Rational design of selective catalysts for ethylene hydroformylation via microkinetic modeling and universal neural network potentials

Abstract

Rapid prediction of the activity and selectivity of catalysts is essential for advancing catalyst development. Computational chemistry offers a promising approach for achieving this aim by enabling such predictions without experimentation. However, traditional methods, such as density functional theory calculations, can be computationally expensive. In this study, we report a descriptor-based microkinetic model that describes the selectivity and activity trends in ethylene hydroformylation on close-packed metal facets using a universal neural network potential. By incorporating adsorbate–adsorbate interactions, our model successfully reproduced the experimental trends in both catalyst activity and selectivity, including the higher production rates of propionaldehyde for Rh and Ir. Furthermore, a high-throughput screening approach was used to evaluate the efficacy of various Rh-based candidates, providing insights into potential improvements through the addition of secondary components. Overall, this study highlights the potential of integrating neural network potentials with microkinetic modeling to advance the rational design of heterogeneous catalysts.