Substitutional effects at A- and B-sites in high-entropy ABO3 perovskites: Insights from machine learning-accelerated simulations and active learning

High-entropy perovskites (HEPs) have emerged as a promising class of functional materials, where the high configurational entropy from multiple principal elements stabilizes complex structures and enables exceptional properties. However, understanding substitutional effects in HEPs remains challenging due to the vast compositional space and intricate element-property interactions. In this study, we present a comprehensive framework that integrates machine learning (ML)-accelerated atomic simulations with active learning to systematically investigate the substitutional effects in HEPs. Using LaCoO 3 as the base structure, both A-site and B-site five-element chemistries are explored, and four key properties are evaluated: formation energy, lattice distortion, atomic distortion, and diffusion coefficient. A multi-target Bayesian Neural Network (BNN) model is developed to predict these properties with quantified uncertainties, while the ParEGO acquisition function guides the active learning process to iteratively refine the models and expand the dataset. The final models enable large-scale property prediction across the compositional space, and Pareto front screening identifies optimal element combinations that balance trade-offs among the target properties. To further facilitate user-friendly exploration, an interactive web application, HEP-Explorer, is developed. This framework provides valuable insights into site-dependent substitutional effects and offers practical guidelines for the experimental design of HEPs with tailored performance for various applications.