\(LU\)-decomposition and numerical structure for solving large sparse nonsymmetric linear systems (Q1609048)

The authors consider a large random sparse linear system and present two direct methods to compute the solution. The first method is an incomplete LU-decomposition method: its technique for reducing fill-ins based on powers of a Boolean matrix is cheap and effective and allows to exploit symmetry. The second method, called numerical structure approach, is based on Cramer's rule, which does not create fill-ins. As expected, it is not efficient, therefore it is not recommended. Moreover the paper presents eight examples which fill ten pages just with numbers. Actually they cannot be easily decoded.