CHAPTER 3 Spectral and -spectral theory of -idempotent...
Transcript of CHAPTER 3 Spectral and -spectral theory of -idempotent...
39
CHAPTER 3
Spectral and -spectral theory of
-idempotent matrices
In this chapter, eigen and -eigen values of a -idempotent matrix are found.
Relations between -eigen values of a matrix and eigen values of the matrix are
analyzed (compare [29]). -diagonalizability and -similarity of -idempotent matrices
are discussed. The spectral and -spectral resolution of a -idempotent matrix (cf.[32]) is
found and all of its oblique projectors are determined as a polynomial in . Relations
among the multiplicity of eigen values of a -idempotent matrix and the matrix
functions such as tr , det and rank are discussed. A set of necessary conditions is
listed for a -idempotent matrix to be reduced to an idempotent matrix. A list comprising
a set of conditions when a -idempotent matrix becomes null, is also determined. An
equivalent condition for -idempotent matrix to be normal is derived.
40
3.1 -eigen value of a matrix
In this section, we shall define a -eigen value of a matrix as a special case of
generalized eigen value problem ( cf.[40] ) for some matrices and . For
that, first we define a permutation function K on the unitary space ℂ .
Let
ℂ then K is defined by,
K
ℂ ,where is the fixed disjoint product of transpositions in .
If is the associated permutation matrix of then it can be easily seen that K .
It is also clear that K K .
i.e., K K
ℂ
Example 3.1.1
Let
ℂ and let
Then K
Also K K
Definition 3.1.2
A -eigen value of a matrix is defined to be a zero of the polynomial
det . This polynomial is known as -characteristic polynomial.
41
Definition 3.1.3
A non-zero vector in ℂ is said to be a -eigen vector of a complex
matrix associated with a -eigen value if it satisfies K where K is as
defined before. This is equivalent to .
Example 3.1.4
is a -idempotent matrix. The -eigen values of are
1,1 and -1. A -eigen vector corresponding to the -eigen value 1 is . It can
be verified that .
i.e.,
Theorem 3.1.5
If is a complex matrix inℂ then
(i) is a ( -eigen value, -eigen vector) pair for if and only if it is an (eigen
value, eigen vector) pair for .
(ii) Every matrix satisfies the - characteristic equation of .
(iii) Any set of -eigen vectors corresponding to distinct -eigen values of a matrix
must be linearly independent.
Proof
(i) If is a ( -eigen value, -eigen vector) pair for then
K
Therefore is a (eigen value, eigen vector) pair for . By retracing the
above arguments, we see that the converse is also true.
(ii) By Cayley-Hamilton theorem, every matrix satisfies its characteristic equation.
That is,
42
det
det det
det det
by (3.1)
Therefore the matrix satisfies the - characteristic equation of . In a similar
manner, it can be easily proved that the matrix satisfies the -characteristic
equation of .
(iii) Any set of -eigen vectors corresponding to distinct -eigen values of a matrix
is the set of eigen vectors corresponding to distinct eigen values of the matrix
by what we have proved above in (i). Hence they are linearly independent.
Remark 3.1.6
If and denote the spectrum and -spectrum of respectively then it is
true that
by theorem 3.1.5 (i)
Theorem 3.1.7
Let ℂ . Then the quadratic form of is the quadratic form for .
Proof
Let be a -eigen vector of .
Then
If KA is a function defined by,
KA , then it is clearly a -quadratic form
in . Equivalently , we have KA .
Hence it is the quadratic form for .
43
Remark 3.1.8
(i) A -hermitian matrix is positive -definite (positive -semi definite) if and only
if all -eigen values of are positive (non-negative).
(ii) It can be seen that, is positive -definite if and only if is positive definite.
(iii) If a -hermitian matrix is positive -definite then is diagonalizable with all
real eigen values of .
Definition 3.1.9
A matrix ℂ is said to be upper [lower] -triangular if whenever
[ ]. This is equivalent to ‘ is upper [lower] -triangular if and only if
is upper [lower] triangular.
Note 3.1.10
The elements are known as -diagonal elements of .
Remark 3.1.11
If a matrix ℂ has linearly independent -eigen vectors, we can form a
matrix in which the columns are linearly independent -eigen vectors of . Hence we
have where is the -diagonal matrix of and it is denoted by
.
The matrix consisting of linearly independent -eigen vectors of in example 3.1.4
is given by
and
Therefore
Definition 3.1.12
Two matrices and in ℂ are said to be -similar if there exists a non-
singular matrix ℂ such that . Equivalently, is -similar to if
and only if is similar to .
44
Example 3.1.13
A matrix
is -similar to the matrix in example 3.1.4.
For a non singular matrix
and
, it can be
verified that .
Note that . That is is similar to .
Note 3.1.14
If and are -similar matrices then they have the same -characteristic
polynomial and hence the same -spectrum.
Remark 3.1.15
Let be the associated permutation matrix of a disjoint product of transpositions
‘ ’. Let be a non-singular matrix such that then we have
that is .
Theorem 3.1.16
Let be a -idempotent matrix.
(i) Let be a matrix similar to then is -idempotent if where
.
(ii) Let be a matrix -similar to then is -idempotent if where
.
Proof
(i)
by remark 3.1.15
. Hence is -idempotent.
(ii)
by remark 3.1.15
45
Therefore is similar to . Hence is -idempotent by what we have proved
above in (i).
Theorem 3.1.17
Let and be two -idempotent matrices. If is -similar to then is
similar to .
Proof
Since is -similar to , we can find a non-singular matrix such that
Hence is similar to .
46
3.2 Spectral characterizations of -idempotent matrices
In this section, the spectral resolution of a -idempotent matrix is determined as
well as the diagonalizability of -idempotent matrices is proved.
Theorem 3.2.1
Let be a -idempotent matrix. Then the eigen values of are zero or cube root
of unity.
Proof
Let be an eigen value of a -idempotent matrix . Then
by
by theorem 2.1.7 (c) and
Since , we have or and where
Example 3.2.2
is a -idempotent.
The eigen values of are .
Theorem 3.2.3
If a matrix ℂ is -idempotent then it is diagonalizable and the spectrum
where
. Moreover, there exist unique disjoint oblique
projectors for such that
47
Proof
Since [ by theorem 2.1.7 (c) ], the polynomial is a multiple
of of and then every root of has multiplicity 1 (cf.[13], pp.11) . Hence the
matrix is diagonalizable.
Moreover, it is clear that . by theorem 3.2.1
We define ’s by the following formula,
Using , we have
In the case that for , we see that . Similarly when
0 .
By spectral theorem [see theorem 1.2.4], we see that the non-zero ’s so obtained are
disjoint oblique projectors (i.e., and for ) to satisfy the
decompositions and . The uniqueness of the decompositions can be proved as
follows:
48
Suppose if possible, let ’s be non-zero disjoint oblique projectors such that
, for complex numbers and
. We wish to prove that this is
actually identical with and except for notations and order of terms. First, it is
proved that ’s are precisely the eigen values of the matrix .
Since , there exists a non-zero vector in the range of such that and
for .
Therefore is an eigen value of [ i.e., . Conversely, if is an eigen
value of then
Since ’s are disjoint, we can find at least one among the non-zero vectors for
which is linearly independent. Hence, it follows that for some . These
arguments show that the set of ’s equals the set of eigen values of . By suitably
changing the order of terms, we have .
Since the expression for is unique in terms of , we have for .
Hence the decompositions and are unique.
Example 3.2.4
Consider the -idempotent matrix in example 2.1.2. The oblique
projectors of are found to be
49
It can be easily verified that the above projectors are disjoint (i.e., for ) and
satisfy the decompositions and .
Remark 3.2.5
By theorem 3.2.3, we observe that every -idempotent matrices are
diagonalizable. Let denote the multiplicity of eigen values ,1 and
respectively of a -idempotent matrix . Note that denotes the multiplicity of eigen
values as well as because ‘imaginary roots occur in pairs’. By spectral theorem, it
can be seen that a -idempotent matrix can be reduced to the following diagonal form (all
the entries off the main diagonal are understood to be zero),
That is we can find a matrix such that
50
For example, consider -idempotent matrix in example 2.1.2. The eigen values
of are and . It is an easy exercise to find such that
that is
Theorem 3.2.6
Let denote the multiplicity of eigen values , 1 and
respectively of a -idempotent matrix of order . Then
(i) tr
(ii) det
(iii) rank tr
Proof
By diagonalizing the matrix , we can find a matrix such that satisfying .
It is obvious that
(i) tr
51
(ii) If then we have det .
Otherwise, det
(iii) Clearly, rank
by ]
tr
Example 3.2.7
(i) Consider the -idempotent matrix in example 2.1.2. The eigen values
of are and . That is
tr
det
rank
(ii) Consider the -idempotent matrix in example 3.2.2. The eigen values
of are and . That is
tr
det
rank
52
3.3 -spectral characterizations of -idempotent matrices
A -eigen value of a -idempotent matrix is found and an equivalent condition for
normality of a -idempotent matrix is also determined in this section.
Theorem 3.3.1
Let be a -idempotent matrix. Then the -eigen values of are and .
Proof
Pre multiplying by , we have
using
Pre multiplying by , we have
by theorem 2.1.7 (c)
using
Since , we have .
Example 3.3.2
The -eigen values of example 3.2.2 are .
Theorem 3.3.3
If a matrix ℂ is -idempotent then . Moreover, there
exist unique disjoint oblique projectors for such that
Proof
By and theorem 3.3.1, it is clear that .
53
We define ’s by the following formula
Then
In the case that for , we have . It can be proved that the
non-zero ’s so obtained are unique disjoint oblique projectors such that satisfying the
decompositions and analogous to the proof of theorem 3.2.3.
Example 3.3.4
Consider the -idempotent matrix in example 2.1.2. The oblique
projectors of are found to be
.
It can be easily verified that the above projectors are disjoint and satisfy the
decompositions and .
Remark 3.3.5
The diagonalizability of (where is a -idempotent matrix) implies the
-diagonalizability of . Let denote the multiplicity of -eigen values
of then we can find a matrix such that
54
For example, consider the -idempotent matrix in example 2.1.2. The
-eigen values of are and . It can be found that
It can be easily verified that .
That is
.
Theorem 3.3.6
Let be a -idempotent matrix. Then holds if and only if
or .
Proof
Suppose then by theorem 3.3.3
Therefore, by
But using
i.e.,
From and , we have .
In a similar manner, it can be proved that whenever .
55
Example 3.3.7
Consider the -idempotent matrix in example 2.2.6. It can be easily
determined that the -eigen values of are and . That is . We observe
that . But the -eigen values of the matrix in example 2.1.2 are and .
Clearly .
Theorem 3.3.8
Let be a -idempotent matrix. Then holds [i.e., reduces to idempotent
(cf. lemma 2.2.5)] if and only if any one of the following condition holds.
(i) is idempotent
(ii) is -idempotent
(iii) is square hermitian
(iv) is hermitian
(v) is -hermitian
(vi) is -square hermitian
(vii) is cube hermitian
(viii) is -cube hermitian
(ix) is -cube hermitian
(x) is skew cube hermitian
(xi) is skew -cube hermitian
(xii) is skew square hermitian
(xiii) is skew -square hermitian
Proof
(i) Since is idempotent, we have
Therefore by
Hence
By theorem 3.3.6,
(ii) Since is -idempotent, we have by theorem 3.3.3
Since is -idempotent, we have by theorem 3.2.3
Hence
That is
Therefore
56
By theorem 3.3.6, .
(iii)
Pre multiplying by , we have
Post multiplying by , we have
Since [by remark 2.1.4], we have
(iv) Since is hermitian, the eigen values of are real.
Therefore ℝ
That is .
Since is diagonalizable [by theorem 3.2.3], we have is idempotent.
(v)
. Hence is hermitian, which is (iv).
(vi)
. Hence is hermitian, which is (iv).
(vii) Since is cube hermitian
Pre and post multiplying by , we have
From and , we have
57
(viii)
. Hence is cube hermitian, which is (vii).
(ix)
by theorem 2.1.7 (e)
. Hence is hermitian, which is (iv).
(x) It is true that
Since is skew cube hermitian, we have
(xi)
. Hence is skew cube hermitian, which is (x).
(xii)
Pre multiplying by , we have
Post multiplying by , we have
Since [by remark 2.1.4], we have
Therefore and hence it can be proved as before (x) that .
(xiii)
As before (xii), it can be proved that .
Theorem 3.3.9
Let be a -idempotent matrix. Then becomes null (i.e., if and only if any
one of the following condition holds.
58
(a) is skew hermitian
(b) is skew hermitian
(c) is skew -hermitian
(d) is skew -hermitian
(e) is skew square hermitian
(f) is skew -square hermitian
(g) is skew cube hermitian
(h) is skew -cube hermitian
Proof
(a) Since is skew hermitian, the eigen values of are pure imaginary.
That is ℝ
It follows that .
Therefore .
(b) Since is skew hermitian, we have
ℝ
It follows that and hence .
(c)
Hence is skew hermitian, which is (b).
(d)
Hence is skew hermitian, which is (a).
(e)
Hence is skew -hermitian, which is (c).
59
(f)
Hence is skew hermitian, which is (a).
(g)
by theorem 2.1.7 (e)
Hence is skew hermitian, which is (b).
(h)
Hence is skew hermitian, which is (a).
Theorem 3.3.10
Let be a -idempotent matrix. Then the following are equivalent
(1) is normal
(2) is square hermitian
(3) is normal
(4) is hermitian
Proof
Suppose if is null then we observe that the theorem is trivially true. Assume that
.
(1) (2) :
By theorem 3.2.3, the matrix can be written as
, not all
Since is normal, ’s are orthogonal projectors (i.e., )
60
Hence becomes
But it is an easy exercise to prove that
Thus we have and hence is square hermitian.
(2) (3) :
by
by
From and , we have
Hence is normal.
(3) (4) :
By theorem 3.3.3, the matrix can be written as
, not both
Since is normal, we have
Hence is hermitian.
61
(4) (1) :
Therefore is square hermitian and it implies that is normal.