Constraint Schur complement preconditioners for nonsymmetric saddle point problems (Q960295)

The paper deals with constraint preconditioners for general saddle point systems. The system matrix \(\mathbf M\in\mathbb R^{(n+m)\times (n+m)}\) is considered in the form of a \(2\times 2\) block matrix: \(\mathbf A\in\mathbb R^{n\times n}\); \(\mathbf B,\mathbf C\in\mathbb R^{m\times n}\); \(-\mathbf D\in\mathbb R^{m\times m}\) and similarly does the constraint preconditioner \(\mathbf Q\) with blocks \(\mathbf G,\mathbf B,\mathbf C,-\mathbf D\). The matrices \(\mathbf A,\mathbf G\) are supposed to be non-singular. The constraint preconditioner is derived from a splitting of the blocks of \(\mathbf M\). The author shows that the fast convergence of the preconditioned iterative methods with the constraint Schur complement preconditioners is mainly determined by the quality of the splitting and on the efficiency of the solution for the Schur complement systems which arise from the implementation of the constraint preconditioners. Results concerning the spectrum and the form of the eigenvectors of the preconditioned matrix \(\mathbf Q^{-1}\mathbf M\) are determined. An upper bound on the degree of its minimal polynomial is given which provides an upper bound on the maximum requested number of iterations of a Krylov subspace method. All principal steps and conclusions in the theoretical part are formulated in form of theorems and corollaries together with adequate proofs. In the second part two particular problems motivated by computational fluid dynamics are presented and the applicability of the previous theoretical results is demonstrated. The first problem is related to a stabilized mixed finite element discretization of the Navier-Stokes system, where \(\mathbf D\neq 0\) and \(\mathbf A\neq\mathbf A^T\) such that \(\mathbf M\) is non-symmetric. The second one originates from a spectral approximation of the incompressible Stokes problem, where \(\mathbf B\neq\mathbf C\) such that \(\mathbf M\) is non-symmetric. The investigation of particular problems represents an immediate application of eigenvalue bounding. Extensive numerical experiments illustrate that reasonable choices for splitting and effectively solving for the Schur complement systems yield good convergence. Although the eigenvalue bounds may not be sharp, they indicate a good eigenvalue clustering for reasonable choices for the splitting procedures mentioned above. The list of references is large enough to represent a sufficient overview of resources including an evaluation of their strengths and weaknesses.