Table 1 Summary of the training sets to train DP for Ti.

From: Specialising neural network potentials for accurate properties and application to the mechanical response of titanium

Dataset type

Number of datasets

Initialisation datasets

1469

DP-GENa bulk

4354

DP-GENa surface

1525

γ-line datasets from specialisation

4600

Total

11,948

  1. aDP-GEN applies trial DPs to perturbed bulk/surface structures in finite temperature MD simulations to generate configurations which are then input into DFT to calculate atomic forces, energy, and virial tensors on candidate structures (that go into training sets).