Machine learning interatomic potentials: A new tool for developing high-entropy perovskites for solid oxide fuel cells, protonic ceramic fuel cells and beyond

High-entropy perovskites (HEPs) have emerged as promising cathode candidates for solid oxide fuel cells (SOFCs) and protonic ceramic fuel cells (PCFCs) as they combine thermodynamic stabilization through configurational entropy with tunable transport and electrocatalytic properties. However, the vast compositional space renders empirical trial-and-error discovery prohibitively slow. In this perspective, we argue that an integrated computational pipeline, comprising machine-learning interatomic potential (MLIP) coupled with Bayesian optimization for multi-objective compositional search, can significantly accelerate the rational design of SOFC and PCFC cathodes. We describe how MLIPs can be leveraged to obtain defect energetics for HEP cathodes. Then, we outline how Bayesian optimization navigates the resulting high-dimensional property landscape to identify Pareto-optimal compositions balancing key functional properties. Finally, we propose promising research directions for catalyzing rapid and reproducible discovery of cathodes in fuel cells and beyond.