Ch. Guo, INCOMPLETE BLOCK FACTORIZATION PRECONDITIONING FOR LINEAR-SYSTEMS ARISING IN THE NUMERICAL-SOLUTION OF THE HELMHOLTZ-EQUATION, Applied numerical mathematics, 19(4), 1996, pp. 495-508
The application of the finite difference method to discretize the comp
lex Helmholtz equation on a bounded region in the plane produces a lin
ear system whose coefficient matrix is block tridiagonal and is some (
complex) perturbation of an M-matrix. The matrix is also complex symme
tric, and its real part is frequently indefinite. Conjugate gradient t
ype methods are available for this kind of linear systems, but the pro
blem of choosing a good preconditioner remains. We first establish two
existence results for incomplete block factorizations of matrices (of
special type). In the case of the complex Helmholtz equation, specifi
c incomplete block factorization exists for the resulting complex matr
ix and its real part if the mesh size is reasonably small. Numerical e
xperiments show that using these two incomplete block factorizations a
s preconditioners can give considerably better convergence results tha
n simply using a preconditioner that is good for the Laplacian also as
a preconditioner for the complex system. The latter idea has been use
d by many authors for the real case.