MACE Preoptimisation for FHI-aims

[GabrielBram]

A workflow intended for pre-optimising structures and NEB transition state paths with the MACE family of forcefields.

Functionality invoked through the standard aims.optimise and aims.ts_search commands, with a user supplied MACE pre-optimiser to the ReactAims object, eg.,

mace_params = {'model': 'large',
               'dispersion': False,
               'default_dtype': 'float64',
               'device': 'cpu'}

reactor = ReactAims(params, basis_set, hpc)
reactor.MaceReact_Preoptimiser = ReactMACE(mace_params, force_field="mace_mp")

atoms = reactor.aims_optimise(atoms, 0.02, relax_unit_cell=False, restart=True, mace_preopt=True)[0]

[ikowalec]

This currently breaks the restarts on task-farming

[GabrielBram]

Dw, was just seeing how it behaved in the test infrastructure. By the looks of it, it's not recognising the restart as converged and attempts to run an FHI-aims simulation even though it can't. Why it's doing that is another matter.

...It also breaks itself.

Hmm. Will request the review when everything is in working order.

[GabrielBram]

Magical - it works on my machine but not on the remote CI tests. Something stupid must have happened.

[codecov]

Codecov Report

Attention: Patch coverage is 98.91304% with 1 lines in your changes are missing coverage. Please review.

Project coverage is 89.15%. Comparing base (4ed96da) to head (15a6303).

Files	Patch %	Lines
carmm/run/workflows/react.py	98.24%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #148      +/-   ##
==========================================
+ Coverage   88.23%   89.15%   +0.91%     
==========================================
  Files          80       81       +1     
  Lines        3265     3355      +90     
==========================================
+ Hits         2881     2991     +110     
+ Misses        384      364      -20     

Flag	Coverage Δ
unittests	89.15% <98.91%> (+0.91%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[GabrielBram]

Hello! This is somewhat ready - I do not like some of the solutions herein (e.g., the approach of re-attaching calculators after the MACE run), and some of the restart logic needs re-working. However, just to get the branch out there and to field suggestions, I've submitted a request for review.