General Toeplitz matrices and the Takenaka-Malmquist basis Adhemar - - PowerPoint PPT Presentation

▶
general toeplitz matrices and the takenaka malmquist basis
SMART_READER_LITE
LIVE PREVIEW

General Toeplitz matrices and the Takenaka-Malmquist basis Adhemar - - PowerPoint PPT Presentation

Apr 07, 2023 28 likes •200 views

General Toeplitz matrices and the Takenaka-Malmquist basis Adhemar Bultheel 1 and Pierre Carrette 2 August 2003 1 Department Computer Science 2 Shell Oil Company K.U.Leuven, Belgium Houston, Texas USA adhemar.bultheel@cs.kuleuven.ac.be

slide-1
SLIDE 1

General Toeplitz matrices and the Takenaka-Malmquist basis

Adhemar Bultheel1 and Pierre Carrette2 August 2003

1Department Computer Science

K.U.Leuven, Belgium adhemar.bultheel@cs.kuleuven.ac.be http://www.cs.kuleuven.ac.be/∼ade/

2Shell Oil Company

Houston, Texas USA pierre.carrette@shell.com

Orthogonal Functions and Related Topics, Røros, Norway, Augist 12-16, 2003

slide-2
SLIDE 2

1/16

Takenaka-Malmquist basis

Orthogonal rational functions w.r.t. Lebesgue measure on T ϕn(z) =

  • 1 − |αn|2

1 − αnz

n−1

  • k=0

z − αk 1 − αkz

  • Bn(z)

, n ≥ 0, αk ∈ D, α0 = 0, ϕ−n = ϕn(1/z) {ϕn}k∈Z complete in L2(T) iff (1 − |αk|) = ∞ ⇒ Bn(z) → 0 Assumption |αk| ≤ c < 1, α0 = 0 .

Orthogonal Functions and Related Topics, Røros, Norway, Augist 12-16, 2003

  • A. Bultheel
slide-3
SLIDE 3

2/16

Convergence in Cq

2π, q > 2

f(ω) := F(eiω) ∈ Cq

2π, q > 2

Classical theory = all αk = 0: denote with a hat. ˆ ϕk(eiω) = eikω, ˆ ck = f, ˆ ϕk =

1 2π

2π f(ω) ˆ ϕk(eiω)dω Thm: f(ω) =

k ˆ

ck ˆ ϕk(eiω) absolutely convergent Thm: f ∈ Cq

2π, q > 2 then |k|<p ckϕk(eiω) → f(ω),

ck = f, ϕk cvg uniform in ω; rate of cvg is at least 1/pq−1 Proof is technical, based on |c0| ≤ f1 and |c±k| ≤ Cf (q)1/(ǫk)q, k ≥ 1 C and ǫ depend on q, c, k

Orthogonal Functions and Related Topics, Røros, Norway, Augist 12-16, 2003

  • A. Bultheel
slide-4
SLIDE 4

3/16

Example

The function is f(ω) = (cos 2ω + 2)/(4 cos 2ω + 5). f(ω) = ReF(eiω), F(z) = (z2 + 2)−1. Poles at ±i √ 2.

1 2 3 4 5 6 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1 2 3 4 5 6 2 4 6 8 10 12 14 x 10

−3

Figure: left: f and f2, random zeros; right: f − f4 and f − f6 for poles αk = (−1)k0.6.

Orthogonal Functions and Related Topics, Røros, Norway, Augist 12-16, 2003

  • A. Bultheel
slide-5
SLIDE 5

4/16

Example convergence

The function is f(ω) = (cos 2ω + 2)/(4 cos 2ω + 5). f(ω) = ReF(eiω), F(z) = (z2 + 2)−1. Poles at ±i √ 2.

2 4 6 8 10 12 14 16 18 20 10

−8

10

−7

10

−6

10

−5

10

−4

10

−3

10

−2

10

−1

10 2 4 6 8 10 12 14 16 18 20 10

−8

10

−7

10

−6

10

−5

10

−4

10

−3

10

−2

10

−1

10

Figure: left: f −fp∞ for (1) αk random, (2) αk = 0, (3) αk = (−1)k0.6, right: Fourier coefs for case (3)

Orthogonal Functions and Related Topics, Røros, Norway, Augist 12-16, 2003

  • A. Bultheel
slide-6
SLIDE 6

5/16

Convergence in C2π

Here convergence need not be uniform ⇒ Ces` aro means. Example: The function is f(ω) = |ω − π|.

1 2 3 4 5 6 0.05 0.1 0.15 0.2 0.25 1 2 3 4 5 6 0.05 0.1 0.15 0.2 0.25 0.3 0.35

Figure: left: |f − fn|, n = 2, 4, 6 for poles αk = 0. right: Ces` aro sum for n = 6.

Orthogonal Functions and Related Topics, Røros, Norway, Augist 12-16, 2003

  • A. Bultheel
slide-7
SLIDE 7

6/16

Ces` aro sums and Toeplitz matrix

ˆ Mp(f) = 1 2π 2π eikωf(ω)eilωdω p−1

k,l=0

= 1 2π 2π ˆ Γp(ω)f(ω)ˆ Γ∗

p(ω)dω

ˆ Γp(ω) = [1, e−iω, . . . , e−i(p−1)ω]∗ = [ ˆ ϕ0(eiω), . . . , ˆ ϕp−1(eiω)]∗ Note ˆ Γ∗

p(µ)ˆ

Γp(σ) =

p−1

  • k=0

ˆ ϕk(eiσ) ˆ ϕk(eiµ) = ˆ Kp(µ, σ) ˆ γp = ˆ Γ∗

p(ω)ˆ

Γp(ω) = ˆ Kp(ω, ω) = p Recall ˆ Mp(f) = 1 2π 2π ˆ Γp(ω)f(ω)ˆ Γ∗

p(ω)dω

Orthogonal Functions and Related Topics, Røros, Norway, Augist 12-16, 2003

  • A. Bultheel
slide-8
SLIDE 8

7/16

Set ˆ Gp(ω) = ˆ Γp(ω)/

  • ˆ

γp. Then ˆ G∗

p(ω) ˆ

Mp(f) ˆ Gp(ω) =

  • |k|<p
  • 1 − |k|

p

  • ˆ

ck ˆ ϕk(eiω) This is Ces` aro sum and lim

p→∞

ˆ G∗

p(ω) ˆ

Mp(f) ˆ Gp(ω) = f(ω) lim

p→∞

ˆ G∗

p(ω) ˆ

Mp(f) ˆ Mp(g) ˆ Gp(ω) = f(ω)g(ω)

Orthogonal Functions and Related Topics, Røros, Norway, Augist 12-16, 2003

  • A. Bultheel
slide-9
SLIDE 9

8/16

Generalized Ces` aro sums

Assume f ∈ Cq

2π with q > 4.

Generalized Toeplitz: Mp(fi) =

1 2π

2π Γp(ω)fi(ω)Γ∗

p(ω)dω

Γp(ω) = [ϕ0(eiω), . . . , ϕp−1(eiω)]∗ Gp(ω) = Γp(ω)

  • γp(ω)

, γp(ω) = Γ∗

p(ω)Γp(ω)

G∗

p(σ)T(Mp(f1), . . . , Mp(fn))Gp(µ) =

  • T(f1, . . . , fn),

σ = µ 0, σ = µ T(x1, . . . , xn) analytic function in n variables. Rate of convergence as fast as 1/p if σ = µ or ln p

p if σ = µ.

Depends on analysis of the generalized Toeplitz matrix Mp(f).

Orthogonal Functions and Related Topics, Røros, Norway, Augist 12-16, 2003

  • A. Bultheel
slide-10
SLIDE 10

9/16

Spectral properties

Classical case lim

p→∞

1 p

p−1

  • i=0

λi( ˆ Mp(f)) = 1 2π 2π f(ω)dω Generalized to f ∈ Cq

2π, q > 4

lim

p→∞

1 p

p−1

  • i=0

T (λi(Mp(f))) = 1 2π 2π T

  • f(χ−1(ω))
  • dω

χ(ω) = limp→∞

χp(ω) p

, χp(ω) = phase of Bp(ω).

Orthogonal Functions and Related Topics, Røros, Norway, Augist 12-16, 2003

  • A. Bultheel
slide-11
SLIDE 11

10/16

Example

2 4 6 8 10 12 14 16 18 20 0.57 0.575 0.58 0.585 0.59 0.595 0.6 0.605

f(ω) = cos 2ω + 2 4 cos 2ω + 5 lim

p→∞

1 p

p−1

  • k=0

λk(Mp(f)) = 1 2π 2π f(˜ χ−1(ω))dω = 2π f(µ)˜ χ(µ)dµ left-hand side for p = 20 is 0.6045, right-hand side integral = 0.6098

Orthogonal Functions and Related Topics, Røros, Norway, Augist 12-16, 2003

  • A. Bultheel
slide-12
SLIDE 12

11/16

Quadrature formulas

The basic idea in the proofs is to approximate integrals by quadratures. Mp(f) = 1 2π 2π Γp(ω)f(ω)Γ∗

p(ω)dω

Use 1 2π 2π g(ω)dω ≈

p−1

  • k=0

Hkg(ωk). Then Mp(f) ≈ Wp(f) = ΥpFpΥ∗

p

Fp = diag(f(ω0), . . . , f(ωp−1)), Υp = [˜ Γp(ω0), . . . , ˜ Γp(ωp−1)], ˜ Γp(ωk) = Γp(ωk)

  • Hk

Orthogonal Functions and Related Topics, Røros, Norway, Augist 12-16, 2003

  • A. Bultheel
slide-13
SLIDE 13

12/16

The quadrature formula is exact in Lp · L−p = span{ϕk(eiω)ϕl(eiω) : k, l = 0, . . . , p − 1} if ωk are zeros of para-orthogonals ϕp(eiω) − ηBp(eiω)ϕp(eiω), η ∈ T and Hk = 1/γp(ωk). The classical case = Szeg˝

  • quadrature: ωk equidistant, Hk = 1/p. Then

Υp is the (unitary) FFT matrix. Mp(f) ≈ Wp(f) = ΥpFpΥ∗

p

EVD of Wp(f) ≈ Mp(f), so f(ωk) ≈ fp(ωk) approximate the eigenvalues

  • f Mp(f).

Orthogonal Functions and Related Topics, Røros, Norway, Augist 12-16, 2003

  • A. Bultheel
slide-14
SLIDE 14

13/16

The general case = rational Szeg˝

  • : ωk ∈ Ωp(θ), Hk = 1/γp(ωk).

Ωp(θ) = {ω ∈ [0, 2π) : χp(ω) = θ mod 2π; η = eiθ; Bp(eiω) = eiχp(ω)}.

1 2 3 4 5 6 −3 −2 −1 1 2 3 1 2 3 4 5 6 −3 −2 −1 1 2 3

Figures: χ10(ω), left αk = ±0.6i, right αk = 0.9eikπ/10

Orthogonal Functions and Related Topics, Røros, Norway, Augist 12-16, 2003

  • A. Bultheel
slide-15
SLIDE 15

14/16

Spectral approximation

2 4 6 8 10 12 14 16 18 20 0.05 0.1 0.15 0.2 0.25 0.3 1 2 3 4 5 6 0.4 0.5 0.6 0.7 0.8 0.9 1

f(ω) = (cos 2ω+2)/(4 cos 2ω+5). Left: 2-norm of the vector of differences λi(Mp(f)) − f(ωi) as a function of p. Right diag(Υ∗

pMp(f)Υp) (circles)

and f(ω).

Orthogonal Functions and Related Topics, Røros, Norway, Augist 12-16, 2003

  • A. Bultheel
slide-16
SLIDE 16

15/16

Spectral approximation

0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.1 0.2 0.3 0.4 0.5 0.6 0.7

cumulative distribution of f(ωk) for ωk ∈ Ω20(1) (dashed) and cumulative distribution of λk(M20(f)) (solid).

Orthogonal Functions and Related Topics, Røros, Norway, Augist 12-16, 2003

  • A. Bultheel
slide-17
SLIDE 17

16/16

Reference

[1] A. Bultheel, P. Carrette. Algebraic and spectral properties of general Toeplitz matrices. SIAM J. Control Optim., 41 (2003), pp. 1413–1439.

Orthogonal Functions and Related Topics, Røros, Norway, Augist 12-16, 2003

  • A. Bultheel