Posted by Evgeniy on November 26, 2008 at 14:42:33:
In Reply to: Re: On details of CASPT2 geometry optimization posted by Valera on November 24, 2008 at 19:55:15:
:
: : 1) For a parallel CASPT2 geometry optimization run to be efficient is the number of CPUS should equal 6N+1 where N is the number of atoms?
: No. The optimization is done in internal coordinates, not in cartesians. This number is printed at the beginning of numerical gradient code.
:
: : 2) Is it possible to speed up the job further giving for instance more processors or/and memory?
: each individual job runs in serial. More memory clearly will help - since rasscf and caspt2 modules use disk a lot, and more memory will increase caching of I/O.
Concerning 2) I meant that each individual job (plus and minus perturbations of geometry to differentiate numerically) could benefit from running in parallel, i.e. similar as for normal serial job which would benefit if it runs in parallel. Will it work out to calculate numerical gradients faster?