Integrating first-principles calculations and diffusion-based generative models to unveil optimal metal-doped oxides for syngas conversion

Abstract

Oxide–zeolite bifunctional catalysts (OX–ZEO) represent a promising strategy for the direct conversion of syngas into light olefins (C₂–C₄), yet the configurational space of the oxide remains vast for experiments, especially for metal-doping. Here, we coupled systematic DFT investigation with a diffusion-based generative model to discover highly active, metal-doped oxides for CO activation, the rate-limiting step. A DFT database of more than 100 surface models was calculated based on a range of host oxides (In₂O₃, ZnO, t-ZrO₂, ZnCr₂O₄, ZnAl₂O₄ and ZnGa₂O₄) doped with diverse metal cations. Taking oxygen vacancies (OVs) into account, we show that the Al dopant can most effectively lower the CO activation barrier and Al-doped ZnCr₂O₄ being the most active surface in the presence of sufficient OVs. Trained on these data, the diffusion model iteratively generated and screened a cluster of low-barrier candidates. The Al-doped ZnCr₂O₄ remains the most active surface; notably, a slight modification in the lattice constant significantly enhances its intrinsic catalytic activity. Across all systems, the barrier correlates linearly with the summed C and O adsorption energies, yielding separate BEP relations for simple and complex oxides. This combined DFT and generative model workflow offers valuable guidance for the rational design of efficient catalysts for syngas conversion.