Sparse approximation of singularity functions (Q1772221)

The author analyzes the approximation of singular functions built from powers of distance functions to corners, edges, etc., in anisotropically refined sparse tensor product spaces. It is shown that anisotropic sparse tensor product spaces of piecewise polynomials of degree \(p\) allow approximations of corner and edge singularities in dimension 3 at a rate of \({\mathcal O}(N^{-p} (\log_2 N)^s)\) with \(s= 2p+ 3/2\) in the \(H^1\) norm and where \(N\) denotes the number of degrees of freedom in the sparse anisotropic tensor product. This estimate shows that the reduction of the convergence rate due to the higher dimension and a low Sobolev regularity can be eliminated up to a logarithmic factor. The estimation applies to any dimension \(d\) and uses sparse tensor product wavelet bases with local refinement. The result yields upper bounds on nonlinear and adaptive approximation schemes in anisotropic wavelet bases for elliptic singularities.