• No results found

Matrix Transformations and Disk of Convergence in Interpolation Processes

N/A
N/A
Protected

Academic year: 2020

Share "Matrix Transformations and Disk of Convergence in Interpolation Processes"

Copied!
11
0
0

Loading.... (view fulltext now)

Full text

(1)

Volume 2008, Article ID 905635,11pages doi:10.1155/2008/905635

Research Article

Matrix Transformations and Disk of Convergence

in Interpolation Processes

Chikkanna R. Selvaraj and Suguna Selvaraj

Department of Mathematics, Pennsylvania State University, Shenango Campus, 147 Shenango Avenue, Sharon, PA 16146, USA

Correspondence should be addressed to Chikkanna R. Selvaraj,[email protected]

Received 23 May 2008; Accepted 22 July 2008

Recommended by Huseyin Bor

Letdenote the set of functions analytic in|z|< ρbut not on|z|ρ1< ρ <∞. Walsh proved that the difference of the Lagrange polynomial interpolant offzand the partial sum of the Taylor polynomial offconverges to zero on a larger set than the domain of definition off. In 1980, Cavaretta et al. have studied the extension of Lagrange interpolation, Hermite interpolation, and Hermite-Birkhoffinterpolation processes in a similar manner. In this paper, we apply a certain matrix transformation on the sequences of operators given in the above-mentioned interpolation processes to prove the convergence in larger disks.

Copyrightq2008 C. R. Selvaraj and S. Selvaraj. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

1. Introduction

Ifx xkis a complex number sequence andA ankis an infinite matrix, thenAxis the

sequence whosenth term is given by

Axnk0

ankxk. 1.1

A matrixAis calledXY matrix ifAxis in the setYwheneverxis inX. For 0≤α <∞, let

x: lim sup|xk|1/k α. 1.2

For various values ofα, this sequence space has been studied extensively by several authors

(2)

Theorem 1.1. An infinite matrixAis aGuGvmatrix if and only if for each numberwsuch that

0< w <1/v, there exist numbersBandssuch that 0< s <1/uand

|ank|wnBsk 1.3

for allnandk.

Letdenote the collection of functions analytic in the disk {z ∈ C :|z| < ρ}

for some 1 < ρ <∞and having a singularity on the circle|z| ρ. InSection 2, we state the results proved by Cavaretta Jr. et al.6on the Lagrange interpolation, Hermite interpolation, and Hermite-Birkhoffinterpolation of fz ∈ in the nth roots of unity, which will be

required. Main results are given inSection 3and deal with the application of a certain matrix to the various polynomial interpolants in the above results. Interestingly, we are able to show that under the matrix transformation the difference of the interpolant polynomials and the corresponding Taylor polynomials converges to zero in a larger region.

2. Preliminaries

Throughout the paper, we assume thatfz with 1< ρ < ∞andfzhas the Taylor

series expansion

fz k0

akzk. 2.1

For each integern≥ 0, letLnz;fdenote the unique Lagrange interpolation polynomial of

degreenwhich interpolatesfzin then1st roots of unity, that is,

Lnω;f fω, 2.2

whereω is anyn1st root of unity. SettingSnz;f

n1

k0akzk, the well-known Walsh

equiconvergence theorem7states that

lim

n→ ∞Lnz;fSnz;f 0, ∀|z|< ρ

2, 2.3

with the convergence being uniform and geometric on any closed subset of |z| < ρ2.This

theorem has been extended in various ways by several authors6,8–10. In all that follows, we state some of the results of6which are needed for our main results.

Under Lagrange interpolation, letting

Sn,jz;f n

k0

akjn1zk, j0,1, . . . , 2.4

(3)

Theorem 2.1. For each positive integerl,

lim

n→ ∞

Lnz;fl−1

j0

Sn,jz;f

0, ∀|z|< ρl1 2.5

and this result is best possible.

In the proof ofTheorem 2.1, it has been shown by the authors that

Lnz;fl−1

j0

Sn,jz;f

1 2πi

Γ

fttn1zn1

t−ztn11tln1dt, 2.6

whereΓis any circle|t|Rwith 1< R < ρ.

In6the authors have studied the Hermite interpolation also in a similar way. For a fixed positive integerr ≥2, lethrn1−1z;fbe the unique Hermite polynomial interpolant

tof, f, . . . , fr−1in then1st roots of unity, that is,

hjrn11ω;f fjω, j0,1, . . . , r−1, 2.7

whereωis anyn1st root of unity. Setting

Hrn1−1,0z;f rn1−1

k0 akzk,

Hrn1−1,jz;f βjz n

k0

akn1rj−1·zk, j1,2, . . . ,

2.8

where

βjz r−1

k0

rj−1

k

zn11k

, j 1,2, . . . , 2.9

in6, Theorem 3, page 162the authors proved the following result.

Theorem 2.2. For each positive integerl,

lim

n→ ∞

hrn1−1z;fl−1

j0

Hrn1−1,jz;f

0, ∀|z|< ρ1l/r 2.10

(4)

In the proof ofTheorem 2.2, it was shown by the authors in6, page 165that

hrn1−1z;fl−1

j0

Hrn1−1,jz;f

1 2πi

Γ ft

t−zKt, zdt, 2.11

whereΓis any circle|t|Rwith 1< R < ρand

|Kt, z| ≤M

Rn1|z|n1|z|1r−1

Rrln1 . 2.12

Under Hermite-Birkhoffinterpolation, the authors in6established several results for different cases. We consider here only the0, mcase. Letnandmbe integers withnm≥0. Letb0,m2n1z;fbe the unique Hermite-Birkhoffpolynomial of degree 2n1 which interpolates

f in then1st roots of unity and whosemth derivative interpolatesfm in then1st roots of unity, that is,

b0,m2n1ω;f fω, b2n10,mω;fmfmω, 2.13

whereωis anyn1st root of unity. Setting

B0,m2n1,0z;f

2n1

k0 akzk,

B2n1,ν0,m z;f

n

j0

ajν1n1·zjqj,νz, ν1,2. . . ,

2.14

whereqj,νzis a polynomial of degreen1given by

qj,νz zn1

ν1n1 jm−n1jm

n1jm−jm z

n11, j0,1, . . . , n, 2.15

with the following conventional notation

jm

⎧ ⎨ ⎩

jj−1· · ·j−m1, ifmj,

0, ifm > j,

2.16

(5)

Theorem 2.3. For each positive integerl,

lim

n→ ∞

b2n10,mz;f

l−1

j0

B2n1,j0,mz;f

0, for|z|< ρ1l/2 2.17

and this result is best possible.

In the proof, it was shown in6, Theorem 4, page 171that

b0,m2n1z;f

l−1

j0

B0,m2n1,jz;f 1 2πi

ΓftKn,lt, zdt, 2.18

whereΓis any circle|t|Rwith 1< R < ρand|Kn,lt, z|is bounded on the circle|t|R <|z|

by

|Kn,lt, z| ≤

|z|n1|t|n1|z|n1 |tz||t|l1n1|tn1−1|

|z|n11|z|n1Rn1

|z| −RRl2n1 . 2.19

3. Main results

Our aim in this section is to apply aGuGvmatrix to the polynomial sequences of operators

in each of the above three theorems given in Section 2 and prove that the difference of transformed sequences in each case converges to zero in a lager disk Dρ. To simplify, let

us denoteLnz;fand

l1

j0Sn,jz;finTheorem 2.1byLnandSn,l, respectively.

Theorem 3.1. LetfzAρand letΓbe any circle|t|Rwith 1< R < ρ. For anyρ > ρ , choose u >ρ/R and 0< v <1. LetAbe aGuGvmatrix and define

λnz

k0

ankLk, σn,lz

k0

ankSk,l 3.1

forn0,1, . . . .Then, for eachl,

lim

n→ ∞λnz−σn,lz 0,zDρ. 3.2

Proof. Using the integral representation given in2.6, for eachl1,2, . . . ,we have

λnz−σn,lz

k0

ankLkSk,l

k0

ank 1

2πi

Γ

fttk1zk1 t−ztk11tlk1dt

k0

ank

1 2πi

Γ

ft t−ztlk1

1− z

t

k1

tk1 tk1−1dt.

(6)

Since|t|R >1, we have

λnz−σn,lz

k0 ank 1 2πi Γ ft t−ztlk1

1− z

t

k1

q0 1 tk1 q dt 1 2πi Γ ft t−z

q0 1 tlqk0 ank 1− z

t k1 1 tlq k dt. 3.4

The interchange of the integral and the summations is justified by showing that the series

k0 ank 1 tlq k ,k0 ank z

t

k 1

tlq

k

3.5

converge absolutely for eachqas follows. From 1.3, for any 1 < w < 1/v we have that

|ank|wnBsk, wheres <1/u < R/ρ < 1.Thus, for eachq, l, andn, we have

k0

|ank| 1 |t|lq

kB wnk0 s Rlq k B wn Rlq

Rlqs, 3.6

becauses/Rlq< R/ρR lq <1 and similarly for|z|

k0 |ank|z

t

k1 |t|1lq

k

wnR

k0 Rlq1 k

wnR

Rlq1

Rlq1 , 3.7

becausesρ/R lq1< R/Rlq1<1. Therefore, from3.6and3.7, identities3.4become

|λnz−σn,lz| ≤ 2 B πwn

Γ |ft| |tz|

q0 1 Rlq Rlq Rlqs

ρRlq Rlq1

dt

B

2πwn

Γ |ft| |tz|

q0

1

Rlqs

ρ Rlq1

dt.

3.8

It can be easily proved that the two series on the right side of the above expression converge by using the ratio test. Assuming thatftis bounded onΓ, w >1 implies that

lim

n→ ∞λnz−σn,lz 0, ∀|z|<ρ. 3.9

Next, we prove a similar result of convergence on a larger disk in the case of the Hermite interpolation. To simplify, let us denote

hrn1−1z;f, l−1

j0

Hrn1−1,jz;f 3.10

(7)

Theorem 3.2. LetfzAρand letΓbe any circle|t|Rwith 1< R < ρ. For anyρ > ρ, choose u >ρ/R and 0< v <1. LetAbe aGuGvmatrix and define

λnz

k0

ankhk, σn,lz

k0

ankHk,l 3.11

forn0,1, . . . .Then, for eachl,

lim

n→ ∞λnz−σn,lz 0,zDρ. 3.12

Proof. Using the integral representation given in2.11, for eachl1,2, . . ., we have

λnz−σn,lz

k0

ankhkHk,l

k0 ank 1

2πi

Γ ft

tzKt, zdt. 3.13

From2.12we obtain that

|λnz−σn,lz| ≤

k0 |ank| 1

2π

Γ |ft| |tz|

MRk1|z|k1|z|1r−1

Rrlk1 dt

M|z|1r1

2π

Γ |ft| |tz|

k0 |ank|

Rk1|z|k1 Rrlk1 dt.

3.14

The interchange of the integral and the summation is justified by showing that the series

k0

|ank| Rrl−1k1,

k0

|ank| |z| Rrl

k1

3.15

converge as follows. From1.3we have that for any 1 < w < 1/v, |ank|wn Bsk,where s <1/u < R/ρ < 1.

Thus for any fixed positive integerr≥2 and for eachlandn

k0

|ank| 1 Rrl−1

k1B

wnRrl

k0 s Rrl−1

k

B

wnRrl−1s, 3.16

becauses/Rrl−1<1/ρR rl−2<1, and similarly for|z|<ρ,

k0

|ank| |z| Rrl

kl

B|z| wnRrl

k0 s|z| Rrl

k B|z|

wn

1

Rrls|z|, 3.17

(8)

Therefore, using3.16and3.17in3.14we get that

|λnz−σn,lz| ≤MB|z|

1r1 2πwn

Γ |ft| |tz|

1

Rrl−1s |z| Rrls|z|

dt. 3.18

Assuming thatftis bounded onΓ, w >1 implies that

lim

n→ ∞λnz−σn,lz 0, for|z|<ρ. 3.19

Thus in the Lagrange case, theGu−Gvmatrix transformation of the sequence operators

produces new sequences such that the difference between the transformed sequences of polynomialsLnandSn,lconverges to zero in an arbitrarily large diskby choosingu >ρ/R .

Similarly, in the Hermite case also, the domain of convergence to zero is arbitrarily large underGuGv matrix transformation by choosingu > ρ/R . But as we see in the following

theorem, in the case of Hermite-Birkhoffinterpolation, the domain of convergence to zero of the difference of the transformed polynomials ofTheorem 2.3is arbitrarily large only if we chooseu >ρ2/R.

To simplify, let us denoteb0,m2n1z;fand lj0−1B2n1,j0,mz;fofTheorem 2.3bybn and Bn,l, respectively.

Theorem 3.3. LetfzAρand letΓbe any circle|t|Rwith 1< R < ρ. For anyρ > ρ , choose u >ρ2/Rand 0< v <1. LetAbe aG

uGvmatrix and define

λnz

k0

ankbk, σn,lz

k0

ankBk,l 3.20

forn0,1, . . . .Then, for eachl,

lim

n→ ∞λnz−σn,lz 0,zDρ. 3.21

Proof. Using the integral representation given in2.18, for eachl1,2, . . . ,we have

λnz−σn,lz

k0 ank 1

2πi

ΓftKk,lt, zdt. 3.22

From2.19we obtain that

|λnz−σn,lz| ≤

k0 |ank| 1

2π

Γ|ft||Kk,lt, z|dt

≤ 1

2π

Γ |ft| |tz|

k0

|ank||z|k1|t|k1|z|k1 |t|l1k1|tk1−1| dt

1

2π

Γ |ft| |z| −R

k0

|ank||z|k11|z|k1Rk1 Rl2k1 dt.

(9)

The interchange of the integral and the summations is justified by showing that the two series in3.23converge as follows. From1.3we have that for any 1 < w <1/v, |ank|wn Bsk,

wheres <1/u < R/ρ2<1. Since|t|Rand|z|<ρ, using the same method used in3.4we

write the first summation as

k0

|ank||z|k1 |t|l1k1

|t|k1|z|k1 Rk11

k0

Bsk|z|k1 wnRl1k1

1 |z|

R

k1

q0 1 Rk1 qB wnq0 ρ Rlq1 k0 Rlq1 kk0 Rlq1 k ρ R k1 B wnq0 ρ Rlq1 Rlq1 Rlq1sρ

ρ R

Rlq2 Rlq2sρ2

,

3.24

becausesρ/R lq1 <1/ρR lq <1 and2/Rlq2<1/Rlq1<1 for eachlandq. Now, for the

second sum in3.23we have

k0

|ank||z|k11|z|k1Rk1 Rl2k1 ≤

B wnk0 sk Rl2k1

ρ2k2ρk1 ρR k1Rk1

B wn

ρ2 Rl22

ρ Rl2

ρ Rl1

1

Rl1s

,

3.25

because for eachl,

2 Rl2 <

1

Rl1 <1,

Rl2 <

1

ρRl1 <1,

Rl1 <

1

ρRl <1,

s Rl1 <

1

ρ2Rl <1.

(10)

Using3.24and3.25, the expression3.23becomes

|λnz−σn,lz| ≤ B wn2π

Γ |ft| |tz|

q0

ρ Rlq1

ρ2 Rlq22

dt

B

2πwn

Γ |ft| |z| −R

ρ2 Rl22

ρ Rl2

ρ Rl1

1

Rl1s

dt. 3.27

It can be easily proved that the two series on the right side of the above expression converge by using the ratio test. Assuming thatftis bounded onΓ, w >1 implies that

lim

n→ ∞λnz−σn,lz 0 3.28

for each|z|.

In two of the above three theorems, for anyRandρsatisfying the stated conditions, we have chosenu >ρ/R > 1 and 0< v <1. In the case of the last theorem, we have chosen

u >ρ2/R >1 and 0< v <1. Now, to see if there exists such aGuGvmatrix, we give below

an obvious example. Define a matrixAby

ank v n

pk, p > u. 3.29

Then for anywsuch that 0< w <1/v, we have that

|ank|wn vw n

pk <

1

p

k

, 3.30

where 1/p <1/u. Hence byTheorem 1.1,Ais aGuGvmatrix.

Acknowledgment

The authors are thankful to the referee for his valuable remarks, in particular his suggestion on improving the presentation ofTheorem 3.3.

References

1 G. H. Fricke and J. A. Fridy, “Matrix summability of geometrically dominated series,” Canadian Journal of Mathematics, vol. 39, no. 3, pp. 568–582, 1987.

2 R. T. Jacob Jr., “Matrix transformations involving simple sequence spaces,” Pacific Journal of Mathematics, vol. 70, no. 1, pp. 179–187, 1977.

3 C. R. Selvaraj and S. Selvaraj, “Matrix transformations of classes of geometric sequences,” The Rocky Mountain Journal of Mathematics, vol. 23, no. 3, pp. 1099–1106, 1993.

(11)

5 P. C. Tonne, “Matrix transformations on the power-series convergent on the unit disc,” Journal of the London Mathematical Society, vol. 4, pp. 667–670, 1972.

6 A. S. Cavaretta Jr., A. Sharma, and R. S. Varga, “Interpolation in the roots of unity: an extension of a theorem of J. L. Walsh,” Resultate der Mathematik, vol. 3, no. 2, pp. 155–191, 1980.

7 J. L. Walsh, Interpolation and Approximation by Rational Functions in the Complex Domain, vol. 20 of American Mathematical Society Colloquium Publications, American Mathematical Society, Providence, RI, USA, 5th edition, 1969.

8 Y. R. Lou, “Extensions of a theorem of J. L. Walsh on the overconvergence,” Approximation Theory and Its Applications, vol. 2, no. 3, pp. 19–32, 1986.

9 T. E. Price Jr., “Extensions of a theorem of J. L. Walsh,” Journal of Approximation Theory, vol. 43, no. 2, pp. 140–150, 1985.

References

Related documents