Multimesh and multiresolution analysis for mesh adaptive interpolation (Q5961728)

The authors introduce the term ``highly heterogeneous function'' to denote a function which is either discontinuous or is smooth despite significant small scale variation. They are interested in approximating such functions on intervals using piecewise continuous polynomials defined on meshes adapted to the function at hand. For the current study, the authors focus on one space dimension, although they declare their intention to extend the results to several space dimensions. Two alternative methods are presented for adapting a mesh to interpolation of discontinuous functions. One method is based on a nonlinear spring analogy with the size of the jump of the function playing a key role. The second is based on the difference between interpolants on two of a sequence of nested meshes. In this case, these differences can be related to the error of interpolation by breaking the interval into subintervals over which the function is either concave or convex. In each case, adaptive interpolation can yield quadratic convergence despite the presence of discontinuities. Numerical results are presented to illustrate the method, although the very severe example \(u(x)=\sqrt{|x-1/2|}\) does not seem to exhibit quadratic convergence in the figures.