Problem with RASSCF concergence in parallel run

Posted by Marcin Andrzejak on July 12, 2009 at 11:07:00:

I am trying to run CASPT2 geometry optimization on a machine containing two 4-core processors (so in total there are 8 cores available simultaneously)

The RASSCF part converges easily when run on one core. Whenever I specify a parallel job (from 2 to 8 cores) the RASSCF calculations converge only if the starting orbitals are already almost converged. If the number of iterations is larger than six or seven, the energu begins to oscillate and finally the job fails to converge altogether.

Is this a known problem of parallel RASSCF? In that case, what can I possibly do to avoid/circumvent it?

Or is it more likely to be a problem of mpich environment/multcore processors/incorrect compilation?

Thanks for any help!
Marcin

Follow Ups:

Post a Followup

Name:
E-Mail:

Subject:

if B is 1s22s22p1, what is Li?

Passfield:

Comments:
: I am trying to run CASPT2 geometry optimization on a machine containing two 4-core processors (so in total there are 8 cores available simultaneously) : The RASSCF part converges easily when run on one core. Whenever I specify a parallel job (from 2 to 8 cores) the RASSCF calculations converge only if the starting orbitals are already almost converged. If the number of iterations is larger than six or seven, the energu begins to oscillate and finally the job fails to converge altogether. : Is this a known problem of parallel RASSCF? In that case, what can I possibly do to avoid/circumvent it? : Or is it more likely to be a problem of mpich environment/multcore processors/incorrect compilation? : Thanks for any help! : Marcin

[ Follow Ups ] [ Post Followup ] [ Molcas user's WWWBoard ]