I'm trying to run Abinit in parallel, since in sequential mode it takes too much time. I have followed all steps in the abinit tutorials to run it in parallel, and they say to use OpenMPI, but I couldn't manage to get it working (it just runs on one of my 8 cores). I also followed this tutorial, but it stills run on 1 core only

Looking at the logs, I finally found something that could be the problem:

Your architecture is not able to handle 16, 8, 4 or 2-bytes FORTRAN file record markers!
You cannot use ABINIT and MPI/IO.
MPI_ERROR_STRING: Unknown error. Please file a bug report.
application called MPI_Abort(MPI_COMM_WORLD, 13) - process 0

Googling doesn't help; my architecture should be able to run in parallel because I can run in parallel with the Windows version of abinit.

What can I do to make it work?

  • Which distribution and version of linux are you using?
    – nozimica
    Oct 28, 2011 at 16:31

1 Answer 1


Several things can cause this:

  1. incompatible versions of mpi, between the library you linked with, the one you have loaded at execution time, and some recent versions of openmpi are a bit buggy.
  2. missing input file for wavefunctions or density. There is no clear error message, but this can get the mpi io into a fit and hang the job
  3. incompatible number of k-points/bands/fft wrt the number of processors you provide, which should be a divisor of nkpt in the default parallelization, and must be an exact divisor of np_kpt*np_band*np_fft in the general paral_kgb case.

Try running in sequential or with fewer processors to see where the problem might be, or link to an older version of mpi (1.2.6 is usually ok) You should also check the tmp_XXX_LOG_P001 002 003 files, which log the errors from each processor. They will give you additional hints.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.