CONTENTS

CHAPTER NO. TITLE PAGE NO.

1 INTRODUCTION

2 PREMLIMINARIES

3 DIRECTED GRAPHS

4 RELATIONS

5 GROUPS

6 TREES

7 CONCLUSION

8 REFERENCES

1
 Algebraic and Matrix form
Introdution:

Linear algebra is a very useful subject and its basic concepts arose and were
used in different areas of mathematics and its applications. It is therefore not
simplifying that the subject had its roots in such diverse field as number theory,
geometry, abstract algebra, analysis and physics. Among the elementary concepts of
linear algebra are linear equations, matrices determinants, linear transformations,
linear independence, dimensions, bilinear forms, quadratic forms and vector space.
Since these concepts are closely interconnected, several usually appear in a given
context and it’s often impossible to disengage them.

Linear algebra is the branch of the mathematics concerning vector spaces and
linear mappings between such spaces. It includes the study of lines, planes and
subspaces but is also concerned with properties of points with coordinates that
satisfy a linear equation form a hyper plane in an n- dimensional space. The
condition under which a set of n hyper planes intersect in a single point is an
important focus of study in linear algebra.

In chapter 1 contains the definition and basic properties of the type of

algebraic system where properties of the set of solutions of a homogeneous system
of equation s motivate and illustrate the notation of a vector space.

In chapter 2, linear and quadratic forms are satisfied, we examine “forms” that
is functions from a vector space V to its field, which are either linear or quadratic.
The linear forms comprise the dual space of V.

In chapter 3, bilinear, quadratic and hermitian forms are satisfied in their

natural setting, as representation of certain types of function on a linear vector space.

2
 In chapter 4, introduce and discuss the problem of finding canonical form s
for matrices with respect to similarity via representation of linear transformation.

In chapter 5, definition of linear transformation on a vector space followed by

the correlation of this notation with matrices. The ensuring discussion leads to the
introduction of two further types of algebraic system groups and rings.

In chapter 6, appear all the standard theorems pertaining to real symmetric

and hermitian matrices with what are believed to be new and greatly simplified
proofs, all of which stem for a single argument.

3
 CHAPTER - I

BASIC DEFINITION

Definition

A function f on V x V to F, where V is a vector space over a field F, is called

bilinear if it is linear in each argument.

That is,

( , )= ( , )+ ( , )

( , , ,

(c , , , ∀ and c F

Definition

If in a bilinear function defined on a vector space V we equate the two

arguments, the resulting function of one variable is called a quadratic function.

(x) = XT AX = ∑ , ij xi xj

The function is a quadratic form.

Definition

A square matrix is called symmetric the transpose is AT = A and aij = aji. A-

1
is also symmetric skew – symmetric matrix k, the transpose is –k, since kij
= -kji. Eigen values are pure imaginary, Eigen vectors are orthogonal. ekt is an
orthogonal matrix.

Example

1 2
( x1, x2 ) = (x1)2 + 6x1x2 + 2(x2)2
4 2

4
Definition

A C1 transformation is an elementary row transformation of type I followed

by the corresponding column transformation. A CIII transformation is type – III
transformation Rij(c) with i > j followed by the corresponding column
transformation.

Definition

A principal sub matrix of order k of an nth- order matrix A = (aij) is a sub

matrix obtained by deleting n-k rows and the corresponding columns of A.

a11 a12 … … a1k

Dk(A) = det a21 a22 … … a2k

⋮ ⋮ …… ⋮

ak1 ak2 … … akk

k = 1, 2,…,n

LINEAR TRANSFORMATION ON A VECTOR SPACE

Definition

A set G = { , , … } together with a composition in G is called a group, if

the following conditions

G1 Associative law :

∀ , , in G

5
G2 Unit element:

G contains at least one element 1, called a right unit element with the property
1 = ∀ in G.

G3 Inverse element:

For each unit element 1 and every element in G the equation = 1 has a
solution in G.

Definition

A subgroup N of a group G is normal in G iff N = N ∀ in G. Here N for

a fixed in G, denotes the set of all elements where ranges over N such a subset
of G is called a coset of N.

Example

Consider the set of six functions { f1, f2, . . . , f6 } of a complex variable Z

where,

f1(Z) = Z, f2(Z) = , f3(Z) = , f4(Z) = , f5(Z) = 1 , f6(Z) =

With the composition fi fj introduced earlier for functions,

(fifj) (Z) = fi (fj(z))

6
The multiplication tables for the set follows,

f1 f2 f3 f4 f5 f6

f1 f1 f2 f3 f4 f5 f6

f 2 f1 f2 f3 f4 f5 f6

f 3 f1 f2 f3 f4 f5 f6

f 4 f1 f2 f3 f4 f5 f6

f 5 f1 f2 f3 f4 f5 f6

f 6 f1 f2 f3 f4 f5 f6

The group is non- commutative and has N = { f1, f2 , f3 }

The cosets of N in G are N and f4N, which form a cyclic group of two
elements.

Definition

If A is a linear transformation on V and S is a subspace V, the set of all vectors

of the form A , where is in S, is denoted by AS. The particular set AV of this type
is called the range of A and denoted by R(A).

The set of all vectors in V such that A = 0 is called the null space of A and
denoted by N(A).

7
Definition

A ring is a system consisting of a set D and two binary composition in D called

addition (+) and multiplication (.) such that

(i) D together with addition is a commutative group.

(ii) Multiplication is associative:
a (b c) = (a b) c
(iii) The following distributive laws
a(b+c)=ab+ac
( a + b ) c = a c + b c.

If multiplication in D is commutative D is called a commutative ring.

Definition

The selection of a basis { 1, 2, ..., n } for a vector space V over F and the
designation of vectors by there – coordinates, the linear transformation →A
on V assumes the concrete form X → AX, where A = ( A ; )

Now X → AX may be interpreted as a linear transformation on Vn(F).

V X Vn(F)

A AX

8
Definition

A subspace S of a vector space V is called an invariant space of the linear

transformation A on V iff AS ⊆ S.

Definition

The matrix PAP-1 is called the transform of A by P. A matrix B so related to

A for some P is called similar to A : A ∼ B.

Definition

A vector ≠ 0 is called a characteristics vector of a linear transformation A if

there exists a scalar C such that A = C . A scalar C is called a “characteristics
value” of A if there exists a vector ≠ 0 such that A =C .

Definition

A relation for an A – generating system { 1, 2, ..., t } is a column matrix

f1( , f2( ,… , ft( T

, f i( in F(λ) such that,

f”( 1+ A2(A) 2+ …+ At(A) t = 0

A matrix M = ( ) such that each column is a relation for { 1, 2, ..., t } is

called a relation matrix for { 1, 2, ..., t }.

Definition
m-1
The companion matrix of a monic polynomial f( ) = a0 + a1 +...+ am-1 +
m
is the matrix

9
 0 0 0 …0
1 0 0 …0
0 1 0 …0
C(f) = ……………………
……………………
……………………
0 0 0 …1

Which has the negatives of the coefficient of f in the last column, i’s in the
diagonal just below the principal diagonal and 0’s elsewhere.

10
 CHAPTER – II

LINEAR AND QUADRATIC FORMS

Definition:

Let V be a vector space over K. A linear form on V is a linear map from V to

K, where K is regarded as a 1- dimensional vector space over K.

f(V1+V2) = f(V1) + f(V2)

f(c V) = c f(V) ∀ V1,V2∈ V and c ∈ K

If dim(V) = n, then a linear form is represented by a 1 n matrix over K.

(i.e) a row vector of length n over K.

f = ( a1,a2,. . ., an )

V = ( x1,x2, . . . , xn)T

f(V) = (a1,a2, ….,an) ⋮

f(V) = a1x1 + a2x2 +….. + anxn

Definition:

Linear forms can be added and multiplied by scalars in the obvious way

(f1+f2) V = f1(V) + f2(V)

(c f) V = c f(V)

So they form a vector space, which is called the dual space of V and is denoted
by V*.

11
Proposition

If V is finite- dimensional, then so is V* and dim(V*) = dim(V).

Proof:

If (V1,. . . ,Vn) is a basis for V and a1, . . . an are any scalars, then there is a
unique linear map f with the property that f(Vi) = ai for i=1,2, . . . n.

f(c1V1 + . . . + cnVn) = a1c1 + . . . +ancn

It is represented by the row vector ( a1, . . . , an ) and its action on Kn is by

matrix multiplication.

Let fi be the linear mp define by

1
fi(Vj) =
0

Then ( f1,. . . . fn) form a basis for V* , the linear form f defined in the preceding
is a1f1 + . . . +anfn.

Then basis is called the dual basis of V* corresponding to the given basis for
V.

∴ It has n elements.

Dim(V*) = n = dim(V)

Definition:

The Kronecker delta ij for i,j ∈ { 1,2, . . . n } is defined by the rule that

1
ij =
0

12
2.1.1 Adjoint

Definition:

Let : V → W be a linear map. There is a linear map *

: W*→ V* defined by
( *(f)) (V) = f( (V))
*
The map is called the adjoint of .

Proposition:

Let : V → W be a linear map. Choose bases B for V and C for W and let A
be the matrix representing relative to the bases. Let B* and C* denote the dual
bases of V* and W* corresponding to B and C.
*
Then the matrix representing relative to the basis C* and B* is the transpose
of A, that is AT.

2.1.2 Change of basis

Proposition:

Let B and B' be bases for V and B* and (B)* the dual bases of the dual space.
Then

*, (B')* = ( , B')-1

Proof:

If P = PB,B' has (i,j) entry Pi.j and Q = *, (B')* has (i,j) entry Qi.j

We’ve =∑

=∑

And so = ( )

13
 = ∑ ∑

=∑ ∑

=∑

Now, qkj is the (j, k) entry of QT.

We’ve I = QTp

QT = p-1

Q = (p -1)T

Q = (pT)-1 as required.

2.2 Quadratic forms

Definition:

A quadratic form in n variables x1, . . ,xn over a field k is a polynomial

∑ ∑

In the variables in which every term has degree two.

Proposition:

A basis change with transition matrix P replaces the symmetric matrix A

representing a quadratic form by the matrix PTAP.

Definition:

Two symmetric matrices A . A' over a field k are congruent if A' = PTAP for
some invertible matrix P.

14
Proposition:

Two symmetric matrices are congruent iff they represent the same quadratic
form with respect to different bases.

2.2.2. Reduction of quadratic forms

Theorem:

Let q be a quadratic form in n variables x1, . . . ,xn over a field k whose

characteristics is not 2. Then by a suitable linear substitution to new variables

y1, . . . yn We can obtain

q = c1 + c2 + c3 + . . . + cn For some c1 … cn∈ k.

Proof:

A quadratic form in n variables x1,…,xn over a field k is a polynomial

∑ ∑

In the variables in which every term has degree two.

A form in one variable is certainly diagonal.

Now, assume that the theorem is true for forms in n-1 variables,

q( x1, x2, . . .xn) = ∑ ∑

Where aij = aji for i≠j

Case (i):

Assume that aij ≠ 0 for some i.

By a permutation of the variables. We can assume that a11 ≠ 0

15
 y1 = x1 + ∑ ⁄

Then we’ve

a11 = a11 +2∑

Where q' is a quadratic form in x2…xn

q(x1, x2, . . . xn) = + q’’ (x2, x3, . . .xn)

Where q'' is the part of q not containing x1 minus q'.

By induction,

q’’ (x2, x3, . . .xn) = ∑

Case (ii)

All aii are zero, but aij ≠ 0 for some i ≠ j

= )

So that, = ( = (

We obtain a new form for q which does contain a non-zero diagonal term.

Case (iii)

All aij are zero. Now q is the zero form and there is nothing to prove

Take c1= c2=. . . = cn = 0

Example

Consider the quadratic form

q(x, y, z) = x2 + 2xy + 4xz + y2 + 4 z2

16
(x+y+2z)2 = x2 + y2 + 4z2 + 2xy + 4yz + 4xz.

q(x, y, z) = x2 + 2xy + 4xz + y2 + 4z2 + 4yz – 4yz.

= (x+y+2z)2- 4yz

= (x+y+2z)2 – (y2 + z2 + 2yz – y2 – z2 + 2yz)

= (x+y+2z)2 – ( (y+z)2 – (y2+z2-2yz)

= (x+y+2z)2 – ( (y+z)2 – (y-z)2)

= (x+y+2z)2 – (y+z)2 + (y-z)2

= u2 + v2 – w2

u = x +y + 2z

v =y–z

w =y+z

The matrix representing the quadratic form namely,

aii = coefficient of xi2

aij = ( coefficient of xi yj)

⇒ x2 + y2 + 2xy + 4xz + 4z2

a11 = coefficient of x2 =1

a22 = coefficient of y2 =1

a33 = coefficient of z2 =4

a12 = ( coefficient of xy ) = = 1

17
 a13 = ( coefficient of xz ) = = 2

a21 = ( coefficient of yx ) = = 1

a23 = ( coefficient of yz ) = 0

a31 = ( coefficient of zx ) = = 2

a32 = ( coefficient of zy ) = 0

1 1 2
A= 1 1 0
2 0 4
1 1 2
~ 0 0 2 R2 → R2 – R1 , R3 → R3 – 2R1
0 2 0
1 2 1
~ 0 2 0 C3 ~ C2
0 0 2
1 2 1
~ 0 1 0 2 , R3 → 2
0 0 1
1 0 0
~ 0 1 0 C3 → C2 – 2C1 , C3 → C3 – C1
0 0 1
1 0 0
A' ~ 0 1 0
0 0 1

Then the invertible matrix P such that PTAP = A1

Thus any quadratic form can be reduced to the diagonal shape

18
 2 2
1 x1 +...+ nxn

By a linear substitution. But this is still not a “canonical form for congruence”.

2.2.3 Quadratic and bilinear forms

Definition:

a) Let b : V V → k be a function of two variables from V with values in k.

We say that b is a bilinear form if it is a linear function of each variables
when the other is kept constant.

b(V, w1+w2) = b(V,w1) + b(V,w2)

b(V,cw) = cb(V,w)

With two similar equations involving the first variables

b is symmetric
b(v,w) = b(w,v) ∀ v,w ∈ V
b) Let q : V → k be a function, q is a quadratic form,
-q(cv) = c2q(v) ∀ c ∈ k, v ∈ V

The function b defined by

b(v,w) = ( q(V+w) –q(v) – q(w))

Is a bilinear form on V.

Remarks:

The bilinear form in the second part is symmetric, and the division by 2 in the
definition is permissible because of our assumption that the characteristics of k is
not 2. The 1st equation
19
 (cx)2 = c2x2

The second has the form

((x+y)2 – x2-y2 ) = xy

And xy is the prototype of a bilinear form. Then linear function x and y is

constant.

Note:

The polarization formula

b(x,y) = (q(x+y) –q(x) – q(y))

2.2.4. Canonical forms for complex and real forms

Theorem:

Any n n complex symmetric matrix A is congruent to a matrix of the form

0
0 0

For some r. Moreover r = rank(A) and so A is congruent to two matrix of this

form then they have the same value for r.

Proof:

Let A is congruent to a matrix of this form.

If P is invertible, then so is PT.

r = rank (PTAP)

= rank (A).

Hence the proof.

20
 CHAPTER - III

BILINEAR AND QUADRATIC FUNCTIONS AND FORMS

BILINEAR FUNCTION AND FORMS

Definition;

A function f on V x V to F, where V is a vector space over a field F, is

called bilinear if it is linear in each argument.

That is,

( , ) = ( , )+ ( , )

( , , ,

(c , , , ∀ and c F

Theorem:

Let f be a bilinear function on V V to F, where V is vector space over F.

Let { 1, 2,… n} and { 1, 2, … n } be any bases of V. Then relative to these
bases for V. Then relative to these bases ∃ a representation of f, that is a
function on Vn (f) x Vn (f) to F such that ( x, y) = f ( , where has –
coordinates X and has - coordinates Y, moreover

, ( X,Y)

= ∑, ij xi yj

= xT Ay

Where aij = i, j) A = aij such a form of degree 2 in x and y is

called a bilinear form.

21
Proof:

Let be the isomorphism of V to Vn (F) determined by the basis.

Define as follows

X= Y=

Then X, Y ) = ,

Thus, , X, Y)

=f(∑ i
i, ∑ j
j)

=∑, i
+( i, j) yj

=∑ i
i j

A direct computation shows that this value of can be written as the

matrix product XTAY.

Definition:

Two bilinear forms are equivalent iff they are representation of one and the
same bilinear function.

Theorem:

Two bilinear forms are equivalent iff their respective matrices A and B are
equivalent ( A ~ B) or alternatively iff the forms have the same rank.

Proof:

Suppose that is the representation of relative to the , bases,

is the representations relative to the , bases.

22
 According to if { 1, 2, … n } is a basis for a vector space V, then the
components of the row matrices A, where ={ 1, 2, … n } form a basis for
V iff A has an inverse.

∃ a non singular matrices P and Q such that

= PT = Q

X = P TU Y = QV

( , ) = , (X, Y)

= XT AY

= (PT U)T A (Q V)

= P UT AQV

= UT (PAQ) V

= (U, V)

We conclude that B = PAQ is equivalent to A

Two m x n matrices A and B over F are rationally equivalent iff B = PAQ

for suitable nonsingular matrices P and Q here P (Q) is obtained by applying in the
some order to I those row ( column ) transformation used to reduce A to B.

In particular, if r (A) = r ∃ matrices P, Q such that PAQ = .

Conversely, Suppose that f and g are the two bilinear forms with

f (x,y) = XTAY

g (x,y) = XTBY

B = PAQ

23
 where P and Q are non-singular.

Then f may be regarded as a representation of itself relative to the , -

bases and g regarded as a representation of f relative to the PT, Q - bases.

Thus f and g are equivalent since both are representation of the bilinear
function f.

Hence the proof.

Theorem:

Corresponding to a bilinear function f of rank r on V, there are basis , on

V relative to which , (x,y) = xiyj. This representation , we regard as the
canonical representation of f.

Proof:

The matrix of , , namely, diag ( Ir , 0 ) has rank r and consequently , is

a representation of f.

We have learned that accompanying a change of basis in a vector space over

F, there is a change in the coordinates of a vector which is described by a system of
linear equation with a non-singular matrix B expressing one set of coordinated X in
terms of the other set U.

X = BU ∴ B is non singular.

Let X = (x1,x2, . . . xn)T in a function f by new variables U = ( u1,u2,. . . un)T

related to the xi ‘ s as above is described as a non-singular substitution on X.

Thus the transition from a given bilinear from f to an equivalent from

amounts to a non-singular substitution on the two sets of variables.

24
Theorem:

With a suitable non-singular substitution X = PT U, Y = QV a bilinear form

of rank r reduces to the form whose value at U,V is ui vi

The co - efficient in the substitution can be restricted to any field that

contains the co efficient of the matrix of the original form.

Example

Reduce the form f = 2x'y' – 3x'y2 + x'y3 – x2y' + 5x2y3 – 6x3y' + 3x3y2 +
19x3y3 to the form uivi by a substitution having rational coefficient.

The matrix of f is

2 3 1
A= 1 0 5
6 3 19

Let us find non – singular matrices P,Q such that

PAQ = diag(Ir,0)

Since the substitution X = PTU, Y = QV will reduce XTAY to ui vi

To find a P and Q we proceed as in

To computation are listed below

1 0 5
≃ 2 3 1 R1 ⟷ R2
6 3 19

1 0 5
≃ 0 3 11 R2 → R2 – 2R1, R3 → R3 – 6R1
0 3 11

25
 1 0 5
≃ 0 3 11 R1 → R1 (-1), R3 → R3 + R2
0 0 0

1 0 5
≃ 0 11
1 3 R2 ⟶ 3
0 0 0

1 0 0
≃ 0 1 11
3 C3 → C3 + 5C1
0 0 0

1 0 0
≃ 0 1 0 C3 → C3 + 11 3 C2
0 0 0
0 1 0
P⟶ 1 0 0
0 0 1
0 1 0
≃ 1 0 0 R2 → R2 – 2R1
0 0 1
0 1 0
≃ 1 2 0 R3 → R3 – 6R1
0 6 1
0 1 0
≃ 1 2 0 R1 → R1 (-1)
0 6 1
0 1 0
≃ 1 2 0 R3 → R3 + R2
1 4 1

0 1 0
≃ 1 2
3 3 0 R2 ⟶ 3
1 4 1

26
 1 0 5
Q:I⟶ 0 1 11
3
0 0 1

Then the substitution X = PTU; Y = QV reduces XTAY to u1v1 + u2v2.

QUADRATIC FUNCTIONS AND FORMS

Theorem:

A square matrix A is the sum of a symmetric matrix and Skew – symmetric

matrix. This representation is unique.

Proof:

A = ½ ( A + AT ) + ½ ( A - AT )

= B +C

Where B = ½ ( A + AT )

C = ½ ( A - AT )

∴ (A AT ) T = A AT

It is clear that B is symmetric and C is skew-symmetric.

If A = B1 + C1 is another such decomposition.

We’ve AT = B1T + C1T

=B1 - C1

And it follows that, A + AT = B1 + C1 + B1 - C1

= 2B1

27
And A – AT = B1 + C1 - B1 + C1

= 2C1

Hence the theorem

Lemma:

Let A denote a square matrix and B a matrix obtained from A by a C1

transformation. The set of principal minors of B is the same as that of A. If B is
obtained from A by a C111 transformation then

Dk ( B )= Dk ( A ) ∀k

Proof :

The set of n x n symmetric matrices over F.

Since the matric congruent to a symmetric matric is symmetric

( A = AT then (PAPT)T = PATPT = PAPT )

The set of all n x n matrices.

Consequently,

The definition of the equivalences relations induced in the set of n x n

symmetric matrix by congruence.

A ≃ B iff there exist a non-singular P such that B = PA P T

Without regard to the features indicated of C – transformation.

It sis easily seen that they are adequate to reduce a symmetric matric of
positive rank to an essential diagonal from which, although not unique, will play
the role of the preferred form in our exposition.

28
Theorem:

Each symmetric matrices of rank r is congruent to diagonal matrix having r

non-zero diagonal elements.

Proof:

We may assume that r > 0.

Then according to symmetric matrix A of positive rank can be reduced with

C transformation to a congruent matrix.

A* = diag ( A1 , A2, … , AS, 0 )

Where each Ai has the form

0
(a) Or A(a) = ,a≠0
0

This reduction may be effected with a sequence of C1 transformation

followed by a sequence of C111 transformation.

It is only necessary to show that the matrix A (a) has his property.

The row transformation R12 (1) and its column analogue, followed by R21
( 1 2) and its column analogue.

Hence the Proof.

FURTHER PROPERTIES OF SYMMETRIC MATRICES

Theorem:

A symmetric matric A of rank r >0 has at least one non-zero principal minor
of order r.

29
Proof:

According to a symmetric matrix A of a positive rank can be reduced with C

transformation to congruent matrix

A* = diag (A1, A2, … , AS, 0 )

Where each A, has the form.

0
A(a) = a≠0
0

This reduced may be effected with a sequence of C1 transformation

followed by a sequence of C111 transformation.

A may be reduced to A* in using a sequence of C1 transformation followed

by a sequence of C111 transformation.

Hence A* may be reduced to A using sequence of C111 transformation

followed by a C1 transformation.

Since Dr ( A* ) 0.

This assertion follows from A denote a square matrix and B a matrix

obtained from A by a C1 transformation.

Then the set of principal minors of B is the same as that of A.

If B is obtained from A by a C111 transformation then.

Dk ( B )= Dk ( A ) for all K.

Hence the Proof.

30
Theorem:

If a symmetric matric A has a principal sub matrix P of order r with non-zero

determinant and every principal sub matrix of order r + 1, r + 2 respectively, which
contains P has zero determinant, then r ( A ) = r.

Proof :

The first hypothesis together with A symmetric matrix A of positive rank

can be reduced with C transformation to a congruent matrix.

A* = diag ( A1 , A2, … , AS, 0 )

Where each A,

0
A(a) = 0
0

This reduction may be effected with a sequence of C1 transformation

followed by a sequence of C111 transformation.

∃ a sequence of C1 transformation which reduces A to a matrix B such that,

Dr (B) 0.

And only C111 transformation are necessary to reduce B to A* in (2)

Then r (A*), hence that A is r

Dr (A*) 0 and if

r (A*) > r

At least one of Dr+1 (A*), Dr+2 (A*) is non zero

Which contradicts the second hypothesis.

Hence the Proof.

31
QUADRATIC FUNCTIONS AND FORMS OVER THE REAL FIELD

Theorem:

Let V denote a vector space over R* and q a quadratic function of rank r on

V to R*. Then q can be represented by the form

(v1)2 + (v2)2 + . . . + (vv)2 – (vv+1)2 - . . . – (vr)2 (1)

Where the number P of positive squares is uniquely determined by q.

We call the canonical representation of a real quadratic function. Started for

quadratic forms, any quadratic form over R* can be reduced by a non singular
substitution over R* to a uniquely determined from of the type finally started for
symmetric matrices a real symmetric matrix A of rank r is congruent to a uniquely
determined diagonal matrix of the type

Proof:

Q is represented by a form UTDU

Where D = diag(d1,d2, … dr,0,…0) relative to a suitable basis for V.

With a change in of the basis vector we may assume d1,d2, . . . dP > 0 and
dP+1 . . . dr < 0

P = PT

= diag ,… , ,… ,1…1 (2)

√ √ √

To prove,

The uniqueness, suppose that, besides (5), q is also represented by

(w1)2 + (w2)2 + . . . + ( )2 - ( +1 2
) - . . . – (w1)2

32
 Hence that (1) and (2) are equivalent forms.

Assume that P' ⊂ P

Now,

q( 0 For those 's corresponding to n- tuples ( v1,v2, … , vn)T such

that vv+1 = . . . = vr = 0.

These 's constitute an n – ( r – p ) dimensional subspace S1 of V

Similarly from (2),

q( ) < 0 for each ≠ 0 corresponding to an n – tuple (w1,w2,. . . wn)T such

that

w1 = . . . = = =...= =0

These condition determine an ( r – P’ ) dimensional subspace S2 of V.

Now d(S1) + d(S2) = n + (P-P') > n which means that S1 S2 0.

Hence S1 and S2 have a non-zero vector in common. For this common

vector q( ) ≥ 0.

The assumption P' > P would lead to a similar contradiction and so P = P′

Hence the Proof.

33
 CHAPTER - IV

LINEAR TRANSFORMATION ON A VECTOR SPACE

Theorem:

The one – one corresponding between the set of all linear function on V to
W and the set of all m x n matrix over F. [ where m = d (w), n = d (V) ] that is
defined when basis are chosen for v and w is an isomorphism between the vector
space of linear function and that of matrices.

Proof:

If A and B are linear function on V to W and W to U respectively.

The product BA of A by B is the function to V to U defined by the equation

(BA) = B (A )

Show that BA is linear and moreover that if the matrices A and B

corresponds to A and B respectively.

When basis are chosen for the spaces involved, then BA corresponds to BA
relative to the same bases.

We’ve begin this chapter with an introduction to linear function on one

vector space to a second vector space in order to decomonstrate the correlation
between such function and m x n matrices.

Such a linear function A is usually called a linear transformation or linear

operator on V and the function value A the image of under A

The word “operator” has the origin in the relation AX = Y of A

A applied to X products Y

34
 The word “image” stems from the interpretation of a linear transformation of
the vector into the vector

At every stage the correspondence of transformation with matrices will be

stressed.

Hence the Proof.

Theorem:

If A is non-singular, the linear transformation A on V defined by AX = Y

relative to any basis a one – one correspondence on V into V.

Conversely, if the linear transformation A on V is one – one, then (A : ) is

non singular. Thus it is appropriate to call the one – one linear transformation on V
non-singular

Proof:

If A is defined by AX = Y relative to the – basis for V.

Then A = iff

= X, = Y

To show that,

A is one – one when A is non-singular.

(i) Given in V ∃ a such that A =

(ii) 1 ≠ 2 implies A 1 ≠A 2

For, (i) suppose that = Y

Define = A-1Y

35
 Then A =

(ii) A 1 =A 2

Where i = Xi , i = 1,2, . . .

Then AX1 = AX2, which gives X1 = X2 or 1= 2 upon multiplication by A-1

Conversely,

If A is one – one and (A, ) = A

Then AX = Y has a unique solution for each choice of Y, which forces A to

be non-singular.

Hence the Proof.

Theorem:

A linear transformation A on V is one – one iff ∃ a linear transformation B

such that AB = BA = I. This formal inverse is then the inverse in the sense of
equation.

Proof:

If such a B exists, then ∀ and

A (B ) = (AB)

=I =n

B (A = (BA)

= (I ) =

The 1st result asserts that any vector is the image A of some vector = B

36
 The 2nd result asserts that if A = then B = , which ⇒ that is the
image of only one vector.

The vector such that B =

∴ A is one – one.

AX = Y with A non-singular where it’s used in conjunction with the

equation = A

If in a vector space the basis is replaced by the – basis, the

coordinates X of a vector are changed to

Y=AX

We call this the alias interpretation of AX=Y, the vector with co-ordinates X
is re-labelled with co-ordinates Y.

When used in conjunction with a basis for V, AX=Y define a non-singular

linear transformation A.

Where (A; ) =A which associates with the vector X the vector Y.

We call this the alibi interpretation of AX=Y, the vector with co-ordinates X
is carried into the vector with co-ordinates Y=AX.

Similarly, AX= Y, with A is non-singular

Basis and basis are given a linear function Aon V is defined by the
equation A =

The image of a vector X is X under A

A(

Hence the proof.

37
Groups of transformation in general.

Theorem: (Cayley)

For every group G there is an isomorphic group of correspondence.

Proof:

The set S of objects upon which the correspondence shall be defined the set
G of group elements {a, b,. . . }.

Show that the group G is isomorphic to a group of correspondences on the

set G.

Consider the correspondence ∅a on G define in terms of the group elements a as

follows

ax = ax all x in G

G2 and G3 for a group is seen to be a one – one correspondence on G onto

itself

To show that, the set Γ of all a’s is a group.

( a b)x = a( b x)

= a (bx)

= a (bx)

= (ab) x

= ab x

So that the closure property holds

The unit element e of G ensures the presence of an identity correspondence in

38
 a x = ex = x ∀ x in G.

The presence in G of an inverse a-1 for each small a in G ensures the presence in of
an inverse for a.

( a a
-1
) x = (a a -1) x = x = e x ∀ x in G.

Thus Γ is a group and we content that G ≅ Γ under the correspondence

a→ a

The correspondence is one – one

a= b

⇒a e =b e (or)

⇒a=b

Show that the correspondence is preserved under multiplication

a b= ab

Hence proved

Singular linear transformation:

Theorem:

The sum of the rank and nullity of a linear transformation A on V is equal to

the dimension of V

r(A) +n(A) =d(V)

39
Proof:

Let a denote the nullity of A, so that N (A) has a basis { 1, 2, 3, ... S} of s

elements

Extend this set to a basis for V by adjoining { 1, 2, 3, . . . t}

Then s +t =d(V)

∴ A i=0 The vectors A i span the range R (A) of A

Actually {A 1,A 2,A 3, . . .A i} is a basis for R(A)

∵ Ci(A i)=0

A Ci i=0 (or)

Ci I is in N(A)

Hence C1=C2=C3= . . . =Ct=0

It follows that S +t is the nullity + the rank of A

Hence proved

Theorem:

If A is a linear transformation V. Then the following three statements are

equivalent

i) A is non-singular.
ii) R(A) =V.
iii) N(A) = 0

40
Proof:

Its sufficient to show that (i) ⇒(ii)

The every element of V occurs as an image under A

Moreover, distinct elements have distinct images since

A 1 =A 2

A( 1 - 2)=0

Consequently

(ii) ⇒(iii)

1 - 2=0

⇒ 1 = 2

Hence A is a one – one or non-singular transformation on V

Suppose that the vector space V is decomposed into a direct sum

V= S1⊕S2

of two subspaces. So that every vector has a unique representation in the form

= 1 + 2 i in Si

We defined a transformation E on V by the rule E = 1 and call E the

projection on S1 along S2

A direct computation show that E is linear

R(E)= S1 and N(E) =S2

Show that V is the direct sum of the range and nullity of E

41
 It is accidental that

V = R(E) ⨁ N(E)

In general it need not even be true that these sub spaces are disjoint

Hence the theorem

Theorem:

A linear transformation A on V is a projection on some subspace S1 iff it is

idempotent: A2=A

Proof:

If A is a projection on S1 along S2 and = 1+ 2 is the decomposition of

with i in Si.

The decomposition of is 1+0, so that

A2 = AA

= A (A )

= A ( 1)

= 1

=A

And A is idempotent.

Conversely, suppose A2=A

We define S1 to be the set of all vectors ∋ A and S2=N(A)

Now, S1 and S2 are subspaces and we shall prove that

42
 V= S1⊕S2

∴ S1 ∩ S2 =0, from the definition of S1 and S

= A + (I-A)

= 1+ 2

Let us say, we’ve

A 1 = A2 = A = 1

A 2 = A (I-A)

= (A-A2)

=0

Thus is in S.

V= S1+S2

Hence V= S1⊕S2

A is projection on S1 along S2.

Hence the proof.

Theorem:

Let A denote a linear transformation on V. ∃ a non-singular linear

transformation Ai on V for which A1A is a projection E. hence A is a projection
followed by a non-singular transformation.

Proof:

Let r= r(A) and { 1 2,… n } a basis for R(A).

43
Extend this to a basis { 1 2,… n } for E.

∴ 1 2,… r are in R (A), ∃ a vectors 1, 2,…, r such that

= i i=1,2,…,r

Finally, we choose a basis which we may denote by { r+1, r+2,…, n} for

N (A). We shall prove that { 1, 2,…, n } is a basis for V.

It is linearly independent

If ci i = 0,

Then A ci i = ci i = 0.

So that, c1=c2=….=cr=0

Consequently, ci I = 0 which forces the remaining ci to be zero.

A1 i = i i=1,2,3…n.

then, A1A i = A1 i = i i =1,2,3,…r.

= A10 = 0 i = r+1, r+2, … , n

So that A1A is a projection E

A1 is non-singular, we obtain a finally, that A = E.

0 1
0

44
 Thus A1 is an interchange of the two basis vectors, and A1D is the projection
followed by A1-1=A1

The transformation is illustrated, where (a,b)T represents the vector a+bt

45
 CHAPTER - V

CANONICAL REPRESENTATION OF A LINEAR TRANSFORMATION

Lemma

If f and g are any two members of F [ ] such that g/f, then N (g) ⊆ N (f)

Proof:

By assumption there exists a polynomial n in F [ ] such that f = gh.

Hence for any vector in N (g),

f(A) = (h(A) g(A))

= h(A) (g(A))

=0

So that is in N (f)

Lemma

Let { f1,f2,…,fr } denote a finite set of polynomial in F [ ] and d their g.c.d,

then N(d) = N(f1) ∩ N(f2) ∩ … ∩ N(fr)

Proof:

Let D denote the right – hand member.

Then N (d) ⊆ D

∴ d/f ⇒ N(d) ⊆ N(fi) i = 1,2,…,r

If f and g are any two members of F [ ] such that g/f, then N (g) ⊆ N(f)

Then there exist polynomial g1,g2,…,gr such that d = g if i.

46
 If is any vector in D.

d(A) = g1(A) f1(A) + … + gr(A) fr(A)

So that is in N (d).

Lemma:

Let { f1,f2,…,fr } denote a finite set of polynomial in F[ ] and h a l.e.m

Then N (h) = N (f1) + … + N(fr)

Proof:

Let S denote the right – hand member, which is the subspace of V consisting
of all sums.

=∑ i i= N (fi)

Then N (h) ⊇ S.

∴ , N(fi) ⊆ N(h) i = 1,2,…r

Hence N (h) contains all sums of the form.

Since

h = hifi i = 1, 2,…r.

Replacing by A in this equation gives the following representation of the identity

transformation I on V.

∑ i (A) hi (A) = I

This implies that for all in V

I =

47
 =∑ i (A) hi (A))

To verify this, observe that,

fi(A) (Si (A) hi (A)) = Si (A) [ hi (A) fi (A) ]

= Si (A) (h (A))

=0

Thus N (h) ⊆ S.

Hence the lemma.

Theorem

Let f = f1 f2 … fr be a decomposition of the polynomial f in F [ ] into

factors fi relatively prime in pairs. Then N (f) is the direct sum of the null spaces
N (fi)

N (f) = N(f1) ⨁ N(f2) ⨁ … ⨁ N(fr)

Proof:

The assumption imply that f is a l.e.m of { f1 f2 … fr }

So that N (f) is the sum of the various null spaces N (fi). Let
{ f1 f2 … fr } denote a finite set of polynomial in F[ ] and h a l.e.m then

N (h) = w(f1) + … + w(fr)

The g.c.d of fk and is 1 k = 1,2,…,r

The polynomial in A corresponding to the polynomial 1 in is I and

N(I) = 0.

0 = N(fk) ∩ N( f i) or

48
 0 = N(fk) ∩ N(fi)

Where N ( fi) as N(fi)

Let { f1,f2,…,fr } denote a finite set of polynomial in F [ ] and h a l.e.m.

then N (h) = N(f1) + … + N(fr)

If there can be found a polynomial f, such that N (f) = V equation determines

a decomposition of V into a direct sum of invariant spaces of A, that is the spaces
N ( fi ) completely reduces V.

If d is a non-zero vector, ∃ a least positive integer m ≤ d [V] such that

{ , 1 …. Am } is a linearly dependent set.

∑ iA
i
=0

Consequently the polynomial g( ) = ci i has the property that

If such a polynomial is found for each member I of a basis of V and f

denotes a l.c.m of this set.

Then f (A) i =0 i = 1,2,…n and f(A) is linear

f(A) ai I =0

That is f (A) = 0

Theorem

There exist a unique monic polynomial of least degree in F [ ] called the

minimum function of A and designated by m [ ], such that m (A) = 0. If f ( ) is
any member of F ( ), f (A) = 0 iff m/f.

49
Proof:

The set of polynomial f ( ) in F [ ] such that f (A) = 0 is non empty.

The remaining requirements for an ideal are quickly verified so that every
ideal E in F ( ) is a principal ideal. If E ≠ (0), the generator of E may be
characterized within associates as a polynomial of least degree in E.

The application of f = f1 f2 … fr be a decomposition of the polynomial f in

F [ ] into factors fi relatively prime in pairs.

N (f) = N (f1) ⊕ N (f2) ⊕ … ⊕ N (fr) to the minimum functions of A gives

the following important results.

Theorem:

The set of characteristics value of A coincides with the set of distinct roots
in F of the minimum function, m ( ) of A.

Proof:

Let C denote a characteristics value of A and a corresponding

characteristics vector.

Then A or

(A-c I) =0

m( ) in the form m( ) = q( ) ( – c) + m(C)

⇒ 0 = m (A)

= m (C)

hence that m(C)=0

50
Conversely,

If C is a root in F of m ( ),

m( ) = q( ) ( -c)

Where ∃ a vector such that

= q(A) 0

The equation,

0 = m(A) = (A-cI) q(A)

= (A-cI)

Then demonstrates that C is a characteristics value.

Lemma:

Suppose that g ( ) is a proper divisor of a divisor f ( ) of the minimum

function. m ( ) of a linear transformation A. then N (g) is properly contained in
N (f).

Proof:

By assumption, m = fh and k = gh is a proper divisor of m

Consequently, Then ∃ a vector such that K (A) = g (A) [ h (A) ] 0.

But then h (A) , which is a member of N (f), is not contained in N (g).

Theorem:

A polynomial matrix M of rank is equivalent to a uniquely determined

diagonal matrix diag ( h1,h2,…,hr,0,…,0 ) such that each hi is a monic polynomial

51
which divides its successor, hi is precisely the ith invariant factor of M. Two
polynomial matrices are equivalent iff they have the same invariant factors.

Proof:

If N is equivalent to the polynomial matrix M of rank r then

⇒ N ≃ M, then their invariant factors agree.

Conversely,

If the invariant factors of M and N coincide, they are equivalent to one and
the same diagonal matrix which displays three factors, and hence equivalent to
each other.

Hence the Proof.

Theorem:

If M is equivalent to D = diag ( f1,f2,…,fr,0,…,0 ) where each fi is a monic

polynomial, then the prime power factors of the fi’ s are the elementary divisors of
M.

Proof:

Let P denote any irreducible factor of any one of the fi’s are arrange the fi’s
according to ascending power of P.

g1 ... gr

Where k1 ≤ k2 ≤ … ≤ k. and each g is prime to P.

Then the highest power of P which divides di (M) = di (D) has exponent
k1 + k2 + … + ki

52
 Hence the highest power of P which divides hi = di / di-1 is pki so that for each
I such that ki > 0. Pki is an elementary divisor of D and M.

A repetition of this argument for each prime which divides some fi shows
that every prime power factor of every fi is an elementary divisor of M.

Moreover that all elementary divisor of M are obtained in this way.

Theorem:

A linear transformation A on V over F is represented by the direct sum of

the companion matrices of its non-trivial similarity factors hk, hk+1, … hn :

(R1) c(hk) ⨁ c(hk+1) ⨁ … ⨁c(hn)

a matrix A with non-trivial similarity factors hk, hk+1, … hn is similar to the matrix
(R1)

Proof:

We shall prove that ( R1 ) represents A relative to the basis for V that is

indicated in Si = i, … i

Let us denote an arbitrary I and its index polynomial hi by simply and

m
h = a0 + a1 + … + in order to simply the notation.

A (A' ) = Ai+1 0 ≤ j < m-1

A (Am-1 ) = Am
m-1
=(- 0 I– 1A -…- m-1A )
m-1
= (- 0 1A -…- m-1A

Since h(A) = 0 + 1A + … + Am =0

53
 Thus in space S=( 1A 1 … Am-1 )

A is represented by c( h).

Which we shall enumerate after the next example.

Hence the Proof.

Theorem

In terms of the notation of this section hi ( ) is both the minimum and

characteristics function of Ai, the transformation induced in Si by A,
i = k, k+1, … , n

Proof:

Let hi ( ) is the index polynomial of i

Hence in particular,

hi(A) i = hi(Ai) I =0

It follows that hi (Ai) = 0 on Si. Since hi (Ai) is seen to carry each basis
vector Aij I for Si into zero.

Thus, if mi denotes the minimum function of Ai, it follows that mi/hi.

On the other hand mi (ai) = 0 and hence ⁄ .

∴ mi = hi

Since c (hi) represents Ai the characteristics function of Ai is

det ( I – c (hi)).

It will follow that hi is the characteristics function of Ai.

If we can prove that

54
 det ( I – c(h)) = h( )

For any monic polynomial

h( =∑

⋯⋯⋯
⋯⋯⋯
⋯⋯⋯
det ( I – c(h)) = det ⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯
⋯⋯⋯⋯
⋯⋯

times each row is added to the preceding one, beginning with the last row
the desired conclusion is immediate.

Theorem

The last similarity factor, hn( of A is the minimum function of A. In

metrical language, the last invariant factor of I – A is the minimum function of
A.

Proof:

Since each similarity factor of A divides hn , hn (A) = 0 on each space Si and

consequently on V.

If m ( ) denotes the minimum function of A, we conclude that m/ hn.

Conversely,

Since m (A) = 0 on Sn, m is divisible by the minimum function hn on Sn.

Hence m = hn.

hkhk+1…hn = f.

55
 CHAPTER – VI

SYMMETRIC AND HERMITIAN MATRICES

Definition

Let V be a real inner product space, and U a subspace of V. the orthogonal

complement of U is the set of all vectors which are orthogonal to everything in U:

= { w ∈ V : w.u = 0 ∀ u ∈ U }

Proposition

If V is an inner product space, and U a subspace of V. with dim(V) = n and

dim(U) = r, then is a subspace of V, and dim ( ) = n - r. Moreover
V = U⊕

Proof:

Proving that is a subspace is straight forward from the properties of the

inner product. If w1,w2 ∈ , then w1.u = w2.u = 0 for all u ∈ U, so (w1+w2).u= 0
for all u ∈ U, whenever w1+w2 ∈ . The argument for scalar multiplies is similar.

Now choose a basis for U and extend it to a basis for V. Then apply the Gram-
Schmidt process to this basis to obtain an orthogonal basis (v1… vn). Since the
process only modifies vectors by adding multiples of earlier vectors, the first r
vectors in the resulting basis will form an orthogonal basis for U. The last n-r vectors
will be orthogonal to U, and so lie in ; and they are clearly linearly independent.
Now suppose that w ∈ and w = civi where (v1…vn) is the orthogonal basis we
constructed. Then ci = w.vi = 0 for i = 1,…,r; so w is a linear combination of the last
n-r basis vectors, which thus form a basis of . Hence dim( ) = n - r as required.

Since we have a basis for V which is a disjoint union bases for U and

56
Definition:

Let V be an inner product space. A linear map ∶ → V is an orthogonal

projection if
2
a) is a projection, that is, =
b) is a self – adjoint that is, 2
= (where *(v) w = v (w) ∀ v,w ∈
V).

Proposition:

If is an orthogonal projection, then Ker( = Im(

Proof:

We know that V = Ker( ) ⊕ Im( we only have to show that these two
subspaces are orthogonal. So take v ∈ Ker( , so that (v) = 0 and w ∈ Im( , so
that w = (u) for some u ∈ V. Then

v.w = v. (u) = *(v) . u = (v).u = 0 as required.

Theorem

If is a self-adjoint linear map on a real inner product space V, then the

eigenspaces of form an orthogonal decomposition of V. Hence there is an
orthogonal bases of V consisting of eigenvectors of . Moreover, there exists
orthogonal projection 1,… n satisfying 1+…+ r = 1 and i j = 0 for i≠j such that
= 1 1+ ...+ r r where 1… r are the distinct eigenvalues of .

Theorem:

Let A be a real symmetric matrix. Then there exists an orthogonal matrix P

such that P-1AP is diagonal. In other words, any real symmetric matrix is
orthogonally similar to a diagonal matrix.

57
Proof:

Let the transition from one orthonormal bases to another is an orthonormal

matrix. It suffices to find an orthonormal bases of eigenvectors, since all the rest
follows from our remarks about projection, together with what we already know
about diagonalizable maps.

The proof will be by induction on n = dim(V). There is nothing to do if n = 1.

So we can assume that the theorem holds for (n-1) dimensional spaces.

Choose an orthonormal bases: then is represented by a real symmetric

matrix A. Its characteristics polynomial has a root over the complex numbers. We
temporarily enlarge the field from to . Now we can find a column vector v ∈ n

such that Av = v. Taking the complex conjugate, remembering that A is real, we

have A ̅ = ̅ ̅ .

If v = [z1 z2 . . . zn , then we have

̅ ( |z1|2 + |z2|2 + . . . + |zn|2 ) = ̅ ̅ T v

= (A ̅ )T v

= ̅ Av

= ̅ T ( v)

= (|z1|2 + |z2|2 + . . . + |zn|2 ),

So ( ̅ ) (|z1|2 + |z2|2 + . . . + |zn|2 ) = 0. Since v is not the zero vector, the

second factor is positive, so we must have ̅ , that is, is real.

Now since has a real eigenvalues, we can choose a real eigenvectors v, and
we can assume that |v| = 1.

58
 Let U be the subspace = { u ∈ V : v.u = 0 }. This is a subspace of V of
dimension n-1. We claim that : U → U. for take u ∈ U. then

v. (u) = *(v).u = (v).u = v.u = 0.

Where we can use the fact that is self-adjoint. So (u) ∈ U.

So is a self-adjoint linear map on the (n-1) dimensional inner product space

U. By the induction hypothesis, U has an orthonormal bases consisting of
eigenvectors of . They are all orthogonal to the unit vector v: so, adding v to the
bases, we get an orthogonal bases for V.

Remark:

The theorem is almost a canonical form for real symmetric relations under the
relation of orthogonal congruence. If we require that the eigenvalues occur in
decreasing order down the diagonal, then the result is a true canonical form each
matrix is orthogonally similar to a unique diagonal matrix with this property.

Corollary:

If is self-adjoint, then eigenvectors of corresponding to distinct

eigenvalues are orthogonal.

Proof:

This follows from the theorem, but is easily proved directly. If (v) = v and
(w) = w, then

v.w = (v).w = *(v) .w = v. (w) = v.w,

So, if then v.w = 0.

59
Example

10 2 2
Let A = 2 13 4
2 4 13

The characteristics polynomial of A is

10 2 2
2 13 4 = (x-9)2 (x-18)
2 4 13

So the eigenvalues are 9 and 18.

For eigenvalues 18 the eigenvalues satisfy

10 2 2 18
2 13 4 = 18
2 4 13 18

So the eigenvectors are multiples of [ 1 2 2 ]T. Normalizing, we can choose a unit

eigenvector [ ]T.

For the eigenvectors satisfy

10 2 2 9
2 13 4 = 9
2 4 13 9

That is, x+2y+2z = 0. Thus the eigenvectors is 2 – dimensional. We need to

choose an orthonormal bases for it. This can be done in many different ways: for
example, we could choose

[ 0 1⁄√2 1⁄√2 ]T and [ -4/3√2 1⁄3√2 1⁄3√2 T.

Then we have an orthonormal bases of eigenvectors. We conclude that, if

60
 1⁄3 0 4⁄3√2
P = 2⁄3 1⁄√2 1⁄3√2
2⁄3 1⁄√2 1⁄3√2

Then P is orthogonal, and

18 0 0
T
P AP = 0 9 0
0 0 9

You might like to check that the orthogonal matrix in example in the last
chapter of the notes also diagonalises A.

Proposition:

The rank of a real symmetric matrix is equal to the number of non-zero

eigenvalues and the signature is the number of positive eigenvalues minus the
number of negative eigenvalues.

Proof:

Given a real symmetric matrix A, there is an orthogonal matrix P such that

PTAP is diagonal, with diagonal entries 1,…, n. suppose that 1,…, n are positive
s+1,…, s+tare negative and the reminder are zero. Let D be a diagonal matrix with
diagonal entries

1⁄√ 1, … , 1⁄√ S, 1⁄√ S+1, … , 1⁄√ S+1,1,…1.

Then

(PD)T APD = DTPTAPD =

SIMULTANEOUS DIAGONALISATION

61
Theorem:

Let A and B be real symmetric matrices, and suppose that A is positive

definite. Then there exists an invertible matrix P such that PTAP = 1 and PTBP is
diagonal. Moreover, the diagonal entries of PTBP are the roots of the polynomial
det(xA-B) = 0.

Proof:

A is a real symmetric matrix, so there exists an invertible matrix P1such that

P1T AP1 is in the canonical form for congruence. Since A is positive definite, this
canonical form must be I; that is, P1TAP1 = I.

Now consider P1TBP = C, This is a real symmetric matrix; so, according to

the spectral theorem, we can find an orthogonal matrix P2 such that P2TCP2 = D is
diagonal. Moreover, P2 is orthogonal, so P2TP2 = I.

Let P = P1P2 then

PTAP = ( AP1) P2 = IP2 = I,

And

PTBP = ( BP1) P2 = CP2 = D,

As required.

The diagonal entries of D are the eigenvalues of C, that is, the roots of the
equation det (x I - C) = 0. Now we have

det ( ) det (x A- B) det (P1) = det ( (x A - B) P1) = det (x A P1 - B P1 )

= det (x I - C).

And det (P1T) = det (P1) is non-zero; so the polynomials det (x A - B) and
det (x I - C) are non-zero multiples of each other and so have the same roots.
62
Theorem:

a) Let and be self-adjoint maps on an inner product space V, and suppose

that . Then there is an orthogonal bases for V which consists of
vectors which are simultaneous eigenvalues for and .
b) Let A and B be real symmetric matrices satisfying AB = BA. Then there
is an orthogonal matrix P such that both PTAP and PTBP is diagonal.

Proof:

Statement (b) is just a translation of (a) into matrix terms; so we prove (a).

Let 1,… r be the distinct eigenvalues of . By the spectral Theorem, have an

orthogonal decomposition

V = U1⊕ … ⊕ Ur,

Where Ui is the i- eigenspace of .

We can claim that maps Ui to Ui. for take (u) = iu. Then

iu )= I ,

So (u) is also an eigenvector of with eigenvalue i. Hence (u) ∈ Ui, as

required.

Now is a self- adjoint linear map on the inner product space Ui, and so by
the spectral theorem again, Ui has an orthonormal bases consisting of eigenvectors
of . But these vectors are also eigenvectors of , since they belong to Ui.

63
 Finally, since we have an orthogonal decomposition, putting together all these
bases gives us an orthonormal bases of V consisting of simultaneous eigenvectors
of and .

Remark:

This theorem easily extends to an arbitrary set of real symmetric matrices such
that any two commute. For a finite set, the proof is by induction on the number of
matrices in the set, based on the proof just given. For an infinite set, we use the fact
that they span a finite-dimensional subspace of the space of all real symmetric
matrices; to diagonalise all the matrices in our set, it suffices to diagonalise the
matrices in a bases.

64
Conclusion

The literature review involves a qualitative content analysis of available

information already published in the some form. It can be a study of the research
objects alone, with the aim of collecting information about its structure, process and
relationships., increasing of familiarity of the researcher with the research object
And establishing and credibility of the project. In addition it can consider previous
research, attempting to link it with the stud currently planned. It may also be geared
towards a historical or comparative analysis of the issue in question so the current
study can be placed in a historical context.

Finally, it may review a theory of methods and techniques most suitable for
the study, simply by looking at the topic and by evaluating their suitability and
effectiveness.

65
 Bibliography

1. Robert R. Stoll, linear algebra and matrix theory, 1952.

2. Roger A. Hom and Charles matrix analysis, 1985.
3. Evan D. Newig, linear algebra and matrix theory, second edition, 1969.
4. D. H. Griffel, volume 1 a, first course of liner algebra and its
applications.
5. W.B. Vasantha Kandhasamy, smarandache linear algebra and its
applications 2003.
6. Peter J. Cameron , Notes on linear algebra 2008.
7. Kenneth Kuthler, Linear algebra I matrices and row operations 2012.

66