Integrating Physical Principles with Machine Learning for Predicting Field-Enhanced Catalysis.

Field-dipole interactions can tune the energetics of polarized species over catalyst nanoparticles (NPs) for sustainable technologies. This can boost the energy efficiency of desired reactions by several orders of magnitude compared with conventional heating. However, the local electric field accumulation over the NPs sharp points and field-dependent adsorption over NPs are not well studied, and the associated computational expense is immense. To address this challenge, we introduce an innovative approach that combines density functional theory (DFT) calculations, DFT-based CO vibrational Stark effects, and physics principles enhanced machine learning (ML). This approach enables precise mapping of local electric fields and integrates the physical principles of the first-order Taylor expansion as a training input into the ML model for predicting field-dependent adsorption, facilitating rapid prediction of field-dependent adsorption energetics with acceptable accuracies, particularly when training data sets are limited. Our methodology reveals the dominant roles of external electric field (EEF), the generalized coordination number (GCN), and NP size in determining the local electric field (LEF) strength. Low-coordinated sites and small NPs size enhanced the LEF by about 4-fold compared to the flat surfaces. Using ML models, we can predict the field-driven adsorption energetics at a given adsorption site of the NPs with high accuracy and efficiency. The integration of modeling and ML algorithms offers exceptional possibilities to facilitate catalyst development and create the opportunity to enter a new paradigm in field-enhanced catalysis design based on fundamentals rather than trial and error.