Eqnorm MPtrj

Version: 0.1.0 Added: 2025-05-26 Published: 2025-05-26 1.31M parameters Missing preds: 0
Repo Paper Docs DOI Files PR Checkpoint
Predictions
κSRMEGeometry OptimizationDiscovery

Convex hull distance prediction errors projected onto elements

1 H 0.246
2 He 0
3 Li 0.0437
4 Be 0.0835
5 B 0.112
6 C 0.0989
7 N 0.133
8 O 0.146
9 F 0.148
10 Ne 0
11 Na 0.0513
12 Mg 0.0637
13 Al 0.102
14 Si 0.107
15 P 0.107
16 S 0.13
17 Cl 0.14
18 Ar 0
19 K 0.0572
20 Ca 0.0647
21 Sc 0.0686
22 Ti 0.0766
23 V 0.107
24 Cr 0.126
25 Mn 0.149
26 Fe 0.138
27 Co 0.0845
28 Ni 0.0817
29 Cu 0.0635
30 Zn 0.072
31 Ga 0.0916
32 Ge 0.0971
33 As 0.094
34 Se 0.136
35 Br 0.131
36 Kr 0
37 Rb 0.0617
38 Sr 0.0595
39 Y 0.0704
40 Zr 0.0834
41 Nb 0.104
42 Mo 0.0845
43 Tc 0.0799
44 Ru 0.114
45 Rh 0.0946
46 Pd 0.0858
47 Ag 0.0568
48 Cd 0.0588
49 In 0.0936
50 Sn 0.0872
51 Sb 0.0872
52 Te 0.156
53 I 0.12
54 Xe 0
55 Cs 0.0627
56 Ba 0.0594
57 La 0.06
58 Ce 0.0655
59 Pr 0.0603
60 Nd 0.0584
61 Pm 0.0519
62 Sm 0.0599
63 Eu 0.0786
64 Gd 0.064
65 Tb 0.0602
66 Dy 0.0638
67 Ho 0.0616
68 Er 0.0629
69 Tm 0.065
70 Yb 0.0667
71 Lu 0.0643
72 Hf 0.0853
73 Ta 0.135
74 W 0.0876
75 Re 0.0889
76 Os 0.109
77 Ir 0.116
78 Pt 0.0984
79 Au 0.0961
80 Hg 0.0558
81 Tl 0.064
82 Pb 0.0873
83 Bi 0.0789
84 Po 0
85 At 0
86 Rn 0
87 Fr 0
88 Ra 0
89 Ac 0.057
90 Th 0.0917
91 Pa 0.0939
92 U 0.105
93 Np 0.157
94 Pu 0.363
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. Yuzhuo Chen  Zhejiang Lab  Zhejiang Lab logo    
  2. Lyuwen Fu  Zhejiang Lab  Zhejiang Lab logo  
  3. Shuxiang Yang  Zhejiang Lab  Zhejiang Lab logo  
  4. Lipeng Chen  Zhejiang Lab  Zhejiang Lab logo  

Trained By

  1. Yuzhuo Chen (Zhejiang Lab)

Model Info

  • Model Version 0.1.0
  • Model Type UIP
  • Targets EFSG
  • Openness OSOD
  • Train Task S2EFS
  • Test Task IS2RE-SR
  • Trained for Benchmark Yes

Training Set

MPtrj: 1.58M structures from 146k materials

Description

eqnorm is a graph neural network model designed for predicting the energy, forces, and stresses of materials. The model utilizes a combination of invariant and equivariant layers to effectively capture the symmetries present in material structures.

Hyperparameters

  • max_force: 0.02
  • max_steps: 500
  • ase_optimizer: "FIRE"
  • cell_filter: "FrechetCellFilter"
  • loss: "Huber"
  • loss_weights: {"energy":20,"force":20,"stress":320}
  • optimizer: "AdamW"
  • weight_decay: 0.001
  • clip_grad_norm: 100
  • ema_decay: 0.999
  • max_learning_rate: 0.01
  • min_learning_rate: 0.000001
  • learning_rate_schedule: "warmcosine"
  • warmup_factor: 0.2
  • epochs: 100
  • batch_train: 128
  • n_layers: 4
  • num_embedding_features: 128
  • num_bessel_basis: 8
  • invariant_layers: 2
  • invariant_neurons: 64
  • poly_p: 6
  • graph_construction_radius: 6
  • max_neighbors: null
  • irreps_hidden: "128x0e+64x1o+32x2e+32x3o"
  • irreps_sh: "1x0e+1x1o+1x2e+1x3o"
  • energy_shift: "per_species"
  • energy_scale: "force_rms"
  • shift_trainable: false
  • scale_trainable: false

Dependencies