A high-dimensional neural network potential for Co₃O₄.

Area of Science:

• Materials Science
• Computational Chemistry
• Catalysis

Background:

• Cobalt oxide (Co3O4) spinel is crucial for oxidation catalysis.
• Accurate simulation of Co3O4 properties requires precise atomic interaction descriptions.
• Existing methods struggle to model Co3O4 complexity, limiting simulation scales.

Purpose of the Study:

• To develop a machine learning potential (MLP) for Co3O4 spinel.
• To enable large-scale molecular dynamics simulations of Co3O4.
• To investigate the temperature-dependent properties of Co3O4.

Main Methods:

• Developed a high-dimensional neural network potential (HDNNP) for Co3O4.
• Trained the MLP using density functional theory (DFT) calculations.
• Validated the MLP by computing structural, vibrational, and dynamical properties.

Main Results:

• Successfully constructed and validated an MLP for bulk Co3O4 spinel.
• Simulations revealed key temperature-dependent properties, including thermal expansion.
• The MLP accurately captures the complex interactions in Co3O4.

Conclusions:

• MLPs offer a powerful approach to simulate complex materials like Co3O4.
• This work provides a validated tool for studying Co3O4 catalysis under realistic conditions.
• The findings advance computational materials science for catalytic applications.