Abstract

General sparse elimination is designed to take maximum advantage of the sparsity of an N x N matrix A. Only the nonzeros of A are stored, along with some extra integer overhead to identify the nonzero matrix elements. This extra integer storage may be avoided for the triangular factors generated by an LDU decomposition, generally without increasing the order of complexity. In addition to permanent storage for the nonzero elements of the factors, our procedure requires at most 5N temporary integer storage.