(582c) Electronic Environment-Based Correction Scheme Using DFT and Wft-Level Energies to Accurately Predict Energetics of Catalytic Interfaces
AIChE Annual Meeting
Data Science & Machine Learning Approaches to Catalysis III: Applications of Machine Learning to Heterogeneous Catalysis: From Porous Materials to Cluster Catalysis
Thursday, November 17, 2022 - 8:36am to 8:54am
For numerous applications of DFT, the calculations of adsorption energies have errors that are substantially larger than âchemical accuracyâ. It has been shown that GGA functionals such as PBE and PW91 predict chemisorption energies that differ from experimental values by as much as 1 eV, which can lead to quantitatively and qualitatively incorrection conclusions in analysis of surface reaction systems. These functionals tend to be inaccurate for systems such as molecules and solids with localized electrons but are commonly used for modeling adsorption at catalytic interfaces due to their low cost. In this work, we propose a data-driven framework to correct the PBE gas-phase energies by using Legendre polynomial multipole descriptors that describe the electronic structure and extend the scheme to predict adsorption and transition state energies of catalytic interfaces. This correction scheme can in principle be applied to any system-level energy obtained from DFT and WFT-level of theory. We train the machine learning (ML) model to the difference in DFT and CCSD(T) model which accounts for the error due to the choice of exchange correlation (XC). The framework is devised such that correction for XC energy density contributions for different points across the system are predicted by the ML model. This is an improvement upon the existing schemes as it leverages the data from larger sets of organic molecules and predicts corrections based on electronic environments rather than chemical groups, enabling it to be applied to systems where bonds are breaking or forming.