No label defined (Q4358724)

The use of wavelets in numerical analysis is at first glance attractive for two basic reasons. First, the wavelet approach and its associated multiresolution analysis provide nice approximation spaces \(V_j\) and \(W_j\) suitable for the computation of an approximate solution to problems in which small-scale structures are localized in space and whose location may vary in time. Second, the multiresolution spaces lead to fast hierarchical algorithms with \(O(N\log N)\) operation count. These two properties alone are not sufficient to fully exploit the potential of wavelets. One must understand the action of operators and their inverses on wavelets to insure that the implicit systems of equations that are a part of many discrete numerical methods have the required nice properties for efficient inversion. These include diagonal dominance, low condition numbers (or the existence of good preconditioners), and banded structure. In this paper, we review recently developed algorithms devoted to the numerical solution of partial differential equations (PDEs) that are fully based on the wavelet approach. This is in contrast to various other methods, which although they make use of wavelets, do not intrinsically depend on them. Such methods include finite-element type methods which use scaling functions as basis functions, and methods which use wavelets as a preconditioner. Although these alternatives may be efficient, they are not considered here. After a short review of basic results related to the approximation of functions, we focus on the approximation of operators and on the design of efficient wavelet-based numerical algorithms for various model PDEs. Finally, examples based on Burgers equation and the heat equation will be presented. For the entire collection see [Zbl 0871.00029].