ECTS: An ultra-fast diffusion model for exploring chemical reactions with equivariant consistency

Abstract Unveiling reaction mechanisms through the exploration of reaction paths — including the identification of transition states (TS), prediction of activation barriers (Eₐ), and mapping of reaction pathways — is central to the study of chemical reactions. However, this process typically relies on extensive and computationally intensive quantum chemical calculations. In this work, we present an Equivariant Consistency generative model for Transition States (ECTS), an ultra-fast diffusion framework that seamlessly integrates TS generation, energy prediction, and pathway exploration within a unified architecture. ECTS demonstrates a computational efficiency improvement of at least two orders of magnitude over conventional diffusion-based models. The generated TS structures exhibit excellent accuracy, with a root-mean-square deviation (RMSD) of only 0.12Å from reference geometries. Furthermore, by progressively refining energy barrier predictions during the denoising process, ECTS achieves a median error of just 2.4 kcal/mol, without requiring any post-DFT corrections. Notably, ECTS also produces reaction pathways that are generally consistent with true reaction paths, demonstrating its potential as a powerful tool for reaction mechanism exploration.