| Abstract Scope |
Addressing future energy demands will require materials that embrace chemical complexity. AI-driven approaches are accelerating this search across catalysis and nuclear waste immobilization. We integrate machine learning, computation, synthesis, and characterization in a closed discovery loop to efficiently navigate high-dimensional composition spaces. Recent work shows that structural accuracy in machine-learned interatomic potentials does not ensure electronic fidelity, motivating hybrid ML-DFT workflows for reliable property prediction. Using this framework, we identify platinum-free high-entropy catalyst candidates and discover a new waste form for iodine immobilization with promising initial performance. These results demonstrate how data-driven design can leverage disorder as a controllable principle for complex materials discovery. |