Universal Multi-fidelity M3GNet for material simulations

The goal of this repository is to showcase the trainable data embedding approach presented in our paper [1] for training atomistic machine learning models on multiple reference methods simultaneously. This repository contains a TorchScript model for evaluating an implementation of the M3GNet [2] model that was trained on two different fidelity levels found in the MatPES dataset [3] (PBE and r2SCAN). In addition, we provide an ASE calculator.

A scalable multi-GPU version of this model and other machine learning force fields will be available in our massively parallel software package Tremolo-X, soon. News will be announced here and on the website. Currently, the website is expecting major changes and will be updated in the next weeks.

Installation:

The model can be installed with pip:

pip install git+https://github.com/Fraunhofer-SCAI/VMDatomistic.git

Dependencies are torch>=2.1.0, ase, pymatgen and optionally torch_geometric for batched inference. You can install torch_geometric automatically with:

pip install git+https://github.com/Fraunhofer-SCAI/VMDatomistic.git#egg=VMDatomistic[pyg]

Examples:

The folder examples contains example scripts, including single sample inference and batched inference. For batched inference, we rely on the mini-batching concept from PyTorch Geometric. Although batched inference is possible without installing torch_geometric, its installation is recommended for efficient handling of graph data.

ASE calculator

The fidelity for the calculator can be changed between r2SCAN (default) and PBE when loading the calculator instance.

from ase.io import read
import torch
from VMDatomistic.calculators import PeriodicVMDCalculator

dtype = torch.float64
device = torch.device('cuda')
atoms = read('path/to/structure')
calculator = PeriodicVMDCalculator.multifidelity_m3gnet(fidelity="r2SCAN", dtype=dtype, device=device)
atoms.calc = calculator
# perform MD simulation or geometry optimization in the following:
...

Citation

When using the model for your work, please cite our work with:

@misc{OSH2025, 
title={Multi-Fidelity Learning for Atomistic Models via Trainable Data Embeddings}, 
DOI={10.26434/chemrxiv-2025-vx7nx-v2}, 
url={https://doi.org/10.26434/chemrxiv-2025-vx7nx-v2}, 
author={Oerder, Rick and Schmieden, Gerrit and Hamaekers, Jan}, 
year={2025},
note={ChemRxiv Preprint.}} 

Please cite [2] and [3], as well.

References

[1] Oerder, R., Schmieden, G., & Hamaekers, J. (2025). Multi-Fidelity Learning for Atomistic Models via Trainable Data Embeddings. ChemRxiv Preprint. DOI: 10.26434/chemrxiv-2025-vx7nx-v2.

[2] Chen, C., & Ong, S.P. (2023). A universal graph deep learning interatomic potential for the periodic table. Nature Computational Science, 2, 718–728. DOI: 10.1038/s43588-022-00349-3.

[3] Kaplan, A. D., Liu, R., Qi, J., Ko, T. W., Deng, B., Riebesell, J., Ceder, G., Persson, K. A., & Ong, S. P. (2025). A Foundational Potential Energy Surface Dataset for Materials. arXiv:2503.04070. DOI: 10.48550/arXiv.2503.04070.

License

The content of this repository is licensed under CC BY-NC-SA 4.0