(569az) Computational Design of Catalyst of Desired Adsorption Energy Using Machine Learning

Catalyst design is the key to many chemical reaction engineering problems. Adsorption is usually an important step in the catalytic cycles, and obtaining materials with desired adsorption energies is crucial to optimizing catalyst performance. Conventional methodologies for the evaluation of adsorption energies are frequently laborious and often constrained to known materials or molecules. They have limited the capacity to explore adsorption behaviors in new compounds.

In this work, we used computational chemistry combined with machine learning to design catalyst materials with desired adsorption energy. We focused on the adsorption of CO molecules on iron porphyrins and their derivatives, which has important applications in CO₂ reduction reaction systems.¹ We explored a large design space of porphyrins, with varying metal centers, atoms connected to the metal center and surrounding ligands. Quantum chemical (QM) calculations were done on a small set of materials, and machine learning models were built to establish the structure-property relationship. Both convolutional neural network and graph convolutional neural networks² were used, which demonstrated good predictive performance. The models were then used to predict the adsorption energy for all the materials in the design space and the information was used for identifying materials with desired adsorption energy. The model predictions were further confirmed using QM calculations. This work provides insights for precise customization of material properties through computational chemistry and machine learning methods.