Evaluation of Machine Learning Interatomic Potentials for Gold Nanoparticles-Transferability towards Bulk.

Area of Science:

• Materials Science
• Computational Chemistry
• Nanotechnology

Background:

• Accurate modeling of gold nanoparticles is essential for various applications.
• Machine learning interatomic potentials (ML-IPs) offer a promising approach for simulating large systems.
• Challenges remain in understanding the transferability and data requirements for ML-IPs.

Purpose of the Study:

• To analyze the efficacy of ML-IPs for modeling gold (Au) nanoparticles.
• To explore the transferability of ML models to larger systems.
• To establish simulation time and size thresholds for accurate ML-IPs.

Main Methods:

• Compared energies and geometries of large Au nanoclusters using VASP and LAMMPS.
• Determined VASP simulation timesteps needed for ML-IPs to reproduce structural properties.
• Investigated minimum training set atomic size for accurate ML-IPs using LAMMPS.

Main Results:

• Identified necessary simulation timesteps and training set sizes for reliable ML-IPs.
• Demonstrated that ML-IPs can be adapted for different gold nanocluster systems with minor adjustments.
• ML-IPs show potential for accurate replication of structural properties in large nanoclusters.

Conclusions:

• ML-IPs are effective for modeling gold nanoparticles.
• Transferability of ML-IPs can be achieved with minor modifications.
• This work provides insights into developing accurate ML-IPs for nanoparticle simulations.