Automated reaction discovery using ab initio molecular dynamics (AIMD) is a field in computational and theoretical chemistry. It aims to identify various reaction pathways originating from a defined molecular system and to build comprehensive reaction networks. A central and crucial requirement for constructing these networks is the reliable automated reaction detection from resulting trajectories. Conventional distance-based criteria and simple bond order schemes are often insufficient to accurately identify bond breaking and bond forming events, as both neglect the underlying electronic structure. This limitation becomes even more pronounced in transition metal systems, where the complex and flexible nature of metal ligand interactions further challenges such simplified approaches. A more sophisticated description of bonding is therefore essential for reliable reaction discovery. In this study, we present a systematic comparison of various bond order analysis methods to evaluate their performance in automated reaction detection. We investigate Mayer, Wiberg, Mulliken, Fuzzy and Laplacian bond orders, as well as approaches derived from Intrinsic Bond Orbitals (IBOs) and Natural Bond Orbitals (NBOs). The main focus is on evaluating the consistency, sensitivity, and interpretability of each bond order scheme in identifying bond breaking and bond formation events. Special attention is given to challenging systems involving transition metal complexes. The results illuminate the strengths and inherent limitations of the respective approaches, they serve as a well-founded guideline for selecting the optimal bond order scheme for robust, automated reaction discovery workflows in organic and organometallic chemistry.
Tillmann Wigger