MINIMAL DEGREE UNIVARIATE PIECEWISE POLYNOMIALS WITH PRESCRIBED SOBOLEV REGULARITY

Similar documents
Stability of Kernel Based Interpolation

Lecture 4: Fourier Transforms.

Vectors in Function Spaces

Scattered Data Interpolation with Polynomial Precision and Conditionally Positive Definite Functions

Kernel B Splines and Interpolation

MATH 590: Meshfree Methods

MATH 590: Meshfree Methods

EXAMPLES OF REFINABLE COMPONENTWISE POLYNOMIALS

MATH 590: Meshfree Methods

On stable inversion of the attenuated Radon transform with half data Jan Boman. We shall consider weighted Radon transforms of the form

Approximation by Conditionally Positive Definite Functions with Finitely Many Centers

Stability constants for kernel-based interpolation processes

Multivariate Interpolation with Increasingly Flat Radial Basis Functions of Finite Smoothness

APPROXIMATING BOCHNER INTEGRALS BY RIEMANN SUMS

We have to prove now that (3.38) defines an orthonormal wavelet. It belongs to W 0 by Lemma and (3.55) with j = 1. We can write any f W 1 as

MATH 590: Meshfree Methods

INVERSE AND SATURATION THEOREMS FOR RADIAL BASIS FUNCTION INTERPOLATION

MATH 590: Meshfree Methods

Proceedings of the 5th International Conference on Inverse Problems in Engineering: Theory and Practice, Cambridge, UK, 11-15th July 2005

We denote the space of distributions on Ω by D ( Ω) 2.

Boundary Integral Equations on the Sphere with Radial Basis Functions: Error Analysis

Math 660-Lecture 15: Finite element spaces (I)

A method for creating materials with a desired refraction coefficient

AN ELEMENTARY PROOF OF THE OPTIMAL RECOVERY OF THE THIN PLATE SPLINE RADIAL BASIS FUNCTION

Ring-like structures of frequency domains of wavelets

Continuity. MATH 161 Calculus I. J. Robert Buchanan. Fall Department of Mathematics

Orthogonality of hat functions in Sobolev spaces

Wavelet Bi-frames with Uniform Symmetry for Curve Multiresolution Processing

Solving the 3D Laplace Equation by Meshless Collocation via Harmonic Kernels

MATH 5640: Fourier Series

1 Brownian Local Time

INEQUALITIES FOR SUMS OF INDEPENDENT RANDOM VARIABLES IN LORENTZ SPACES

JASSON VINDAS AND RICARDO ESTRADA

Radial Basis Functions I

An inverse scattering problem for short-range systems in a time-periodic electric field. François Nicoleau

MATH 590: Meshfree Methods

Linear Independence. Linear Algebra MATH Linear Algebra LI or LD Chapter 1, Section 7 1 / 1

Exponential stability of operators and operator semigroups

Heisenberg's inequality for Fourier transform

CHAPTER 3 Further properties of splines and B-splines

RKHS, Mercer s theorem, Unbounded domains, Frames and Wavelets Class 22, 2004 Tomaso Poggio and Sayan Mukherjee

arxiv: v1 [math.co] 8 Dec 2013

Numerical Solutions to Partial Differential Equations

NECESSARY CONDITIONS FOR WEIGHTED POINTWISE HARDY INEQUALITIES

ISSN: [Engida* et al., 6(4): April, 2017] Impact Factor: 4.116

1 Lecture 6-7, Scribe: Willy Quach

Splines which are piecewise solutions of polyharmonic equation

Construction of Biorthogonal Wavelets from Pseudo-splines

AN ALGORITHM FOR COMPUTING FUNDAMENTAL SOLUTIONS OF DIFFERENCE OPERATORS

Applied and Computational Harmonic Analysis

be the set of complex valued 2π-periodic functions f on R such that

Time-Dependent Statistical Mechanics A1. The Fourier transform

Reducing subspaces. Rowan Killip 1 and Christian Remling 2 January 16, (to appear in J. Funct. Anal.)

Michael J. Johnson. Department of Mathematics and Computer Science Kuwait University P.O. Box: 5969 Safat Kuwait

Fourier Transform & Sobolev Spaces

Reproducing Kernels of Generalized Sobolev Spaces via a Green Function Approach with Differential Operators

c 2000 Society for Industrial and Applied Mathematics

1. Introduction. A radial basis function (RBF) interpolant of multivariate data (x k, y k ), k = 1, 2,..., n takes the form

HOMEOMORPHISMS OF BOUNDED VARIATION

Time Series Analysis. Solutions to problems in Chapter 7 IMM

Optimal Approximation by sk-splines on the Torus

Construction of Multivariate Compactly Supported Tight Wavelet Frames

ON A CERTAIN GENERALIZATION OF THE KRASNOSEL SKII THEOREM

UNIVERSITY OF WISCONSIN-MADISON CENTER FOR THE MATHEMATICAL SCIENCES. A characterization of the approximation order of multivariate spline spaces

Applied Numerical Analysis Quiz #2

MATH 590: Meshfree Methods

MATH 590: Meshfree Methods

SINGULAR MEASURES WITH ABSOLUTELY CONTINUOUS CONVOLUTION SQUARES ON LOCALLY COMPACT GROUPS

Viewed From Cubic Splines. further clarication and improvement. This can be described by applying the

SOME MEASURABILITY AND CONTINUITY PROPERTIES OF ARBITRARY REAL FUNCTIONS

f(x) cos dx L L f(x) sin L + b n sin a n cos

Construction of scaling partitions of unity

REMARKS ON THE VANISHING OBSTACLE LIMIT FOR A 3D VISCOUS INCOMPRESSIBLE FLUID

On orthogonal polynomials for certain non-definite linear functionals

Cubic Splines MATH 375. J. Robert Buchanan. Fall Department of Mathematics. J. Robert Buchanan Cubic Splines

NEW BOUNDS FOR TRUNCATION-TYPE ERRORS ON REGULAR SAMPLING EXPANSIONS

INTRODUCTION TO REAL ANALYSIS II MATH 4332 BLECHER NOTES

Application of modified Leja sequences to polynomial interpolation

Reproducing Kernel Hilbert Spaces

Measure Theory and Lebesgue Integration. Joshua H. Lifton

Overlapping additive Schwarz preconditioners for interpolation on the unit sphere by spherical radial basis functions

The Finite Spectrum of Sturm-Liouville Operator With δ-interactions

Consistency Estimates for gfd Methods and Selection of Sets of Influence

Numerische Mathematik

Chapter One. The Calderón-Zygmund Theory I: Ellipticity

Wiener Measure and Brownian Motion

Microlocal Methods in X-ray Tomography

Properties of Entire Functions

MGA Tutorial, September 08, 2004 Construction of Wavelets. Jan-Olov Strömberg

1.5 Approximate Identities

LOCAL TIMES OF RANKED CONTINUOUS SEMIMARTINGALES

NEWMAN S INEQUALITY FOR INCREASING EXPONENTIAL SUMS

1. Fourier Transform (Continuous time) A finite energy signal is a signal f(t) for which. f(t) 2 dt < Scalar product: f(t)g(t)dt

MATH 409 Advanced Calculus I Lecture 16: Mean value theorem. Taylor s formula.

Chapter 3 Continuous Functions

M412 Assignment 9 Solutions

Periodic solutions of weakly coupled superlinear systems

LECTURE 3 Functional spaces on manifolds

FORMAL GROUPS OF CERTAIN Q-CURVES OVER QUADRATIC FIELDS

An Overview of the Martingale Representation Theorem

Transcription:

MINIMAL DEGREE UNIVARIATE PIECEWISE POLYNOMIALS WITH PRESCRIBED SOBOLEV REGULARITY Amal Al-Rashdan & Michael J. Johnson* Department of Mathematics Kuwait University P.O. Box: 5969 Safat 136 Kuwait yohnson1963@hotmail.com* Abstract. For k {1, 2, 3,... }, we construct an even compactly supported piecewise polynomial ψ k whose Fourier transform satisfies A k (1 + ω 2 ) k b ψ k (ω) B k (1 + ω 2 ) k, ω R, for some constants B k A k >. The degree of ψ k is shown to be minimal, and is strictly less than that of Wendland s function φ 1,k 1 when k > 2. This shows that, for k > 2, Wendland s piecewise polynomial φ 1,k 1 is not of minimal degree if one places no restrictions on the number of pieces. 1. Introduction A function Φ L 1 (R d ) is said to have Sobolev regularity k > if its Fourier transform Φ(ω) := (2π) d/2 R d Φ(x)e ıx ω dx satisfies A(1 + ω 2 ) k Φ(ω) B(1 + ω 2 ) k, ω R d, for some constants B A >. Such functions are useful in radial basis function methods since the generated native space will equal the Sobolev space W2 k(rd ). The reader is referred to Schaback [3] for a description of the current state of the art in the construction of compactly supported functions Φ having prescribed Sobolev regularity. Wendland (see [4] and [5]) has constructed radial { functions Φ d,l (x) = φ d,l ( x ), where φ d,l is a piecewise polynomial of the form φ d,l (t) =, p being a polynomial. For d {1, 2, 3,...} p( t ), t 1, t > 1 and l {, 1, 2,...}, with the case d = 1, l = excluded, Φ d,l has Sobolev regularity k = l+(d+1)/2 and the degree of the piecewise polynomial φ d,l, namely d/2 +3l+1, is minimal with respect to this property. A natural question to ask is whether the degree of φ d,l would still be minimal if we considered functions of the form Φ(x) = φ( x ) where φ is a piecewise polynomial having bounded support. In this note, we answer this question in the univariate case d = 1. Specifically, we construct a compactly supported even piecewise Key words and phrases. positive definite function, compactly supported, B-spline. 1 Typeset by AMS-TEX

2 PIECEWISE POLYNOMIALS WITH PRESCRIBED SOBOLEV REGULARITY polynomial ψ k, with Sobolev regularity k (see Theorem 2.8), and we show that the degree of ψ k, namely 2k, is minimal (see Theorem 2.1). In comparison with Wendland s function Φ 1,k 1 (which has Sobolev regularity k when k > 1), we see that deg ψ k = deg φ 1,k 1, if k = 2, while deg ψ k = 2k < 3k 2 = deg φ 1,k 1 when k > 2. 2. Results Wendland s piecewise polynomial φ d,l can be identified as a constant multiple of the B-spline having l + 1 knots at the nodes 1 and 1 and d/2 + l + 1 knots at. This can be verified simply by observing that φ d,l and the above-mentioned B-spline have the same degree, d/2 + 3l + 1, and satisfy the same number of continuity conditions across each of the nodes 1,, 1, namely d/2 + 2l + 1 at 1, 1 and 2l + 1 at. It is well understood in the theory of B-splines that multiple knots are to be avoided if one wishes to keep the degree low, and with this in mind, we define ψ k as follows. For k = 1, 2, 3,..., let ψ k be the B-spline having knots k,..., 2, 1, ;, 1, 2,..., k (note that is the only double knot). For easy reference, we display ψ k (t) (normalized) for t [, k] and k = 1, 2, 3: { 8 24t ψ 1 (t) = (1 t) 2 2 + 24t 3 7t 4, t [, 1], ψ 2 (t) = (2 t) 4, t (1, 2] 198 27t 2 + 27t 4 18t 5 + 37t 6, t [, 1] ψ 3 (t) = 153 + 27t 945t 2 + 9t 3 45t 4 + 9t 5 8t 6, t (1, 2] (3 t) 6, t (2, 3] We begin by mentioning several salient facts about the B-spline ψ k which can be found in [1, pp. 18 131]. The piecewise polynomial ψ k is supported on [ k, k], positive on ( k, k), even and of degree 2k. Furthermore, it is 2k 1 times continuously differentiable on R\{} and 2k 2 times continuously differentiable on all of R. It follows from this that the 2k 1 order derivative, D 2k 1 ψ k, is a piecewise linear function which is supported on [ k, k] and is continuous except at the origin where it has a jump discontinuity. Consequently, the 2k order derivative has the form D 2k ψ k = 2πa δ + 2πaj (χ [j 1,j) + χ [ j,1 j) ), for some constants a, a 1, a 2,..., a k and where δ is the Dirac δ-distribution defined by δ (f) = f(). We can thus express the Fourier transform of D 2k ψ k as ( D 2k ψ k ) (ω) = a + 2 a j sin(jω) sin((j 1)ω) ω = a + 2(a j a j+1 ) sin(jω), ω with a k+1 :=, whence it follows that (2.1) ψk (ω) = (ıω) ( 2k D 2k ( 1) ψ k ) (ω) k = a ω + 2(a j a j+1 ) sin(jω).

A. AL-RASHDAN & M.J. JOHNSON 3 Lemma 2.2. Let β R. Then there exist unique scalars c 1, c 2,..., c k R such that (2.3) β + c j cos(jω) = O( ω 2k ) as ω. Proof. Define g(w) = β + k i=1 c i cos(iω). Since g C (R) is even, (2.3) holds if and only if D 2l g() = for l =, 1, 2,..., k 1. These conditions form the system of linear equations [c 1, c 2,..., c k ]A = [ β,,,..., ], where A is the k k matrix given by A(i, j) = ( 1) j 1 i 2j 2. Writing A(i, j) = ( i 2 ) j 1, we recognize A as a nonsingular Vandermonde matrix, and therefore, (2.3) holds if and only if [c 1, c 2,..., c k ] = [ β,,,..., ]A 1. Theorem 2.4. Let β, c 1, c 2,..., c k R be such that (2.3) holds. Then (2.5) β + c j cos(jω) = βα k (1 cos ω) k, ω R, where α k > is defined by 1 α k = 1 π (1 cos ω)k dω. Proof. Since cos j ω span{1, cosω, cos2ω,..., coskω} for j =, 1,..., k, it follows that there exist b j R such that (1 cos ω) k = b + k b j cos(jω). Note that < 1 = 1 α k π (1 cos ω) k dω = 1 π b dω + b j 1 π cos(jω) dω = b, and hence βα k (1 cos ω) k = β + k βα kb j cos(jω). Since βαk (1 cos ω) k = O( ω 2k ) as ω, it follows from the lemma that c j = βα k b j for j = 1, 2,..., k, and therefore (2.5) holds. Corollary 2.6. Let a be as in (2.1). Then ( 1) k a > and (2.7) ψk (ω) = ( 1)k a α k ω (1 cos t) k dt, ω. Proof. It follows from (2.1) that ψ k (ω) = ( 1)k f(ω), where f(ω) := a ω + k 2(a j a j+1 ) sin(jω). Since ψ k is supported on [ k, k] and positive on ( k, k), it follows that ψ k is continuous (in fact entire) and ψ k () >. Consequently, f(ω) = O( ω 2k+1 ) as ω. Since f is infinitely differentiable, it follows that f (ω) = a + k 2j(a j a j+1 ) cos(jω) = O( ω 2k ) as ω, and so by Theorem 2.4, f (ω) = a α k (1 cos ω) k. Since f() =, we can write f(ω) = ω f ω (t) dt = a α k (1 cos t)k dt, and hence obtain (2.7). That ( 1) k a > is now evident since < ψ k () = lim ψ ω + k (w). Remark. At this point, it is also easy to show that ψ k (ω) = ( 1)k a (ω + b j sin(jω)), ω, where the scalars {b j } are determined by the fact that ψ k is continuous at.

4 PIECEWISE POLYNOMIALS WITH PRESCRIBED SOBOLEV REGULARITY Theorem 2.8. For k {1, 2, 3,...}, ψ k has Sobolev regularity k; that is, there exist constants B k A k > such that (2.9) A k (1 + ω 2 ) k ψ k (ω) B k (1 + ω 2 ) k, ω R. Proof. As in the proof of Corollary 2.6, let us write ψ k (ω) = ( 1)k f(ω), where f(ω) := a ω+ k 2(a f(ω) j a j+1 ) sin(jω). Since lim ω ω = a, it follows that lim w w 2k ψ k (w) = ( 1) k a. Since ( 1) k a > (by Corollary 2.6), it follows that there exists N > such that (2.9) holds for ω N. That ψ k (ω) > for all ω > follows easily from Corollary 2.6, and since ψ k is continuous and ψ k () >, we see that (2.9) holds for ω N. We finally conclude that (2.9) holds for all ω R since ψ k is an even function. We now show that the degree of ψ k is minimal. Theorem 2.1. If ψ is an even, compactly supported, piecewise polynomial satisfying (2.9), then the degree of ψ is at least 2k. Proof. Let ψ be an even, compactly supported piecewise polynomial satisfying (2.9) and let the l-th derivative of ψ be the first discontinuous derivative of ψ (if ψ is itself discontinuous then l = ). Then D l+1 ψ can be written as (2.11) D l+1 ψ = g + n 2πcj δ xj, where g L 1 (R) and c j is the height (possibly ) of the jump discontinuity at x j. We can then express the Fourier transform of ψ as ψ(ω) = (ıω) l 1 ( D l+1 ψ ) (ω) = (ıω) l 1 (ĝ(ω) + Θ(ω)), where Θ(ω) = n c je ıxjω. Since Θ is bounded and ĝ(ω) has limit as ω (by the Riemann-Lebesgue lemma), it follows that ψ(ω) = O( ω l 1 ) as ω. With the left side of (2.9) in view, we conclude that l + 1 2k. Since Θ is a non-trivial almost periodic function (see [2, pp.9 14]), it follows that Θ(ω) o(1) as ω, and with the right side of (2.9) in view, we see that l + 1 2k. Therefore, l + 1 = 2k and we conclude that ψ k is 2k 2 times continuously differentiable. Since ψ k is compactly supported (ie. ψ k is not a polynomial), it follows that ψ k has degree at least 2k 1 (see [1, pp. 96-12]). In order to show that the degree of ψ k is at least 2k, let us assume to the contrary that the degree equals 2k 1. In this case the l = 2k 1 derivative of ψ k is piecewise constant and hence g = and ψ(ω) = ( 1) k ω 2k Θ(ω). Since ψ is continuous at, it follows that Θ() =. Since Θ is an almost periodic function, there exist values ω 1 < ω 2 < such that lim n ω n = and lim n Θ(ω n ) = ; but this contradicts the left side of (2.9). Therefore, ψ has degree at least 2k.

A. AL-RASHDAN & M.J. JOHNSON 5 References 1. C. de Boor, A practical guide to splines, Applied Mathematical Sciences 27, Springer-Verlag, New York, 1978. 2. C. Corduneanu, Almost periodic functions, Chelsea Publishing Company, New York, 1989. 3. R. Schaback, The missing Wendland functions, Adv. Comp. Math. 34 (211), 67 81. 4. H. Wendland, Piecewise polynomial, positive definite and compactly supported radial functions of minimal degree, Adv. Comp. Math. 4 (1995), 389 396. 5. H. Wendland, Scattered Data Approximation, Cambridge Monographs on Applied and Computational Mathematics, Cambridge University Press, Cambridge, 25.