A new nonlinear Uzawa algorithm for generalized saddle point problems (Q2493719): Difference between revisions

The authors discuss nonlinear Uzawa algorithms for solving generalized saddle point problems. Combining ideas from the papers by \textit{J. H. Bramble, J. E. Pasciak} and \textit{A. T. Vassilev} [SIAM J.~Numer.~Anal.~34, No. 3, 1072--1092 (1997; Zbl 0873.65031)] and \textit{Z.-H. Cao} [Appl.~Numer.~Math.~46, No. 2, 157--171 (2003; Zbl 1032.65029)] new nonlinear Uzawa methods are proposed. The new methods use nonlinear approximations of the inverse of the matrix \(A\) and the inverse of the Schur complement \(BA^{-1}B^T+C\) or of the inverse of \(A\) and the inverse of the approximate Schur complement \(B{\hat A}^{-1}B^T + C\), where \({\hat A}\) is defined, e.g., by an incomplete Cholesky factorization of \(A\). For both algorithms convergence theorems are given and proved. Additionally, the convergence factors of the methods of Bramble et al. and of Cao, and the new methods are analysed. Finally, these methods are applied to solve systems of equations arising from a finite element discretization with local and global stabilization of the incompressible steady state Stokes problem. This example shows that the new method has better convergence properties than the other ones.