![[2602.20210] Multimodal Crystal Flow: Any-to-Any Modality Generation for Unified Crystal Modeling](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.20210] Multimodal Crystal Flow: Any-to-Any Modality Generation for Unified Crystal Modeling
Summary
The paper presents Multimodal Crystal Flow (MCFlow), a unified model for crystal generation tasks that enhances performance by integrating various modalities and leveraging compositional and crystallographic priors.
Why It Matters
This research addresses the limitations of existing task-specific models in crystal structure prediction and generation, providing a more versatile framework that can improve the efficiency and accuracy of crystal modeling in materials science. Its implications extend to various applications, including drug discovery and materials engineering.
Key Takeaways
- MCFlow offers a unified approach to crystal modeling across multiple tasks.
- The model incorporates strong compositional and crystallographic priors.
- Experimental results show MCFlow's competitive performance against existing models.
Computer Science > Machine Learning arXiv:2602.20210 (cs) [Submitted on 23 Feb 2026] Title:Multimodal Crystal Flow: Any-to-Any Modality Generation for Unified Crystal Modeling Authors:Kiyoung Seong, Sungsoo Ahn, Sehui Han, Changyoung Park View a PDF of the paper titled Multimodal Crystal Flow: Any-to-Any Modality Generation for Unified Crystal Modeling, by Kiyoung Seong and 3 other authors View PDF HTML (experimental) Abstract:Crystal modeling spans a family of conditional and unconditional generation tasks across different modalities, including crystal structure prediction (CSP) and \emph{de novo} generation (DNG). While recent deep generative models have shown promising performance, they remain largely task-specific, lacking a unified framework that shares crystal representations across different generation tasks. To address this limitation, we propose \emph{Multimodal Crystal Flow (MCFlow)}, a unified multimodal flow model that realizes multiple crystal generation tasks as distinct inference trajectories via independent time variables for atom types and crystal structures. To enable multimodal flow in a standard transformer model, we introduce a composition- and symmetry-aware atom ordering with hierarchical permutation augmentation, injecting strong compositional and crystallographic priors without explicit structural templates. Experiments on the MP-20 and MPTS-52 benchmarks show that MCFlow achieves competitive performance against task-specific baselines across multi...