Thermal dynamics and coalescence of gold nanoclusters: Insights from machine learning-accelerated simulations - Maryam Sabooni Asre Hazer

Understanding the finite-temperature dynamics of ligand-protected metal nanoclusters is essential for understanding their catalytic activity and stability, yet remains challenging due to the prohibitive computational cost of extended-timescale density functional theory (DFT) simulations. Here, I present a study employing machine learning-derived interatomic potentials, the Atomic Cluster Expansion (ACE) framework trained on DFT data to perform molecular dynamics simulations of Au₁₄₄(SR)₆₀ nanoclusters at temperatures ranging from 300 to 550 K over timescales extending to 0.12 microseconds, representing a five-order-of-magnitude improvement over conventional ab initio methods. Our simulations reveal temperature-induced structural evolution proceeding via layer-by-layer mobility enhancement, leading to spontaneous formation and detachment of polymeric gold-thiolate chains and cyclic motifs at the cluster surface that produce experimentally observed cluster compositions. We also capture the complete coalescence of two clusters, yielding structures remarkably similar to experimental results. These findings provide atomistic insights into thermal activation of nanoclusters for catalysis and demonstrate how machine learning is enabling rational design of nanomaterials by efficiently exploring vast chemical spaces previously inaccessible to both experiment and theory.