Deep potential for a face-centered cubic Cu system at finite temperatures

Area of Science:

• Materials Science
• Computational Chemistry
• Condensed Matter Physics

Background:

• Neural network potentials are crucial for accurate molecular dynamics simulations.
• The quality of these potentials depends heavily on training data and processes.
• Comprehensive verification is needed to ensure model accuracy.

Purpose of the Study:

• To develop a highly accurate neural network potential for face-centered cubic (FCC) copper.
• To validate the potential's performance across different temperatures and conditions.
• To assess the potential's generalization capabilities for predicting material properties.

Main Methods:

• Utilized first-principle calculations to generate comprehensive training datasets for FCC Cu.
• Employed Deep Potential Molecular Dynamics (DeePMD) for the neural network potential training process.
• Verified the potential's accuracy through reproduction of known properties and prediction of new ones.

Main Results:

• The developed neural network potential accurately reproduces various properties of FCC Cu at 0 K.
• The potential demonstrates good performance at finite temperatures, predicting elastic constants and melting point.
• The potential exhibits enhanced generalization capacity, predicting grain boundary energy without specific training data.

Conclusions:

• The developed neural network potential for FCC Cu is accurate and reliable for molecular dynamics simulations.
• The method shows applicability under more practical conditions, including finite temperatures and complex structures.
• This work highlights the effectiveness of machine learning potentials for materials science research.