Consistent sparse factorization (Q1205913)

A preconditioned conjugate gradient method (PCG) for solving \(Au=f\) is considered for which the preconditioner \(B\) is a sparse factorization of local accuracy \(O(h^ 2)\). It is therefore called consistent with typical five-point-difference or nine-point finite element discretizations on a rectangular grid. \(B\) captures the low-mode behaviour of \(A\), higher mode error components are damped efficiently by CG. As numerical example Poisson's equation with smooth and oscillatory true solutions are considered. For problems with smooth solutions the combination of the consistent sparse factorization (CSF) with Jacobi iteration only shows marginal improvement over pure CSF iteration, whereas compound iteration is to be preferred for problems with high mode solution components. CSF is not compared with other PCG methods.