Preconditioning linear saddle point problems (Q4356928)

The authors study the construction of preconditioners for linear saddle point problems which are designed to inherit convergence rates from underlying continuous systems which are hence independent of discretization parameters. The approach is illustrated by a Stokes problem, a mixed formulation of a second-order elliptic equation and a scattered data approximation. The authors introduce an abstract class of linear saddle point problems which contains the continuous problems to be studied as well as their discretizations. Properties that render preconditioners useful are formulated on that level. Later it is shown that the differential systems satisfy the required conditions. The presented numerical examples (only for the mixed second-order problem on unit square, constant as well as variable coefficients) confirm the stated independence of the stepsize of discretization. For the entire collection see [Zbl 0863.00021].