![[2602.16020] MolCrystalFlow: Molecular Crystal Structure Prediction via Flow Matching](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.16020] MolCrystalFlow: Molecular Crystal Structure Prediction via Flow Matching
Summary
MolCrystalFlow introduces a novel flow-based generative model for predicting molecular crystal structures, addressing challenges in computational chemistry by disentangling intramolecular complexity from intermolecular packing.
Why It Matters
This research is significant as it advances the field of molecular crystal structure prediction, which has implications for materials science and drug discovery. By integrating machine learning with traditional methods, it enhances the efficiency and accuracy of predicting complex molecular arrangements.
Key Takeaways
- MolCrystalFlow effectively separates intramolecular and intermolecular complexities in crystal structures.
- The model utilizes Riemannian manifolds for representing molecular orientations and centroid positions.
- Benchmarking shows MolCrystalFlow outperforms existing generative models for periodic crystals.
- Integration with universal machine learning potential accelerates the prediction process.
- This approach paves the way for data-driven discovery in molecular crystals.
Computer Science > Machine Learning arXiv:2602.16020 (cs) [Submitted on 17 Feb 2026] Title:MolCrystalFlow: Molecular Crystal Structure Prediction via Flow Matching Authors:Cheng Zeng, Harry W. Sullivan, Thomas Egg, Maya M. Martirossyan, Philipp Höllmer, Jirui Jin, Richard G. Hennig, Adrian Roitberg, Stefano Martiniani, Ellad B. Tadmor, Mingjie Liu View a PDF of the paper titled MolCrystalFlow: Molecular Crystal Structure Prediction via Flow Matching, by Cheng Zeng and 10 other authors View PDF HTML (experimental) Abstract:Molecular crystal structure prediction represents a grand challenge in computational chemistry due to large sizes of constituent molecules and complex intra- and intermolecular interactions. While generative modeling has revolutionized structure discovery for molecules, inorganic solids, and metal-organic frameworks, extending such approaches to fully periodic molecular crystals is still elusive. Here, we present MolCrystalFlow, a flow-based generative model for molecular crystal structure prediction. The framework disentangles intramolecular complexity from intermolecular packing by embedding molecules as rigid bodies and jointly learning the lattice matrix, molecular orientations, and centroid positions. Centroids and orientations are represented on their native Riemannian manifolds, allowing geodesic flow construction and graph neural network operations that respects geometric symmetries. We benchmark our model against state-of-the-art generative model...
