Machine Learning Design of Perovskite Catalytic Properties

Ryan Jacobs,Jian Liu,Harry Abernathy,Dane Morgan

2023-11-03

Abstract:Discovering new materials that efficiently catalyze the oxygen reduction and evolution reactions is critical for facilitating the widespread adoption of solid oxide fuel cell and electrolyzer (SOFC/SOEC) technologies. Here, we develop machine learning (ML) models to predict perovskite catalytic properties critical for SOFC/SOEC applications, including oxygen surface exchange, oxygen diffusivity, and area specific resistance (ASR). The models are based on trivial-to-calculate elemental features and are more accurate and dramatically faster than the best models based on ab initio-derived features, potentially eliminating the need for ab initio calculations in descriptor-based screening. Our model of ASR enables temperature-dependent predictions, has well calibrated uncertainty estimates and online accessibility. Use of temporal cross-validation reveals our model to be effective at discovering new promising materials prior to their initial discovery, demonstrating our model can make meaningful predictions. Using the SHapley Additive ExPlanations (SHAP) approach, we provide detailed discussion of different approaches of model featurization for ML property prediction. Finally, we use our model to screen more than 19 million perovskites to develop a list of promising cheap, earth-abundant, stable, and high performing materials, and find some top materials contain mixtures of less-explored elements (e.g., K, Bi, Y, Ni, Cu) worth exploring in more detail.

Materials Science

What problem does this paper attempt to address?

The main objective of this paper is to predict the catalytic performance of perovskite materials using machine learning (ML) methods, particularly for applications in solid oxide fuel cells and electrolyzers (SOFC/SOEC) technology. Specifically, the research team developed machine learning models based on simple, computable elemental features to predict key catalytic properties, including oxygen surface exchange, oxygen diffusion rate, and area-specific resistance (ASR). These models outperform traditional ab initio methods (such as the O p-band center obtained from density functional theory calculations) in terms of prediction accuracy and have significant advantages in computational speed.

Additionally, the paper demonstrates that its ASR model can perform temperature-dependent predictions and provides well-calibrated uncertainty estimates. Using a time cross-validation method, the model was shown to effectively identify potential high-performing materials before they were first discovered. Finally, the research team used the model to screen approximately 19 million perovskite compounds, identifying a series of highly catalytic, inexpensive, and stable material candidates. Some of these materials contain less-explored elemental combinations (such as potassium, bismuth, yttrium, nickel, copper, etc.), which are worth further detailed study. Overall, this paper aims to accelerate the design and discovery of efficient perovskite catalysts through machine learning methods.