Preconditioners for Generalized Saddle-point Problems Preconditioners for Generalized Saddle-point Problems *

We examine block-diagonal preconditioners and efficient variants of indefinite preconditioners for block two-by-two generalized saddle-point problems. We consider the general, nonsymmetric, nonsingular case. In particular, the (1,2) block need not equal the transposed (2,1) block. Our preconditioners arise from computationally efficient splittings of the (1,1) block. We provide analyses for the eigenvalue distributions and other properties of the preconditioned matrices. We extend the results of [de Sturler and Liesen 2003] to matrices with non-zero (2,2) block and to allow for the use of inexact Schur complements. To illustrate our eigenvalue bounds, we apply our analysis to a model Navier-Stokes problem, computing the bounds, comparing them to actual eigenvalue perturbations and examining the convergence behavior.