Machine-Learning-Guided Discovery and Statistical Analysis of Multielemental Reverse Water–Gas Shift Catalysts

Machine-learning (ML)-guided experimentation enabled the discovery of a reverse water–gas shift (RWGS) catalyst with low-loading precious group metals (PGMs; Pt, Pd, Ir, Ru, Rh, and Au) that surpasses both a commercial benchmark and the best design from our earlier high-PGM work (Pt(3)/Rb(1)–Ba(1)–Mo(0.6)–Nb(0.2)/TiO2_P25; numbers in parentheses denote wt %). By optimizing PGM and additive compositions as well as support under ≤1 wt % PGM loading and iterating between prediction and experiment over 35 cycles, the workflow identified the best catalyst, Pt(0.8)–Ru(0.1)–Pd(0.1)/Mo(0.8)–Ba(1)–Re(1)–Ho(1)–Na(0.4)/TiO2_STR100N. Comparative post hoc statistical analysis of the 3 wt % dataset from our previous study revealed both common beneficial factors and features specific to the low-PGM regime. Collectively, these findings not only accelerate catalyst discovery but also deepen fundamental understanding of the RWGS reaction and provide chemically interpretable design principles for future catalyst development.