From combinatorial explosion to targeted optimization: a hybrid strategy for high-entropy catalyst discovery
The vast compositional space of high-entropy materials presents a fundamental challenge for catalyst discovery. Considering 21 candidate elements at a 1% atomic resolution, this combinatorial explosion exceeds 10 billion (>10^10) possibilities, rendering direct experimental exploration impractical. Furthermore, purely data-driven approaches often struggle to comprehend the intrinsic chemical roles of discrete elemental identities, yet they excel at mapping continuous concentration gradients. Recognizing this distinction, we transform this combinatorial explosion into a targeted optimization problem by decoupling elemental selection from compositional ratio refinement. Ultrafast carbon thermal shock (CTS) is first employed to screen viable elemental combinations and establish an optimal quinary framework. Machine learning (ML) is subsequently applied to optimize compositional ratios within this reduced space, where statistical modeling efficiently navigates the remaining high-dimensional landscape. Targeting the oxygen evolution reaction (OER) as a proof-of-concept, our hybrid framework pruned the search space from over 10^10 possible compositions down into 13 systems, ultimately identifying high-entropy oxide (HEO)-Fe17.57Co28.45Ni31.27Mo10.57Zr12.14 as the optimal catalyst. The optimized high-entropy oxide exhibits an overpotential of 240 mV at 10 mA cm−2 and sustains stable operation at 1 A cm−2 for over 600 h in 1 M KOH. Mechanistic analysis reveals that Mo electronically tunes oxygen-intermediate adsorption, while Zr enhances structural robustness, collectively enabling high activity and durability. This work demonstrates that bridging discrete physical screening with continuous data-driven optimization provides an efficient and generalizable pathway for navigating high-dimensional material frontiers.