Machine Learning and Scaling Laws for Prediction of Accurate Adsorption Energy

Abstract

Finding an "ideal" catalyst is a matter of great interest in the communities of chemists and material scientists, partly because of its wide spectrum of industrial applications. Information regarding a physical parameter termed "adsorption energy", which dictates the degrees of adhesion of an adsorbate on a substrate, is a primary requirement in selecting the catalyst for catalytic reactions. Both experiments and in silico modeling are extensively being used in estimating the adsorption energies, both of which are an Edisonian approach, demand plenty of resources, and are time-consuming. In this paper, employing a data-mining approach, we predict the adsorption energies of monoatomic and diatomic gases on the surfaces of many transition metals (TMs) in no time. With less than a set of 10 simple atomic features, our predictions of the adsorption energies are within a root-mean-squared error (RMSE) of 0.4 eV with the quantum many-body perturbation theory estimates, a computationally expensive method with a good experimental agreement. Based on the important features obtained from machine learning models, we construct a set of mathematical equations using the compressed sensing technique to calculate adsorption energy. We also show that the RMSE can be further minimized up to 0.10 eV using the precomputed adsorption energies obtained with the conventional exchange and correlation (XC) functional by a new set of scaling relations.