Can Machine Learning Predict the Reaction Paths in Catalytic CO2 Reduction on Small Cu/Ni Clusters?

Abstract

In this paper, we explore the catalytic CO2 reduction process on 13-atom bimetallic nanoclusters with icosahedron geometry. As copper and nickel atoms may be positioned in different locations and either separated into groups or uniformly distributed, the possible permutations lead to many unnecessary simulations. Thus, we have developed a machine learning model aimed at predicting the energy of a specific group of bimetallic (CuNi) clusters and their interactions with CO2 reduction intermediates. The training data for the algorithm have been provided from DFT simulations and consist only of the coordinates and types of atoms, together with the related potential energy of the system. While the algorithm is not able to predict the exact energy of the given complex, it is able to select the candidates for further optimization with reasonably good certainty. We have also found that the stability of the complex depends on the type of central atom in the nanoparticle, despite it not directly interacting with the intermediates.

From Characterization to Discovery: Artificial Intelligence, Machine Learning and High-Throughput Experiments for Heterogeneous Catalyst Design

“…214 However, the works cited focus on the discovery of new catalysts. Stottko et al (2023) 225 describe a DFT-trained ETR algorithm for predicting energy interactions within 13atom bimetallic clusters (Cu and Ni) and CO 2 reduction intermediates to identify promising catalyst candidates without exhaustive simulations. The data included atom coordinates and types within the clusters, along with the corresponding system energy.…”