Bayesian Optimization-Guided Design of Silica-Supported Nickel Catalysts from Nickel Phyllosilicates

Here, we report a Bayesian optimization (BO)-guided approach to optimize Ni/SiO₂ catalyst synthesis from the reduction of nickel phyllosilicate (rNiPS). Key synthesis parameters─calcination temperature, calcination time, reduction temperature, and reduction time─were tuned to maximize the concentration of exsolved Ni^x+ species and Ni⁰ nanoparticles, which are active sites for levulinic acid (LA) hydrogenation to γ-valerolactone (GVL). Using 15 initial samples of differently synthesized rNiPS catalysts (rNiPS-1 to rNiPS-15) to initiate the BO with Gaussian process regression, we rapidly identified synthesis conditions after three iterations, which increase the combined concentrations of Ni^x+ (x = ∼1.66) and Ni⁰/10 by ∼14% (rNiPS-18) compared to the benchmark. The optimized catalyst’s physicochemical properties, including porosity, crystallinity, reducibility, surface acidity, and local Ni geometry, were analyzed, revealing higher Ni^x+ and lower Ni⁰ concentrations than the benchmark catalyst. Additionally, the turnover frequency of rNiPS-18 for LA hydrogenation to GVL increased nearly 50% compared to that of the benchmark, underscoring BO’s effectiveness in designing Ni catalysts enriched with Ni^x+ and Ni⁰.