Structural Origin for the Enhanced Photocatalysis of Zr–Ti Oxide Composites from Machine Learning

Multielement oxides are notorious for their structural complexity, offering ample room for structure engineering toward applications. Here we demonstrate, via comprehensive machine learning (ML) based techniques to explore millions of likely structures on the Zr–Ti oxide global potential energy surface, that the Zr–Ti oxide with varied Zr:Ti ratios can cover a full range of composite structure types, including the cation-disordered solid solution, the cation-ordered homogeneous crystals, and heterophase junctions. We show that the three-phase junctions are the global minima at x = 0.375 and 0.625 of Zr x Ti 1– x O 2, constituted by a monoclinic ZrO 2 phase, a thin layer of orthorhombic ZrTi 2 O 6 phase, and an anatase TiO 2 phase. This three-phase junction has an orientation relationship (OR) of (100) M //(100) O //(112) A;[001] M //[001] O //[111̅] A and low interfacial energy (0.21 J/m 2 with respect to monoclinic ZrO 2 and anatase TiO 2 ), and more importantly, it possesses a unique electronic structure of Type II band features, where the valence band maximum (VBM) is mainly located at the interface O lattice sites near the monoclinic-ZrO 2 phase and the conduction band minimum (CBM) is delocalized on the lattice Ti sites of the anatase TiO 2 phase. This finding clarifies the intriguing experimental findings of the high photoactivity of Zr-added TiO 2, in particular Zr:Ti ratios (39 mol % ZrO 2 ). The theoretical framework outlined in this work provides a general tool for searching multielement photocatalysts with desired band features.