Reducing complexity of algebraic multigrid by aggregation. (Q2829109)

At first the authors give a short overview on the basic principles of algebraic multigrid methods (AMG). Especially, the classical and the aggregation coarsening approaches are described. A combination of aggregation and classical AMG is proposed, i.e. a fixed number of levels aggregation is used and on the rest of the levels the classical AMG with a short-range interpolation. For the aggregation a new variant is proposed which is similar to the aggregation algorithm described in the paper by \textit{P. Vaněk} et al. [Computing 56, No. 3, 179--196 (1996; Zbl 0851.65087)]. The efficiency of the presented algorithm is shown by academic examples (isotropic and anisotropic Poisson problems in a cube) and by a problem arising in reservoir simulations. Hereby, the application of parallel computers is considered.