Bond-Order Time Series Analysis for Detecting Reaction Events in Ab Initio Molecular Dynamics Simulations

Ab initio molecular dynamics is able to predict novel reaction mechanisms by directly observing the individual reaction events that occur in simulation trajectories. In this article, we describe an approach for detecting reaction events from simulation trajectories using a physically motivated model based on time series analysis of ab initio bond orders. We found that applying a threshold to the bond order was insufficient for accurate detection, whereas peak finding on the first time derivative resulted in significantly improved accuracy. The model is trained on a reference set of reaction events representing the ideal result given unlimited computing resources. Our study includes two model systems: a heptanylium carbocation that undergoes hydride shifts and an unsaturated iron carbonyl cluster that features CO ligand migration and bridging behavior. The results indicate a high level of promise for this analysis approach to be used in mechanistic analysis of reactive AIMD simulations more generally.