A new paradigm for parallel adaptive meshing algorithms (Q2706443)

In finite element computations on parallel computers, the distributions of subdomains to the processors is usually based on a split of the coarsest mesh. When the meshes are adaptively refined, a redistribution may be necessary in order to improve the load balancing. This requires a large amount of communication that is severe on workstation clusters. NEWLINENEWLINENEWLINEA new concept by the authors avoids this problem. Each processor does not only have the information about the subdomain associated to it. It gets the elliptic problem for the coarsest grid on the whole domain. The processor performs the adaptive refinement with the well-known procedure, it only sets the error indicators to zero for the elements outside its subdomain. NEWLINENEWLINENEWLINEThis concept is elaborated. The analysis uses old and new interior error estimates. Moreover, arguments from additive multigrid algorithms and the mortar method enter. The implementation with PLTMG is discussed.