Machine Learning-Accelerated First-Principles Study of Atomic Configuration and Ionic Diffusion in Li₁₀GeP₂S₁₂ Solid...

Area of Science:

• Materials Science
• Solid-state Chemistry
• Computational Materials Science

Background:

• Lithium germanium phosphorus sulfide (LGPS) is a key solid electrolyte for all-solid-state batteries.
• LGPS properties depend on ground-state structures, but numerous configurations arise from site disorder and fractional occupancy.
• Current methods for identifying stable structures are computationally intensive.

Purpose of the Study:

• To efficiently predict the most stable atomic configuration of LGPS.
• To investigate the influence of different configurations on Li ion diffusion.
• To evaluate the LAsou method for materials discovery.

Main Methods:

• Utilized the machine learning- and active-learning-based LAsou method combined with first-principles calculations.
• Employed ab initio molecular dynamics to study Li ion diffusion from 500-900 K.
• Screened candidate structures using electrostatic energy criteria.

Main Results:

• Successfully predicted the most stable LGPS configuration.
• Demonstrated that atomic configurations and Li ion distribution significantly affect Li ion diffusion.
• Observed variations in Li ion diffusion coefficients across different structures.

Conclusions:

• The LAsou method is effective for predicting stable solid electrolyte structures.
• LGPS structure significantly influences Li ion mobility, impacting battery performance.
• LAsou accelerates theoretical calculations and aids in designing new solid electrolytes.