Multiscale radial kernels with high-order generalized Strang-Fix conditions (Q2200787)

For multivariable approximation, quasi-interpolation especially when using radial basis functions is a very useful and highly accurate scheme. The basis of analysing the approximation power (e.g. by polynomial precision) is formed by the celebrated Strang and Fix theorem which shows which polynomials in any dimension we may recover by semi-continous convolution with shifts of kernels. Therefore, an essential tool when proving approximation estimates with the important method of quasi-interpolation with radial basis functions is forming kernels that obey these conditions. In this article, the authors find kernels made by scalings of radial basis functions so as to improve the degree by which the Strang and Fix conditions can be applied. These kernels are developed and it is shown that recursive methods can be employed to calculate them. The radial basis function kernels used can be applied for gridded or meshless data.