mat-disorder

Disordered Material

Goal

To generate clean, ordered atomic configurations from disordered starting structures (e.g., experimental structures with fractional occupancies). These ordered candidates can be used for ground-state property calculations, phase stability analysis, or as starting points for MLIP training.

Instructions

1. Identify Disordered Structures: Ensure your input structure (typically a CIF file) contains fractional occupancies or partial site occupancies.

2. Generate Ordered Candidates: Use the ranking and sampling strategy based on Ewald energy to pick configurations that satisfy stoichiometry while minimizing electrostatic repulsion.

   # Env: base-agent
   python .agents/skills/mat-disorder/scripts/run_ordering.py disordered.cif \
       --n_structures 50 --target_atoms 50 --output_dir ordered_results
   

Strategy: Ewald Energy Ranking

The script uses pymatgen's OrderDisorderedStructureTransformation with a fast Ewald-based solver (ALGO_FAST). It generates a large pool of candidates, ranks them by Ewald energy, and samples across the spectrum to ensure both low-energy (ground-state-like) and higher-energy (excited-state-like) configurations are captured.

Supercell Expansion

For structures with very few atoms per cell or complex stoichiometry, the script automatically searches for a supercell expansion that:

1. Is close to the --target_atoms (default: 50).
2. Maintains valid stoichiometry (total counts must be integers).
3. Is as cubic as possible to avoid long, thin cells.

• Limit: Avoid setting --target_atoms too high (>120) if you plan to follow up with DFT calculations.

Standalone Usage (Python API)

from .agents.skills.mat_disorder.scripts.order_disorder_sampler import OrderDisorderSampler
from ase.io import read

atoms = read("disordered.cif")
sampler = OrderDisorderSampler(
    atoms=atoms,
    n_structures=20,
    target_atoms=60,
    include_perturbation=1
)
ordered_structures = sampler.sample()