Matrix-dependent prolongations and restrictions in a blackbox multigrid solver (Q2638751): Difference between revisions

If one applies standard multigrid methods to solve linear systems resulting from the 9-point discretization of a linear second-order elliptic partial differential equation with discontinuous coefficients or dominating first-order terms, the rate of convergence often deteriorates. To improve the convergence behaviour of the multigrid methods in these cases the author develops a special multigrid code, in which matrix- dependent prolongations and restrictions are used. By hard numerical examples it is shown that this code is more robust and more efficient (for these hard problems) than a standard multigrid code based on the usual prolongation and restriction obtained by linear interpolation.