Negative eigenvals. when parallel solving

[gluege]

Hello!

I have the following problem: Jobs that run nicely on a single machine produce massive negative eigenvalues and fail when submitted to a parallel environment involving multiple machines.

I am using a user material subroutine, that does not contain any MPI-related functions. I thought ABQ includes the corresponding code in the main prog. at compiling and linking time. I think the calc. wouldn't even start if this was missing.

Has someone experienced similar problems?

Greetings

 

[gluege]

sorry, i forgot to mention that the only difference between the nicely running job and the neg. eigenvalue party is that i (un)comment the mpi_host_list in the abaqus_v6.env file. so a compiler and abaqus-version conflict can be discarded.

 

[gluege]

OK, i got it.

If one does not state explicitely that the Domain-Decomposition-Solver should be used, the direct solver is invoked, leading to the behaviour described above.

The problems i had stem from using input files from an older ABQ-version + Umat. With input-files generated with the version i use, the preprocessor notes that parallelisation is not possible with the direct solver, and aborts.

greetings!