Numerical linear algebra algorithms and software (Q1591199)

This paper gives an excellent survey of the modern methodology for solving linear equations. Dense algorithms are first discussed including LU factorization and block methods. The influence of computer architecture on performance is next examined, and software libraries are analyzed. Sparse methods are then discussed including the reduction of bandwidth by graph theory, and nested dissections. Among the iteration methods special attention is given to stationary processes, to the Krylov subspace methods, to preconditioners including incomplete factorization. Libraries and standards in sparse methods are described at the end of the paper.