Constructing nested bases approximations from the entries of non-local operators (Q444067)

The authors propose a method for constructing nested bases approximations to large-scale fully populated matrix discretizations of Fredholm type integral operators. The purpose is to reduce the storage complexity of the matrix to logarithmic-linear ones. Essentially, the obtained method is an improvement of the adaptive cross approximations (ACA) technique from a computational complexity point of view such that only few of the matrix entries for approximating the whole matrix are used. The efficiency of this nested approximation method in comparison with the standard ACA is tested in two numerical experiments: in the approximation of single layer potentials and in the approximation of the solution of Dirichlet problems. An improvement in the memory requirements for nested ACA is observed.