Decarbonizing aviation with sustainable drop-in fuels is essential for credible net-zero pathways. However, assembling energy-dense jet fuel hydrocarbons from C1 feedstocks demands multifunctional catalysts whose performance arises from nonlinear, high-dimensional interactions─beyond single-descriptor design rules. Here we present a dual-engine artificial intelligence framework that couples closed-loop active learning with interpretable machine learning, demonstrated for syngas conversion to sustainable aviation fuel. The approach autonomously explores vast catalyst spaces while distilling human-interpretable principles. We identify previously unreported compositions and a general rule: on a stable spinel backbone, placing a d-block metal at the tetrahedral (A)-site and an early lanthanide at the octahedral (B)-site creates cooperative d-f interactions that enable π-back-donation into the π* orbitals of oxygenated intermediates, strengthening adsorption and lowering formation barriers to accelerate intermediate generation and C-C coupling to jet-range aromatics. Guided by this rule, active sites such as Zn-Ce/Sm, Fe-Pr/La, and Ni-Ce achieve >75% selectivity to jet-fuel-range aromatic hydrocarbons with high space-time yields. Overall, the dual-engine approach not only accelerates discovery but also yields transparent, experimentally validated design rules─a generalizable blueprint for interpretable, AI-enabled catalyst design in complex sustainable chemistries.
Tian et al. (Mon,) studied this question.