Machine learning potential-accelerated multiscale dynamical simulations of nanodiamond structural reconstruction

Abstract

Atomistic understanding of structural transformations in nanodiamonds (NDs) is vital for manipulating their physicochemical properties, yet remains limited due to the inherent trade-off between simulation accuracy and scale. Here, we develop a machine learning potential (MLP) with density functional theory accuracy and implement it within the deep potential molecular dynamics framework to enable large-scale simulations of NDs comprising 103–104 atoms over nanosecond timescales. Our simulations reveal that the transformation dynamics are governed by morphology, surface facets, particle size, and temperature. We identify a multistage transformation pathway, sequentially characterized by outward-in graphitization, inward-out atomic migration, and a subsequent self-healing process, driven by surface energy minimization and internal stress relaxation. These results provide atomistic insight into the evolution of NDs and demonstrate the power of MLP-based approaches for modeling complex, multiscale structural transformations in nanocarbon materials.