Dynamical Origin of Rebound versus Dissociation Selectivity during Fe-Oxo-Mediated C–H Functionalization Reactions

Abstract

The mechanism of catalytic C−H functionalization of alkanes by Fe-oxo complexes is often suggested to involve a hydrogen atom transfer (HAT) step with the formation of a radical-pair intermediate followed by diverging pathways for radical rebound, dissociation, or desaturation. Recently, we showed that in some Fe-oxo reactions, the radical pair is a nonstatistical-type intermediate and dynamic effects control rebound versus dissociation pathway selectivity. However, the effect of the solvent cage on the stability and lifetime of the radical-pair intermediate has never been analyzed. Moreover, because of the extreme complexity of motion that occurs during dynamics trajectories, the underlying physical origin of pathway selectivity has not yet been determined. For the reaction between [(TQA_Cl)Fe IV O] + and cyclohexane, here, we report explicit solvent trajectories and machine learning analysis on transition-state sampled features (e.g., vibrational, velocity, and geometric) that identified the transferring hydrogen atom kinetic energy as the most important factor controlling rebound versus nonrebound dynamics trajectories, which provides an explanation for our previously proposed dynamic matching effect in fast rebound trajectories that bypass the radical-pair intermediate. Manual control of the reaction trajectories confirmed the importance of this feature and provides a mechanism to enhance or diminish selectivity for the rebound pathway. This led to a general catalyst design principle and proof-of-principle catalyst design that showcases how to control rebound versus dissociation reaction pathway selectivity.

Running Wild through Dirhodium Tetracarboxylate-Catalyzed Combined CH(C)-Functionalization/Cope Rearrangement Landscapes: Does Post-Transition-State Dynamic Mismatching Influence Product Distributions?

“…The results of our quantum chemical calculations, both static and dynamicwhich add to a growing number of quantum chemistry-based dynamics simulations on metal-catalyzed reactions ,,,− have revealed exceedingly complex reaction networks for Davies’ Rh 2 L 4 -promoted C–H­(C) functionalization/Cope rearrangements. These results confirm Davies et al’s original proposal of a flat potential energy surface with a PTSB and showcase the even greater complexity of the reaction network tied to this surface and the dynamical behavior of the molecules faced with traversing it.…”

From Data to Chemistry: Revealing Causality and Reaction Coordinates through Interpretable Machine Learning in Supramolecular Transition Metal Catalysis

“…To the best of our knowledge there is only a single study of the reaction dynamics of a transition-metal complex with explicit solvent with repeated sampling of the reaction step. 28 This study on Fe-oxomediated C-H functionalization reactions by Joy et. al used kinetic energies, quantum numbers and velocities to distinguish between two different dynamic reaction pathways.…”

“…al used kinetic energies, quantum numbers and velocities to distinguish between two different dynamic reaction pathways. 28 While they also used ML for feature selection, their focus is on physical chemical factors that impact reactivity. Our focus lies on the analysis of structural changes to ease interpretation and to facilitate causality analysis, which has never been attempted for chemical reactions but opens a completely new angle on how to understand reactivity.…”

“…By tracking the motion of the nuclei during a reaction, we may observe a different reaction pathway, 25,26 The dynamic effects are well documented for organic molecules in implicit solvents, 27 yet rarely investigated in transition metal catalysis. 25,27,28 While the inclusion of dynamics in mechanistic studies makes a model more realistic, it adds complexity and significantly increases computational cost. This limitation can be addressed by relying on multiscale methods such as quantum mechanics/molecular mechanics (QM/MM) which describe the catalytic centre and substrates at a QM level, while the surrounding environment is treated with MM.…”

“…Machine learning approaches can be used to evaluate the data and extract condensed results which contain insight into the reaction coordinate. 28 While neural network approaches 32 have garnered a lot of attention recently, they are not very well suited for data analysis approaches where understanding of the process is required, due to their black box nature, which makes them very difficult to evaluate. 33 Instead, interpretable machine learning techniques, such as decision trees, random forests or logistic regression, offer good performance in extracting relevant information from large datasets and presenting them in easily understandable ways.…”