Wrenformer

Version: v0.1.0 Added: 2022-11-26 Published: 2021-06-21 5.17M parameters Ensemble of 10 models Missing preds: 5 (0.002%)
Repo Paper DOI Files PR Checkpoint
Predictions
Discovery

ML vs DFT Formation Energies

ML vs DFT Convex Hull Distance

Convex hull distance prediction errors projected onto elements

1 H 0.56
2 He 0.00
3 Li 0.14
4 Be 0.26
5 B 0.44
6 C 0.31
7 N 0.43
8 O 0.61
9 F 0.73
10 Ne 0.00
11 Na 0.21
12 Mg 0.19
13 Al 0.28
14 Si 0.27
15 P 0.28
16 S 0.41
17 Cl 0.47
18 Ar 0.00
19 K 0.25
20 Ca 0.21
21 Sc 0.22
22 Ti 0.22
23 V 0.28
24 Cr 0.29
25 Mn 0.30
26 Fe 0.34
27 Co 0.23
28 Ni 0.21
29 Cu 0.21
30 Zn 0.23
31 Ga 0.23
32 Ge 0.25
33 As 0.25
34 Se 0.35
35 Br 0.40
36 Kr 0.00
37 Rb 0.24
38 Sr 0.21
39 Y 0.22
40 Zr 0.26
41 Nb 0.27
42 Mo 0.28
43 Tc 0.14
44 Ru 0.31
45 Rh 0.25
46 Pd 0.22
47 Ag 0.16
48 Cd 0.19
49 In 0.26
50 Sn 0.23
51 Sb 0.23
52 Te 0.32
53 I 0.33
54 Xe 0.32
55 Cs 0.25
56 Ba 0.21
57 La 0.20
58 Ce 0.18
59 Pr 0.19
60 Nd 0.19
61 Pm 0.15
62 Sm 0.19
63 Eu 0.17
64 Gd 0.19
65 Tb 0.21
66 Dy 0.21
67 Ho 0.21
68 Er 0.21
69 Tm 0.22
70 Yb 0.22
71 Lu 0.19
72 Hf 0.27
73 Ta 0.33
74 W 0.25
75 Re 0.25
76 Os 0.31
77 Ir 0.31
78 Pt 0.25
79 Au 0.25
80 Hg 0.15
81 Tl 0.18
82 Pb 0.22
83 Bi 0.17
84 Po 0.00
85 At 0.00
86 Rn 0.00
87 Fr 0.00
88 Ra 0.00
89 Ac 0.13
90 Th 0.26
91 Pa 0.21
92 U 0.27
93 Np 0.25
94 Pu 0.37
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

  1. Janosh Riebesell  University of Cambridge, Lawrence Berkeley National Laboratory  
  2. Rhys Goodall  University of Cambridge  
  3. Rokas Elijošius  University of Cambridge  

Trained By

  1. Janosh Riebesell  University of Cambridge, Lawrence Berkeley National Laboratory  

Model Info

  • Model Version v0.1.0
  • Model Type Transformer
  • Targets E
  • Openness OSOD
  • Train Task RP2RE
  • Test Task IP2E
  • Trained for Benchmark Yes

Training Set

MP v2022.10.28: 155k structures

Description

Wrenformer is a standard PyTorch Transformer Encoder trained to learn material embeddings from composition, space group, Wyckoff positions in a structure. Model workings A ML–powered materials discovery workflow using Wrenformer's Wyckoff string inputs to predict formation energies for candidate materials in an enumerated library of Wyckoff representations (shapes are used to denote different Wyckoff positions and colors to denote different element types). Predicted formation energies are then compared against the known convex hull of stability. Structures satisfying the required symmetries are relaxed for materials predicted to be stable.

Long

It builds on Roost and Wren, by being a fast structure-free model that is still able to distinguish polymorphs through symmetry.

Dependencies