Elucidating oxide-ion and proton transport in ionic conductors using machine learning potentials.

Area of Science:

• Materials Science
• Computational Chemistry
• Electrochemistry

Background:

• Developing solid electrolytes for efficient ion transport is key for solid oxide fuel cells operating below 600°C.
• Atomistic modeling and machine learning accelerate the design and understanding of ionic conductors.

Purpose of the Study:

• To develop accurate moment tensor potentials (MTPs) for Ba7Nb4MoO20 and Sr3V2O8 using machine learning.
• To validate MTPs against ab initio calculations and experimental data for oxide-ion and proton transport.

Main Methods:

• Utilized passive and active learning techniques to create MTPs.
• Performed ab initio molecular dynamics and density functional theory calculations.
• Compared MTP predictions with experimental data for diffusion coefficients and conductivities.

Main Results:

• MTPs accurately reproduced ab initio molecular dynamics data and DFT results for forces, energies, and stresses.
• Predicted diffusion coefficients and conductivities for oxide ions and protons with excellent agreement with experiments.
• Accurately estimated migration barriers, demonstrating MTP robustness and transferability.

Conclusions:

• Machine learning-derived MTPs offer a computationally efficient and accurate approach for simulating ion transport.
• These MTPs are valuable tools for designing next-generation solid oxide fuel cells.