Molecular dynamics study of corrugation in low-defect graphene using machine learning potential

Abstract Machine learning has significant promise to improve inter-atomic potentials for accurate, efficient, and more general atomic simulations. This study investigates the effect of low vacancy concentrations (0.05%–0.8%) on graphene corrugation using molecular dynamics simulations enhanced by a neural network potential. Our results demonstrate that the presence and quantity of vacancy defects significantly affect the formation of corrugation patterns on graphene surfaces. Specifically, we found that the type and concentration of defects play crucial roles; for instance, double vacancies result in more pronounced corrugation than single vacancies. Furthermore, maintaining an appropriate distance between the vacancy defects can reduce lattice corrugation. Interestingly, while it is generally expected that an increase in defect concentration correlates with heightened corrugation, our results indicate a counterintuitive trend at low defect concentrations (0.05%–0.2%). Specifically, within this range, an increase in defect concentration leads to a decrease in corrugation formation on graphene layers. This reduction may explain experimental observations, such as the increased Young’s modulus and altered Thermal Expansion Coefficient (TEC) reported in graphene with similar defect densities. However, at higher defect concentrations, a subsequent increase in corrugation is observed. These findings offer valuable insights for designing and enhancing graphene-based materials across various technological applications.