Rev. Anal. Num´er. Th´eor. Approx., vol. 33 (2004) no. 1, pp. 73–78 ictp.acad.ro/jnaat
COMBINED SHEPARD OPERATORS WITH CHEBYSHEV NODES
CRISTINA O. OS¸AN∗and RADU T. TRˆIMBIT¸ AS¸†
Abstract. In this paper we study combined Shepard-Lagrange univariate inter- polation operator
SL,mn,µ(Y;f, x) :=Sn,µL,m(f, x) =
n+1
P
k=0
|x−yn,k|−µ(Lmf)(x, yn,k)
n+1
P
k=0
|x−yn,k|−µ
,
where (yn,k) are the interpolation nodes and (Lmf)(x;yn,k) is the Lagrange interpolation polynomial with nodesyn,k,yn,k+1,yn,k+2,. . . ,yn,k+m, when the interpolation nodes (yn,k)k=1,nare the zeros of first kind Chebyshev polynomial completed with yn,0 = −1 and yn,n+1 = 1. We give a direct proof for error estimation and some numerical examples.
MSC 2000. 65D05, 41A05.
Keywords. Shepard interpolation, Chebyshev nodes.
1. INTRODUCTION
Let Y = yn,i∈[−1,1]; i= 0, n+ 1; n∈N be an infinite matrix where each row is a set of distinct points in [−1,1]. Forf ∈Cm([−1,1]) the Shepard- Lagrange operator is defined by
(1) Sn,µL,m(Y;f, x) :=Sn,µL,m(f, x) =
n+1
P
k=0
|x−yn,k|−µ(Lmf)(x, yn,k)
n+1
P
k=0
|x−yn,k|−µ
,
wherem∈N,m < nis prescribed.
The Shepard-Lagrange operator was treated in [1] and [6]. Its most impor- tant properties are: it preserves polynomials of degree m, i.e.,
Sn,µL,m(ej;x) =ej(x), j= 0, m, whereej(x) =xi. Also,
Sn,µL,m(f;yn,k) =f(yn,k), k= 1, n.
∗“Babe¸s-Bolyai” University, Faculty of Mathematics and Computer Science, str. M.
Kogˇalniceanu 1, 400084 Cluj-Napoca, Romania, e-mail: [email protected].
†“Babe¸s-Bolyai” University, Faculty of Mathematics and Computer Science, str. M.
Kog˘alniceanu 1, 400084 Cluj-Napoca, Romania, e-mail: [email protected].
In this paper we give an error estimation analogous to that given in [6], but the present proof is direct and exploits the properties of Cebyshev nodes and Lagrange interpolation polynomials with such nodes.
2. ERROR ESTIMATION
If f ∈ Cp([a, b]), p ∈ N, p < m, and (Lmf) is the m-th degree Lagrange interpolation polynomial with nodes x0, x1, . . . , xm ∈ [a, b], the following error estimation holds (see [3])
(2) kf −Lmfk∞≤Cp(b−a)p logmpmω f(p);m−pb−a, whereCp >0.
Let (yn,k)k=0,n+1 be the set of the roots of first kindn-th degree Chebyshev polynomial, completed with yn,0 = −1 and yn,n+1 = 1. For k = 1, n, yn,k = cosθn,k, where θn,k = (2k−1)π/(2n).
Remark 1. We have
|θn,k−θn,k+1|= πn.
The following theorem holds:
Theorem 1. If f ∈ Cp([−1,1]) and µ > p+ 1, we have (3) f(x)−Sn,µL,m(f;x)≤ (1−xnµ−12)p
Z 1 n−1
ω(f(p);t√ 1−x2) tµ−p dt.
Proof. Letyn,d be the closest point to x. We share the nod set (yn,k) into three classes: k= 0, d−m;k=d−m+ 1, d;k=d+ 1, n+ 1.
Equations (1) and (2) imply
f(x)−Sn,µL,m(f;x)≤
≤
n+1
X
k=0
x−yn,d x−yn,k
µ|f(x)−(Lmf)(x, yn,k)|
≤Cplogmpm
"d−m X
k=0
x−yn,d
x−yn,k
µ|x−yn,k|pω f(p);|x−ym−pn,k|
+
d
X
k=d−m+1
x−yn,d x−yn,k
µ|yn,k+m−yn,k|pω f(p);|yn,k+mm−p−yn,k|+
+
n+1
X
k=d+1
x−yn,d
x−yn,k
µ|x−yn,k+m|pω f(p);|x−ym−pn,k+m|
:=Cplogm
mp (S1+S2+S3).
We also have
|x−yn,d| ≤πnp1−x2, (4)
|x−yn,k| ≥πn|k−d|p1−x2, (5)
|x−yn,k+m| ≤ |x−yn,k|+πmn. (6)
We shall also use the following properties for modulus of continuity δ2 ≥δ1 ⇒ ω(f;δδ 2)
2 ≤ ω(f;δδ 1)
1 , (7)
ω(f;λδ)≤(1 +λ)ω(f;δ) if λ∈R+, (8)
δ1 ≤δ2 ⇒ω(f;δ1)≤ω(f;δ2), (9)
ω(f;δ1+δ2)≤ω(f;δ1) +ω(f;δ2).
(10)
Now, it follows the estimations for S1,S2, and S3. S1=
d−m
X
k=0
x−yn,d x−yn,k
µ|x−yn,k|pω f(p);|x−ym−pn,k|.
From (7) forδ1:= πn|k−d|√
1−x2 and δ2 :=|x−yn,k|we have
ω(f(p);|x−yn,k|)
|x−yn,k| ≤ 2n(π+1)
π|k−d|√ 1−x2. We obtain
S1≤c1
√1−x2 n
p
·
d−m
X
k=0 1
|k−d|µ−p ω f(p);|k−d|n p1−x2 with
c1 =2πp(π+ 1)1 +m−p1 , S2 =
d
X
k=d−m+1
x−yn,d
x−yn,k
µ|yn,k+m−yn,k|pω f(p);|yn,k+mm−p−yn,k|.
Since |yn,k+m−yn,k| ≤ mπn √
1−x2 we have
ω f(p);|yn,k+mm−p−yn,k|≤ 1 +m−p1 (mπ+ 1)ω f(p);
√1−x2 n
. We obtain
S2 ≤c2
√1−x2 n
p
ω f(p);
√1−x2 n
with
c2 = (mπ)p(mπ+ 1)p(m−1) 1 +m−p1 ,
S3 =
n+1
X
k=d+1
x−yn,d
x−yn,k
µ|x−yn,k+m|pω f(p);|x−ym−pn,k+m|.
ForS3 we remark that
|x−yn,k+m|p≤(|x−yn,k|+|yn,k+m−yn,k|)p
=
p
X
r=0 p r
|x−yn,k|p−r|yn,k+m−yn,k|r. Now, we have
S3 ≤ |x−yn,d|µ
n+1
X
k=d+1 p
X
r=0 p r
|yn,k−yn,k+m|r
|x−yn,k|µ−p+r ω f(p);|x−ym−pn,k+m|
≤πp
√1−x2 n
p p X
r=0 p r
mr n+1
P
k=d+1 1
|k−d|µ−p ω f(p);|x−ym−pn,k+m|
= 1+m−p1 πp(m+1)p
√1−x2 n
p n+1
X
k=d+1 1
|k−d|µ−p ω f(p);|x−yn,k+m|. We remark that for µ−p >1,Pn+1k=d+1|k−d|p−µis bounded byM.
From this observation and (10), (7) we obtain ω f(p);|x−yn,k+m|≤
≤M(mπ+ 1)ω f(p);
√1−x2 n
+ 2(π+ 1)ω f(p);|k−d|n p1−x2. In conclusion, we obtain forS3 the following inequality
S3≤
√1−x2 n
p
c3ω f(p);
√1−x2 n
+c4 n+1
X
k=d+1
ω f(p);|k−d|n √ 1−x2
|k−d|µ−p
with the constants
c3=πp(m+ 1)pM(mπ+ 1) 1 + m−p1 , c4= 2πp(m+ 1)p(π+ 1) 1 + m−p1 . Since
ω f(p);
√1−x2 n
≤ nµ−p−11
Z 1
n−1
ω f(p);t√
1−x2 tµ−p dt and
S1+S2+S3 ≤
≤
√1−x2 n
ph
C(c2, c3)·ω f(p);
√1−x2 n
+C(c1, c4)·ω f(p);|k−d|n p1−x2i
(3) follows.
3. EXAMPLES AND GRAPHS
Let us consider the functionf : [−1,1]→R, f(x) = sinπx.
Its graph appears in figure 1, together with the Shepard-Lagrange approxima- tion functions forµ∈ {2,4},n= 16 andm∈ {1,2}.
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
−1.5
−1
−0.5 0 0.5 1 1.5
(a) Graph off.
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
−1.5
−1
−0.5 0 0.5 1 1.5
(b) Shepard-Lagrange, n = 16, µ= 2,m= 1.
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
−1.5
−1
−0.5 0 0.5 1 1.5
(c) Shepard-Lagrange, n = 16, µ= 2,m= 2.
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
−1.5
−1
−0.5 0 0.5 1 1.5
(d) Shepard-Lagrange, n = 16, µ= 4,m= 1.
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
−1.5
−1
−0.5 0 0.5 1 1.5
(e) Shepard-Lagrange, n = 16, µ= 4,m= 2.
Fig. 1. Graph off and various Shepard-Lagrange interpolants.
REFERENCES
[1] Gh. Comanand R. Trˆımbit¸as¸,Combined Shepard univariate operators, East Journal on Approximations,7, no. 4, pp. 471–483, 2001.
[2] G. CriscuoloandG. Mastroianni,Estimates of the Shepard interpolatory procedure, Acta. Math. Hung.,61, nos. 1–2, pp. 79–91, 1993.
[3] M. Crouzeix and A. L. Mignot, Analyse num´erique des ´equations diff´erentielles, 2e´edition, Masson, Paris, 1989.
[4] B. Della Vecchia and G. Mastroianni, Pointwise estimates of rational operators based on general distribution of knots, Facta Universitatis (Niˇs), Ser. Math. Inform.,6, pp. 63–78, 1991.
[5] B. Della Vecchia and G. Mastroianni, Pointwise simultaneous approximation by rational operators, J. Approx. Theory,65, pp. 140–154, 1991.
[6] R. Trˆımbit¸as¸, Univariate Shepard-Lagrange interpolation, Kragujevac J. Math., 24, pp. 85–94, 2002.
[7] D. Shepard,A two dimensional interpolation function for irregularly spaced data, Proc.
23rd Nat. Conf. ACM, pp. 517–523, 1968.
Received by the editors: August 5, 2003.