Origin of the surface facet dependence in the thermal degradation of the diamond (111) and (100) surfaces in vacuum investigated by machine learning molecular dynamics simulations

We perform machine learning molecular dynamics simulations to gain an atomic-level understanding of the dependence of the graphitization and thermal degradation behavior of diamond to the (111) and (100) surface facets. The interatomic potential is constructed using graph neural network model, trained using energies and forces from spin-polarized van der Walls-corrected density functional theory calculations. Our results show that the C(111) surface is more susceptible to thermal degradation, which occurs from 2850 K through synchronized bilayer exfoliation mechanism. In comparison, the C(100) surface thermally degrade from a higher temperature of 3680 K through the formation of sp1 carbon chains and amorphous sp2-sp3 carbon network. Due to the dangling bonds at the step edges, the stepped surfaces are more susceptible to thermal degradation compared to the corresponding flat surfaces, with the stepped C(111) and C(100) surfaces thermally degrading from 1810 K and 3070 K, respectively. We propose potential applications of this study in diamond tool wear suppression, diamond polishing, and production of graphene directly from the diamond surface.