Spherical harmonics based descriptor for neural network potentials: Structure and dynamics of Au₁₄₇ nanocluster

Area of Science:

• Computational chemistry
• Materials science
• Nanotechnology

Background:

• Accurate prediction of nanocluster structures is crucial for understanding their properties.
• Traditional methods like Density Functional Theory (DFT) are computationally expensive for large clusters.

Purpose of the Study:

• To develop a highly efficient and accurate method for fitting the potential energy surface of nanoclusters.
• To determine the global minimum structure and dynamics of large gold clusters (Au147).

Main Methods:

• Integration of a spherical harmonics-based descriptor with an artificial neural network.
• Global optimization and molecular dynamics simulations for Au147.
• Validation of results using Density Functional Theory (DFT).

Main Results:

• Achieved accuracy comparable to quantum mechanics with the speed of empirical potentials.
• Computational time for Au147 energy and force calculations reduced to ~1.7 seconds.
• Identified a global minimum structure for Au147 that is 4 eV lower in energy than the icosahedral structure.
• The global minimum structure consists of 105 surface atoms and 42 core atoms.
• Au147 exhibits a dynamic surface, indicating significant fluxionality.

Conclusions:

• The proposed method offers a significant speed-up for accurate nanocluster simulations.
• The global minimum structure of Au147 is not an icosahedron, challenging previous assumptions.
• The dynamic surface of Au147 opens new avenues for studying its reaction dynamics.