Table 1 Training datasets and fitting errors of the constructed deep potentials

From: Predicting hydrogen diffusion in nickel–manganese random alloys using machine learning interatomic potentials

Category

Subsets

RMSEE (meV/atom)

RMSEF (meV/Å)

Nframe

a. Initial dataset

1. Ni with 0–2 H atoms

0.81

24.67

568

 

2. Ni–12.5 at.% Mn with 0–2 H atoms

1.32

32.99

729

 

3. Ni–25.0 at.% Mn with 0–2 H atoms

5.30

57.21

756

b. GeNNIP4MD dataset

1. Ni with 0–2 H atoms

0.81

24.84

596

 

2. Ni–12.5 at.% Mn with 0–2 H atoms

4.63

50.15

2438

 

3. Ni–25.0 at.% Mn with 0–2 H atoms

6.15

74.27

12,690

All

 

5.70

68.41

17,777

  1. The training dataset consists of the initial dataset and additional dataset generated through GeNNIP4MD. For each subset, the number of structures (Nframe), root mean square error of energy (RMSEE), and root mean square error of atomic forces (RMSEF) are shown.
Back to article page