Formation energy calculator

[lbluque]

Implement automatic formation energy predictions.

1. Element references are pulled from hugging face and added to the predict unit (same as the isolated atom refereces)
2. Formation energy calculations are enabled by using a context manager that decorates the calculate method. This way I did not touch any of the calculator code and its usage is explicit and descriptive to users.
3. OMat24 formation energies will default to apply appropriate MP2020 style corrections. This requires fairchem.data.omat to be installed.
4. Adapted the inorganic bulk relaxation tutorial to show how to predict formation energies.

Usage:

from ase.build import bulk
from fairchem.core.calculate.ase_calculator import FAIRChemCalculator, set_predict_formation_energy

atoms = bulk('MgO', crystalstructure='rocksalt', a=4.21, cubic=True)
atoms.calc = FAIRChemCalculator.from_model_checkpoint("uma-s-1p1", task_name="omat")

with set_predict_formation_energy(atoms.calc, apply_corrections=True):
    corrected_formation_energy = atoms.get_potential_energy()

total_energy = atoms.get_potential_energy()

Tests need #1603 to pass

[Andrew-S-Rosen]

@lbluque: Thanks for putting this together! Do you have an ETA for when this PR might be merged? We are hoping to submit a manuscript next month where we think the results of this PR could significantly improve our results and would be interested in trying it out before we send off the paper. I was meeting with @zulissi today and he suggested I reach out.

[lbluque]

Hi @Andrew-S-Rosen 👋 This is mainly just pending getting the DFT calculations ready and uploaded to HF. I'll see if I can push on this to have it ready by next week!

[Andrew-S-Rosen]

@lbluque: That would be great, thanks! We'll be sure to include the results in our paper then and be the first to use it!

Tagging @aryannsaha. This is what I will have you try once it's merged.

[zulissimeta]

Relatively minor comments here. The major question in my mind is whether using set_predict_formation_energy to modify a calculator is the right approach here.

[zulissimeta]

Can we add a numerical comparison to the MP value here? Just mention what MP says and include a link to the MP structure.

[lbluque]

good idea!

[codecov-commenter]

Codecov Report

❌ Patch coverage is 26.53061% with 36 lines in your changes missing coverage. Please review.
✅ Project coverage is 65.86%. Comparing base (d8d88a7) to head (460acdc).
⚠️ Report is 3 commits behind head on main.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
src/fairchem/core/calculate/ase_calculator.py	10.25%	35 Missing ⚠️
src/fairchem/core/calculate/pretrained_mlip.py	88.88%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1538      +/-   ##
==========================================
- Coverage   66.49%   65.86%   -0.64%     
==========================================
  Files         131      131              
  Lines       10781    10607     -174     
==========================================
- Hits         7169     6986     -183     
- Misses       3612     3621       +9     

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