Identifying correlations between certain properties of heterogeneous systems and their catalytic performance is crucial for the development of efficient screening methods to generate new leads for catalysts with superior properties. Using machine-learning neural networks trained on density functional theory simulations, Wang et al. show that the electric dipole and its related parameters can accurately predict the key parameters of surface-molecular adsorbate interactions: molecular adsorption energy and transferred charge. The predictive power of the proposed neural network models is comparable to the widely used d-band theory. The transferability of these models between different heterogeneous systems suggests that the electric dipole moment is a promising new type of catalytic descriptor.
J. Am. Chem. Soc. 142, 7737 (2020).