Reproduce “ReCurveflow: A Flow Matching Framework that Learns Curved Reaction Trajectories to Predict Transition State Geometries” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Seungheun Baek, Mogan Gim, Jaewoo Kang
RECURVEFLOW is a flow matching framework that learns the intrinsically curved paths of chemical reactions, predicting transition state geometries and generating physically plausible reaction trajectories. Its dual off-path correction scheme enhances robustness, leading to a 1.3x improvement in NEB optimization efficiency and enabling reliable activation barrier ranking for competing pathways.
Key insights
- Supervising flow matching models with continuously curved reference paths, derived from full NEB image bands, improves the physical fidelity of learned reaction trajectories and TS predictions.
- Explicitly training the generative model to correct for deviations from the reference path, especially those encountered during self-generated rollouts, is crucial for mitigating exposure bias and enhancing inference robustness.
- Accurate generation of full reaction pathways provides superior initializations for quantum-chemical NEB optimization, significantly accelerating convergence.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.