Data-driven discovery of lunar Fe-bearing oxides for solar-wind-derived water production

The exploration and discovery of lunar water reservoirs, particularly in sunlit regions, remain challenging due to an incomplete understanding of water formation mechanisms via hydrogen retention. Here, a data-driven screening workflow is proposed to identify lunar-based oxides (Si-Al-Mg-Ca-Fe-Ti-Na-O) that can retain solar-wind-implanted H and promote H 2 O production. Machine-learning (ML) models trained on spin-polarized electronic-structure metrics, local geometries, and magnetic moments predict reaction energies for H insertion (HIR) and H 2 O formation (WFR) with high accuracy. The results reveal a decisive role of high-/mixed-valence Fe-bearing oxides, defined by the oxidation states from stoichiometry, in enabling both H retention and H 2 O production. In contrast, in-situ Fe 2+ -bearing lunar minerals are generally unfavorable for H insertion and/or H 2 O production. Applying four material screening criteria, i.e., HIR energy, WFR energy, H + diffusion length, and fugacity ratio of H 2 O to H 2 , identifies eleven high-/mixed-valence Fe-bearing oxides with lunar magnetite as promising reservoirs for solar-wind-derived H 2 O production. This work demonstrates how the electronic-structure-informed ML approach can accelerate the discovery of lunar oxide hosts for implantation-driven hydrogen chemistry and lunar in situ resource utilization.