model_name: DPA-4.0-Pro-MPtrj model_key: dpa-4.0-pro-mptrj model_version: v2026.05.20 date_added: '2026-05-20' date_published: '2026-05-20' authors: - name: Tiancheng Li affiliation: AI for Science Institute, Beijing; Peking University email: LTC201806070316@outlook.com github: https://github.com/OutisLi - name: Duo Zhang affiliation: AI for Science Institute, Beijing orcid: https://orcid.org/0000-0001-9591-2659 corresponding: true - name: Linfeng Zhang affiliation: AI for Science Institute, Beijing; DP Technology email: linfeng.zhang.zlf@gmail.com corresponding: true - name: Han Wang affiliation: Beijing Institute of Applied Physics and Computational Mathematics (IAPCM) email: wang_han@iapcm.ac.cn corresponding: true trained_by: - name: Tiancheng Li affiliation: AI for Science Institute, Beijing; Peking University email: LTC201806070316@outlook.com github: https://github.com/OutisLi repo: https://github.com/deepmodeling/deepmd-kit url: https://github.com/deepmodeling/deepmd-kit doi: https://github.com/deepmodeling/deepmd-kit # DPA-4.0 preprint not yet available, will be released soon paper: https://github.com/deepmodeling/deepmd-kit # DPA-4.0 preprint not yet available, will be released soon pr_url: https://github.com/janosh/matbench-discovery/pull/324 checkpoint_url: https://figshare.com/files/64733586 license: code: LGPL-3.0 code_url: https://github.com/deepmodeling/deepmd-kit/blob/master/LICENSE checkpoint: CC-BY-4.0 checkpoint_url: https://creativecommons.org/licenses/by/4.0/legalcode openness: OSOD trained_for_benchmark: true train_task: S2EFS test_task: IS2RE-SR targets: EFS_G model_type: UIP model_params: 32_724_908 n_estimators: 1 training_set: [MPtrj] training_cost: Nvidia H20 GPUs: {amount: 8, hours: 288} hyperparams: # Matbench Discovery relaxation settings used for the submitted WBM IS2RE results. max_force: 0.02 max_steps: 500 ase_optimizer: FIRE cell_filter: FrechetCellFilter graph_construction_radius: 6.0 max_neighbors: 384 architecture: type: DPA4/SeZM n_blocks: 8 lmax: 5 mmax: 1 channels: 96 n_radial: 16 so2_layers: 4 radial_so2_mode: none radial_so2_rank: 0 n_focus: 1 n_atten_head: 1 ffn_blocks: 1 sandwich_norm: [true, false, true, false] s2_activation: [false, true] lebedev_quadrature: false activation_function: silu glu_activation: true precision: float32 learning_rate_schedule: type: WSD loss: loss_func: mae f_use_norm: true loss_prefactor: energy: 20 force: 20 virial: 5 optimizer: Muon weight_decay: 0.001 training: numb_steps: 3_000_000 batch_size: 500 gradient_max_norm: 5 requirements: deepmd-kit: 3.1.4 torch: 2.11.0+cu128 ase: 3.28.0 notes: Description: | DPA-4.0-Pro-MPtrj is a DeePMD-kit universal interatomic potential in the DPA4/SeZM architecture family, trained only on the MPtrj dataset for this Matbench Discovery submission. Architecture: | DPA4 uses local-frame SO(2)-equivariant message passing with envelope-gated attention over invariant channels. The submitted model has 8 interaction blocks, lmax=5, mmax=1, 96 channels, 16 radial basis functions, and a 6.0 Å cutoff with up to 384 selected neighbors. Training: | The model was trained for 3,000,000 steps using the Muon optimizer, WSD learning-rate schedule, MAE energy/force/virial loss weights 20/20/5, with AMP, TF32, and compile enabled. metrics: phonons: kappa_103: κ_SRME: 0.2414 κ_SRE: 0.1323 pred_file: models/deepmd/dpa-4.0-pro-mptrj/2026-05-14-kappa-103-FIRE-dist=0.03-fmax=1e-4-symprec=1e-5.json.gz pred_file_url: https://figshare.com/files/64733580 geo_opt: pred_file: models/deepmd/dpa-4.0-pro-mptrj/2026-05-12-wbm-IS2RE-FIRE.jsonl.gz pred_file_url: https://figshare.com/files/64733670 struct_col: dpa4_structure symprec=1e-2: rmsd: 0.0694 # unitless n_sym_ops_mae: 2.3791 # unitless symmetry_decrease: 0.2207 # fraction symmetry_match: 0.6502 # fraction symmetry_increase: 0.1039 # fraction n_structures: 256963 # count analysis_file: models/deepmd/dpa-4.0-pro-mptrj/2026-05-12-wbm-IS2RE-FIRE-symprec=1e-2-moyo=0.7.9.csv.gz analysis_file_url: https://figshare.com/files/64735083 symprec=1e-5: rmsd: 0.0694 # unitless n_sym_ops_mae: 6.818 # unitless symmetry_decrease: 0.6561 # fraction symmetry_match: 0.2497 # fraction symmetry_increase: 0.0674 # fraction n_structures: 256963 # count analysis_file: models/deepmd/dpa-4.0-pro-mptrj/2026-05-12-wbm-IS2RE-FIRE-symprec=1e-5-moyo=0.7.9.csv.gz analysis_file_url: https://figshare.com/files/64735089 discovery: pred_file: models/deepmd/dpa-4.0-pro-mptrj/2026-05-12-dpa-4.0-pro-mptrj-preds.csv.gz pred_file_url: https://figshare.com/files/64733577 pred_col: e_form_per_atom_dp full_test_set: F1: 0.832 # fraction DAF: 4.943 # dimensionless Precision: 0.848 # fraction Recall: 0.816 # fraction Accuracy: 0.943 # fraction TPR: 0.816 # fraction FPR: 0.03 # fraction TNR: 0.97 # fraction FNR: 0.184 # fraction TP: 35987.0 # count FP: 6439.0 # count TN: 206432.0 # count FN: 8105.0 # count MAE: 0.029 # eV/atom RMSE: 0.076 # eV/atom R2: 0.82 # dimensionless missing_preds: 2 # count unique_prototypes: F1: 0.85 # fraction DAF: 5.628 # dimensionless Precision: 0.86 # fraction Recall: 0.841 # fraction Accuracy: 0.954 # fraction TPR: 0.841 # fraction FPR: 0.025 # fraction TNR: 0.975 # fraction FNR: 0.159 # fraction TP: 28052.0 # count FP: 4554.0 # count TN: 177560.0 # count FN: 5322.0 # count MAE: 0.03 # eV/atom RMSE: 0.078 # eV/atom R2: 0.823 # dimensionless missing_preds: 0 # count most_stable_10k: F1: 0.982 # fraction DAF: 6.307 # dimensionless Precision: 0.964 # fraction Recall: 1.0 # fraction Accuracy: 0.964 # fraction TPR: 1.0 # fraction FPR: 1.0 # fraction TNR: 0.0 # fraction FNR: 0.0 # fraction TP: 9641.0 # count FP: 359.0 # count TN: 0.0 # count FN: 0.0 # count MAE: 0.033 # eV/atom RMSE: 0.122 # eV/atom R2: 0.698 # dimensionless missing_preds: 0 # count