A new family of preconditioned iterative solvers for nonsymmetric linear systems (Q1917423)

The authors introduce a new family of iterative methods, the so-called EN-type methods, for solving unsymmetric linear equation systems \(Ax = b\). These methods include a method due to \textit{T. Eirola} and \textit{O. Nevanlinna} [Linear Algebra Appl. 121, 511-520 (1989; Zbl 0683.65018)], which is related to the well-known Broyden family, differing in that the approximation to the inverse Jacobian of \(A\) is improved by a rank-one update whilst simultaneously improving the current approximation \(x_k\) to the solution of the linear system. Although the methods are described as applied to dense systems, the main interest is in their application to sparse systems. The computational complexities for a variety of such methods are derived and compared with those of other popular methods such as CGS and BiCGS. Convergence of Broyden and EN-type methods is discussed and comments on preconditioning made. The paper concludes with numerical experiments on some iteration matrices resulting from partial differential equations as well as some from the Boeing-Harwell collection.