Machine Learning of Atomic Dynamics and Statistical Surface Identities in Gold Nanoparticles

It is known that metal nanoparticles (NPs) may be dynamic and atoms may move within them even at fairly low temperatures. Characterizing such complex dynamics is key for understanding NPs’ properties in realistic regimes, but detailed information on, e.g., the stability, survival, and interconversion rates of the atomic environments (AEs) populating them are non-trivial to attain. Here we show a machine learning approach that allows us to reconstruct the complex atomic dynamics of metal NPs from high-dimensional data extracted from molecular dynamics simulations. Using different-shape gold NPs as a representative example, an AEs’ dictionary allows us to label step-by-step the individual atoms in the NPs, identifying the native and non-native AEs and populating them along the MD simulations at various temperatures. Tracking the emergence, annihilation, lifetime, and dynamic interconversion of the AEs, our approach permits estimating a “statistical equivalent identity” for metal NPs based on the intrinsic atomic dynamics present within them.