Residual reduction algorithms for nonsymmetric saddle point problems (Q611835)

Paralleling the work of Bramble-Pasciak-Vassilev and also based on earlier work by the first author, two versions of nonsymmetric Uzawa algorithms are investigated using spectral properties of the Schur complement. This leads to the derivation of optimal iteration parameters. The first method can be seen as a 2-parameter gradient iteration where for one parameter an interval is given, for the other the optimal value. The result applies to the finite and the infinite dimensional case. The second method is developed to improve the first one in the case of large Reynolds numbers and reduces, in every step of the iteration, the residual calculating both parameters optimally. The methods are applied to the steady-state incompressible Navier-Stokes equations led iteratively to Oseen equations. Numerical results are given for the driven cavity problem and show that even unstable finite element spaces combined with multilevel on uniform refinements can be used (though the reached accuracy then is much less than for stable spaces).