Iterative methods for solving \(Ax=b\), GMRES/FOM versus QMR/BiCG (Q675681): Difference between revisions

The convergence behaviour of the generalized minimal residual method (GMRES), the full orthogonal method (FOM), the quasi-minimal residual method (QMR), and the bi-conjugate gradient method (BiCG) is studied. At first, the author briefly describes these methods. Then, it is shown that any type of residual norm behaviour which can be obtained by using the BiCG method on the equation \(Ax = b\) can also be obtained by applying FOM to some other problem with the same eigenvalues. A similar result is obtained for the QMR and GMRES methods but involving the GMRES residual norms and the QMR quasi-residual norms. Two sets of numerical experiments were carried out. In the first one, the effect of increasing nonnormality on the convergence of the QMR and GMRES methods is studied. The second set of experiments is performed to track effects of the eigenvalue distribution on the convergence of the QMR method for normal and nonnormal matrices.