[gmx-users] segmentation fault - extending trajectory uwing gmx4.0.4 on 16 processors with PME

[Mark Abraham]

Ángel Piñeiro wrote:
> Dear all,
> I have some membrane protein systems running in our cluster using
> gmx4.0.4. The minimizations finish with reasonable potential energy
> values (of the order of 10^-5 with maximum forces of the order of
> 10^-3). Then I am trying to run several 1-ns-long equilibration MD
> trajectories with progressive relaxation of the protein positional
> restraints decreasing the force constant from 1000 kJ/mol/nm^2 to 0, all
> using PME on 16 processors. The PME load is about 25% in all cases. Some
> systems run with no problem but others abort with final lines in the
> mdrun.log file like these:
> MPI Application rank 4 killed before MPI_Finalize() with signal 11
> MPI Application rank 15 killed before MPI_Finalize() with signal 11

Those messages from your MPI system are not helpful in diagnosing why 
GROMACS is apparently aborting. Look at the stderr/stdout and the .log 
file to see what GROMACS thinks is going on.

> It is worth to mention that some of the simulations that present this
> segmentation fault are just extensions of trajectories that finished
> successfully, with the only change of reducing the force constant in 200
> KJ/mol/nm^2, for example from 800 to 600 or from 1000 to 800. Other
> simulations present this problem just during the first equilibration
> step with a force constant of 1000.
> 
> IMPORTANT: A copy of the same topol.tpr files run on my local computer
> (on one processor) with no problem, so I do believe that it is not a
> problem of the initial configuration.

That's possible, but not conclusive.

> Anyone knows the source of this problem?

Not without more diagnostics.

Mark