--- library_name: mace license: other tags: - chemistry - materials-science - molecular-dynamics - interatomic-potential - machine-learning-force-field --- # MACE-MH-1: Multi-Head Foundation Model for Atomistic Materials Chemistry GitHub ## Highlights **MACE-MH-1** is a foundation machine-learning interatomic potential (MLIP) that bridges molecular, surface, and materials chemistry through cross-domain learning: - **Unified cross-domain capability** spanning inorganic crystals, molecular systems, surface chemistry, and reactive organic chemistry with a single model - **State-of-the-art performance** across materials, molecular crystals, surfaces, and molecular benchmarks with a global performance score of 0.862 - **Enhanced MACE architecture** with improved weight sharing across chemical elements and non-linear tensor decomposition in the product basis ![image](https://cdn-uploads.huggingface.co/production/uploads/630df4308df86f1e5bec63c3/sqXxh5DrLSXj8elKnJJS8.png) ## Model Overview **MACE-MH-1** has the following features: - **Type**: E(3)-equivariant graph neural network for interatomic potentials - **Architecture**: MACE - **Interaction Blocks**: Non-linear - **Training Stages**: Pre-training on OMAT-24 (100M inorganic crystals) + Multi-head fine-tuning on diverse datasets - **Hyper-Parameters**: 512 node channels, 128 edge channels, L=1, max_ell=3, 2 layers - **Chemical Coverage**: 89 elements - **Cutoff Radius**: 6 Å - **Multiple Heads**: OMAT PBE (main), OMOL (ωB97M-VV10), OC20 (surfaces), SPICE, RGD1, MPTraj, Matpes (r2scan) For more details, please refer to the [paper](https://arxiv.org/abs/2510.25380), [GitHub repository](https://github.com/ACEsuit/mace), and [MACE foundations](https://github.com/ACEsuit/mace-foundations). ## Performance ### Materials Benchmarks (PBE+U Reference) | Benchmark | Metric | MACE-MH-1 | ORB-v3 | UMA-S-1.1 | |-----------|--------|-----------|---------|-----------| | Phonon BZ | MAE (K) | **5** | 15 | 9 | | Phonon ωavg | MAE (K) | **3** | 5 | 4 | | Phonon ωmin | MAE (K) | **11** | 29 | 21 | | Phonon ωmax | MAE (K) | 12 | 12 | **11** | | Entropy (300K) | MAE (J/mol·K) | 8 | 13 | **7** | | Helmholtz Free Energy (300K) | MAE (kJ/mol) | **2** | 3 | **2** | | Heat Capacity | MAE (J/mol·K) | **3** | 4 | **3** | | Bulk Modulus | MAE (GPa) | 12.49 | **7.18** | 14.33 | | Shear Modulus | MAE (GPa) | **7.95** | 8.03 | 8.18 | | Thermal Conductivity | RMSE (W/mK) | 0.24 | 0.21 | **0.20** | ### Molecular Crystal Benchmarks | Benchmark | Metric | MACE-MH-1-OMAT-D3 | ORB-v3 | UMA-S-1.1-OMAT-D3 | |-----------|--------|-------------------|---------|-------------------| | X23 Formation Energy | MAE (kJ/mol) | **15.82** | 28.76 | 27.99 | | Ice Polymorphs (DMC) | MAE (meV) | **11.23** | 138.44 | 310.82 | ### Surface Benchmarks | Benchmark | Metric | MACE-MH-1-OMAT-D3 | ORB-v3-D3 | UMA-S-1.1-OMAT-D3 | |-----------|--------|-------------------|-----------|-------------------| | S24 Adsorption | MAE (eV) | **0.095** | 0.174 | 0.329 | | OC20 Adsorption | MAE (eV) | **0.138** | 0.159 | 0.172 | | OC20 Correlation | Pearson's r | **0.98** | 0.974 | 0.97 | ### Molecular Benchmarks | Benchmark | Metric | MACE-MH-1-OMAT-D3 | ORB-v3-D3 | UMA-S-1.1-OMAT-D3 | |-----------|--------|-------------------|-----------|-------------------| | Wiggle150 | MAE (kcal/mol) | **4.80** | 7.65 | 6.60 | | GMTKN55 Overall | WTMAD (kcal/mol) | **11.23** | 22.30 | 30.83 | | PLF547 (proteins) | MAE (kcal/mol) | **0.626** | 1.829 | 2.935 | | S30L (host-guest) | MAE (kcal/mol) | **10.13** | 13.64 | 15.14 | ### Physicality Tests | Test | Metric | MACE-MH-1 | ORB-v3 | UMA-S-1.1 | |------|--------|-----------|---------|-----------| | Slab Extensivity | Δ (meV) | **0.0** | -709.7 | -453.8 | | H-Atom Additivity | max \|ΔF\| (meV/Å) | **0.0** | 61.65 | 969.2 | | Diatomic Force Flips | Mean count | **2.09** | 2.91 | 10.73 | | Diatomic Minima | Mean count | **1.42** | 1.62 | 4.82 | ## Training Data ### Pre-training - **OMAT-24**: 100M configurations of inorganic crystals (PBE/PBE+U) spanning 89 elements ### Multi-Head Fine-tuning - **OMAT Replay**: 10M configurations (10% subset) to prevent catastrophic forgetting - **MPTraj**: 1.5M configurations from Materials Project with PBE+U - **SPICE-1**: ~1M organic molecules (ωB97M-D3(BJ)/def2-TZVP) - **OC20**: 2M metal surface slabs and adsorbate complexes (PBE) - **OMOL-1%**: 1.2M diverse organic and organometallic configurations (ωB97M-VV10) - **RGD1**: 300K organic reaction intermediates and transition states (B3LYP/6-31G*) - **MATPES R2SCAN**: 400K inorganic crystals (r²SCAN) ## Installation and Usage ### Installation ```bash pip install mace-torch ``` ### Basic Usage (Python) ```python from mace.calculators import mace_mp from ase import Atoms # Load the MACE-MH-1 model (using the OMAT/PBE head) calc = mace_mp(model=path, default_dtype="float64", device="cuda", head="omat_pbe") # Create an example structure atoms = Atoms('H2O', positions=[[0, 0, 0], [0, 0, 1], [0, 1, 0]]) atoms.calc = calc # Calculate energy and forces energy = atoms.get_potential_energy() forces = atoms.get_forces() print(f"Energy: {energy} eV") print(f"Forces:\n{forces}") ``` ## Available Model Heads MACE-MH-1 contains multiple task-specific heads trained on different levels of theory: | Head Name | Level of Theory | Best For | Access | |-----------|----------------|----------|--------| | **omat_pbe** (default) | PBE/PBE+U | General materials, balanced performance across tasks | Specify in model | | **omol** | ωB97M-VV10 | 1% of OMOL data: Molecular systems, organic chemistry, Organometallic | Specify in model | | **spice_wB97M** | ωB97M-D3(BJ) | Molecular systems and organic chemistry | Specify in model | | **rgd1_b3lyp** | B3LYP | Reaction chemistry | Specify in model | | **oc20_usemppbe** | PBE | Surface catalysis, adsorbates | Specify in model | | **matpes_r2scan** | r²SCAN meta-GGA | High-accuracy materials | Specify in model | By default, the **OMAT head** (PBE) is used, which provides the best cross-domain performance. ## Best Practices 1. **For fine-tuning**: Use OMAT head first. Test other heads if needed. 1. **For materials**: Use OMAT head. Use D3 corrections for systems with dispersions. Test matpes_r2scan head if r2scan better reference. 2. **For molecules**: Consider using OMOL head (ωB97M-VV10) for improved intramolecular interactions. OMAT head good for condensed phase molecular systems, test it too. 3. **For surfaces**: OMAT head provides excellent performance; OC20 head available for specialized applications ## Citation If you use MACE-MH-1 in your research, please cite: ```bibtex @article{batatia2025crosslearning, title={Cross Learning between Electronic Structure Theories for Unifying Molecular, Surface, and Inorganic Crystal Foundation Force Fields}, author={Batatia, Ilyes and Lin, Chen and Hart, Joseph and Kasoar, Elliott and Elena, Alin M. and Norwood, Sam Walton and Wolf, Thomas and Cs{\'a}nyi, G{\'a}bor}, journal={arXiv preprint arXiv:2510.25380}, year={2025} } @article{batatia2022mace, title={MACE: Higher order equivariant message passing neural networks for fast and accurate force fields}, author={Batatia, Ilyes and Kovacs, David Peter and Simm, Gregor and Ortner, Christoph and Cs{\'a}nyi, G{\'a}bor}, journal={Advances in Neural Information Processing Systems}, volume={35}, pages={11423--11436}, year={2022} } ``` ## License This model is released under the ASL License. ## Acknowledgments This work was supported by computational resources from: - Jean Zay HPC (Grand Challenge GC010815458) - Isambard-AI and Sovereign AI ## Contact - **GitHub**: [ACEsuit/mace](https://github.com/ACEsuit/mace) - **Foundations**: [mace-foundations](https://github.com/ACEsuit/mace-foundations) - **Issues**: Please report issues on the GitHub repository