Posted by Grigory Shamov on February 05, 2010 at 00:35:58:
Dear Molcas Developers,
I'm using the DFT in the MOLCAS code; with the RI/Cholesky it is nice and efficient tool. I've also found that it is quite modular, and dft_util allows for adding other functionals easily. I've added some simple exchange functionals (like, Gill96, OPTX, KT) to it, and it seems to be working. Also combinations like BPBE or B2PLYP are easy to create, which is nice.
Still, I have some questions regarding DFT in MOLCAS, could you please answer to them?
1. Is it possible, and how to use DFT non-selfconsistently? Like, to have a SCF in one DF, and then post-SCF computation with another DF? Can (should?) be gradients computed in such a setup as well?
2. What does the code expect as gamma_ab? I've succeed in adding exchange functionals that depend on gamma_aa, gamma_bb only ( eX=f(rho_sigma, gamma_sigma); then one takes sqrt of the latter parameter to obtain abs(grad_rho). Local correlation functionals (like, Wigner) also work; but somehow in my hands GGA correlation functionals do not work.
Am I correct in understanding that derivatives stored in dF_dRho(ipGaa, ipGab, ipGbb) are wrt to gamma_aa, gamma_ab, gamma_bb such as abs(grad_rho_total)=sqrt(gamma_aa + gamma_bb + 2*gamma_ab)? If not, could you please correct me, perhaps I confused things somehow. There is a comment in the code that "may be derivatives wrt to these should be taken instead" -- does that mean that they now aren't so?
3. If I want to contribute the files to dft_utils, (would it be of interest? They are quite simple functionals, admittedly) -- how does one create and upload a patch? Can I just mail the files/diffs to an administrator instead?
Thank you very much in advance,
--
Grigory Shamov,
University of Manitoba