Computational Prediction of Proton-Conducting Perovskite Properties Under Realistic Conditions

Abstract

Listen

Translate article

Measurements of proton-conducting Perovskite properties under extreme conditions remain challenging due to the limitation of experimental techniques. As a powerful tool, molecular dynamic (MD) simulation can provide abundant information on well-defined systems under realistic conditions. However, quantum-chemistry-based MD simulations are computationally expensive, while traditional potentials lack accuracy and training efficiency, especially for multi-element systems like Perovskite. Recently, machine learning potential (MLP) has attracted much attention since this tool empowers MD simulations with accuracy and efficiency. In this presentation, we will discuss the properties prediction of INL-developed benchmarking Perovskite, PrNi1-δCoδO3 (PNC), as an example based on density functional theory (DFT) and a well-trained sparse Gaussian process (SGP) potential. Properties, including expansion, diffusivity, and mechanical properties, which significantly influence the electrochemical cell efficiency and durability, were predicted under extreme conditions. The established protocol provides a versatile avenue for future proton-conducting perovskite investigations.