Hello Timo, Thanks for your detailed reply giving many possible reasons.
There are no constraints on the equation. It is a first order time dependent problem, so there are just initial conditions that are taken care of separately. The code is for 3D brick elements with 9 dofs per node. I just use dof_hanlder.distribute_dofs() function to number the nodes. Can this result in weird couplings ? The matrix is non-singular at all points in time. I am pretty sure about this. I have tested the code rigorously for different parameters. I am almost sure that this is not a special case. Also, the number of locally_owned_dofs on each processor was between 800 - 1100. Well, I might try with changing the settings in MUMPS. But I need to figure out what value should I set ICNTL(14) to. I am taking a look at the manual. The function you specified will help. Also, does deal.ii support superLU_dist ? I mean I know Petsc does but searching for superLU + deal.ii doesn't give any useful results to me. Thanks a lot again for your help. I really appreciate it. On Monday, February 13, 2017 at 12:59:02 PM UTC-5, Timo Heister wrote: > > Rajat, > > > The issue is that I get an error (mentioned below, Info(1) = -9, Info(2) > = > > 188672) when I run the code on greater than 2 nodes ( ~40 processors). > There > > is no error on running the code on 2 nodes or less. I looked in the > manual > > to see what the error Info(1) = -9 means. Also, the time at which the > error > > happens is random. > > I haven't seen this error before. I can see several potential reasons for > this: > 1. You are doing something that causes a large number of entries per > row (very large number of couplings, maybe some weird constraints?) > 2. We have a bug in deal.II that could be related to not communicating > things correctly to PETSc or within PETSc how it hands stuff to MUMPS. > This could be likely to happen if a single processor has no own DoFs. > You can experiment with a tiny test problem and many processors and > see if that triggers it. > 3. Your matrix is not invertible but you only trigger a problem with > MUMPS in certain cases. > 4. You actually need to change the settings for MUMPS. It looks like > this is easy to do: > > https://urldefense.proofpoint.com/v2/url?u=https-3A__www.mcs.anl.gov_petsc_petsc-2Dcurrent_docs_manualpages_Mat_MATSOLVERMUMPS.html&d=DwIBaQ&c=Ngd-ta5yRYsqeUsEDgxhcqsYYY1Xs5ogLxWPA_2Wlc4&r=4k7iKXbjGC8LfYxVJJXiaYVu6FRWmEjX38S7JmlS9Vw&m=EsoKswmf7wHHqLsosvyPZ-wn9Gx1iDt69A7APCDil-Y&s=9hrcHEAv7BOgDg1kjclulz4bnJ6WHbMG4QpsHbAVhDc&e= > > > -- > Timo Heister > http://www.math.clemson.edu/~heister/ > -- The deal.II project is located at http://www.dealii.org/ For mailing list/forum options, see https://groups.google.com/d/forum/dealii?hl=en --- You received this message because you are subscribed to the Google Groups "deal.II User Group" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
