Machine Learning–Driven Insights Into the Self‐Assembly of Gold–Thiolate Nanoclusters

ABSTRACT Thiolate‐protected gold nanoclusters occupy a unique domain between molecular complexes and metallic nanoparticles, exhibiting size‐specific electronic structure, ligand‐mediated stability, and emergent collective behavior during synthesis. This work introduces and validates a machine learning‐driven strategy to predict atomistic coalescence mechanisms of self‐assembly of gold–thiolate nanoclusters across a significant size range of products (–) and thermal conditions (500–700 K). We used an atomic cluster expansion interatomic potential trained on density functional theory data, enabling molecular dynamics simulations up to 0.1 µs timescale. By systematically simulating both homo‐ and heterocoalescence reactions, we identify a hierarchical reaction network in which ligand dynamics, transient metal exposure, and geometric compatibility jointly govern selection of fusion pathways. Generating and analyzing 87 distinct coalescence products, we identify size‐dependent fusion behaviors and ligand‐gated reactive windows that govern hierarchical cluster growth. Quantitative comparison with experimentally identified metal–ligand stoichiometries validates that our simulations reproduce experimentally observed surface‐to‐volume scaling laws and ligand coverage trends spanning more than a 20‐fold range of cluster sizes. The robustness of these predictions is demonstrated through statistical analysis of reaction trajectories and structural motif evolution. Our strategy can be generalized to other ligand‐protected metal nanoclusters and nanoparticle systems, provided sufficient training data is generated, offering a predictive framework for rational design of size‐focused synthesis, controlled aggregation, and hierarchical assembly in nanocluster‐based materials.