(364j) Assessing Accuracy and Improving Prediction of Chemical Reaction Barriers Using Density Functional Theory and Machine Learning Approaches

Accurate prediction of chemical reaction barriers is essential for advancements in fields such as catalysis, materials science, and molecular design. Computational methods such as density functional theory (DFT), empirical forcefields and machine learning forcefields are pivotal in these predictions. However, Ab initio methods have limitations in studying chemical reactions due to their computational costs. Empirical forcefields are computationally inexpensive but lack the necessary bonding terms to model chemical reactions effectively. Bond-order potentials like ReaxFF enable the sampling of bond-breaking and bond-formation within a system. However, these forcefields lack accuracy and require reparameterization for new systems. Machine learning interatomic potentials (MLIPs) can be used to explore numerous reactions at costs comparable to empirical forcefield methods and with near-ab initio accuracy. Another issue is the unphysical errors in commonly used DFT methods. DFT suffers from self-interaction errors that compromises its accuracy in predicting important chemical reaction barriers. My research aims to address the limitations of existing methods and enhance the predictive capabilities of computational techniques through self-interaction corrections in a DFT method and machine learning approaches.

To address the limitations of ab-initio methods and empirical forcefields, we have developed an active learning scheme that uses MLIPs and transition-state finding techniques to generate highly informative reactive datasets. This scheme does not require information of reaction pathways or knowledge of final products, thus minimizing human bias. Our MLIPs are trained using the DeePMD formalism with this reactive active learning scheme. We have accurately determined reaction pathways and transition state barriers for gas phase ammonia synthesis and the reaction of methylene imine with water molecules. We found our method to be data efficient as compared to conventional MLIP active learning schemes. Our scheme can determine new minimum energy pathways for various reactions, making it a robust tool for efficiently screening reactions in small molecule design and discovery.

Parallelly, my research focuses on advancing the accuracy and efficiency of computational methods for predicting chemical reactions and properties. Traditional DFT faces limitations due to self-interaction errors, impacting the prediction of reaction barriers and other properties. To address this, we employ the Fermi-Löwdin Self-Interaction Correction (FLOSIC) method to implement the Perdew-Zunger self-interaction correction (PZSIC) energy, which can only be implemented on an orbital-by-orbital basis. Our goal is to assess the accuracy of reaction barrier heights for uncatalyzed and catalyzed chemical reactions. We found that the FLOSIC method significantly improves the prediction of chemical reaction barrier heights for uncatalyzed reactions. The orbitals directly involved in bond-breaking and bond-making events, which are typically stretched bond orbitals in transition states, make the largest contribution to the self-interaction corrections. For catalyzed reactions, we have studied the impact of the FLOSIC method on several reactions involved in the selective catalytic reduction of NOx on a zeolite catalyst, Cu-SSZ-13. We found that FLOSIC improves barrier heights for several of the transition-metal based chemical reactions. However, in some cases, it predicts smaller reaction barriers. We have extended our work to studying important chemical reactions involving metal organic frameworks.