Speaker
Description
The activity and selectivity of realistic heterogeneous catalysts can be altered noticeably by small changes in a multitude of factors such as bulk composition, dopants, defects, reaction conditions, etc. Their effects are furthermore interrelated in non-trivial ways. As an important first step to rationally disentangle them, we here aim to understand their influences on the evolution of local atomic-scale structural motifs presented by the catalyst. Specifically, we do this for the M1 structural modification of (Mo,V)O$_{𝑥}$ and (Mo,V,Te,Nb)O$_{𝑥}$ as an active catalyst for oxidative dehydrogenation of ethane to ethylene. The large primitive cell of the M1 catalyst challenges a detailed study of all surface terminations by means of predictive-quality first-principles calculations. To this end, we deconstruct the primitive cell into ‘rod-like structures’ of surface motifs with various oxygen content. A machine-learned Gaussian Approximation Potential (GAP), trained against this structural library faithfully reproduces experimental data from electron microscopy [1]. MD simulations of M1 catalyst hk0 prismatic faces with the iteratively improved GAP help to rationalize the influence of vanadium and niobium doping on the active surface structure.
References
[1] L. Masliuk et al., J. Phys. Chem. C 121, 24093 (2017).
| Keywords | Machine Learning, Material Science, Heterogeneous Catalysts, M1 Catalysts, Surface Science, DFT |
|---|
