Does "pair_style pace" support thermal conductivity calculations?

Does anyone know if the ACE potentials implemented in pair_style pace (package ML-PACE) support thermal conductivity calculations?

My understanding is that for Green-Kubo, if they support a site energy and the site stress defined as the derivative of this site energy w.r.t. deformation gradient, then LAMMPS will be able to use those ingredients to compute the heat flux. Can someone confirm that this is in fact necessary and sufficient?

I guess there’s always Muller-Plathe as an option, as long as a site energy is defined, right?

For an authoritative answer it would be best to ask the ACE developers directly.

Thanks Axel. I’ll contact them, but it’d be nice to be able to tell them what they actually need to ensure for it to work.

Do you know who can explain the actual requirements for a potential so it can support Green-Kubo heat flux? Is what I said about the site energy and site stress correct, or is there some other constraint? I guess that’s a question for whoever coded up the heat flux, and I assumed that was the core LAMMPS developers, but don’t know for sure.

Looks like the Boone paper cited in the lammps docs discusses that. Does anyone else want to weigh in on whether that’s the definitive word on computing the heat flux from the local stress, or if there’s some other reference I should be looking at?

You can usually find the names of who coded a feature in the source code.

Thanks - the other references for the centroid expression (in particular the Surblys et al. PRE) are a good starting point for what I think I need. I’ll follow up here with my conclusions or any further questions, once I’ve tried to do some derivations that might be helpful.