A rational RBF interpolation with conditionally positive definite kernels (Q2240692)

When conditionally positive definite radial basis functions are used for approximation, there is a choice of approximation methods that are useful: routine pointwise interpolation from shifts of the kernels (in this paper, these would be non-equally spaced shifts), quasi-interpolation (a universally useful approach, especially when smoothing is required) or rational interpolation. The authors choose the latter method and emphasize the usefulness of meshless interpolation. The coefficients in the denominator expression are chosen so as to maximize certain eigenvalues of the interpolation matrices, and particular attention is paid to approximants with singularities. Moreover, the scalability of the polyharmonic radial basis function kernels are considered where a partition of unity is formed by the bases. Several numerical examples are presented in some detail to illustrate the results.