The Thiolate Shell as a Dynamic Gold Reservoir: Size-Dependent Thermal Restructuring of Gold Nanoclusters from MACE-Based Machine-Learning Molecular Dynamics

Atomically precise thiolate-protected gold nanoclusters are used in catalysis, sensing and nanomedicine, where thermal activation, reaction conditions or local heating can change their atomic structure. Whether the gold–thiolate shell remains a fixed protective layer at elevated temperature, or takes part in the restructuring of the cluster, is still an open question, because first-principles molecular dynamics is limited to small clusters and short times. Here we train MACE-AuSCH, an equivariant machine-learned interatomic potential, on a published density functional theory dataset and validate it against 16 crystallographically characterised clusters (cohesive-energy RMSE of 2.1 meV/atom). We then follow seven Aun(SCH3)m clusters, from Au25 to Au279 and with icosahedral and FCC-derived cores, between 8 and 800 K. In all clusters, the effective stiffness of the Au–S bonds is 3.3 to 7.5 times that of the Au–Au bonds, and the metallic Au–Au network reorganises first while most of the Au–S connectivity is kept. At 500 K, the largest core displacements occur in the FCC-derived Au52 and Au92. At 800 K, Au atoms move in both directions between the metallic core and the staple shell, and the loss of Au–S connectivity generally grows with cluster size, from 3% for Au25 to about 40% for Au144. These results describe the protecting shell as a dynamic reservoir of Au atoms and point to an intracluster route that may precede ligand and metal exchange, fragmentation and sintering.