Abstract
The identification of key parameters that correlate with catalytic performance through a combination of experiments and model calculations can accelerate the development of improved catalysts and reveal the relevant underlying processes. However, the analysis of correlations in heterogeneous catalysis is often hindered by inconsistent data. Besides, nontrivial, yet unknown relationships may be important, and the intricacy of the various processes may be significant. Here, we address these challenges for perovskite-catalyzed CO oxidation by linking systematic experiments and artificial intelligence (AI). For this purpose, 14 AMn(1-x)CuxO3 phase-pure perovskites with A = Pr, La were synthesized, characterized, and tested according to rigorous experiments. To the so-generated consistent dataset, we applied the symbolic-regression SISSO approach and identified a descriptor for CO consumption rates as an analytical expression containing the bulk and surface A content, the copper surface content, and the deviation (D) of the normalized lattice constants from the cubic root of the normalized cell volume. Crucially, D reflects a crystallographic distortion that depends on the element at the A site. Thus, in addition to the relative abundance of redox-active species on the surface, the species at the A-sites also influence the performance by modulating the properties of the material and the surface.