Identifying Limitations in Screening High-Throughput Photocatalytic Bimetallic Nanoparticles with Machine-Learned Hydrogen Adsorptions

Abstract

The Sabatier principle is of fundamental importance to computational catalyst discovery, saving researchers time and expense by predicting catalytic activity in silico at scale. However, as polycrystalline and nanoscale catalysts increasingly dominate industry, computational screening tools must be adapted to these uses. In this work, we demonstrate the effectiveness of computational adsorption energy screening in nanocatalysis by comparing a multisite adsorption energy prediction workflow against a large experimental dataset of hydrogen evolution activities over bimetallic nanoparticles. Comparing 16 million hydrogen adsorption energy predictions with the hydrogen evolution activity of 5,300 experiments across 84 monometallic and bimetallic systems, we discover that favorable adsorption energies are a necessary condition for experimental activity, but other factors often determine trends in practice. About half of the bimetallic search space can be excluded from experimental screens using hydrogen adsorption predictions, but these tools may become significantly more powerful when combined with other screening tools.