Posted by Valera Veryazov on September 29, 2009 at 19:16:28:
In Reply to: Some questions on RICD (aCD) posted by Evgeniy on September 29, 2009 at 18:14:42:
A believe that a parallel calculation (any, not only Molcas) will give the same numbers
as a serial one - is a common, but a serious mistake.
Especially, if this number is coming from iterative scheme with many iterations (like NUMERICAL_GRADIENT).
I would started to worry about validity of an installation, if parallel and serial versions gives really different values for energies. A fact that reference numbers,
computed for a serial run can't be reproduced with 'required for serial' thresholds - is absolutely normal.
In addition to that, implementation of RICD and aCD in molcas is different for serial and parallel codes.