. In this case, we can write

q (x) = xt Ax
= (α1 u1 + α2 u2 + ... + αn un )t A (α1 u1 + α2 u2 + ... + αn un )
= α12 ut1 Au1 + α22 ut2 Au2 + ... + αn2 utn Aun
= λ1 α12 ut1 u1 + λ2 α22 ut2 u2 + ... + λn αn2 utn un (since Aui = λi ui , i = 1, 2, ..., n)
≤ λmax α12 ut1 u1 + α22 ut2 u2 + ... + αn2 utn un

= λmax α12 + α22 + ... + αn2 (since uti ui = 1, i = 1, 2, ..., n)

= λmax (since α12 + α22 + ... + αn2 = 1).

Hence, q (x) ≤ λmax .

Now, for each x ∈ Rn we put

x
u= ; i.e., kuk2 = 1.
kxk2

Since q (u) ≤ λmax , it follows that

 
x 1
q = q (x) ≤ λmax .
kxk2 kxk2

Therefore,
q (x) ≤ λmax kxk2 = λmax x21 + x22 + ... + x2n .

This completes the proof.

Exercise 15.
1. Let A be a hermitian matrix, and let

f : Cn × Cn → C
(x, y) 7→ xt Ay.

Prove that f is a hermitian form.

2. Let f be a sesquilinear Hermitian form over a vector space E. Show that

∀ x, y, y 0 ∈ E, ∀ α, β ∈ C : f (x, αy + βy 0 ) = αf (x, y) + βf (x, y 0 ) .

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 83

Notice that if f : E × E → C is linear on the left and semilinear on the right then f is
called ”sesquilinear form” ; i.e., ∀ x, x0 , y, y 0 ∈ E, ∀ λ ∈ C :

• f (λx + x0 , y) = λf (x, y) + f (x0 , y)

• f (x, λy + y 0 ) = λf (x, y) + f (x, y 0 ) .

A hermitian sesquilinear form is a sesquilinear form f over E satisfying

f (x, y) = f (y, x) for all x, y ∈ E.

1. Let A be a hermitian matrix. We prove that the mapping

f : Cn × Cn → C
(x, y) 7→ xt Ay

is a hermitian form. (i.e., f is hermitian sesquilinear form ). In fact, for every x, x0 , y ∈

Cn and λ ∈ C, we have

t
f (λx + x0 , y) = (λx + x0 ) Ay
t
= λxt Ay + (x0 ) Ay
= λf (x, y) + f (x0 , y) .

Thus, f is linear from the left. Similarly, for every x, y, y 0 ∈ Cn and λ ∈ C, we have

f (x, λy + y 0 ) = xt A(λy + y 0 )
= λxt Ay + xt A(y)0
= λf (x, y) + f (x, y 0 ) .

Thus, f is semi-linear from the right.

Moreover, for each x, y ∈ Cn , we have

f (x, y) = xt Ay
= (xt Ay)t (since xt Ay ∈ C)
= (y)t At x
= y t (At )x
= y t A∗ x
= y t Ax (since A is hermitian)
= f (y, x) .

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 84

2. Let f be a sesquilinear hermitian form over a vector space E. We show that

∀x, y, y 0 ∈ E, ∀α, β ∈ C : f (x, αy + βy 0 ) = αf (x, y) + βf (x, y 0 ) .

In fact, we have

∀ x, y, y 0 ∈ E, ∀ α, β ∈ C : f (x, αy + βy 0 ) = f (αy + βy 0 , x)
= αf (y, x) + βf (y 0 , x)
= αf (y, x) + βf (y 0 , x)
= αf (x, y) + βf (x, y 0 ) .

Exercise 16. Let

f : C2 × C2 → C
(x, y) 7→ 4x1 y1 + (2 − i) x1 y2 + (2 + i) x2 y1 − 5x2 y2 .

Show that f is a hermitian sesquilinear form. Calculate f (x, x), where x ∈ C2 .

Solution. We write f in the form

! !
 4  2−i y1
f (x, y) = x1 x2
2 + i −5 y2
! !
x1 y1
= X t AY , where X = and Y = .
x2 y2

Since A∗ = A, then A is Hermitian. Then f is a Hermitian sesquilinear form.

Calculate f (x, x), where x ∈ C2 . In fact, we have

f (x, x) = 4 |x1 |2 + (2 − i) x1 x2 + (2 + i) x2 x1 − 5 |x2 |2 .

Exercise 17.
1. Diagonalize the following Hermitian quadratic forms :
i) q1 = ix1 x2 − ix2 x1 , E = C2 .
ii) q2 = x1 x1 + ix1 x2 − ix2 x1 + x2 x2 , E = C2 .
iii) q3 = x1 x1 + a12 x1 x2 + a21 x2 x1 + a22 x2 x2 .
2. Deduce the signature of the quadratic form given by :

q20 = αx1 x1 + ix1 x2 − ix2 x1 + x2 x2 , α ∈ R.

Solution.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 85

– We can write

q1 = ix1 x2 − ix2 x1
= x1 (ix2 ) + x1 (−ix2 )


= x1 −ix2 + x1 (−ix2 ) (which is of the form z1 z2 + z1 z2 )
1 1 1 1
= |x1 − ix2 |2 − |x1 + ix2 |2 (since z1 z2 + z1 z2 = |z1 + z2 |2 − |z1 − z2 |2 )
2 2 2 2
2 2
= |f1 | − |f2 | ,

where f1 et f2 are linearly independent forms over C2 ; since

1 −i
6= 0.
1 i

The signature of q1 is (1, 1) .

– Likewise, we have

q2 = x1 x1 + ix1 x2 − ix2 x1 + x2 x2
= (x1 − ix2 ) (x1 + ix2 )
= (x1 − ix2 ) (x1 − ix2 )
= |x1 − ix2 |2
= |f1 |2 .

The signature of q2 is (1, 0) .

For the quadratic form q20 = αx1 x1 + ix1 x2 − ix2 x1 + x2 x2 , α ∈ R. We see that

q20 = (α − 1) x1 x1 + q2
= (α − 1) |x1 |2 + |x1 − ix2 |2 .

We deduce that 
 α = 1, the signature is (1, 0) .

α > 1, the signature is (2, 0) .

α < 1, the signature is (1, 1) .


c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 86

– We have

q3 = x1 x1 + a12 x1 x2 + a21 x2 x1 + a22 x2 x2

= (x1 + a21 x2 ) (x1 + a12 x2 ) + (a22 − a12 a21 ) x2 x2
= (x1 + a21 x2 ) (x1 + a21 x2 ) + (a22 − a12 a21 ) x2 x2
= |x1 + a21 x2 |2 + (a22 − a12 a21 ) |x2 |2 .
| {z }
∈R

Exercise 18. Let E be a vector space with dimension 3, and let q be the hermitian
quadratic form over E given by

q = a12 x1 x2 + a13 x1 x3 + a12 x2 x1 + a23 x2 x3 + a13 x3 x1 + a23 x3 x2 .

Give the diagonal form of using Gauss method.

Solution. We have

q = a12 x1 x2 + a13 x1 x3 + a12 x2 x1 + a23 x2 x3 + a13 x3 x1 + a23 x3 x2

= x1 (a12 x2 + a13 x3 ) + x1 (a12 x2 + a13 x3 ) + a23 x2 x3 + a23 x3 x2
= x1 (a12 x2 + a13 x3 ) + a23 x2 x3 + x1 (a12 x2 + a13 x3 ) + a23 x3 x2
a23 a23 a12
= x1 (a12 x2 + a13 x3 ) + x2 (a12 x2 + a13 x3 ) − x2 x2 +
a13 a13
a23 a23 a12
x1 (a12 x2 + a13 x3 ) + x2 (a12 x2 + a13 x3 ) − x2 x2
a13 a13
| {z }
     
a23 a23 a23 a12 a23 a12
= x1 + x2 (a12 x2 + a13 x3 ) + x1 + x2 (a12 x2 + a13 x3 ) − + x2 x2
a13 a13 a13 a13
  2   2  
1 a23 1 a23 a23 a12
= x1 + + a12 x2 + a13 x3 − x1 + − a12 x2 − a13 x3 − 2Re |x2 |2 .
2 a13 2 a13 a13

Exercise 19. Diagonalize the Hermitian quadratic form given by its matrix :
 
0 1−i 0
Mq =  1 + i 0 i .
 

0 −i 0

Here, Mq is the matrix of the Hermitian quadratic form q with respect to the standard basis
of C3 .

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 87

Solution. We have

q = (1 − i) x1 x2 + (1 + i) x2 x1 + ix2 x3 − ix3 x2
= x2 [(1 + i) x1 + ix3 ] + x2 [(1 − i) x1 − ix3 ]
= x2 [(1 − i) x1 − ix3 ] + x2 [(1 − i) x1 − ix3 ] (which is of the form z1 z2 + z1 z2 )
1 1
= |x2 + (1 − i) x1 − ix3 |2 − |x2 − (1 − i) x1 + ix3 |2
2 2
2 2
= |f1 | − |f2 | .

The signature is (1, 1) .

Exercise 20. Let  
1 0 b
B = 0 a+i a  , a, b ∈ C.
 

b b + 1 b − ai
For which values of the parameters a and b is the matrix B Hermitian ? and then find its
Hermitian quadratic form.
Solution. The matrix B is Hermitian if and only, if
   
1 0 b 1 0 b

t
B = B ⇔ 0 a+i a = 0 a−i b+1 
   

b b + 1 b − ai b a b + ai

 a+i∈R

⇔ b+1=a

b − ai ∈ R


 a = α − i, where α ∈ R

⇔ b=α−1−i

α − 1 − i − (α − i) i ∈ R


 a = α − i, where α ∈ R

⇔ b = (α − 1) − i

α − 2 − (1 + α) i ∈ R


 α = −1,

⇔ a = −1 − i

b = −2 − i.


Therefore,  
1 0 −2 − i
B= 0 −1 −1 + i  .
 

−2 + i −1 − i −3

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 88

The Hermitian quadratic form of B is given by

qB = x1 x1 + (−2 − i) x1 x3 − x2 x2 + (−1 + i) x2 x3 + (−2 + i) x3 x1 + (−1 − i) x3 x2 − 3x3 x3 .

Exercise 21 Prove that every real quadratic form q = xt Ax is diagonalizable.

Further, prove that if q is definite positive, then the integral
Z Z Z
I= ... e−q(x1 ,x2 ,...,xn ) dx1 dx2 ...dxn
Rn

converges and calculate its value 1 .

Solution. Let

q = xt Ax
  
a11 a12 · · · a1n x1
x2
  
12 a22 · · ·
  a a2n   
= x1 x2 · · · xn   .. 
.
  
  
a1n a2n · · · ann xn

be quadratic form over Rn . We prove that q is diagonalizable. However, since A is sym-

metric, there exists an orthogonal matrix P such that A = P DP t , where

D = diag {λ1 , λ2 , ..., λn } .

It follows that

t
q = xt Ax = xt P DP t x = xt P D P t x = P t x DP t x.

Setting
 
v1
v2
 
t
 
P x=v= .. ,
.
 
 
vn
R +∞ 2 √
1. Use the following well-known formula : −∞
e−t = π.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 89

Implies

q = v t Dv
  
λ1 v1
 λ2 v2
  
  
= v1 v2 · · · vn  ..
 .. 

 . 
 .


λn vn
= λ1 v12 + λ2 v22 + ... + λn vn2 ,

where (λi )i=1,2,...,n are the eigenvalues of A. Further, suppose that q is definite positive, i.e.,
λi > 0 for every i = 1, 2, ..., n. Then
Z Z Z
I = ... e−q(x1 ,x2 ,...,xn ) dx1 dx2 ...dxn
Z Z Rn
Z
e−(λ1 v1 +λ2 v2 +...+λn vn ) dv1 dv2 ...dvn , where αJ ∈ R∗
2 2 2
= αJ ...
Rn
 Z +∞ n
αJ −t2
= √ e dt
λ1 λ2 ...λn −∞
αJ √
n
= √ π .
λ1 λ2 ...λn

Note that
1
dv1 dv2 ...dvn = dx1 dx2 ...dxn
αJ

Exercise 22. Let q = xt Ax be a quadratic form over the vector space Rn . Prove that

q is nondegenerate ⇔ det (A) 6= 0 (i.e., A is invertible).

Solution. By Definition 3.6, recall that ker f = {x ∈ E; xt Ay = 0 for each y ∈ E}. Then

ker f = {0} ⇔ ∀ y ∈ Rn : xt Ay = 0 ⇒ x = 0
⇔ ∀ y ∈ Rn : y t At x = 0 ⇒ x = 0
⇔ At x = 0 ⇒ x = 0 ; since ∀ y ∈ Rn : y t At x = 0 ⇔ At x = 0

⇔ At ∈ GLn (R)
⇔ A ∈ GLn (R) .

Exercise 23. Let E = R2 [x] be the vector space of polynomials having degree ≤ 2, and

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 90

let

Q : E→R
p 7→ p (0) p (1) .

1. Prove that Q is a quadratic form, and then give its polar form f .
2. Determine MQ (B), where B is the canonical basis of E.
3. Prove that f is degenerate. Is it positive ?, definite positive ?, negative ?, definite
negative ?
Solution.
1. From simple computation, the polar form of Q is given by

f : R2 [x] × R2 [x] → R
1
(p, q) →
7 f (p, q) = (Q (p + q) − Q (p − q))
4
1 1
= p (0) q (1) + p (1) q (0) .
2 2

2. Calculate the matrix MQ (B), where B is the canonical basis of R2 [x]. We have

1
1
f (1, 1) = 1, f (1, x) = , f 1, x2 = , f x, x2 = 0, f x2 , x2 = 0.



2 2

Therefore,
1 1
 
1
 2 2 
 1 
MQ (B) =  0 0 .
 2 
 1 
0 0
2
3. Since det (MQ (B)) = 0, then f is degenerate.
Further, Q neither positive nor negative ; since
(
Q (2x − 1) = (−1) × 1 = −1 < 0,
Q (−x − 2) = (−2) × (−3) = 6 > 0.

1
√ √
Remark 6.1. The eigenvectors of MQ (B) are 2
3 + 12 , 12 − 1
2
3, 0. Then MQ (B) neither
positive nor negative.

Exercise 24. Let (E, h., .i) an inner product space (a pre-Hilbert space) and let F be a

⊥
⊥
subspace of E. Prove that F ⊂ F ⊥ and so F = F ⊥ whenever E has finite dimension.

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 91

Solution. We have
(
F ⊥ = {x ∈ E ; hx, yi = 0 for each y ∈ F } ,

⊥ 
F ⊥ = x ∈ E ; hx, yi = 0 for each y ∈ F ⊥ .

⊥
⊥
We prove that F ⊂ F ⊥ . Let x0 ∈ F . Assume that x0 ∈ / F ⊥ , there exists y0 ∈ F ⊥ such
that hx0 , y0 i =
6 0. But, hx, y0 i = 0 for every x ∈ F . A contradiction.

⊥
Next, assume that E is a finite dimension space. Since E = F ⊕ F ⊥ = F ⊥ ⊕ F ⊥ , by
(1.1) we get

⊥
dim F + dim F ⊥ = dim E et dim F ⊥ + dim F ⊥ = dim E.

⊥
⊥
⊥
which gives dim F = dim F ⊥ . Moreover, since F ⊂ F ⊥ , we have F = F ⊥ .
Exercise 25. Let ϕ be the mapping defined on the vector space E = Rn [x] by
Z b
ϕ (P, Q) = P (t) Q (t) dt, where a < b.
a

1. Prove that ϕ is an inner product (a scalar product).

2. For n = 2, calculate Mϕ (B) "this is the matrix of ϕ in the standard basis of R2 [x]".
3. Apply Cauchy-Schwarz’s inequality.
Solution. We prove that ϕ is an inner product. That is, ϕ is a symmetric bilinear form
definite positive.

For each (P, Q, P1 , Q1 ) ∈ E 4 and for each λ ∈ R, we have

Z b
ϕ (λP + P1 , Q) = (λP + P1 ) (t) Q (t) dt
a
Z b
= (λP (t) Q (t) + P1 (t) Q (t)) dt
a
Z b Z b
= λ P (t) Q (t) dt + P1 (t) Q (t) dt
a a
= λϕ (P, Q) + ϕ (P1 , Q) ,

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 92

and also, we have

Z b
ϕ (P, λQ + Q1 ) = P (t) (λQ + Q1 ) (t) dt
a
Z b
= P (t) (λQ (t) + Q1 (t)) dt
a
Z b Z b
= λ P (t) Q (t) dt + P (t) Q1 (t) dt
a a
= λϕ (P, Q) + ϕ (P, Q1 ) .

Then ϕ is a bilinear form. Further, ϕ is symmetric since for each (P, Q) ∈ E 2 , one has
Z b Z b
ϕ (P, Q) = P (t) Q (t) dt = Q (t) P (t) dt = ϕ (Q, P ) .
a a

Rb
For all P ∈ E − {0}, we have ϕ (P, P ) = a
P 2 (t) dt > 0. Then ϕ is definite positive. Hence
ϕ is an inner product.
Now, we calculate Mϕ (B) :
 
ϕ (1, t) ϕ (1, t2 )
ϕ (1, 1)
Mϕ (B) =  ϕ (t, t) ϕ (t, t2 ) 
ϕ (1, t)
 

ϕ (1, t2 )
ϕ (t, t2 ) ϕ (t2 , t2 )
 Rb Rb
Rb 2 
tdt dt
t dt
a
Rb 2 R ab
Rab 3 
=  tdt
t dt a t dt 

Rab 3 R ba 2
Rb
a
t dt a t4 dt
a
t dt
 
b − 1 b2 − a2 b3 − a3
 2 1 2 3 2 3 4 3 4 
 
 b −a b −a b −a 
=  .
 2 3 4
 b3 − a3 b4 − a4 b5 − a5 


3 4 5

– From Cauchy-Schwarz inequality, for each (P, Q) ∈ E 2 , we have

|ϕ (P, Q)|2 = |hP, Qi|2 ≤ hP, P i hQ, Qi

That is,
Z b 2 Z b Z b
2
P (t) Q (t) dt ≤ P (t) dt Q2 (t) dt.
a a a

Exercise 26. Let A be a symmetric matrix with real entries. Prove that the quadratic
form q = xt Ax is definite positive if and only, if the eigenvalues of A are strictly positive.
Solution. Let q = xt Ax be quadratic form definite positive, where A ∈ Sn (R) and let

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 93

(λ, x) be eigenpair of A. Since x 6= 0, it follows that

0 < xt Ax = hx, Axi = hx, λxi = λ hx, xi ⇔ λ > 0 ; because λ ∈ R (A is symmetric).

| {z }
>0 ; since x6=0.

Exercise 27. Let f be a bilinear form on a vector space E. Show that the mapping :

q : E→R
x 7→ f (x, x)

is a quadratic form.
Solution. Let f be a bilinear form over E. Clearly, the mapping

ϕ E×E →R
:
f (x, y) + f (y, x)
(x, y) →
7 ϕ (x, y) =
2

is symmetric bilinear form. Further, for each x ∈ E we have ϕ (x, x) = f (x, x) = q (x).
Then q is a quadratic form over E.
Exercise 28. Let q be a quadratic form over E. Prove that two vectors x and y satisfying
q (x) q (y) < 0 are independent.

Solution. Assume, by the way of contradiction that x, y are dependent. Since x and y
are nonzero (otherwise, if x or y is zero then q (x) q (y) = 0), there exists λ ∈ R∗ such that
y = λx. By (3.5), q (x) q (y) = λ2 (q (x))2 > 0, this contradicts our assumption.
Exercise 29. Diagonalize the quadratic form

q (x, y) = ax2 + 2bxy + cy 2 .

Deduce its signature.

Solution. Note that
! !
  a b x
q (x, y) = x y
b c y

1. Assume that a 6= 0. By (3.11), we put

1
x = x0 − (by 0 ) and y = y 0 .
a

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 94

It follows that

q (x, y) = ax2 + bxy + cy 2

 2  
0 b 0 b 0 0 2
= a x − y + 2b x − y y + c (y 0 )
0
a a
b2
 
0 2
= a(x ) + c − (y 0 )2
a
b2
 
b 2
= a(x + y) + c − y2
a a
b2
 
2
= a · |f1 | + c − · |f2 |2 ,
a

where f1 = x + ab y and f2 = y are two independent linear forms over R2 , since

1 0
b 6= 0.
1
a
2
– If a, c − ba > 0, then the signature of q is (2, 0) .
2
– If a, c − ba < 0, then the signature of q is (0, 2) .
2 2
– If a > 0 and c − ba < 0 or a < 0 and c − ba > 0, then the signature of q is (1, 1).
2. Assume that a = 0 and b 6= 0. There are two cases.
2.1. For c = 0, we let
x = x0 + y 0 and y = x0 − y 0 ,

which implies

q (x, y) = 2bxy = 2b (x0 + y 0 ) (x0 − y 0 )

x+y 2 x−y 2
= 2b(x0 )2 − 2b(y 0 )2 = 2b( ) − 2b( ).
2 2

2.2. For c 6= 0, we let

1
y =u− (bv) and x = v.
c
It follows that

q = 2bxy + cy 2
   2
b b
= 2bv u − v + c u − v
c c
2
b
= cu2 − v 2
c
2
b2

b
= c y + x − x2 .
c c

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 95

Exercise 30. Let A ∈ Mn (R) be a symmetric definite positive matrix. Using two me-
thods, prove that det (A) is strictly positive.
Solution. 1st method. We show that A is definite positive ⇔ ∀ λ ∈ Sp (A) : λ > 0. In
fact, if A is definite positive, for each eigenpair (λ, x) of A we have

0 < xt Ax = hx, Axi = hx, λxi = λ hx, xi ⇔ λ > 0, since λ ∈ R (A is symmetric).

| {z }
>0, since x6=0.

It follows that
Y
det (A) = λ > 0.
λ∈Sp(A)

2nd method. In the case when A is symmetric definite positive, from Theorem 2.5, we
have
A = M t M , where M ∈ GLn (R) .

Hence, det (A) = det (M t M ) = (det (M ))2 > 0 (note that det (M ) = det (M t ) and det M 6= 0
since M is invertible).

Remark 6.2. Let A ∈ Mn (C) be a hermitian definite positive matrix. Then det (A) ∈
R∗+ .

c
2024, University 8 Mai 45 Guelma. Department of Mathematics. Djamel Bellaouar
 Conclusion

Quadratic forms have many applications in cryptography. In the context, it is very

interesting to know large prime numbers which are represented by some special quadratic
forms. For example, it is well known that every prime number of the form 4k + 1 can be
represented by the quadratic form q = x2 + y 2 . Currently, there are many open problems
on the distribution of values of quadratic forms, some others include quadratic forms
involving systems of forms having k-tuple of variables. For more information, see the
paper Ten problems on quadratic forms stated in [7]. In addition, in sesquilinear algebra, the
study of Hermitian spaces is the basic of Hermitian Geometry.

96
 Bibliographie

[1] I. Kaplansky, Linear Algebra and Geometry, A Second Course," Allyn & Bacon, Boston,
1969.
[2] W. Scharlau, Quadratic and Hermitian Forms, Springer-Verlag, Berlin, 1985.
[3] D. Shapiro, Compositions of Quadratic Forms, de Gruyter, Berlin, 2000.
[4] M. Knebusch, Specialization of Quadratic and Symmetric Bilinear Forms (Vol. 11). Sprin-
ger Science & Business Media, 2011.
[5] Kurgalin, Sergei, et al., Bilinear and Quadratic Forms. Algebra and Geometry with Py-
thon (2021), 335-356.
[6] Milnor, J. W., & Husemoller, D. (1973). Symmetric bilinear forms (Vol. 5). Berlin : Sprin-
ger.
[7] K. Szymiczek, Ten problems on quadratic forms. Acta Mathematica et Informatica Uni-
versitatis Ostraviensis, 10(1)(2002), 133-143.

97