My Community

I have a few questions regarding the usage of POTCAR files with and without kinetic-energy density information. For calculations using a regular GGA functional (excluding meta-GGA functionals), which version of the POTCAR is recommended?

In particular, I observed that for BCC Fe, the energies obtained from these two POTCAR versions differ by approximately 60 meV/atom. I also noticed that some researchers use POTCARs with kinetic-energy density information, while others do not. Since I am working on machine learning potential development, ensuring consistent energy values is crucial for my project.

Thank you for your time and assistance. I look forward to your guidance on this matter.

Hello!

Let me start with the most important point which you already mentioned... however, I would like to stress this once again: It is essential for machine learning force fields that the potential energy landscape is consistent throughout the entire data set. Hence, for a given atom type you should never use two or more different POTCAR files when creating training data.

Regarding the choice of POTCAR files I would ask you to have a closer look into our recently updated Wiki pages on this topic. There is a overview of available pseudopotentials here:

wiki/index.php/Available_pseudopotentials

and guidelines how to select between the POTCAR variants here:

wiki/index.php/Choosing_pseudopotentials

Since you mentioned using regular GGA functionals I would without further insight recommend to use the Fe (8 valence electrons, 3d74s1, ENMAX = 267.882eV) from the "Standard PBE potentials (potpaw.64)" list.

Best regards,
Andreas Singraber