Fast approximation by periodic kernel-based lattice-point interpolation with application in uncertainty quantification (Q2068357)

Kernel and radial basis function approximation is nowadays an ubiquitous tool in approximation theory and its applications. This applies for low and for sometimes very high dimensions too. Especially when grids are used for cardinal interpolation, or when grids and lattices are used for the important method of quasi-interpolation, recursive methods offer themselves as evaluation algorithms for the approximants. One of the main technical problems the user is faced with is the computation of the (quasi-)interpolation coefficients, that is, normally a potentially very large linear system of equations has to be solved. To make matters worse, these linear systems of equations are often very badly conditioned, so pre-conditioning is almost always required (in contrast to quasi-interpolation where almost no high-dimensional linear systems of equations need to be solved). The purpose of this article is to offer a solution for this disadvantage of the kernel-based methods. Recursive methods are attractive in this context, too; they are of the essence because they more often than not provide fast and stable schemes for evaluation. This also is used in this article. The authors specifically take Fourier transform methods as a useful approch to this practical problem into focus, and of course this is highly relevant when periodic approximants are to be dealt with. The principal applications the authors have in mind are the numerical solutions of partial differential equations, and they provide theory and algorithms with detailed error estimates and extensive examples.