Activity Descriptors of Mo2C-based Catalysts for C-OH Bond Activation

Deriving structure-activity relationships is crucial for designing efficient catalysts. To aid in this quest, data-driven methods such as machine learning (ML) are emerging. In this work, we incorporate ML tools with accurate density functional theory (DFT) energetics, electronic and geometric features, in the context of designing efficient molybdenum carbide (Mo2C)-based catalysts for biomass conversion. Previously, it was shown that C-OH activation is the rate-determining step in the hydrodeoxygenation (HDO) reaction. Therefore, in this work, DFT was used to obtain accurate barriers (Ea) and reaction energy (∆E) for the C-OH activation in HDO over the most stable (111) and meta-stable facets (010, 101, and 110) of Mo2C and transition metal doped Mo2C. The 101 facet was identified as the most active facet for C-OH activation. While, doping of the active site with Zr and Nb was identified as promising strategy to improve the activity. Further, scikit-learn’s ML models were used to obtain the best primary features correlating with the Ea. Ridge Regression (RR) gives the best ML model for predicting Ea with a test RMSE of 0.21 eV. SHapley Additive exPlanations (SHAP) analysis was then performed which reveals that ∆E and d-band center are the most important features contributing to the activity. Finally, SISSO was used to validate RR model and SHAP results, and to obtain a 2-dimensional physically interpretable generic-descriptor comprising of dopant’s local environment, d-band features, and ∆E.