About the factorization of rational matrix functions Juan C. S - - PowerPoint PPT Presentation

about the factorization of rational matrix functions
SMART_READER_LITE
LIVE PREVIEW

About the factorization of rational matrix functions Juan C. S - - PowerPoint PPT Presentation

About the factorization of rational matrix functions Juan C. S anchez Rodr guez jsanchez@ualg.pt Departamento de Matem atica Universidade do Algarve CEAF-Center of Functional Analysis and Applications QISG - 2012 - OLH AO About


slide-1
SLIDE 1

About the factorization of rational matrix functions

Juan C. S´ anchez Rodr´ ıguez jsanchez@ualg.pt Departamento de Matem´ atica Universidade do Algarve CEAF-Center of Functional Analysis and Applications

QISG - 2012 - OLH˜ AO

slide-2
SLIDE 2

1

About the factorization of rational matrix functions

1 Introduction

Definition of factorization Riemann-Hilbert BVP Fundamental set of solution of Riemann-Hilbert problem

2 Factorization procedure

Riemann-Hilbert problem and difference equations The Z-transform Solution of the difference equation Main Result

3 Examples

First example Second example

4 Historic remarks and references

slide-3
SLIDE 3

2

Section’s Contents

1 Introduction

Definition of factorization Riemann-Hilbert BVP Fundamental set of solution of Riemann-Hilbert problem

2 Factorization procedure 3 Examples 4 Historic remarks and references

slide-4
SLIDE 4

3

Definition of factorization If A is a m × m continuous matrix function on T, by (left) factorization of A in C(T) we mean the following representation: A(t) = A+(t)Λ(t)A−(t), t ∈ T, where Λ(t) = diag [tκ1, . . . , tκm], κi ∈ Z, κi ≥ κj, i < j, i, j = 1, . . . , m, and A±1

+ ∈ C m×m +

(T), A±1

− ∈ C m×m −

(T),

slide-5
SLIDE 5

4

Example of (left) factorization The matrix function A(t) = t + 1 1 t2 t

  • ,

admits the factorization A(t) = A+(t)Λ(t)A−(t), with Λ(t) = diag

  • t,

1

  • and

A+(t) =

  • −1

t + 1 −t + 1 t2

  • ,

A−(t) = 1 1 1

  • .

The integers κ1 = 1 and κ2 = 0, are the partial indices.

slide-6
SLIDE 6

5

Riemann-Hilbert BVP This problem consists in finding the m-vector function Φ(t) = ϕ+(t), t ∈ T+ ϕ−(t), t ∈ T− where ϕ± are analytic in T±, respectively, with the following boundary condition: ϕ+ = Aϕ−, ϕ−(∞) ∈ C, (1) and A is a given m × m matrix function.

slide-7
SLIDE 7

6

Fundamental set of solution Φ(1), . . . , Φ(m) are particular solutions of (1); −κi is the order at infinity of Φ(i), i = 1, . . . , m; Among the solutions of (1), −κ1 is the lowest order at infinity; Among the solutions of (1), which cannot be written in the form p1Φ(1) + · · · + psΦ(s), 1 ≤ s < m, −κs+1 is the lowest order at infinity; Let Φ−(t), t ∈ T−, be the matrix function whose columns are the vector functions Φ(i), i = 1, . . . , m. Then the factors of the factorization of A are A+ = AΦ−, Λ(t) = diag [tκ1, . . . , tκm], A− = (Φ−Λ)−1 .

slide-8
SLIDE 8

7

Section’s Contents

1 Introduction 2 Factorization procedure

Riemann-Hilbert problem and difference equations The Z-transform Solution of the difference equation Main Result

3 Examples 4 Historic remarks and references

slide-9
SLIDE 9

8

Riemann-Hilbert problem and difference equations It is very well known that the factorization of rational matrix-functions is, up to a multiplication by a polynomial, equivalent to the factorization of a polynomial matrix-function. Let A be the m × m matrix-function with polynomial entries aij = a(k)

ij tk + . . . + a(1) ij t + a(0) ij ,

then A admits the following representation A(t) = Aktk + ... + A1t + A0, where As is the square constant matrix As = {a(s)

ij }, s = 0, . . . , k.

slide-10
SLIDE 10

9

Riemann-Hilbert problem and difference equations If ϕ−(t) = ϕ0 +

  • n=1

ϕnt−n, the pair of vector functions ϕ± is a solution of the Riemann-Hilbert problem (1) if and only if enjoy the equality ϕ+ = Aϕ− = Aϕ0 +

k

  • i=1

Ai

i

  • j=1

ϕjtj−i+ +

  • n=1

(Akϕn+k + . . . + A1ϕn+1 + A0ϕn)t−n, and we can assert that the vector function ϕ+ is analytic in T+ if and only if the vector ϕn is a solution of the linear system of difference equations with constant coefficients Akϕn+k + . . . + A1ϕn+1 + A0ϕn = 0.

slide-11
SLIDE 11

10

The Z-transform By Z-transform of xn we mean the scalar function Z(xn) = x =

  • n=1

xnz−n. Some properties of the Z-transform Z(xn+k) = zkx − k

i=1 xizk−i

Z(δ(s)

n ) = 1 zs , where δ(s) n

= { 1, n = s 0, n = s is the Kronecker delta sequence, Z

  • an−1

=

1 z−a, |z| > |a|,

Z

  • nan−1

=

z (z−a)2 , |z| > |a|,

Z

  • (n+s−2)!

(n−1)! an−1

= f (s)

a

(z), |z| > |a|, where f (s)

a

(z) = s!zs−1 (z − a)s , a = 0, s ∈ N.

slide-12
SLIDE 12

11

The Z-transform Applying the Z-transform to the linear system Akϕn+k + . . . + A1ϕn+1 + A0ϕn = 0. and using the Z-transform’s properties we can write the following identity Ak

  • zkϕ −

k

  • i=1

ϕizk−i

  • + . . . + A1(zϕ − ϕ1) + A0ϕ = 0.

where ϕi = (ϕ1i, ..., ϕmi)⊤, i = 1, ..., k, are arbitrary constant vectors.

slide-13
SLIDE 13

12

The Z-transform Then ϕ, the Z-transform of ϕn, enjoy the following properties: ϕ = A−1(z) k

j=1

j

i=1 Ajϕizj−i;

ϕ is a column with rational entries; The poles of ϕ are zeros of det A; ϕ = l

i=1

di

j=1 Cijf (j) zi (z) + d i=1 Kiz−i, where

f (s)

a

(z) = s!zs−1 (z − a)s , a = 0, s ∈ N. ϕ(∞) = 0.

slide-14
SLIDE 14

13

Solution of the difference equation Applying the Z-transform inverse we get the following solution ϕn =

l

  • i=1

di

  • j=1

Cij (n + j − 2)! (n − 1)! zn−1

i

+

d

  • i=1

Kiδ(i)

n

(2)

  • f

Akϕn+k + . . . + A1ϕn+1 + A0ϕn = 0, where the vector-sequence ϕn must satisfy the initial conditions ϕn|n=i = ϕi, i = 1, . . . , k.

slide-15
SLIDE 15

14

Solution of the difference equation Theorem If zi is one of the zeros of det A(t) and |zi| > 1 the series

  • n=1

ϕnt−n with ϕn =

−di

  • j=1

Cij (n + j − 2)! (n − 1)! zn−1

i

, converges in T−, if and only if, Cij = 0, j = 1, ..., di.

slide-16
SLIDE 16

15

Main Result Theorem The pair of functions ϕ− = ϕ0 +

  • n=1

ϕnt−n and ϕ+ = Aϕ− is the general solution of a Riemann-Hilbert problem, if and only if, the coefficients ϕn are given by (2), where the constants ϕsi, i = 1, ..., k; s = 1, ..., m must satisfy the additional conditions: ϕ(∞) = 0; ϕn|n=i = ϕi; Cij = 0, j = 1, ..., di. for each zero zi of det A, such that |zi| > 1.

slide-17
SLIDE 17

16

The factorization procedure Let A(t) = Aktk + ... + A1t + A0, to obtain the factorization of A, we can use the following procedure: Write Akϕn+k + . . . + A1ϕn+1 + A0ϕn = 0. Find the Z-transform of ϕn. Impose the condition ϕ(∞) = 0. Using the Z-transform inverse, obtain the solution ϕn of the linear system of difference equations. Consider the restrictions Cij = 0, j = 1, ..., di, for each zero zi of det A, such that |zi| > 1. Use the initial conditions ϕn|n=i = ϕi. Obtain the vector function ϕ−(t) = ϕ0 + ∞

n=1 ϕnt−n.

Find a fundamental set of solutions of the Riemann-Hilbert BVP. Construct the factors of the left factorization of A.

slide-18
SLIDE 18

17

Section’s Contents

1 Introduction 2 Factorization procedure 3 Examples

First example Second example

4 Historic remarks and references

slide-19
SLIDE 19

18

First example Let A be the 2 × 2 polynomial matrix-function A(t) = t + 1 1 t2 t

  • = A2t2 + A1t + A0,

where A2 = 1

  • ,

A1 = I2, A0 = 1 1

  • .

In order to find the general solution of the Riemann-Hilbert BVP, we need to solve the respective system of difference equations A2ϕn+2 + A1ϕn+1 + A0ϕn = 0.

slide-20
SLIDE 20

19

First example Let ϕi =

  • ai

bi ⊤ , i = 1, 2, be a constant vector. Applying the Z-transform: ϕ(z) =

  • −(a2 + b1)z−1

a1 + a2 + b1 + (a2 + b1)z−1

  • .

Taking into account ϕ(∞) = 0, we have the following equality a1 + a2 + b1 = 0, and consequently ϕ(z) = (a2 + b1)

  • −1

1 ⊤ z−1. Using the inverse of the Z-transform we get ϕn = (a2 + b1)

  • −1

1 ⊤ δ(1)

n .

slide-21
SLIDE 21

20

First example The sequence ϕn must satisfy the initial conditions ϕn|n=1 =

  • a1

b1 ⊤ , ϕn|n=2 =

  • a2

b2 ⊤ , then a1 = −(a2 + b1), a2 = 0, b1 = a2 + b1, b2 = 0, and making use of the obtained equality a1 + a2 + b1 = 0, we have ϕn = a1

  • −1

1 ⊤ δ(1)

n .

slide-22
SLIDE 22

21

First example The general solution of the Riemann-Hilbert BVP is ϕ+ = Aϕ−, where ϕ− = ϕ0 +

  • n=1

ϕnt−n =

  • a0

b0 ⊤ + a1t−1 −1 1 ⊤ . To construct the factorization of A we need to find a particular solution with the lowest order at infinity: Φ(1)(t) = ϕ(1)

− (t) = t−1

−1 1 ⊤ , t ∈ T−, with order −1 at infinity. So κ1 = 1. Next, we need to find another solution with the lowest order at

  • infinity. This solution cannot be written in the form p1Φ1, where

p1 is a polynomial. One such solution, with the desired property is for instance Φ(2)(t) = ϕ(2)

− (t) =

  • 1

⊤ , t ∈ T−, and obviously κ2 = 0.

slide-23
SLIDE 23

22

First example Then Λ(t) = diag

  • t,

1

  • ,

and Φ−(t) = −t−1 1 t−1

  • .

Consequently A+(t) = A(t)Φ−(t) =

  • −1

t + 1 −t + 1 t2

  • ,

A−(t) = (Φ−(t)Λ(t))−1 = 1 1 1

  • ,

are the factors of the left factorization of A(t) = t + 1 1 t2 t

  • .
slide-24
SLIDE 24

23

Second example Next we will consider the factorization of the following polynomial matrix function A(t) = (t − 3)2 (t − 3)2 t3 t3 + t2(3t − 1)2

  • .

(3) In this case, some of the calculations were done with the help of the software Wolfram Mathematica. The matrix function A can be written as follows A(t) = A4t4 + A3t3 + A2t2 + A1t + A0, where A0 = 9 9

  • ,

A1 = −6 −6

  • ,

A2 = 1 1 1

  • ,

A3 = 1 −5

  • ,

A4 = 9

  • .
slide-25
SLIDE 25

24

Second example As we already know, we need to solve the system of difference equations: A4ϕn+4 + A3ϕn+3 + A2ϕn+2 + A1ϕn+1 + A0ϕn = 0, where ϕn =

  • an

bn ⊤. The Z-transform ϕ of ϕn is given by ϕ = A−1(z)[A1ϕ1 + A2(ϕ1z + ϕ2) + A3(ϕ1z2 + ϕ2z + ϕ3)+ +A4(ϕ1z3 + ϕ2z2 + ϕ3z + ϕ4)], with the initial data ϕi =

  • ai

bi ⊤ , i = 1, . . . , 4.

slide-26
SLIDE 26

25

Second example Then, using the software Wolfram Mathematica, we have that a and b, the Z-transforms of an and bn are, respectively a = M1 z + M2 z2 + M3 z − 1

3

+ M4z (z − 1

3)2 +

M5 z − 3 + M6 (z − 3)2 , b = N1 z + N2 z2 + N3 z − 1

3

+ N4z (z − 1

3)2 +

N5 z − 3 + N6 (z − 3)2 , with M5 = 1 256(283a1−5a2+283b1−5b2), M6 = −67 64(3a1−a2+3b1−b2), and N5 = 1 256(−27a1+5a2−27b1+5b2), N6 = − 3 64(3a1−a2+3b1−b2).

slide-27
SLIDE 27

26

Second example M5, M6, N5 and N6 must vanish, then b1 = −a1, b2 = −a2. The new expressions for the coefficients Ni, Mi, i = 1, . . . , 4 are M1 = a1 − 6a3 + 21b3 − 54b4, M2 = a2 − a3 + 5b3 − 9b4, M3 = 9(a3 − 3b3 + 9b4), M4 = −3(a3 − 2b3 + 9b4), Ni = −Mi, i = 1, . . . , 4. and a = M1 z + M2 z2 + M3 z − 1

3

+ M4z (z − 1

3)2 ,

b = N1 z + N2 z2 + N3 z − 1

3

+ N4z (z − 1

3)2 .

slide-28
SLIDE 28

27

Second example Using the inverse of Z-transform we can affirm that an = M1δ(1)

n

+ M2δ(2)

n

+ M3 3n−1 + M4 n 3n−1 , bn = −an. and b3 = −a3, b4 = −a4. So, the final expressions of the coefficients Ni, Mi, i = 1, . . . , 4 are M1 = a1 − 27a3 + 54a4, M2 = a2 − 6a3 + 9a4 M3 = 36a3 − 81a4, M4 = −9a3 + 27a4, and Ni = −Mi, i = 1, . . . , 4.

slide-29
SLIDE 29

28

Second example Finally, the general solution of the Riemann-Hilbert problem is ϕ+ = Aϕ−, ϕ− = ϕ0 +

  • n=1

ϕnt−n, where ϕn = an

  • 1

−1 ⊤ , Then ϕ− =

  • a0

b0 ⊤ + a(t)

  • 1

−1 ⊤ with a(t) = M1 t + M2 t2 + M3 t − 1

3

+ M4 t (t − 1

3)2

and M1 = a1 − 27a3 + 54a4, M2 = a2 − 6a3 + 9a4, M3 = 36a3 − 81a4, M4 = −9a3 + 27a4.

slide-30
SLIDE 30

29

Second example To conclude the factorization of A we need to find the fundamental set of solutions. Among the solutions of the considered Riemann-Hilbert problem, the lowest order at infinity is −4, when ϕ0 = ϕ1 = ϕ2 = ϕ3 = 0, i.e., when a0 = b0 = a1 = a2 = a3 = 0 and a4 = 0. One solution with such order at infinity is, for instance Φ(1)(t) = ϕ(1)

− (t) =

1 t2(3t − 1)2

  • 1

−1 ⊤ , t ∈ T−. Further, the solution Φ2 must have order equal to zero at infinity. Otherwise, Φ2 can be written in the form p1Φ2, where p1 is a

  • polynomial. Among the solutions with the order equal to zero at

infinity we take Φ(2)(t) = ϕ(2)

− (t) =

  • 1

⊤ , t ∈ T−. Now it is possible to assert that κ1 = 4 and κ2 = 0.

slide-31
SLIDE 31

30

Second example Then Λ = diag

  • t4,

1

  • ,

and taking into account that Φ−(t) =

  • 1

t2(3t−1)2

1 −

1 t2(3t−1)2

  • ,

we have that the external factors of the factorization of A are A+(t) = A(t)Φ−(t) =

  • (t − 3)2

−1 t3

  • A−(t) = (Φ−(t)Λ(t))−1 =
  • − (3t−1)2

t2

1 1

  • .
slide-32
SLIDE 32

31

Section’s Contents

1 Introduction 2 Factorization procedure 3 Examples 4 Historic remarks and references

slide-33
SLIDE 33

32

Historic remarks and references As far as we know, there are two methods of constructing the factorization of rational matrices: The splitting of zeros was proposed by Gakhov in the middle

  • f XX century

The second method is based on the space theory for linear input-output system. Bart H., Gohberg I. and Kaashoek M. A. Minimal Factorization

  • f Matrix and Operator Functions. Birkh¨

auser, 1979. Clancey K. F. and Gohberg I. Factorization of Matrix Functions and Singular Integral Operators. Birkh¨ auser, 1981. Litvinchuk G. S. and Spitkovskii I. M. Factorization of Measurable Matrix Functions. Birkh¨ auser, 1987. Rodr´ ıguez, J. S. and Campos, L. F. Factorization of rational matrix functions and difference equations. Submitted for publication.