Accelerated Discovery of Proton-Conducting Perovskites through Density Functional Theory and Machine Learning

Abstract

Hydrogen is an important energy carrier resource in response to limiting greenhouse gas emissions. Proton-conducting perovskite oxide is one of the key materials for highly efficient carbon-neutral hydrogen technologies, such as hydrogen production, CO2 hydrogenation, and ammonia synthesis. Many attempts have been made based on doped perovskites made of well-tested materials, such as BaZrO3, BaCeO3, BaHfO3, BaTiO3, and SrZrO3. However, the resulting perovskites have often suffered stability and conductivity problems. Furthermore, complex phenomena occurring during hydration present challenges for expanding the materials library. Herein, we demonstrate accelerated discovery of proton-conducting perovskites with high conductivity using machine learning (ML) predictions. We constructed consistent training data using density functional theory (DFT) which enable high accuracy of ML model. DFT computations were performed on > 1000 doped perovskite compositions to get their properties of lattice parameters, point defects (e.g., O vacancies, H interstitials), density of states, hydration energy, and proton migration energy. Several ML algorithms including Linear Regression, Bayesian Ridge Regression, Random Forest Regression, Neural networks, and k-Nearest Neighbor were tested for minimum errors and coefficient of determination. The multidimensional relationships between a set of >50 features and conductivity were mapped out using the optimized ML model. We screened a large material space of A-site and B-site doped perovskites to predict potential proton-conducting materials for various energy applications. The outcomes are promising for accelerating the design and applications of proton-conducting perovskite oxides in hydrogen technologies.