AlphaNet-v1-OAM Convex hull distance prediction errors projected onto elements 1 H 0.13
2 He 0.00
3 Li 0.03
4 Be 0.04
5 B 0.06
6 C 0.05
7 N 0.09
8 O 0.15
9 F 0.11
10 Ne 0.00
11 Na 0.03
12 Mg 0.04
13 Al 0.05
14 Si 0.06
15 P 0.06
16 S 0.08
17 Cl 0.10
18 Ar 0.00
19 K 0.04
20 Ca 0.04
21 Sc 0.04
22 Ti 0.04
23 V 0.06
24 Cr 0.09
25 Mn 0.11
26 Fe 0.09
27 Co 0.04
28 Ni 0.04
29 Cu 0.04
30 Zn 0.04
31 Ga 0.04
32 Ge 0.05
33 As 0.06
34 Se 0.09
35 Br 0.08
36 Kr 0.00
37 Rb 0.04
38 Sr 0.04
39 Y 0.04
40 Zr 0.04
41 Nb 0.06
42 Mo 0.05
43 Tc 0.04
44 Ru 0.05
45 Rh 0.05
46 Pd 0.05
47 Ag 0.04
48 Cd 0.03
49 In 0.05
50 Sn 0.05
51 Sb 0.06
52 Te 0.12
53 I 0.07
54 Xe 0.05
55 Cs 0.04
56 Ba 0.04
57 La 0.04
58 Ce 0.04
59 Pr 0.03
60 Nd 0.03
61 Pm 0.03
62 Sm 0.03
63 Eu 0.06
64 Gd 0.04
65 Tb 0.03
66 Dy 0.04
67 Ho 0.03
68 Er 0.03
69 Tm 0.03
70 Yb 0.04
71 Lu 0.03
72 Hf 0.04
73 Ta 0.08
74 W 0.06
75 Re 0.04
76 Os 0.05
77 Ir 0.07
78 Pt 0.06
79 Au 0.06
80 Hg 0.03
81 Tl 0.03
82 Pb 0.06
83 Bi 0.04
84 Po 0.00
85 At 0.00
86 Rn 0.00
87 Fr 0.00
88 Ra 0.00
89 Ac 0.03
90 Th 0.05
91 Pa 0.05
92 U 0.06
93 Np 0.09
94 Pu 0.20
95 Am 0.00
96 Cm 0.00
97 Bk 0.00
98 Cf 0.00
99 Es 0.00
100 Fm 0.00
101 Md 0.00
102 No 0.00
103 Lr 0.00
104 Rf 0.00
105 Db 0.00
106 Sg 0.00
107 Bh 0.00
108 Hs 0.00
109 Mt 0.00
110 Ds 0.00
111 Rg 0.00
112 Cn 0.00
113 Nh 0.00
114 Fl 0.00
115 Mc 0.00
116 Lv 0.00
117 Ts 0.00
118 Og 0.00
57-71 La-Lu Lanthanides
89-103 Ac-Lr Actinides
Model Authors Bangchen Yin Tsinghua University Jiaao Wang University of Texas at Austin Weitao Du DAMO Academy, Alibaba Inc Yuanqi Du Cornell University Chenru Duan Deep Principle, Inc Carla P. Gomes Cornell University Graeme Henkelman The University of Texas at Austin Hai Xiao Tsinghua University Trained By 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 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: 0.03max_steps: 500ase_optimizer: "FIRE"cell_filter: "FrechetCellFilter"optimizer: "Adam"loss: "MAE"loss_weights: {"energy":4,"force":100,"stress":10}batch_size: 256initial_learning_rate: 0.0002learning_rate_schedule: "StepLR(decay_steps=10000, decay_ratio = 0.93)"epochs: 4n_layers: 4n_hidden_channels: 176n_bessel_basis: 8n_heads: 24graph_construction_radius: 5max_neighbors: 50
