A machine learning potential for simulating infrared spectra of nanosilicate clusters

Area of Science:

• Chemical Physics
• Computational Chemistry
• Materials Science

Background:

• Machine learning (ML) models can achieve the accuracy of ab initio methods for interatomic potentials at a lower computational cost.
• Efficient generation of training data is crucial for developing accurate ML interatomic potentials.
• Nanosilicate clusters are important in astrophysical environments, but their properties require accurate simulation methods.

Purpose of the Study:

• To develop a neural network-based ML interatomic potential for nanosilicate clusters.
• To establish an efficient protocol for generating training data for ML potentials.
• To enable the simulation of nanosilicate clusters and extraction of their anharmonic infrared spectra.

Main Methods:

• An accurate and efficient protocol for collecting training data was applied.
• Initial training data were generated using normal modes and farthest point sampling.
• An active learning strategy, employing an ensemble of ML models, was used to extend the training dataset.
• Parallel sampling over structures accelerated the data generation process.
• Molecular dynamics simulations were performed using the developed ML potential.

Main Results:

• A neural network-based ML interatomic potential for nanosilicate clusters was successfully constructed.
• The ML potential achieved high accuracy comparable to ab initio methods with reduced computational cost.
• The protocol efficiently generated necessary training data, including through active learning.
• Molecular dynamics simulations of nanosilicate clusters of various sizes were performed.
• Anharmonic infrared spectra were extracted from the simulations.

Conclusions:

• The developed ML potential accurately describes nanosilicate clusters.
• The efficient data generation protocol, including active learning, is effective for ML potential development.
• The extracted anharmonic infrared spectra provide insights into silicate dust properties in interstellar and circumstellar environments.
• This work facilitates a deeper understanding of astrophysical chemical physics.