A relaxed splitting preconditioner for generalized saddle point problems (Q747191): Difference between revisions

The saddle-point problems in this paper originate from 2D linearized Navier-Stokes equations and the matrix has the form \[ \left[\begin{matrix} A_1 & 0 & B_1^T \\ 0 & A_2 & B_2^T \\ -B_1 & -B_2 & C \end{matrix}\right], \] \(B_1,B_2\) full row rank, \(A_1,A_2\) nonsymmetric positive definite, \(C\) symmetric positive definite. A preconditioner based on generalized relaxed splitting (GRS) is proposed of the form \[ \mathcal{P}_{GRS}= \frac{1}{\alpha}\left[\begin{matrix} A_1 & 0 & 0 \\ 0 & \alpha I & 0 \\ -B_1 & 0 & \alpha I \end{matrix}\right] \left[\begin{matrix} \alpha I & 0 & B_1^T \\ 0 & A_2 & B_2^T \\ 0 & -B_2 & \alpha I +C \end{matrix}\right],\, \alpha>0. \] It is inspired by a related preconditioner of \textit{M. Benzi} et al. [J. Comput. Phys. 230, No. 16, 6185--6202 (2011; Zbl 1419.76433)]. The eigenstructure and the degree of the minimal polynomial (hence the dimension of the Krylov subspace) is analyzed. The role of the relaxation parameter \(\alpha\) is illustrated with numerical examples.