set overlap determines how much each subdomain overlaps the other domains. Generally the larger the overlap the faster the convergence but then each subproblem is larger so the the end result may be slower code. You should try -pc_asm_overlap 2 to see if that helps convergence. Based on your determination that accurate subdomain solves are needed, it is unlikely using more overlap with a poorer subdomain solve is likely to work well.
Incompressible solvers generally have a particular structure that you must leverage in the linear solver or you get slow convergence. You should research how this can be done for your problem. Barry On Dec 12, 2011, at 6:51 PM, Sebastian Skatulla wrote: > Hello > > I am struggling to solve an incompressible solid mechanics problem using > a meshfree code based on Moving Least Square approximations. My > knowledge of solvers and preconditioners is quite limited. But it seems > that Additive Schwarz or Gram Schmidt preconditioners together with > GMRES is a good choice - yet still not good enough in my case. So, I was > experimenting with exact (LU) solves in each subdomain which does the > trick but the memory need is significant. Now, my question is what > purpose serves the function PCASMSetOverlap and could it be used to > improve this setup in any way? Thanks. > > Regards, > Sebastian > -- > Dr Sebastian Skatulla > Department of Civil Engineering > University of Cape Town > Cape Town > South Africa >
