Scattered-data interpolation on \(\mathbb R^n\): Error estimates for radial basis and band-limited functions. (English) Zbl 1081.41014

The approximation and interpolation of band-limited functions of the Paley-Wiener class \(B_{\sigma } \) are investigated. Here \(B_{\sigma } =\{f \in L^2({\mathbb R}^n): \text{supp} (\widehat {f}) \subseteq B(0,\sigma)\} \), where \(\widehat {f}\) is the Fourier transform of the function \(f\) and \(B(0,\sigma )\) is the ball with the radius \(\sigma \) and the centre at the origin. The functions \(f \in B_{\sigma } \) are continuous and \(\lim_{x \to \infty }f(x) = 0\), i.e. \(f\in C_0({\mathbb R}^n)\). Let us introduce the norm in the space \(C_0 \cap L^2 \equiv C_0 ({\mathbb R}^n) \cap L^2({\mathbb R}^n)\) by the equality \(\|f\|_{C_0 \cap L^2} = \max (\|f\|_{\infty } ,\|f\|_2 )\), where \(\|\cdot\|_{\infty } \) (or \(\|\cdot\|_{\infty } \)) is the \(sup\)-norm (or \(L^2\)-norm respectively). The separation radius \(q\) of the set \(X = \{x^{j}\}_{j =1}^{N} \subset {\mathbb R}^n\) is introduced by the equation \(q = {1\over2}\min _{j \neq k}\|x^j-x^k\|_2\), where \(\|\cdot\|\) is Euclidean norm in \({\mathbb R}^n\). The diameter \(\text{diam}(X)\) of the set \(X\) is defined by the equality \(\text{diam}(X)= \max_{j \neq k}\|x^{j} - x^k\|_2 \). The main results of the paper are the following two theorems.
Theorem 3.5. Let \(X = \{ x^{j} \}_{j = 1}^{N} \subset {\mathbb R}^n\) is a set of distinct points with the separation radius \(q\) and \(\text{diam}(X)\leq1\). If \(f \in C_0 \cap L^2\) and \(\sigma \geq \sigma _0 = {24\over q}\{ {\sqrt \pi \over3}\Gamma ( {n+2\over 2} )\}^{2\over n+1}\), then there exists a function \(f_{\sigma }\in B_{\sigma } \) such that \(f|_X = f_{\sigma }|_X \) and \(\|f - f_{\sigma }\|_{C_0 \cap L^2} \leq (5 + 2^{n + 3})\text{dist}_{C_0 \cap L^2} (f,B_{\sigma })\), where \(\text{ dist}_{C_0\cap L^2} (f,B_{\sigma }) = \inf _{g\in B_{\sigma }}\|f - g\|_{C_0 \cap L^2} \). Let \(W_2^k = W_2^k ({\mathbb R}^n)\) is the Sobolev space consisting of all functions \(f \in L^2({\mathbb R}^n)\) with distributional derivatives \(D^{\alpha }f\in C_0({\mathbb R}^n)\), \(|\alpha|=\alpha _1+\dots+\alpha _n\leq k\). Further let \(C_0^k = C_0^k ({\mathbb R}^n)\) be the space of all functions having continuous derivatives \(D^{\alpha }f\in C_0({\mathbb R}^n)\), \(|\alpha|\leq k\).
Theorem 3.8. Let \(k > 0\) be an integer and \(\sigma > 0\). If \(f \in C_0^k \cap W_2^k \), then there exists a constant \(C=C(k,n)>0\) such that \(\text{dist}_{C_0 \cap L^2} (f,B_{\sigma }) \leq C\sigma ^{ -k}\|f\|_{C_0^k \cap W_2^k } \).
Corollary 3.9. Let \(f \in C_0^k \cap W_2^k \) and \(f_\sigma \) be as in Theorem 3.5. Then \(\|f - f_{\sigma }\|_{C_0 \cap L^2} \leq C\sigma ^{- k}\|f\|_{C_0^k \cap W_2^k } \), where \(C = C(k,n)>0\).


41A30 Approximation by other special function classes
41A25 Rate of convergence, degree of approximation
41A05 Interpolation in approximation theory
41A63 Multidimensional problems
42B35 Function spaces arising in harmonic analysis
Full Text: DOI