CrystalDiT: A Diffusion Transformer for Crystal Generation
This addresses the problem of efficient materials discovery for researchers in data-limited scientific domains, showing that simple architectures can outperform complex ones.
The paper tackled crystal structure generation by proposing CrystalDiT, a diffusion transformer that treats lattice and atomic properties as a unified system, achieving a 9.62% SUN rate on MP-20, outperforming recent methods like FlowMM (4.38%) and MatterGen (3.42%).
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 9.62% SUN (Stable, Unique, Novel) rate on MP-20, substantially outperforming recent methods including FlowMM (4.38%) and MatterGen (3.42%). 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.