advice on basis set/pseudo for complex molecule

[Cyrille]

Dear all

I am having difficulties to model some complex molecules and I need your advice.
I have started using a small basis set (FHI SZP) to avoid too lengthy calculation.
However I realized by comparing with Quantum espresso that it was probably not sufficient.
I have switched to a much more complete basis set SG15 (medium) for all the atoms (C,H,B,N,Fe).
The calculation are becoming much heavier and I wonder if I could use SG15 only for the transition metal and keep smaller basis set for C,H,B and N.

I also need to perform DFT+U with VdW (as discussed yesterday).

Let me also add that the python file attached to my message correspond to an isolated molecule but I want to perform calculation for organized layers with a super-cell containing 3 molecules and test many configurations! Therefore I should find the optimal solution in terms of precision/speed.

Thanks a lot for your help.

Cyrille

[Daniele Stradi]

Hi Cyrille,

Yes, you can mix different potentials/basis sets. So it is definetly possible to use SG15-M on Fe and SZP on C,H,B,H.

Maybe you can also try DZP on C,H,B,N,Fe. Compared to SG15-M, the DZP basis is less extended, and therefore will computationally more efficient than a full SG15-M basis on all atoms.

What is the difference w.r.t. QE? Do you have some reference we can compare against?

Regards,
Daniele.

[Cyrille]

Hi Daniele

I knew you would answer this question for chemist:-)
I am now trying SZP for H, C and B. DZP for N (directly connected to Fe) and SG15 (medium) for Fe.

The comparison with QE is not straightforward because this is a very delicate molecule. A so called spin crossover molecule with two spin state: Low spin (S=0) and High Spin (S=2). DFT+U is absolutely mandatory (other wise LS is drastically too low in energy) and VdW is also rather important.
Therefore there are many factors (and parameters) that make the comparison difficult.

However I realized that with QW in SZP I found that the equilibrium configuration was slightly distorted and asymmetric (a kind of Jahn-Teller distorsion) which is not the case with QE . In fact both configurations exist with QE (asymmetric and symmetric) but the symmetric one is the lowest in energy contrary to QW..

By the way, which density_mesh_cutoff do you suggest. I took 150.0*Hartree but maybe it is not necessary.?

Cheers

Cyrille

[Daniele Stradi]

Hi Cyrille,

Regarding the basis set you are using, sometimes SZP may be a little bit too small as a basis set for carbon-based systems (for example, the lattice constant of graphene is too large using SZP). The Jahn-Teller effect might be due to this.

However, these things are very basis-set dependent. I suggest that you try with your basis set, and if you still have problem you try to use:

(i) DZP on all atoms
(ii) DZP on C,N,H,B and SG15-M on Fe

Regarding the Mesh cutoff, 150*Hartree is probably over-converged. Normally we suggest 75*Hartree for the FHI basis and 100*Hartree for the SG15 basis. I have tested a few phonon calculations and the forces are normally converged with these cutoff. I suggest that you either use these cutoff (or maybe 100*Ha for FHI and 125*Ha for SG15 if you really want to stay on the safe side) and then eventually increase it further to double check the total energies once you have converged the geometries.

Regards,
Daniele.

[Cyrille]

Hi Daniele

OK thanks I will try DZP  (with DFT+U and VdW DFT-D2) first for all atoms and see if it cures my Jahn-Teller problem.
These spin crossover molecules are really complicated! Everything is inter-connected...

best
Cyrille