Study of Li atom diffusion in amorphous Li₃PO₄ with neural network potential

Area of Science:

• Materials Science
• Computational Chemistry
• Solid-State Physics

Background:

• Atomic diffusion in amorphous materials is crucial for developing advanced information and energy devices.
• Reliable and computationally efficient theoretical methods are needed to understand this diffusion process.
• Amorphous lithium phosphate (Li3PO4) serves as a key material for these applications.

Purpose of the Study:

• To investigate the application of neural network (NN) potentials for studying atomic diffusion in amorphous materials.
• To benchmark the accuracy and speed of NN potentials against Density Functional Theory (DFT) calculations.
• To characterize Li vacancy diffusion and ion transport properties in amorphous Li3PO4.

Main Methods:

• Application of neural network potentials combined with nudged elastic band, kinetic Monte Carlo, and molecular dynamics.
• Comparison of NN potential results with Density Functional Theory (DFT) calculations.
• Utilizing large supercell models (>1000 atoms) for comprehensive analysis.

Main Results:

• NN potentials exhibit high accuracy, with average errors of 0.048 eV for energy barriers and 0.041 eV for activation energy compared to DFT.
• NN potential simulations are 3-4 orders of magnitude faster than DFT, while maintaining consistency with ab initio molecular dynamics.
• Observation of P2O7 unit formation in amorphous Li3PO4, aligning with experimental findings.
• Estimated Li diffusion activation energy of 0.55 eV, which closely matches experimental values.

Conclusions:

• Neural network potentials provide a reliable and computationally efficient method for studying atomic diffusion in amorphous materials.
• The developed NN potential accurately predicts diffusion behavior and structural properties of amorphous Li3PO4.
• This approach facilitates the design and optimization of novel information and energy devices based on amorphous materials.