Deepmd potentials in pyiron

[kimiaGF]

Has anyone tried to use a deepmd machine learned potential in pyiron? Are there any resources for this?

Thank you!

[jan-janssen]

I guess not yet - but as long as they use the standard pair style and pair coeff commands it should not be too difficult.

[kimiaGF]

it does use the standard commands, but the version of LAMMPS necessary for the potential to be recognized involves the installation of a plug in (see conda page for lammps-dp). Without this plug-in enabled version of LAMMPS, running the scripts results in an "unrecognized pair potential" error.

[jan-janssen]

Now I am confused. I thought the result of the discussion deepmodeling/deepmd-kit-recipes#1 was that you can alternatively install lammps and deepmd-kit both from conda-forge to get the pair_style deepmd

[pmrv]

You can find documentation on how to add custom executables for lammps and custom lammps potentials configurations. If you are interested in adding the fitting of deepmd potentials, we have some infrastructure for fitting potentials in pyiron_contrib. Over there, there are also reference implementations for MTP, ACE and EAM potentials. Hope that helps.