AlphaNet-v1-OMA

Version: v1 Added: 2025-05-12 Published: 2025-05-12 4.65M parameters Missing preds: 0
Repo Paper Docs DOI Files PR Checkpoint
Predictions
κSRMEGeometry OptimizationDiscoveryDiatomics

Convex hull distance prediction errors projected onto elements

1 H 0.13
2 He 0
3 Li 0.0282
4 Be 0.035
5 B 0.0596
6 C 0.0475
7 N 0.0881
8 O 0.145
9 F 0.114
10 Ne 0
11 Na 0.0303
12 Mg 0.04
13 Al 0.0453
14 Si 0.0592
15 P 0.0577
16 S 0.0837
17 Cl 0.101
18 Ar 0
19 K 0.0369
20 Ca 0.0409
21 Sc 0.0352
22 Ti 0.0438
23 V 0.0626
24 Cr 0.0871
25 Mn 0.111
26 Fe 0.091
27 Co 0.0449
28 Ni 0.0411
29 Cu 0.0408
30 Zn 0.043
31 Ga 0.0446
32 Ge 0.0496
33 As 0.0557
34 Se 0.0906
35 Br 0.0843
36 Kr 0
37 Rb 0.0439
38 Sr 0.0364
39 Y 0.041
40 Zr 0.0447
41 Nb 0.0551
42 Mo 0.0524
43 Tc 0.0434
44 Ru 0.0539
45 Rh 0.0524
46 Pd 0.05
47 Ag 0.0361
48 Cd 0.0322
49 In 0.0524
50 Sn 0.0481
51 Sb 0.0583
52 Te 0.117
53 I 0.0689
54 Xe 0.048
55 Cs 0.0432
56 Ba 0.0369
57 La 0.035
58 Ce 0.035
59 Pr 0.0329
60 Nd 0.0303
61 Pm 0.0312
62 Sm 0.0324
63 Eu 0.0584
64 Gd 0.0419
65 Tb 0.0349
66 Dy 0.0351
67 Ho 0.0316
68 Er 0.0326
69 Tm 0.0334
70 Yb 0.0401
71 Lu 0.0337
72 Hf 0.0445
73 Ta 0.0832
74 W 0.0567
75 Re 0.0414
76 Os 0.0501
77 Ir 0.0659
78 Pt 0.0603
79 Au 0.0643
80 Hg 0.0312
81 Tl 0.0321
82 Pb 0.0554
83 Bi 0.0411
84 Po 0
85 At 0
86 Rn 0
87 Fr 0
88 Ra 0
89 Ac 0.0308
90 Th 0.0457
91 Pa 0.0457
92 U 0.0572
93 Np 0.0873
94 Pu 0.198
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. Bangchen Yin  Tsinghua University  Tsinghua University logo  
  2. Jiaao Wang  University of Texas at Austin  
  3. Weitao Du  DAMO Academy, Alibaba Inc  
  4. Yuanqi Du  Cornell University  
  5. Chenru Duan  Deep Principle, Inc  
  6. Carla P. Gomes  Cornell University  
  7. Graeme Henkelman  The University of Texas at Austin  
  8. Hai Xiao  Tsinghua University  Tsinghua University logo  

Trained By

  1. Bangchen Yin (Tsinghua University)

Model Info

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

Training Set

OMat24: 101M structures from 3.23M materials

Subsampled Alexandria: 10.4M structures from 3.23M materials

MPtrj: 1.58M structures from 146k materials

Description

AlphaNet is a local frame-based equivariant model designed to tackle the challenges of achieving both accurate and efficient simulations for atomistic systems. AlphaNet enhances computational efficiency and accuracy by leveraging the local geometric structures of atomic environments through the construction of equivariant local frames and learnable frame transitions.

We changed the RBF function and used a small size model in this version.

Hyperparameters

  • max_force: 0.03
  • max_steps: 500
  • ase_optimizer: "FIRE"
  • cell_filter: "FrechetCellFilter"
  • optimizer: "Adam"
  • loss: "MAE"
  • loss_weights: {"energy":4,"force":100,"stress":10}
  • batch_size: 256
  • initial_learning_rate: 0.0002
  • learning_rate_schedule: "StepLR(decay_steps=10000, decay_ratio = 0.93)"
  • epochs: 4
  • n_layers: 4
  • n_hidden_channels: 176
  • n_bessel_basis: 8
  • n_heads: 24
  • graph_construction_radius: 5
  • max_neighbors: 50

Dependencies