Neural Network Path Optimization for Finding Transition States on a Machine Learned Potential
Published in Submitted to Nature Computational Science, 2025
Recommended citation: K. Hegazy, E. Yuan, et al. (2025). "Neural Network Path Optimization for Finding Transition States on a Machine Learned Potential." Submitted to Nature Computational Science.
A neural-network framework, Popcornn, for finding chemical transition states on machine-learned interatomic potentials.
- First double-ended transition state search method benchmarked at high-throughput scale (O(1000) reactions).
- Consistently outperforms decades-old gold-standard methods (NEB, GSM) with fixed default settings.
- Parameterizes reaction paths as continuous neural networks with GPU-parallelized adaptive integration.