Machine Learning Potential to Model the Diamond Phase Nucleation in Misoriented Bilayer Graphene

Abstract

The machine learning potential (MLP) is proposed based on the representation of the environment through moment tensors to model the diamond phase nucleation in misoriented bilayer graphene. MLP is trained on a set of graphene structures, 2D diamond, and their hydrogenated modifications obtained by density functional theory computations. Learned MLP accurately reproduces energies and strengths of these structures and correctly describes hydrogenation of bilayer graphene and the formation of interlayer bonds. Growth of the diamond phase in bigraphene with a 5° misorientation of layers is studied using MLP. It is found that the formation energy increases with an increase in the number of hydrogen atoms, which indicates hydrogen cluster nucleation on the surface of bilayer graphene. Hydrogenation of the system leads to the growth of the cubic diamond region up to the AA′ stacking promoting the formation of lonsdaleite with the \((10\bar{1}0)\) surface. This fact allows us to draw the conclusion about the adequacy of the potential obtained.