Incomplete block factorization preconditioning for linear systems arising in the numerical solution of the Helmholtz equation (Q1917436)

The paper is concerned with the numerical solution of the Dirichlet boundary value problem in a plane domain for the inhomogeneous Helmholtz equation with a complex coefficient. A five-point finite difference discretization yields a linear system with a complex symmetric coefficient matrix, being block-tridiagonal and also being a complex perturbation of an \(M\)-matrix. It is to be solved by conjugate gradient type methods, here by the biconjugate gradient method, see e.g. \textit{D. A. H. Jacobs} [IMA J. Numer. Anal. 6, 447-452 (1986; Zbl 0614.65028)]. Hereby the problem of choosing an effective preconditioner remains. The suggestion to use preconditioners found to be good for the unperturbed block-tridiagonal matrix also for the perturbed matrices tends to be unsatisfactory when the perturbation is relatively large. For this case, two incomplete block factorizations for the complex system matrix and for its real part are established (in the case of small mesh size). Numerical test results (Dirichlet problem in the unit square) show that the use of these factorizations as preconditioners gives considerably better convergence results than the use of preconditioners used earlier.