Atomistic Details of Nanocluster Formation from Machine-Learned-Potential-Based Simulations

Area of Science:

• Nanoscience and Nanotechnology
• Computational Chemistry
• Materials Science

Background:

• Understanding metal nanocluster formation mechanisms is crucial but challenging.
• Existing computational methods face limitations in accuracy, system size, and simulation time for solution-based nucleation.
• Molecular details of nanocluster nucleation are often inaccessible through experimental methods alone.

Purpose of the Study:

• To develop and validate a computational approach for simulating metal nanocluster nucleation in solution.
• To overcome the limitations of traditional simulation techniques in terms of accuracy and scale.
• To provide molecular-level insights into the nucleation process of a model silver nanocluster.

Main Methods:

• Combined deep neural networks (DNNs) with well-tempered metadynamics (WT-MetaD) for enhanced molecular simulations.
• Employed neural-network-potential-based molecular dynamics to capture dynamic behavior.
• Simulated the nucleation of a silver nanocluster (Ag₆(SCNH₂)₆) in methanol, including scaling to 30 precursors.

Main Results:

• Achieved density-functional-theory (DFT)-level accuracy in modeling nanocluster formation in solution for the first time using DNNs.
• WT-MetaD simulations revealed an almost barrierless nucleation pathway from dispersed precursors to a stable cluster.
• Demonstrated spontaneous nucleation even with a larger system of 30 randomly distributed precursors, highlighting method robustness.

Conclusions:

• The integrated DNN and WT-MetaD approach successfully models metal nanocluster nucleation in solution.
• This computational strategy overcomes significant challenges in simulating complex nucleation processes.
• The study paves the way for advanced computational investigations into nanocluster formation and design.