You are on page 1of 148

CHAPTER 8 Eigenvalues, Diagonalization, and Special Matrices

Outline
- Eigenvalues and Eigenvectors - Diagonalization of Matrices - Orthogonal and Symmetric Matrices - Quadratic Forms - Unitary, Hermitian, and Skew-Hermitian Matrices

Eigenvalues and Eigenvectors


Suppose A is an n*n matrix of real number. If write an n-vector E as a column

then AE is an n*1 matrix, which we may also think of as an n-vector

Vectors have directions associated with them. Depending on A, the direction of AE will generally be different from that of E. It may happen that for some vector E, AE and E are parallel. In this event, there is a number such that AE = E. Then is called an eigenvalue of A, with E an associated eigenvector.
Mathematics for Computer Engineering 4

Eigenvalues contain important information about the solution of systems of differential equations, and in models of physical phenomena.

Mathematics for Computer Engineering

DEFINITION 8.1 Eigenvalues and Eigenvectors

A real or complex number is an eigenvalue of A if there is a nonzero n*1 matrix (vector) E such that AE =E Any nonzero vector E satisfying this relationship is called an eigenvector associated with the eigenvalue .
Mathematics for Computer Engineering 6

Eigenvalues are also known as characteristic values of a matrix, and eigenvectors can be called characteristic vectors.

Mathematics for Computer Engineering

We will typically write eigenvectors as column matrices and think of them n as vectors in R . If an eigenvector has complex components, we may think n of it as a vector in C . Since an eigenvector must be a nonzero vector, at least one component is nonzero.
Mathematics for Computer Engineering 8

If is a non zero scalar and AE =E, then A(E) = (AE) = (E) = (E) This mean that nonzero scalar multiples of eigenvectors are again eigenvectors.

Mathematics for Computer Engineering

Example

Mathematics for Computer Engineering

10

Example

Mathematics for Computer Engineering

11

Mathematics for Computer Engineering

12

Mathematics for Computer Engineering

13

Finding all of the eigenvalues of A AE =E then E - AE = 0 or InE - AE = 0 (In - A)E = 0


Mathematics for Computer Engineering 14

This make E a nontrivial solution of the n*n system of linear equations (In - A)X = 0 This system can have nontrivial solution if and only if the coefficient matrix has determinant zero.
Mathematics for Computer Engineering 15

Thus, is an eigenvalue of A exactly when

|In A| = 0

Mathematics for Computer Engineering

16

When the determinant is expanded, it is a polynomial degree n in , called the characteristic polynomial of A. The root of this polynomial are eigenvalues of A. Corresponding to any root , any nontrivial solution E of (In - A)X = 0 is an eigenvector associated with .

Mathematics for Computer Engineering

17

Theorem 8.1
Let A be an n*n matrix of real or complex numbers. Then, 1. is an eigenvalue of A if and only if |In A| = 0. 2. If is an eigenvector of A, then any nontrivial solution of (In - A)X = 0 is an associated eigenvector.
Mathematics for Computer Engineering 18

DEFINITION 8.2 Characteristic Polynomial The polynomial |In A| is the

characteristic polynomial of A. and is denoted pA().

Mathematics for Computer Engineering

19

Mathematics for Computer Engineering

20

Example

Mathematics for Computer Engineering

21

Mathematics for Computer Engineering

22

Mathematics for Computer Engineering

23

Example

Mathematics for Computer Engineering

24

Mathematics for Computer Engineering

25

Mathematics for Computer Engineering

26

Mathematics for Computer Engineering

27

Mathematics for Computer Engineering

28

Theorem 8.2 Gerschgorin


Let A be an n*n matrix of real or complex numbers. For k = 1,...,n, let

Mathematics for Computer Engineering

29

Let Ck be a circle of radius rk centered at (k,k), where kk = k+ik. Then each eigenvalue of A, when plotted as a point in the complex plane, lies on or within one of the circles C1,...,Cn.

Mathematics for Computer Engineering

30

The circles Ck are called Gerschgorin circle. For the radius of , read across row k and add the magnitudes of the row elements, omitting the diagonal element kk. The center of Ck is kk, plotted as a point in the complex plane. If the Gerschgorin circles are drawn and the disks they bound are shaded, then we have a picture of a region containing all of the eigenvalues of A.
Mathematics for Computer Engineering 31

Example

Mathematics for Computer Engineering

32

It is not clear what the roots of this polynomial are. Form the Gerschgorin circles. Their radii are

Mathematics for Computer Engineering

33

Gerschgorin circles
Mathematics for Computer Engineering 34

Gerschgorin's theorem is not intended as an approximation scheme, since the Gerschgorin circles may have large radii. For some problem, however, just knowing some information about possible locations of eigenvalues can be important.

Mathematics for Computer Engineering

35

Diagonalization of Matrices
The elements aii of a square matrix is called main diagonal elements. All other elements are called off-diagonal elements.

Mathematics for Computer Engineering

36

DEFINITION 8.3 Diagonal Matrix

A square matrix having all offdiagonal elements equal to zero is called a diagonal matrix

Mathematics for Computer Engineering

37

Theorem 8.3
Let

Mathematics for Computer Engineering

38

Then 1.

Mathematics for Computer Engineering

39

2. |D| = d1d2 ... dn 3. D is nonsingular if and only if each main element is non zero 4. If each dj 0,then

Mathematics for Computer Engineering

40

5. The eigenvalues of D are its main diagonal elements. 6. An eigenvector associated with dj

is

with 1 in row j and all other elements zero.

Mathematics for Computer Engineering

41

DEFINITION 8.4 Diagonalizable Matrix

An n*n matrix A is diagonalizable if there exists an n*n nonsingular -1 matrix P such that P AP is a diagonal matrix. When such P exists, we say that P diagonalizes A

Mathematics for Computer Engineering

42

Theorem 8.4 Diagonalizability


Let A be an n*n matrix. Then A is diagonalizable if it has n linearly independent eigenvectors. Further, if P is the n*n matrix having these eigenvectors as -1 columns, then P AP is the diagonal matrix having the corresponding eigenvalues down its main diagonal.
Mathematics for Computer Engineering 43

Suppose 1,...,n are the eigenvalues of A, and V1,...,Vn are corresponding eigenvectors. If these eigenvectors are linearly independent, we can form a nonsingular matrix P using Vj as column j.

Mathematics for Computer Engineering

44

Example

Mathematics for Computer Engineering

45

Mathematics for Computer Engineering

46

Mathematics for Computer Engineering

47

Mathematics for Computer Engineering

48

Mathematics for Computer Engineering

49

Example

Mathematics for Computer Engineering

50

Mathematics for Computer Engineering

51

Mathematics for Computer Engineering

52

Mathematics for Computer Engineering

53

Example

Mathematics for Computer Engineering

54

Mathematics for Computer Engineering

55

Mathematics for Computer Engineering

56

Example

Mathematics for Computer Engineering

57

Eigenvector associated with -3 is

Mathematics for Computer Engineering

58

Find eigenvectors associated with -1.

The general solution

Mathematics for Computer Engineering

59

Two linearly independent eigenvectors associated with eigenvalue 1, for example

Mathematics for Computer Engineering

60

We can form the nonsingular matrix

that diagonalizes A :

Mathematics for Computer Engineering

61

Mathematics for Computer Engineering

62

Mathematics for Computer Engineering

63

Mathematics for Computer Engineering

64

Theorem 8.5
Let A be an n*n diagonalizable matrix. Then A has n linearly independent eigenvectors. Further, -1 if Q AQ is a diagonal matrix, then -1 the diagonal elements of Q AQ are the eigenvalues of A and the columns of Q are corresponding eigenvectors.
Mathematics for Computer Engineering 65

Mathematics for Computer Engineering

66

Mathematics for Computer Engineering

67

Example

Mathematics for Computer Engineering

68

Mathematics for Computer Engineering

69

Theorem 8.6
Let the n*n matrix A have n distinct eigenvalues. Then corresponding eigenvectors are linearly independent.

Mathematics for Computer Engineering

70

Mathematics for Computer Engineering

71

Mathematics for Computer Engineering

72

Corollary 8.1
Let the n*n matrix A have n distinct eigenvalues. Then A is diagonalizable.

Mathematics for Computer Engineering

73

Example

Mathematics for Computer Engineering

74

Mathematics for Computer Engineering

75

Orthogonal and Symmetric Matrices


DEFINITION 8.5 Orthogonal Matrix

A real square matrix A is t t orthogonal if and only if AA =A A = In.


An orthogonal matrix is nonsingular matrix and we find its inverse simply by taking its transpose.

Mathematics for Computer Engineering

76

Example

Mathematics for Computer Engineering

77

Mathematics for Computer Engineering

78

Theorem 8.7
A is an orthogonal matrix if t and only if A is an orthogonal matrix .

Mathematics for Computer Engineering

79

Theorem 8.8
If A is an orthogonal matrix, then |A| = 1.

Mathematics for Computer Engineering

80

A set of vector in R is said to be orthogonal if any two distinct vectors in the set are orthogonal (that is, their dot product is zero. The set is orthonormal if, in addition, each vector has length 1. We claim that the row of the orthogonal matrix form an orthonormal set of vectors, as do the columns.
Mathematics for Computer Engineering 81

From the last example, the row vectors are :

These each have length 1. and each is orthogonal to each other two.
Mathematics for Computer Engineering 82

Similarly, the column vectors are :

Each is orthogonal to the other two, and each has length 1.


Mathematics for Computer Engineering 83

Theorem 8.9
Let A be a real n*n matrix. Then, 1. A is orthogonal if and only if the row vectors form an orthonormal set n of vectors in R . 2. A is orthogonal if and only if the column vectors form an orthonormal n set of vectors in R .
Mathematics for Computer Engineering 84

Mathematics for Computer Engineering

85

Determine all 2*2 orthogonal matrix

What do we have to say about a, b, c, and d to make this an orthogonal matrix?

Mathematics for Computer Engineering

86

First, the two row vectors must be orthogonal and must have length 1.

ac+bd a2+b2 c2+d2

= 0 = 1 = 1

(8.1) (8.2) (8.3)

Mathematics for Computer Engineering

87

Second, the two column vectors must also be orthogonal.

ab+cd

= 0

(8.4)

Final, |Q| = 1

ad-bc = 1
This lead to two cases.
Mathematics for Computer Engineering 88

Case 1 ad-bc = 1 Multiply equation (8.1) by d to get

acd+bd2 = 0
Substitute ad =1+bc into this equation to get

c(1+bc)+bd2 = 0 c+b(c2+d2) = 0
Mathematics for Computer Engineering 89

But from (8.3) c2+d2 = 1 c+b = 0 c = -b


Put this into equation (8.3) to get

ab - bd = 0
Then b = 0 or a = d, leading to two subcases.
Mathematics for Computer Engineering 90

Case 1(a) b = 0 Then c = -b = 0 also, so

But each row vector has length 1, so a2 = d2 = 1. Further, |Q| = ad = 1 in the present case so a = d = 1 or a = d = -1. In these case,

Mathematics for Computer Engineering

91

Case 1(b) b 0
Then a = d, so

Since a2 + b2 = 1, there is some in [0,2 ) such that a = cos() and b = sin(). Then,

This include the 2 results of case 1(a) by choosing = 0 or = 2


Mathematics for Computer Engineering 92

Case 1 ad-bc = -1 By an analysis similar to that just done, we find now that, for some ,

Mathematics for Computer Engineering

93

This two cases give all the 2*2 orthogonal matrices. For example, with = /4 we get the orthogonal matrices.

Mathematics for Computer Engineering

94

and with = /6 we get the orthogonal matrices.

Mathematics for Computer Engineering

95

We can recognize the orthogonal matrices

as a rotation in the plane.

Mathematics for Computer Engineering

96

If the positive x, y system is rotated counterclockwise radian to form a new x', y' system. The coordinates in the two systems are related by

Mathematics for Computer Engineering

97

DEFINITION 8.6 Symmetric Matrix

A square matrix is symmetric if A =A .


This mean that each aij = aji, or that the matrix elements are the same if reflected across the main diagonal. For example,

Mathematics for Computer Engineering

98

Theorem 8.10
The eigenvalues of a real, symmetric matrix are real numbers.

Mathematics for Computer Engineering

99

Mathematics for Computer Engineering

100

Mathematics for Computer Engineering

101

Mathematics for Computer Engineering

102

Mathematics for Computer Engineering

103

Theorem 8.11
Let A be a real symmetric matrix. Then eigenvectors associated with distinct eigenvalues are orthogonal.

Mathematics for Computer Engineering

104

Mathematics for Computer Engineering

105

Example

Mathematics for Computer Engineering

106

Mathematics for Computer Engineering

107

Theorem 8.12
Let A be a real symmetric matrix. Then there is a real, orthogonal matrix that diagonalizes A.

Mathematics for Computer Engineering

108

Example

Mathematics for Computer Engineering

109

Mathematics for Computer Engineering

110

Quadratic Forms
DEFINITION 8.7 A (complex) quadratic form is an symmetric.

in which each ajk and zj is a complex number.


Mathematics for Computer Engineering 111

Mathematics for Computer Engineering

112

Mathematics for Computer Engineering

113

Mathematics for Computer Engineering

114

Example

Mathematics for Computer Engineering

115

Mathematics for Computer Engineering

116

Lemma 8.1
Let A be an n*n matrix of real or complex numbers. Let be an eigenvalue with eigenvector Z. Then.

Mathematics for Computer Engineering

117

Mathematics for Computer Engineering

118

Mathematics for Computer Engineering

119

Mathematics for Computer Engineering

120

Theorem 8.13 Principal Axis Theorem


Let A be a real symmetric matrix with eigenvalues 1,..,n. Let Q be an orthogonal matrix that diagonalizes A. Then the change of variables X = QY transforms to

Mathematics for Computer Engineering

121

Mathematics for Computer Engineering

122

Example

Mathematics for Computer Engineering

123

Mathematics for Computer Engineering

124

Example

Mathematics for Computer Engineering

125

Mathematics for Computer Engineering

126

Unitary, Hermitian, and SkewHermitian Matrices


If U is a nonsingular complex matrix, U-1 exists and is generally also complex matrix. Lemma 8.2

Mathematics for Computer Engineering

127

DEFINITION 8.8 Unitary Matrix An n*n complex matrix U is unitary if and only if

or

Mathematics for Computer Engineering

128

Example

Mathematics for Computer Engineering

129

If U is a real matrix,then the unitary t t condition U =In become UU = In, which makes U an orthogonal matrix. Unitary matrix are the complex analogues of orthogonal matrices. Since the rows(or column) of an orthogonal matrix form an orthonormal set of vectors, we will develop the complex analogue of the concept of orthonormality.

Mathematics for Computer Engineering

130

Mathematics for Computer Engineering

131

Mathematics for Computer Engineering

132

DEFINITION 8.9 Unitary System of Vectors Complex n-vectors F1,...,Fr form a unitary system if Fj Fk = 0 for j k, and each Fj Fj = 1

Mathematics for Computer Engineering

133

THEOREM 8.14
Let U be n*n complex matrix. Then U is unitary if and only if its row vectors form a unitary system.

Mathematics for Computer Engineering

134

Example

Mathematics for Computer Engineering

135

Mathematics for Computer Engineering

136

THEOREM 8.15
Let be an eigenvalue of the unitary matrix U. Then ||= 1.

Mathematics for Computer Engineering

137

Mathematics for Computer Engineering

138

DEFINITION 8.10 1 Hermitian Matrix An n*n complex hermitian if and only if

matrix

is

2 Skew-Hermitian Matrix An n*n complex matrix S is skewhermitian if and only if

Mathematics for Computer Engineering

139

Example

Mathematics for Computer Engineering

140

Mathematics for Computer Engineering

141

THEOREM 8.16
Let

be a complex matrix. Then

Mathematics for Computer Engineering

142

Mathematics for Computer Engineering

143

THEOREM 8.17
1. The eigenvalues of a hermitian matrix are real. 2. The eigenvalues of a skew-hermitian matrix are zero or pure imaginery.

Mathematics for Computer Engineering

144

Mathematics for Computer Engineering

145

Mathematics for Computer Engineering

146

P P-1AP diagonal matrix


5 A= 1 0

0 0 0 3 0 2

]
147

Mathematics for Computer Engineering

0 1 P AP = 1 0

5 1 0

0 3 / 2 1

]
148

Mathematics for Computer Engineering

You might also like