Computational screening of alloy-type H2/O2 recombination catalysts: A hybrid approach using machine-learning interatomic potentials, molecular dynamics, and Bayesian optimization

Hydrogen safety is a critical issue for hydrogen-based energy and industrial systems, where controlling hydrogen accumulation is essential to mitigate the risk of serious explosions. Passive autocatalytic recombiners (PARs) with equipped catalysts address this challenge by catalyzing the conversion of H 2 and O 2 to H 2 O. Although pristine Pt remains the benchmark catalyst in terms of activity, its high cost motivates the search for low-cost alternatives that can operate under severe conditions, including high temperatures, high humidity, and fluctuating reactant concentrations. Here, we computationally screen Pt-based alloy catalysts as cost-effective alternatives to pristine Pt using a hybrid protocol that combines Bayesian optimization with molecular dynamics simulations driven by machine-learning interatomic potentials, aiming to emulate PAR-relevant reactive conditions in an accelerated MD setting. The prescreening prioritized Pt 85 Re 15 , Pt 85 Ag 15 , Pt 85 W 15 , and Pt 70 W 15 Ag 15 as cost–activity trade-off candidates capable of sustaining H 2 O (g) formation with 15–30% reduced Pt content. Although none of the alloys surpassed pristine Pt in catalytic activity, they retained substantial per-Pt catalytic activity. Trajectory analyses indicate that alloying can influence adsorption and activation behavior under the simulated conditions. Consequently, within the MD timescales and conditions examined here, surface intermediate accumulation, especially OH*, becomes an important limitation to sustained reactivity. These results highlight the importance of managing surface-intermediate coverage in the design of PAR catalysts.