Gaussian Process Approach to Constructing Transferable Force Fields for Thiolate-Protected Gold Nanoclusters.

Area of Science:

• Materials Science
• Computational Chemistry
• Nanotechnology

Background:

• Gold nanoparticles possess unique properties like plasmon resonances and photoluminescence.
• High atom counts in nanoparticles lead to prohibitive computational costs for traditional methods like density functional theory (DFT).

Purpose of the Study:

• To develop accurate and efficient machine learning force fields for gold thiolate-protected nanoclusters.
• To reduce the computational expense of simulating gold nanoclusters using molecular dynamics (MD).

Main Methods:

• Utilized the FLARE++ code with an active learning algorithm to construct force fields.
• Trained the force field initially on Au20(SCH3)16 and subsequently retrained with validation data from diverse gold nanoclusters.
• Performed molecular dynamics (MD) simulations to validate the machine learning force field's accuracy.

Main Results:

• The developed force fields accurately predict energies for gold nanoclusters within and outside the training dataset.
• Achieved quantum mechanical level accuracy in key performance metrics for nanocluster simulations.
• Demonstrated the efficacy of active learning in accelerating the development of reliable force fields for nanomaterials.

Conclusions:

• Machine learning force fields, particularly when enhanced by active learning, offer a computationally efficient alternative to DFT for gold nanoclusters.
• The developed force fields enable accurate large-scale molecular dynamics simulations of gold nanoclusters.
• This work paves the way for more extensive computational studies of functionalized gold nanomaterials.