A low-order block preconditioner for saddle point linear systems (Q725823)

The paper considers solution of large sparse saddle point problems. A new preconditioner for Krylov subspace methods (e.g. GMRES) is proposed, based on describing formally the system matrix as a 3 by 3 block structured one. Then, some spectral properties of the preconditioned matrix are derived. In particular, eigenvalue distribution is studied. Furthermore, an upper bound for the degree of the minimal polynomial is provided. Based on the numerical examples, the new preconditioner can outperform other preconditioners such as HSS, RHSS, etc.