Flexible machine-learning interatomic potential for simulating structural disordering behavior of Li₇La₃Zr₂O₁₂ solid...

Area of Science:

• Materials Science
• Computational Chemistry
• Energy Storage

Background:

• Solid-state electrolytes, such as lithium lanthanum zirconium oxide (LLZO), offer enhanced safety and energy density for batteries.
• Atomic disorder at interfaces in LLZO significantly degrades battery performance.
• Machine-learning (ML) interatomic potentials can simulate complex interfaces with high accuracy and scalability.

Purpose of the Study:

• To develop and validate a machine-learning interatomic potential for simulating crystalline, disordered, and amorphous LLZO.
• To enable accurate modeling of atomic disorder and its impact on LLZO properties and performance.
• To accelerate simulations of complex phenomena in solid-state battery materials.

Main Methods:

• Construction of a neural network-based ML potential trained on ab initio data.
• Validation of the ML potential against ab initio simulations for structural, vibrational, elastic, and transport properties.
• Application of the ML potential to simulate grain boundary effects and thermal transitions in LLZO.

Main Results:

• The developed ML potential accurately predicts structural, vibrational, elastic properties, and Li diffusivity of LLZO.
• Simulations show the potential correctly captures grain boundary effects on Li diffusivity.
• The ML potential accurately models thermal transition behavior and enables quantum-accurate simulations thousands of times faster than ab initio methods.

Conclusions:

• The ML potential provides a powerful tool for simulating disordered LLZO systems and interfaces.
• This approach overcomes limitations of traditional methods for studying complex battery materials.
• The developed potential facilitates accelerated discovery and optimization of solid-state battery electrolytes.