Data-driven simulation and characterisation of gold nanoparticle melting.

Area of Science:

• Materials Science
• Computational Chemistry
• Nanotechnology

Background:

• Accurate simulation of nanoparticle thermal stability is crucial for technological applications.
• Developing efficient and transferable force fields is essential for molecular dynamics simulations.
• Density Functional Theory (DFT) provides high-accuracy data but is computationally expensive for large systems.

Purpose of the Study:

• To develop efficient, transferable, and interpretable machine learning force fields for gold nanoparticles.
• To investigate the thermodynamic stability and solid-liquid phase change of gold nanoparticles using molecular dynamics.
• To characterize the melting mechanism and identify key atomic arrangements during phase transitions.

Main Methods:

• Machine learning force field development using DFT-calculated data.
• Molecular dynamics simulations of gold nanoparticles (1-6 nm, up to 6266 atoms).
• Unsupervised learning for categorizing local atomic environments and analyzing phase transitions.

Main Results:

• Predicted melting temperatures for gold nanoparticles show good agreement with experimental data.
• Identified a solid-liquid phase change mechanism through simulation.
• Characterized distinct liquid atomic arrangements at the nanoparticle surface and interior.

Conclusions:

• Machine learning force fields offer an efficient and accurate approach for simulating gold nanoparticle thermal stability.
• Melting of gold nanoparticles initiates at the outer layers, driven by surface phenomena.
• The study provides a data-driven understanding of nanoparticle melting and liquid-like atomic structures.