HIENet

Version: v1.0.1 Added: 2025-07-01 Published: 2025-02-25 7.51M parameters Missing preds: 0
Repo Paper DOI Files PR Checkpoint
Predictions
κSRMEGeometry OptimizationDiscovery

Convex hull distance prediction errors projected onto elements

1 H 0.25
2 He 0
3 Li 0.0468
4 Be 0.0925
5 B 0.115
6 C 0.097
7 N 0.132
8 O 0.143
9 F 0.169
10 Ne 0
11 Na 0.0569
12 Mg 0.0674
13 Al 0.104
14 Si 0.111
15 P 0.103
16 S 0.138
17 Cl 0.162
18 Ar 0
19 K 0.0674
20 Ca 0.0674
21 Sc 0.0656
22 Ti 0.0733
23 V 0.0979
24 Cr 0.128
25 Mn 0.153
26 Fe 0.143
27 Co 0.0873
28 Ni 0.0752
29 Cu 0.0678
30 Zn 0.07
31 Ga 0.0951
32 Ge 0.101
33 As 0.0922
34 Se 0.147
35 Br 0.146
36 Kr 0
37 Rb 0.0685
38 Sr 0.0651
39 Y 0.0749
40 Zr 0.08
41 Nb 0.101
42 Mo 0.0913
43 Tc 0.0701
44 Ru 0.112
45 Rh 0.0955
46 Pd 0.0864
47 Ag 0.059
48 Cd 0.0653
49 In 0.0982
50 Sn 0.0922
51 Sb 0.0881
52 Te 0.164
53 I 0.144
54 Xe 0.064
55 Cs 0.0763
56 Ba 0.0637
57 La 0.0632
58 Ce 0.0661
59 Pr 0.0636
60 Nd 0.0628
61 Pm 0.0555
62 Sm 0.0641
63 Eu 0.0825
64 Gd 0.0693
65 Tb 0.0641
66 Dy 0.0685
67 Ho 0.0648
68 Er 0.0658
69 Tm 0.0687
70 Yb 0.0665
71 Lu 0.0682
72 Hf 0.0829
73 Ta 0.134
74 W 0.0939
75 Re 0.0946
76 Os 0.102
77 Ir 0.122
78 Pt 0.101
79 Au 0.103
80 Hg 0.06
81 Tl 0.0671
82 Pb 0.0936
83 Bi 0.0749
84 Po 0
85 At 0
86 Rn 0
87 Fr 0
88 Ra 0
89 Ac 0.0573
90 Th 0.0938
91 Pa 0.102
92 U 0.101
93 Np 0.172
94 Pu 0.411
95 Am 0
96 Cm 0
97 Bk 0
98 Cf 0
99 Es 0
100 Fm 0
101 Md 0
102 No 0
103 Lr 0
104 Rf 0
105 Db 0
106 Sg 0
107 Bh 0
108 Hs 0
109 Mt 0
110 Ds 0
111 Rg 0
112 Cn 0
113 Nh 0
114 Fl 0
115 Mc 0
116 Lv 0
117 Ts 0
118 Og 0
57-71 La-Lu Lanthanides
89-103 Ac-Lr Actinides

Model Authors

  1. Keqiang Yan  Texas A&M University  Texas A&M University logo
  2. Montgomery Bohde  Texas A&M University  Texas A&M University logo  
  3. Kryvenko Andrii  Texas A&M University  Texas A&M University logo  
  4. Ziyu Xiang  Texas A&M University  Texas A&M University logo  

Model Info

  • Model Version v1.0.1
  • Model Type UIP
  • Targets EFSG
  • Openness OSOD
  • Train Task S2EFS
  • Test Task IS2RE-SR
  • Trained for Benchmark No

Training Set

MPtrj: 1.58M structures from 146k materials

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.

Hyperparameters

  • 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,"force":1,"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: null
  • graph_construction_radius: 5

Dependencies