Fast FIRE Refinement¶
tricor-generated supercells relaxed with the built-in FIRE spring network. FIRE is orders of magnitude faster than MACE-MP0 refinement at lower accuracy, and is the practical option for cells of 100³ Å and larger (see cost and convergence). Each final structure is scored with a MACE-MP0 single point, so the FIRE and MACE pipelines can be compared directly.
The pipeline per (material, regime) is:
1. Voronoi tile (cell.generate(num_steps=0))
2. Orientation refine (cell.refine_initial_orientations)
3. Cleanup (bond_relax + enforce_hard_core)
4. FIRE relaxation (cell.shell_relax), or
thermostatted sampling (cell.thermal_relax) for liquid
Shell targets are MACE-calibrated when the optional mace-torch dependency is installed (shell.calibrate_to_mace()): per-pair bond stiffness, per-triplet angle stiffness, Morse anharmonicity, and the hard-core wall are measured from the MACE-MP0 potential on the reference crystal, which improves accuracy over the hand-tuned defaults. Without mace-torch the pipeline runs identically on the registry weights. Carbon’s composite sp²/sp³ target runs uncalibrated.