Mace polar

[alinelena]

No description provided.

[ElliottKasoar]

https://github.com/ACEsuit/mace/releases/tag/v0.3.16 should makes things a little easier and resolve the thread above

[trdurrant]

I had a quick go building this branch, bumping the requirement to mace-torch==0.3.16 and hit two issues (which you most likely already know about):

• The graph_longrange dependency is needed
• The default model polar-1-s isn't included in mace v0.3.16 release (which has polar-1-m and polar-1-l) My mistake, it looks like this is included in the repo, but not part of the paper or documentation Model details for MACE-POLAR-1-S ACEsuit/mace-foundations#65

With these addressed it looks like it is working (although I don't have anything to test against)

[trdurrant]

The mace_polar charges do not seem to be directly available in the results, but I didn't have any problem fishing them out of the atoms object as given in the example here #673:

charges = atoms.calc.results['charges']

[ElliottKasoar]

Other than maybe being consistent with the Colab install commands, this looks great

[ElliottKasoar]

Couple of small fixes actually.

The return from our SinglePoint.run is a custom dictionary of results corresponding to what the user asked for (e.g. energy, forces, stresses, hessian), not a copy of the calc.results dict, so we need to access this to get the dipole etc.

[alinelena]

Couple of small fixes actually.

The return from our SinglePoint.run is a custom dictionary of results corresponding to what the user asked for (e.g. energy, forces, stresses, hessian), not a copy of the calc.results dict, so we need to access this to get the dipole etc.

one note here i think we shall copy all the calc.results info onto the atoms.info and similar in new ase4 this will have to be done explicity anyhow

[ElliottKasoar]

Looks good, and the tutorials run locally for me, so hopefully a release should work