Towards Rational Design of Confined Catalysis in Carbon Nanotube by Machine Learning and Grand Canonical Monte Carlo Simulations

Abstract The microenvironment is recognized to be as crucial as active sites in heterogeneous catalysis. It was found that the catalytic activity of a set of chemical reactions can be significantly influenced by the confined space of carbon nanotubes (CNTs), with some reactions showing superior activity, while others experience a negative impact. The rational design of confined catalysis must rely on the accurate insights of confined microenvironment. However, the structural complexity of confined catalysts and the interaction with microenvironment hinders the deciphering of chemical origins behind experiments. In this work, Grand‐canonical Monte Carlo (GCMC) simulations are conducted for confined catalysis in CNTs at various reaction atmospheres, accelerated by machine learning potentials. The statistical outcomes of GCMC simulations corroborate a general feature that the electronic interaction (binding energy) inside CNTs is weaker than outside cases. By using Random Forest (RF) model, we ascertain that the shortening of the bond lengths of catalysts within the confined space is the dominant factor, resulting in the weakened binding energy and the downshift of the d ‐band center. Using the bond length variation as a simplified descriptor, our microkinetic models successfully reproduced the seemingly contradictive experiments, namely, the observed enhancement and suppression for the same reaction.