Table 1 Training dataset for machine learning atomic potential in the Fe-H binary system

From: Machine learning interatomic potential reveals hydrogen embrittlement origins at general grain boundaries in α-iron

Subsets

Energy RMSE (meV/atom)

Force RMSE (meV/Å)

Number of structures

a. Initial dataset

1. Equilibrated Fe with dilute H

1.00

25.63

144

2. Perturbed Fe with dilute H

0.86

29.44

595

3. Strained Fe with 1 H

1.83

24.75

99

4. H-vacancy clusters

0.75

36.18

124

5. 2H in neighboring TISs

1.12

17.83

207

6. Single H atom and H₂ molecules

13.49

200.14

24

All initial datasets

1.52

50.12

270

b. DP-GEN dataset

1. Fe with dilute H

2.17

83.23

1212

2. Fe with high concentration H

4.77

132.41

2088

3. H atoms in vacancy

5.36

142.26

901

4. H atoms on surface

6.45

167.85

1408

5. Generalized stacking faults with H atoms

6.16

141.73

1081

6. Perturbed tilt grain boundaries with H atoms

6.19

136.26

4924

7. Self-interstitial atom (\(\left\langle 111\right\rangle\), \(\left\langle 110\right\rangle\), \(\left\langle 100\right\rangle\) dumbbell)

4.61

104.70

5591

8. H atoms in vacancy cluster

2.51

104.35

16531

All datasets

4.09

109.16

36438

  1. The training dataset consists of an initial dataset comprising six subsets and a dataset obtained from concurrent learning comprising eight subsets (DP-GEN dataset). The number of structures included in each subset and the root mean square error (RMSE) for energy and atomic force are also shown.
Back to article page