A non-stationary approximation scheme on scattered centers in \({\mathbb R}^{d}\) by radial basis functions. (Q1811617)

The author develops nonstationary approximation schemes which are local and provide spectral approximation accuracy, i.e. approximation orders do depend only on the smoothness of the approximants. They are applied to scattered data problems and thought as alternative to interpolation methods, which have the disadvantage leading to large ill-conditioned linear systems. The obtained approximation schemes are independent of any uniform grid issue. They can be applied e.g. to the Gaussian kernel and more generally to any basis function whose Fourier transform is positive around the origin. To some extend it is easier to implement than the scheme suggested by the author [ Constructive Approximation 17, 227--247 (2001; Zbl 0994.41019)].