Descriptor-Free Design of Multicomponent Catalysts

Multicomponent alloys and oxides are material systems, offering the great promise of unique yet advantageous catalytic properties through appropriate choice of compositions. However, strategic design of the highly active multicomponent catalyst is challenged by the vast number of potential candidates and complex inter-component effects. Herein, we demonstrate the successful employment of Bayesian optimization (BO) to improve the experimental measured activity as a direct function of compositional variables without educating physical knowledge to the machine. We applied BO in screening spinel CraMnbFecCodNieCufZn3-a-b-c-d-e-fO4 for the decomposition of nitric oxide into environmentally friendly nitrogen. Starting with 30 manually surveyed samples, 35 more samples were measured, and six oxides were discovered with higher specific activities. The best candidate discovered in the current work, Co2.1Cu0.6Zn0.2Mn0.1O4, showed significantly better catalytic performance than the benchmark standard. Although not directly educated about the underlying physical origin of variation in specific activity, the optimization balanced the exploration to locate promising regions and exploitation to survey in the identified region of Mn-containing subspace. The success to directly optimize the experimental measured specific activity in the large compositional space through a small sampled data set demonstrates the great potential of BO in the discovery and design of multicomponent catalysts.

Automated summary

This study demonstrates the successful application of Bayesian optimization to improve the experimental measured catalytic activity as a direct function of compositional variables. Starting with 30 manually surveyed samples, 35 more samples were measured, and six oxides with higher specific activities were discovered.