Firefly and PC GAMESS-related discussion club

 
Learn how to ask questions correctly  
 
 

 

Alex Granovsky
gran@classic.chem.msu.su

>I wanted to set up a run using a 100% PBE exchange and 100% PBE
correlation functional; that is; an X+C pairing akin to the option
available in GAMESS US 2013 (PBE) and Gaussian 09 (PBEPBE).

>Question: Since Firefly is vastly superior to the mentioned programs,
how can I do this in Firefly? Thanks so much!

>Supp. Info: Although the question deals also on "how to generally
pair X+C functionals in Firefly", the problem or confusion of not
being able to compute 100% CPBE and 100% XPBE with Firefly originates
since (1) The DFTTYP=PBE96 or even CPBE in Firefly FF8.0.1.pdf is
written as using *also* the "PW91 local correlation" and (2) 21
benchmarks employing Gaussian and Firefly match *perfectly* when
calculated with B3LYP but not when calculated with PBEPBE and PBE96,
respectively.

>For point (1), one might speculate that since CPBE is actually a
"gradient correction", it makes sense to think as CPW91 as the left
hand side of Eq. 3 in Phys. Rev. Lett. 77, 3867, 1996. However, this
is not clear for the non-expert and in view of the many benchmarks
done I appreciate the guidance!


PBE correlation is a sum of local part and non-local part.
The local part of PBE correlation in its original form is
a PW91L one. The non-local part of PBE correlation depends on
the density gradients and on the local part of correlation.

With Firefly, PBE96 functional consists of XPBE96 and CPBE96.
Here XPBE96 is PBE 96 exchange functional and CPBE96 is PBE 96
correlation functional. The latter includes local part (PW91L)
and non-local part of PBE 96 correlation, so your speculation is
absolutely correct.

This detailed partitioning makes sense as one could use e.g.
VWN1 or VWN5 local functionals in conjunction with gradient
corrected part of PBE 96 correlation.

As to arbitrary X+C pairing, current versions of Firefly
do not support this. As to PBEPBE you are right you need to use
the predefined PBE96 functional name. The subtle difference in
computed energies of Firefly's PBE96 and Gaussian's PBEPBE is most
likely explained by the larger sensitivity of PBE correlation energy
to the parameters of a grid as compared with B3LYP. In addition,
Firefly uses parameters of PW91L functional that are computed to
the full decimal precision while other programs may use their
approximate values.

Hope this helps.

Kind regards,
Alex Granovsky