A two-stage adaptive scheme based on RBF collocation for solving elliptic PDEs (Q2192500)

Radial basis functions of the multiquadric type or Gaussians or Matérnfunctions with parameters are well-known sets of kernel functions for univariate and multivariable approximation on scattered/meshless/gridded data. Their approximation qualities come from their ability to reproduce polynomials of high orders and thus generate highly accurate approximations in any dimension to gridded or scattered data. These approximants are usually formed by quasi-interpolation (a particularly useful algorithm to provide smoothing of data) or interpolation (so-called collocation in the language of partial differential equations). For the latter the so-called multiquadric kernel is especially useful because it guarantees regularity of the interpolation problem without any further conditions. The authors of this article use these methods to approximate the solution of elliptic PDEs and form new numerical schemes by non-symmetric collocation. They use adaptic methods too. The algorithms are tested in numerical examples that are given in great detail in the paper.