On generation of ccECP UPF files

[pi246]

Hi,

I have a few questions regarding the generation of ccECP UPF files.

1. I notice in issue #74 , @aannabe mentioned that the semi-local ccECP pseudo-potentials are recast into the Kleinman-Bylander (KB) separable form for efficiency. Could someone provide more details on this transformation? In particular, is there a script available that performs this step?
2. The choice of the local channel seems to involve some subtleties. What tests or checks are typically used to determine whether the chosen local channel is appropriate?

Thanks in advance for your help!

[aannabe]

Hi @pi246, currently, there should be at least two ways to convert a semi-local (SL) file to a KB form. One is using ppconvert, which is public and included within qmcpack (more info in the documentation). The other is a modified version of OPIUM, which, to my knowledge, was not made public. The ccECPs were SL->KB transformed using this latter tool.

You are right that the choice of local channel (and also atomic configuration) affects the conversion quality. For ccECPs, these were chosen to reduce the SL->KB conversion errors and to avoid ghost states. For example, look at the outputs of these conversions, which are given in the *.rpt files in the library. Even though there is some loss of accuracy from the original SL quality presented in the ccECP paper, the results in solids, say the final DMC cohesive energies of Al/Si solids, have been consistent with experiments.

[pi246]

Hi @aannabe, thank you for your response! Could you please elaborate on how to determine whether ghost states are present? What specific criteria are typically used?

[aannabe]

We actually fully relied on OPIUM to do the ghost testing. All that was changed was that the ccECP SL potential was injected, and the rest was business-as-usual for OPIUM's workflow. So I would suggest looking at their documentation and source. I briefly looked at it: they mention that they follow the procedure of Gonze, Stumpf, and Scheffler.