CASINO pseudopotential bases in DFT

[varelse]

I have problems with convergence of DFT and HF for my molecule (triangle of 33 C atoms with 15 H on edges) using the aug-cc-pVDZ-CDF basis from CASINO website. After some discussions on the QMC conference I suspect that the problem lies in the diffuse D function. Everything works when I simply remove this function for basis. The problem is that I suspect that the functions of such a basis are no longer optimized. How much can it affect my calculations? Is it "safe" to do that? Maybe I should remove all the diffuse functions (also S and P) and turn it into something like a cc-pVDZ basis? And what if I used some "normal" basis (e.g. from EMSL Exchange) instead of the optimized one?

[Mdeible]

Hi varelse,

I don't think that using a "normal" basis set from EMSL would give better results. These "normal" basis sets are designed for all electron calculations, so using them with the Trail-Needs pseudopotential will not produce results as good as with the basis sets listed on the CASINO website, which are designed to be used with the Trail-needs pseudopotentials.

What system are you trying to study? We may be able to follow up with better advice about the basis set if we understand the problem a little better.

Thanks,
Mike

[varelse]

Ok, I attach the picture of the molecule.
Here i post Gaussian input and output and the fchk file made from checkpoint file that I take the geometry from. I also note that the problem arises with total spin 1/2, for spin 3/2 which should be the ground state it converges (the idea is to compare the energies of different total spins, and do the same with the molecule charged with an extra electron, but this part maybe later, we expected that the charge-neutral wave functions will be less correlated than charged).

[Mdeible]

Hi Vareise,

For most systems, it is generally safe to delete the diffuse functions, and just use the cc-pvDZ basis set. We have run many tests that indicate that diffuse functions are not important to the final DMC energy. Note that, for weakly interacting systems, there may be some instances where the diffuse functions are important, but since you're not studying weak interactions, the cc-pvDZ basis set should suit your purpose nicely.

Mike

[varelse]

Ok, thank you, that's good news.

BTW for HF energy I get E= -190.33364 with a diffuse D function and E= -190.29420 without it - so there is a difference of about 0.01 Hartree - so you suspect that the difference will disappear in DMC?

[Yun_Pen_Wang]

The pseudopotentials for GAUSSIAN, CRYSTAL and GAMESS only have the pseudopotential part, where to find the basis set corresponding to one pseudo potential?

best wishes,
Yun-Peng Wang

[Mike Towler]

Dear Yun-Peng,

See my reply in another thread.

Best,
Mike