CrystalDiT: A Diffusion Transformer for Crystal Generation
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866911346113642496 |
|---|---|
| author | Yi, Xiaohan Xu, Guikun Xiao, Xi Zhang, Zhong Liu, Liu Bian, Yatao Zhao, Peilin |
| author_facet | Yi, Xiaohan Xu, Guikun Xiao, Xi Zhang, Zhong Liu, Liu Bian, Yatao Zhao, Peilin |
| contents | We present CrystalDiT, a diffusion transformer for crystal structure generation that achieves state-of-the-art performance by challenging the trend of architectural complexity. Instead of intricate, multi-stream designs, CrystalDiT employs a unified transformer that imposes a powerful inductive bias: treating lattice and atomic properties as a single, interdependent system. Combined with a periodic table-based atomic representation and a balanced training strategy, our approach achieves 8.78% SUN (Stable, Unique, Novel) rate on MP-20, substantially outperforming recent methods including FlowMM (4.21%) and MatterGen (3.66%). Notably, CrystalDiT generates 63.28% unique and novel structures while maintaining comparable stability rates, demonstrating that architectural simplicity can be more effective than complexity for materials discovery. Our results suggest that in data-limited scientific domains, carefully designed simple architectures outperform sophisticated alternatives that are prone to overfitting. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2508_16614 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | CrystalDiT: A Diffusion Transformer for Crystal Generation Yi, Xiaohan Xu, Guikun Xiao, Xi Zhang, Zhong Liu, Liu Bian, Yatao Zhao, Peilin Machine Learning Materials Science I.2.6; I.2.10 We present CrystalDiT, a diffusion transformer for crystal structure generation that achieves state-of-the-art performance by challenging the trend of architectural complexity. Instead of intricate, multi-stream designs, CrystalDiT employs a unified transformer that imposes a powerful inductive bias: treating lattice and atomic properties as a single, interdependent system. Combined with a periodic table-based atomic representation and a balanced training strategy, our approach achieves 8.78% SUN (Stable, Unique, Novel) rate on MP-20, substantially outperforming recent methods including FlowMM (4.21%) and MatterGen (3.66%). Notably, CrystalDiT generates 63.28% unique and novel structures while maintaining comparable stability rates, demonstrating that architectural simplicity can be more effective than complexity for materials discovery. Our results suggest that in data-limited scientific domains, carefully designed simple architectures outperform sophisticated alternatives that are prone to overfitting. |
| title | CrystalDiT: A Diffusion Transformer for Crystal Generation |
| topic | Machine Learning Materials Science I.2.6; I.2.10 |
| url | https://arxiv.org/abs/2508.16614 |