Machine Learning Prediction of the Entropy Formation Ability of Multicomponent REPO ₄ as Environmental Barrier Coatings

ABSTRACT

Multicomponent rare earth phosphates (REPO ₄ ) are promising environmental barrier coating (EBC) materials for SiC‐based ceramic matrix composites (CMCs), but identifying compositions that stabilize the desirable xenotime phase across the vast compositional space remains a major challenge. Conventional geometric descriptors, such as the mean cation radius (MR) and cation size disorder (CSD), provide only limited predictive capability because they do not fully capture the thermodynamic factors governing entropy stabilization. Here, we develop a machine learning (ML) framework trained on high‐throughput DFT‐calculated entropy formation ability (EFA), a computational metric that quantifies the propensity of a composition to form an entropy‐stabilized xenotime phase, for 232 REPO ₄ compositions. The optimized model achieves strong predictive performance and demonstrates good generalization to higher‐order quaternary and quinary compositions outside the training set. Feature analysis reveals that descriptors related to atomic volume, electronegativity variation, and electronic structure play critical roles in stabilizing the xenotime phase beyond purely geometric effects. Experimental synthesis and x‐ray diffraction studies confirm the formation of single‐phase xenotime in high‐EFA compositions, whereas low‐EFA compositions exhibit mixed or monazite phases. This framework enables rapid screening of the vast compositional space and elucidates how atomic size, packing, and bonding characteristics govern the entropy stabilization in multicomponent REPO ₄ , accelerating the discovery of next‐generation EBC materials for SiC‐based CMCs.