VASP was used for structure optimization and JDFTX was used for constant-potential calculation

[moyulyy]

Dear, Shankar,

I know that vasp and jdftx use different pseudopotentials, especially the implicit solvation model required in the constant-potential calculation of jdftx. Therefore, there are slight differences in the calculation of the same structure between the two kinds of software, and I want to clarify this difference.

I have compared several systems with vasp and jdftx, and they do show very small differences in geometry. However, how should this difference be assessed for the broader system?

Specifically, I first obtained a convergent CONTCAR by using VASP optimization calculation, and then carried out a single point calculation by applying a series of potentials using jdftx to obtain the energy and electronic structure characteristics under different voltages. How do you evaluate the error of this method?

Do you advocate such an approach? Or, do you know of any jobs that have used this strategy?

In addition, do you have any suggestions for the calculation of about 400 atomic systems?

best, lyy

[shankar1729]

There are really two questions here: (1) differences between geometry for calculations at the same condition due to different pseudopotentials, and (2) differences due to calculation condition eg. solvation and electrode potential.

I'd expect (1) to not be an issue as long as both pseudopotentials are good enough. However, you are likely more interested in (2). The single-point approach is decent in most cases because you are starting with a converged geometry with zero forces: this makes the first order perturbation zero with respect to changes like (1) or (2), and only leaves behind second order perturbations. In the cases that the effect of potential or solvation is weak, the single-point approach is therefore sufficient.

However, if the impact of potential or solvation fundamentally changes the geometry, e.g. changes the preferred adsorption site or orientation on a surface, it is not a perturbation that can be analyzed as above. In that case, you would want to do self-consistent geometries. I don't think there's a clean general prescription of when the changes would be large, and you would need intuition about the specific systems.

Overall, my recommendation would be do the single-point only if absolutely necessary; converge the geometry self-consistently whenever possible.