Highly Selective Low-Temperature Acetylene Semihydrogenation Guided by Multiscale Machine Learning

Catalytic hydrogenation is the key measure to remove traces of acetylene in ethylene in the petroleum industry. Herein we report a highly selective and stable nanocatalyst, Pd1Ag3 supported on rutile-TiO2 (r-TiO2) annealed at unusually high temperatures (>750 °C), which can purify ethylene mixed with 1% of acetylene at 97.2% selectivity and 100% acetylene conversion below 100 °C. The selectivity is more than 10% higher than that in our previous work. This advance is achieved by a rational catalyst search featuring machine learning to correlate catalyst synthesis conditions with the catalyst performance and a large-scale machine-learning atomic simulation for disclosing composite atomic structures at high temperatures. We show that Pd1Ag3 alloy crystal nanoparticles form until 727 °C and the alloy nanoparticles grow epitaxially on r-TiO2(110) via its {111} facets. The maximum exposure of the alloy {111} surface is the key to the highest selectivity among the different supports tested, as confirmed by high-resolution characterization experiments and microkinetics simulations. Our results demonstrate the power of multiscale machine-learning tools in guiding the catalyst design and clarifying the atomic nature in complex heterogeneous catalysis.