(173af) Machine Learning Prediction of Adsorption Energies over Heterogeneous Catalysts

The increased capacity and availability of computational resources have enabled us to perform the expensive calculations required for broad-scale computational catalysis exploration. Density functional theory (DFT) is already fairly balanced for accuracy and computational cost, but coupling these calculations with machine learning could further decrease computational cost [1]. In this work, we use feature engineering along with statistical learning methods to identify useful features and algorithms for predicting the adsorption reaction/activation energies for a whole reaction pathway from a few benchmark DFT calculations. We use DFT energies provided by Mamun et al. (2019) on the CatApp database for training and validation. Our goal is to develop a model that can calculate these energies at accuracies held as the DFT “gold standard” (~0.1 eV) [2] at speeds orders of magnitude faster than the brute force calculations. Models developed in previous work depend on the type of adsorbate to determine mode of calculation for energy prediction. We propose to develop a single model that can accurately predicts energies for a variety of adsorbates. Our computational findings, along with domain-knowledge, are used to draw conclusions about the phenomena involved with these adsorption energies.

[1] Hansen, K. Biegler, F. Ramakrishnan, R. Pronobis, W. von Lilienfeld, O. A. Müller, K.-R. Tkatchenko, A Machine Learning Predictions of Molecular Properties: Accurate Many-Body Potentials and Nonlocality in Chemical Space. The Journal of Physical Chemistry Letters 2015, 6 (12), 2326–2331.

[2] Fiedler, L.; Shah, K.; Bussmann, M.; Cangi, A. Deep Dive into Machine Learning Density Functional Theory for Materials Science and Chemistry. Physical Review Materials 2022, 6 (4).