Motion (physics)

Algorithm

Physics

Mathematical analysis

Interpolation (computer graphics)

Artificial intelligence

Finite difference

Approximations of π

Machine epsilon

Acoustics

Lattice (music)

Applied mathematics

Mathematics

Computer science

Artificial neural network

Polynomial

Radial basis function

Journal of Computational Physics

Radial basis function-generated finite difference (RBF-FD) approximations generalize classical grid-based finite differences (FD) from lattice-based to scattered node layouts. This greatly increases the geometric flexibility of the discretizations and makes it easier to carry out local refinement in critical areas. Many different types of radial functions have been considered in this RBF-FD context. In this study, we find that (i) polyharmonic splines (PHS) in conjunction with supplementary polynomials provide a very simple way to defeat stagnation (also known as saturation) error and (ii) give particularly good accuracy for the tasks of interpolation and derivative approximations without the hassle of determining a shape parameter. In follow-up studies, we will focus on how to best use these hybrid RBF polynomial bases for FD approximations in the contexts of solving elliptic and hyperbolic type PDEs.

On the role of polynomials in RBF-FD approximations: I. Interpolation and accuracy