High pressure crystals structure relaxations

[JonathanSchmidt1]

Pre-review checklist for PR author

PR authors must check the boxes below when creating the PR.

• I've confirmed the contribution guidelines.

---

Summary

Adds a new high-pressure relaxation benchmark that evaluates MLIP performance in predicting equilibrium structures under hydrostatic pressure.

• 3000 structures from the Alexandria database
• Relaxed at 7 pressure conditions:
  0, 25, 50, 75, 100, 125, 150 GPa
• Results compared against PBE reference calculations

---

Metrics evaluated (at each pressure)

• Volume per atom MAE (ų/atom)
• Enthalpy per atom MAE (eV/atom), calculated as:
  H = E + PV
• Convergence rate (%)

---

Methodology

• Structures relaxed using janus-core's GeomOpt
• ASE FixSymmetry constraint used to preserve crystallographic symmetry (matching DFT protocol)
• FrechetCellFilter with specified scalar pressure
• Force convergence threshold: 0.0002 eV/Å
• Maximum of 500 steps
• Up to 3 consecutive relaxation attempts if not converged
• Obvious outliers based on the predicted energies are marked as non-converged and are part of the convergence percentage metric but not the MAEs

---

Linked issue

Resolves #338

---

Progress

• Calculations
• Analysis
• Application
• App works but I am not sure about the good and bad values for the metrics yet.
• Documentation

---

Testing

Tested on the following models:

• mace-mpa-0
• orb-v3-consv-inf-mpa
• Regression test comparing to results produced with scripts from the original paper for mace-mpa-0

---

New decorators / callbacks

No new decorators or callbacks required.

Note the pull request right now includes a new model

• orb-v3-consv-inf-mpa
  because I wanted to compare to the original paper but this could be removed.
  I am looking for some feedback on the good/bad values for the metrics