Efficient application of nonlinear stationary operators in adaptive wavelet methods—the isotropic case

Abstract

This paper deals with the efficient application of nonlinear operators in wavelet coordinates using a representation based on local polynomials. In the framework of adaptive wavelet methods for solving, e.g., PDEs or eigenvalue problems, one has to apply the operator to a vector on a target wavelet index set. The central task is to apply the operator as fast as possible in order to obtain an efficient overall scheme. This work presents a new approach of dealing with this problem. The basic ideas together with an implementation for a specific PDE on an L-shaped domain were presented firstly in [38]. Considering the approximation of a function based on wavelets consisting of piecewise polynomials, e.g., spline wavelets, one can represent each wavelet using local polynomials on cells of the underlying domain. Because of the multilevel structure of the wavelet spaces, the generated polynomial usually consists of many overlapping pieces living on different spatial levels. Since nonlinear operators, by definition, cannot generally be applied to a linear decomposition exactly, a locally unique representation is sought. The application of the operator to these polynomials now has a simple structure due to the locality of the polynomials and many operators can be applied exactly to the local polynomials. From these results, the values of the target wavelet index set can be reconstructed. It is shown that all these steps can be applied in optimal linear complexity. The purpose of the presented paper is to provide a self-consistent development of this operator application independent of the particular PDE, operator, underlying domain, types of wavelets, or space dimension, thereby extending and modifying the previous ideas from [38].