Data-Efficient Design of High-Entropy Oxygen Carriersfor Chemical Looping Using Active Learning

Abstract

Translate Article

High-entropy materials,first demonstrated in metallic alloys andlater extended to oxides and other systems, unlock a vast compositionalspace with properties suited for catalysis, energy, and structuralmaterials. However, the high compositional complexity makes systematicexploration challenging, and only a small portion of the design spacehas been studied. To address this, we introduce an active learningstrategy that integrates predictive modeling, uncertainty estimation,and iterative sampling to efficiently navigate embedded compositionalmaterial spaces. This approach continuously learns from previous evaluations,focusing subsequent searches on the most promising regions while reducingboth time and data requirements. We demonstrate this methodology inthe search for high-entropy oxygen carriers for chemical looping,where it rapidly accelerates discovery and identifies promising candidatesmore effectively than conventional trial-and-error or grid-searchapproaches. Importantly, this strategy is general and well-suitedto exploring the vast space of multicomponent materials.