model_name: HIENet model_key: hienet model_version: v1.0.1 date_added: '2025-07-01' date_published: '2025-02-25' authors: - name: Keqiang Yan affiliation: Texas A&M University - name: Montgomery Bohde affiliation: Texas A&M University orcid: https://orcid.org/0009-0007-7705-1219 corresponding: true - name: Kryvenko Andrii affiliation: Texas A&M University orcid: https://orcid.org/0009-0000-3395-0967 - name: Ziyu Xiang affiliation: Texas A&M University orcid: https://orcid.org/0000-0003-0925-8705 repo: https://github.com/divelab/AIRS/tree/main/OpenMat/HIENet doi: https://doi.org/10.48550/arXiv.2503.05771 paper: https://arxiv.org/pdf/2503.05771 checkpoint_url: https://github.com/divelab/AIRS/blob/main/OpenMat/HIENet/checkpoints/HIENet-V3.pth pr_url: https://github.com/janosh/matbench-discovery/pull/268 openness: OSOD trained_for_benchmark: false train_task: S2EFS test_task: IS2RE-SR targets: EFS_G model_type: UIP model_params: 7_510_280 n_estimators: 1 hyperparams: max_force: 0.05 max_steps: 500 ase_optimizer: FIRE cell_filter: FrechetCellFilter epochs: 200 optimizer: AdamW loss: Huber - delta=0.01 loss_weights: energy: 1.0 force: 1.0 stress: 0.01 batch_size: 48 initial_learning_rate: 0.01 learning_rate_schedule: CosineWarmupLR - warmup_factor=0.2, warmup_epochs=0.1, lr_min_factor=0.0005 weight_decay: 0.001 lmax: 3 num_invariant_conv: 1 inv_features: [384, 384] irreps: 384x0e -> 512x0e+128x1e+64x2e -> 512x0e+128x1e+64x2e+32x3e -> 512x0e radial_basis: bessel n_radial_bessel_basis: 8 cutoff_function: poly_cut - p_value=6 activation_gate: silu/tanh activation_scalar: silu/tanh dropout: 0.04 dropout_attention: 0.08 conv_denominator: 35.989574 ema_decay: 0.999 forces_rms_scale: 0.799 max_neighbors: .inf graph_construction_radius: 5.0 license: code: GPL-3.0 code_url: https://github.com/divelab/AIRS/blob/main/OpenMat/HIENet/LICENSE checkpoint: GPL-3.0 checkpoint_url: https://github.com/divelab/AIRS/blob/main/OpenMat/HIENet/LICENSE requirements: torch: 2.1.2 torch-geometric: 2.6.1 numpy: 1.26.4 ase: 3.25.0 braceexpand: 0.1.7 e3nn: 0.5.6 pymatviz: 0.16.0 pyyaml: 6.0.1 torch-scatter: 2.1.2 scikit-learn: 1.7.0 pymatgen: 2025.6.14 wandb: 0.20.1 torch-ema: 0.3.0 training_set: [MPtrj] training_cost: Nvidia A100 GPUs: { amount: 8, hours: 361 } notes: Description: | HIENet is a hybrid invariant-equivariant graph neural network interatomic potential that combines E(3) invariant and O(3) equivariant message passing layers for materials discovery. The model uses physics-informed gradient-based predictions to ensure all outputs satisfy key physical constraints including force conservation and rotational symmetries, enabling accurate prediction of energy, forces, and stress for crystalline materials. metrics: phonons: kappa_103: κ_SRME: 0.6423 pred_file: models/hienet/hienet/2025-07-01-kappa-103-FIRE-dist=0.01-fmax=1e-4-symprec=1e-5.json.gz pred_file_url: https://figshare.com/files/55909451 κ_SRE: 0.4898 geo_opt: pred_file: models/hienet/hienet/2025-07-01-wbm-geo-opt-IS2RE-FIRE.json.gz struct_col: hienet_structure pred_file_url: https://figshare.com/files/55909424 symprec=1e-2: rmsd: 0.0795 # unitless n_sym_ops_mae: 1.8758 # unitless symmetry_decrease: 0.0755 # fraction symmetry_match: 0.8002 # fraction symmetry_increase: 0.1163 # fraction n_structures: 256963 # count analysis_file: models/hienet/hienet/2025-07-01-wbm-geo-opt-IS2RE-FIRE-symprec=1e-2-moyo=0.4.2.csv.gz analysis_file_url: https://figshare.com/files/55909445 symprec=1e-5: rmsd: 0.0795 # unitless n_sym_ops_mae: 2.6208 # unitless symmetry_decrease: 0.3493 # fraction symmetry_match: 0.4521 # fraction symmetry_increase: 0.1532 # fraction n_structures: 256963 # count analysis_file: models/hienet/hienet/2025-07-01-wbm-geo-opt-IS2RE-FIRE-symprec=1e-5-moyo=0.4.2.csv.gz analysis_file_url: https://figshare.com/files/55909448 discovery: pred_file_url: https://figshare.com/files/55909421 pred_file: models/hienet/hienet/2025-07-01-wbm-IS2RE.csv.gz pred_col: e_form_per_atom_hienet full_test_set: F1: 0.766 # fraction DAF: 4.361 # dimensionless Precision: 0.748 # fraction Recall: 0.784 # fraction Accuracy: 0.918 # fraction TPR: 0.784 # fraction FPR: 0.055 # fraction TNR: 0.945 # fraction FNR: 0.216 # fraction TP: 34551.0 # count FP: 11622.0 # count TN: 201249.0 # count FN: 9541.0 # count MAE: 0.039 # eV/atom RMSE: 0.082 # eV/atom R2: 0.791 # dimensionless missing_preds: 2 # count most_stable_10k: F1: 0.973 # fraction DAF: 6.194 # dimensionless Precision: 0.947 # fraction Recall: 1.0 # fraction Accuracy: 0.947 # fraction TPR: 1.0 # fraction FPR: 1.0 # fraction TNR: 0.0 # fraction FNR: 0.0 # fraction TP: 9469.0 # count FP: 531.0 # count TN: 0.0 # count FN: 0.0 # count MAE: 0.033 # eV/atom RMSE: 0.069 # eV/atom R2: 0.894 # dimensionless missing_preds: 0 # count unique_prototypes: F1: 0.777 # fraction DAF: 4.932 # dimensionless Precision: 0.754 # fraction Recall: 0.801 # fraction Accuracy: 0.929 # fraction TPR: 0.801 # fraction FPR: 0.048 # fraction TNR: 0.952 # fraction FNR: 0.199 # fraction TP: 26723.0 # count FP: 8721.0 # count TN: 173393.0 # count FN: 6651.0 # count MAE: 0.041 # eV/atom RMSE: 0.084 # eV/atom R2: 0.793 # dimensionless missing_preds: 0 # count