A note on the sparse representation of discrete integral operators (Q1899323)

The author compares the applicability of the collocation and a Galerkin method to solve rather general integral equations which include both regular and singular kernels in one and two dimensions. Presenting a number of Galerkin approximations for some integral operators using standard bases he demonstrates that a sparse matrix representation appears very often on the contrary to a common point of view that it is necessary to use specialized nontraditional bases to produce the sparsification. He shows how to give sparse matrix representations for a number of integral operators in one and two dimensions for smooth kernels as well as for some singular ones. On this ground, the author suggests that the traditional preference of the collocation over the Galerkin method has to be re-examined in the light of exploiting sparseness in a number of known algorithms. The problem of constructing quadrature rules preserving sparsening is also discussed. Some theoretical results on the growth rates of various matrix elements obtained by using specific bases for the Galerkin method are presented.