Capturing Dynamic Core Reconstruction and Ligand Desorption of Atomically Precise Ag Nanoclusters with Machine Learning Force Field

Atomically precise silver nanoclusters (NCs) protected by alkynyl ligands represent an emerging class of electrocatalysts demonstrating high activity and selectivity in reactions, such as CO₂ electroreduction. However, their dynamic structural evolution mechanisms under electrochemical operating conditions remain elusive. Conventional experimental characterization faces a grand challenge to resolve atomic-scale dynamic processes, while ab initio molecular dynamics (AIMD) simulations are solely confined to picosecond time scales, insufficient for capturing the dynamics of evolution over longer times. Combining multiscale constant potential simulations and a deep potential molecular dynamics (DPMD) scheme, here we developed a high-accuracy machine learning force field within the deep-learning framework to elucidate the electrochemical structural evolution in all-alkynyl-protected Ag₁₅ NC and its doped systems (Ag₈Au₇, Ag₉Cu₆, and Ag₁₄Cl NCs). We found that the metal cores of all NCs undergo a transition from octahedral to disordered, accompanied by partial or complete cleavage of surface alkynyl ligands. The dopants critically modulate the stability by regulating desorption pathways, with Ag₉Cu₆ NC exhibiting exceptional resistance to dissociation due to robust Cu-C bonding. Our nanosecond-level DPMD simulations based on trained machine learning force fields further confirmed that doping dramatically affects the number of desorbed alkyne ligands (4 for Ag₁₅, 6 for Ag₈Au₇, and 8 for Ag₁₄Cl) and the degree of core ordering, and a long-term simulation of >2000 ps was crucial for capturing the dynamic electrochemical interface. This study established the first quantitative correlation between electrochemical interface dynamics and doping effects, providing a theoretical paradigm for designing highly stable atomically precise catalysts.