MLIP-MC

ASE framework for Monte Carlo simulations with universal Machine-Learned Interatomic Potentials (MLIP).

Overview

MLIP-MC is a Python package for performing Monte Carlo simulations of gas adsorption in porous materials using machine-learned interatomic potentials. The package integrates seamlessly with the ASE (Atomic Simulation Environment) framework and supports MLIP models from FAIRChem, MACE-Torch and Orbital backends.

Installation

Backend Selection

MLIP-MC supports three MLIP backends. You must install one of them:

• FAIRChem: For models trained with FAIRChem (e.g., OC20, OC22 models)
• MACE-Torch: For MACE models (e.g., MACE-MP models)
• Orbital: For Orbital models (e.g. orb_v3_conservative_inf_omat)

Install from PyPI (Recommended)

Install MLIP-MC with your preferred backend directly from PyPI:

With FAIRChem backend:

pip install "mlip-mc[fairchem]"

With MACE-Torch backend:

pip install "mlip-mc[mace-torch]"

With Orbital backend:

pip install "mlip-mc[orb-models]"

ROCm users: Install the ROCm version of PyTorch first before installing MLIP-MC:

pip install torch==2.8.0 torchvision==0.23.0 torchaudio==2.8.0 --index-url https://download.pytorch.org/whl/rocm6.4

Install from Source

Clone the repository and install locally:

git clone https://github.com/jackevansadl/MLIP-MC.git
cd MLIP-MC
pip install ".[fairchem]"   # or ".[mace-torch]" or ".[orb-models]"

Development mode (includes test tooling):

pip install -e ".[BACKEND_OF_YOUR_CHOICE,dev]"

Usage

Command-Line Interface

GCMC Isotherm Simulation

# Multiple pressure points (auto-distributed across GPUs)
mlip_mc \\
    --mode gcmc \\
    --adsorbent framework.xyz \\
    --adsorbate-molecule CO2 \\
    --temperature 298.0 \\
    --pressures 0.1,0.5,1.0,2.0,5.0,10.0,20.0 \\
    --n-equil 10000 \\
    --n-prod 20000 \\
    --model models/model.pt \\
    --output-dir results \\
    --checkpoint-interval 10000 \\
    --write-trajectory \\
    --trajectory-interval 100

Widom Insertion

# Basic Widom insertion calculation
mlip_mc \\
    --mode widom \\
    --adsorbent framework.xyz \\
    --adsorbate-molecule CO2 \\
    --temperature 298.0 \\
    --n-trials 10000 \\
    --model models/model.pt \\
    --output-dir widom_results

Command-Line Arguments:

• --mode: Simulation mode: gcmc (Grand Canonical Monte Carlo) or widom (Widom insertion) (default: gcmc)
• --adsorbent: Path to adsorbent structure file (.xyz, .cif, etc.) (required)
• --adsorbate-path: Path to adsorbate structure file (optional). The chemical formula will be automatically extracted to match with the fugacity table.
• --adsorbate-molecule: Molecule name (e.g., CO2, CH4) if not using file. This name will be used to match with the fugacity table.
• --temperature: Temperature in Kelvin (required)
• --pressures: Comma-separated pressures in bar, or single number (required for GCMC mode)
• --n-equil: Number of equilibration steps for GCMC (default: 10000)
• --n-prod: Number of production steps for GCMC (default: 20000)
• --n-trials: Number of Widom insertion trials (default: 10000)
• --checkpoint-interval: Interval for saving history checkpoints in GCMC (default: 10000)
• --model: Path to MLIP model file (required). Can be a local path or a Hugging Face URI (e.g., hf://your-org/your-repo or hf://your-org/your-repo:model.pt). When using the hf:// scheme, files are automatically downloaded and cached. The model format should match your installed backend (FAIRChem .pt files or MACE .model files).
• --output-dir: Output directory (default: results)
• --write-trajectory: Specifies to write out simulation trajectory every --trajectory-interval steps (default: False)
• --trajectory-interval: Interval for saving structures into trajectory file (default: 100)
• --hf-token: Hugging Face access token for downloading private models or bypassing interactive login (optional)
• --gpu-id: GPU device ID to use (for Widom mode, default: 0, use -1 for CPU)
• overwrite_checkpoints: Specifies to overwrite checkpoint files every --checkpoint-interval steps (default: False)

Model caching: When using the hf:// scheme, downloads are cached under ~/.cache/mlip-mc/<repo>/<filename> (or a custom directory set via the MLIP_MC_CACHE environment variable). Subsequent runs reuse the cached file even when launched from different working directories.

Note: The adsorbate name for EOS (fugacity) calculation is automatically determined:

• If --adsorbate-molecule is provided, that name is used to match with the fugacity table
• If --adsorbate-path is provided, the chemical formula is extracted from the structure file using ASE
• If the name/formula doesn't match any entry in the fugacity table, the simulation falls back to ideal gas approximation

Python Interface

You can also use the package programmatically for more control and integration into your workflows:

GCMC Isotherm

from mlip_mc import run_gcmc

# Run GCMC simulation
results = run_gcmc(
    adsorbent_path="framework.xyz",
    adsorbate_molecule="CO2",
    temperature=298.0,
    pressure_points=[0.1, 1.0, 5.0],
    n_equilibration_steps=10000,
    n_production_steps=20000,
    model_path="models/model.pt",  # or use hf://your-org/your-repo
    output_dir="results",
    write_trajectory=True,
    trajectory_interval=100,
    checkpoint_interval=10000
)

# Access results
print(f"Pressures: {results['pressures']}")
print(f"Uptakes: {results['uptakes']}")
print(f"Temperature: {results['temperature']} K")

Widom Insertion

from mlip_mc import run_widom

# Run Widom insertion calculation
results = run_widom(
    adsorbent_path="framework.xyz",
    adsorbate_molecule="CO2",
    temperature=298.0,
    n_trials=10000,
    model_path="models/model.pt",  # or use hf://your-org/your-repo
    output_dir="widom_results"
)

Output Files

Simulations generate output files in the specified output directory (default: results/):

• GCMC Isotherm (using run_gcmc()):

  • isotherm_data.json: Complete isotherm data (pressures, uptakes, energies, etc.)
  • log_{pressure}bar.bin: Binary log file containing all iteration data with trajectory (step, uptake, interaction_energy, total_energy)
  • traj_{pressure}bar.xyz: Simulation trajectory
  • restart/restart_{pressure}bar.xyz and .json: Restart information (updated every step for crash recovery)
  • checkpoints_{pressure}bar/checkpoint_{step}/: History checkpoints saved at intervals specified by --save-interval
    • traj.xyz: Snapshot trajectory
    • results.json: Snapshot results (n_iter, uptake, interaction_energy, total_energy)

• GCMC (direct class usage):

  • log_{pressure}bar.bin: Binary log file with all iteration data
  • restart/restart_{pressure}bar.xyz and .json: Restart information
  • checkpoints_{pressure}bar/checkpoint_{step}/: History checkpoints
  • traj_{pressure}bar.xyz: simulation trajectory

• Widom Insertion:

  • widom_results.json: Adsorption energies and calculated properties (Henry's constant, weighted average energy, etc.)
  • log_widom.bin: Binary log file containing all valid insertions with trajectory data (trial number, adsorption energy, total energy, atomic structure) - one record per valid insertion
  • widom_trajectory.xyz: Last valid insertion structure saved at the end of simulation
  • restart/restart_widom.xyz and .json: Restart information (updated every step for crash recovery)

Isotherm Data Format

The isotherm_data.json file contains:

{
    "temperature": 298.0,
    "pressures": [0.1, 0.5, 1.0, ...],
    "uptakes": [0.5, 1.2, 2.1, ...],
    "uptake_stds": [0.1, 0.2, 0.3, ...],
    "adsorption_energies": [-0.15, -0.18, -0.20, ...],
    "unit_cell_volume_A3": 1234.5,
    "unit_cell_volume_cm3": 1.234e-21,
    "adsorbent_file": "tests/zif8.xyz",
    "n_equilibration_steps": 10000,
    "n_production_steps": 20000
}

Supported Compounds

The Peng-Robinson EOS supports the following compounds (via PREOS.from_name()):

H2 (hydrogen) He (helium) NH3 (ammonia) H2O (water)
CH4 (methane) N2 (nitrogen) O2 (oxygen) Ar (argon)
CO (carbon monoxide) CO2 (carbon dioxide) C2H2 (acetylene) C2H6 (ethane)
C3H8 (propane) C4H10 (butane) C6H6 (benzene) C6H14 (n-hexane)

Citation

If you use this code in your research, please cite:

@software{mlip_mc,
  title = {MLIP-MC: Monte Carlo Simulations with Machine-Learned Interatomic Potentials},
  author = {Edwards, Connor W. and Yang, Fengxu and Stracke, Konstantin and Evans, Jack D.},
  year = {2025},
  license = {MIT}
}

License

This project is licensed under the MIT License - see the LICENSE file for details.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Acknowledgments

Much of the code in this repository is based on or derived from the work published at:

• Zenodo: 10.5281/zenodo.7904959

Additional acknowledgments:

• Built on the ASE framework
• Supports MLIP models from FAIRChem, MACE-Torch and Orbital