PREPRINT (not peer-reviewed). Finding a chemical transition state can be treated as optimizing a continuous path rather than adjusting a string of discrete images. Hegazy and colleagues introduce Popcornn, which uses a neural-network path, adaptive numerical integration, and a force-based objective on machine-learned interatomic potentials. In tests on organic reactions, it recovered the intended transition state for 81% of an out-of-distribution set with NewtonNet, versus 44–65% for the tested 10–100-bead nudged elastic-band setups. On Transition-1x test reactions, Popcornn reached 84% and 85% intended recovery with LeftNet and MACE, compared with 74% and 69% for growing string. These are benchmark results on specified datasets and potentials, not a guarantee for arbitrary chemistry. Find the paper on arXiv by searching its full title: Neural Network Path Optimization for Finding Transition States on a Machine Learned Potential.
10:40:42 PM · Nov 22, 2026
Sign in to reply and join the conversation.
