Screening of Metal Catalysts for CO2 Conversion via Machine Learning and Molecular Simulations

Abstract

This study's primary objective is to improve catalyst discovery by assessing earth-abundant metal catalysts for the conversion of CO2 to methane through the use of machine learning (ML) and molecular dynamics (MD) simulations. The highest CO2 binding energy on 61 metals was determined to be -9.75 eV for nickel (Ni), -8.7 eV for copper (Cu), and -7.75 eV for carbon (C). Various ML models were developed to predict binding energies on the metallic surfaces. Easily accessible properties of the metals and features obtained from molecular simulations were used as input features. RANSACRegressor, LinearSVR, HuberRegressor, OrthogonalMatchingPursuit CV, and LarsCV models exhibited high prediction accuracy with R-squared values of 0.99 and RMSE ranging from 0.18 to 0.40. Feature significance analysis revealed that density (D) is among the most significant structural features affecting binding energy. This work offers a dependable, high-throughput method for identifying efficient CO2 conversion catalysts, advancing sustainable technologies.