ABSTRACT Given that the present yield of propylene (C 3 H 6 ) is relatively low, developing an efficient direct dehydrogenation of propane (PDH) catalyst that exhibits a high yield of C 3 H 6 is therefore immensely beneficial for the chemical industry. By extracting physical meaning, machine learning, combined with an interpretable descriptor, is an effective method for speeding up catalyst design. Existing descriptors, however, are only capable of predicting selectivity, rendering the direct prediction of the yield of C 3 H 6 challenging. Furthermore, the current descriptors primarily screen transition metal elements, thereby neglecting other potentially high‐performance metals beyond the transition metal zone. Hence, constructing a universal descriptor that is capable of screening all metallic elements and predicting the yield of C 3 H 6 is highly desirable but challenging. Herein, a universal descriptor that incorporates frontier orbital occupancy to describe the electronic contributions of metals from different blocks is proposed, thereby enabling the comprehensive screening of all metallic elements, concurrently predicting C 3 H 6 yield. The screening process predicted that potential catalysts, IrGa@NC, composed of both post‐transition and transition metals, achieved the C 3 H 6 yield of 51.1%, outperforming state‐of‐the‐art catalysts. This work provides insights into the complicated PDH reaction and the development of catalysts with full‐period metallic elements.
Gao et al. (Mon,) studied this question.