PC GAMESS/Firefly-related discussion club

Learn how to ask questions correctly

Alex Granovsky
gran@classic.chem.msu.su

with Firefly, it is possible to perform really large CASSCF,
MCQDPT2, and XMCQDPT2 calculations. You can look at the
"Performance" section for data on our standard CASSCF benchmark.
The page for Core i7 has link to recent benchmarks of MCQDPT2
using resolvent fitting (aka table-driven) approach,
as well as using more standard technique as implemented in Firefly.

XMCQDPT2 require virtually the same amount of memory, disk, and CPU
time as if it were plain MCQDPT2 calculations.

You can find more information at:

http://classic.chem.msu.su/gran/gamess/mcscf.html

http://classic.chem.msu.su/gran/gamess/mcqdpt2.html

http://classic.chem.msu.su/gran/gamess/xmcqdpt2.html

I've been routinely performing XMCQDPT2 calculations with up to ca. 2000 basis functions.

The main limitation is not the number of basis functions actually,
but rather the size of active space. What we were capable to explore
with current code is CAS of up to 14 electrons on 16 orbitals,
however, one may need to use CSF selection procedure for such a
large active spaces.

Another problem is how to get good starting orbitals for CASSCF.
You may try NBOs, natural orbitals of MP2, or (preferred) state-
averaged NOs of CIS. The procedure how to compute the latter
using Firefly is a little bit tricky, so please post request
for additional information on this if you need help.

Please let me know if you need more information or some help with input files.

Regards,
Alex


On Tue Feb 3 '09 2:27pm, Pedro Silva wrote
------------------------------------------
>I have been performing TDDFT computations on a 1023 basis function system. However, the predicted wavelengths for the transition I am interested in are much shorter than the experimentally observed ones. Therefore, I am thinking of running higher level computations. The system is quite large, so I fear I cannot run CI or CASSCF. Would XMCQDPT be able to do the trick in reasonable CPU time?

>Pedro