PC GAMESS/Firefly-related discussion club

Learn how to ask questions correctly

Alex Granovsky
gran@classic.chem.msu.su

this is impossible but there are still some hints:

For geometry optimization, one can run single point energy +
gradient job first. This will be the rough (usually upper) estimate
of the each geometry optimization step... and most likely there
will be several dozens of them. The same is true for saddle
point location, etc...

Some post-SCF computational methods (e.g., MP3 and above, CCSD,
etc... ) require more or less well defined number of floating
point operations to complete. Usually, this is the rather
complex power-series like expansion depending on the number
of orbitals of various kinds (frozen core, active, virtual).
E.g., for MP4-SDTQ the most demanding step is usually
const*n³*V³*(n+V) where n is the number
of active orbitals and V is the number of virtual ones.
As mentioned by Igor, this information can be used to estimate
CPU time using information on smaller jobs.

regards,
Alex

On Thu Oct 8 '09 11:41pm, Thomas wrote
--------------------------------------
>Hi,

>Is there any possibility to predict the approx. computing time of the used system before starting a calculation or to check the computing progress (e.g. in percent) during a calculation?

>Thanks a lot – also for the opportunity for using PCGamess,

>Thomas