Mechanism-Level Chemical Reaction Simulation with Electron Bookkeeping Transformer

Modeling the mechanisms of chemical reactions is central to the prediction and design of chemical reactivity. We present Electron Bookkeeping Transformer (Bookkeeper), a model that simulates chemical reactions at mechanistic resolution by generating electron movement trajectories. Bookkeeper explicitly models each electron as a token and predicts its movement sequentially, following the arrow pushing formalism in chemistry. This ensures mass and charge conservation during both elementary step and reaction pathway predictions by construction. Bookkeeper achieves higher accuracy in predicting both individual mechanistic steps and full reaction pathways, shows strong generalization to out-of-distribution reaction types, and runs significantly faster than previous flow matching-based and large language model-based methods. The accuracy and efficiency make Bookkeeper a powerful framework for virtual screening of chemical reactions across diverse reagents, for validating computer-assisted synthesis planning predictions, and for providing mechanistic interpretability of reactivity; the utility of Bookkeeper as a tool will most likely grow as does its training corpus.