Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
Crystalite, a novel diffusion Transformer, is introduced for efficient crystal modeling by incorporating Subatomic Tokenization, a compact chemical representation, and a Geometry Enhancement Module (GEM) that injects periodic minimum-image pair geometry into the attention mechanism. This approach maintains the efficiency of standard Transformers while better capturing crystalline structure. Crystalite achieves state-of-the-art results on crystal structure prediction and de novo generation, demonstrating superior S.U.N. discovery scores and faster sampling compared to geometry-heavy alternatives.
Crystalite proves you can ditch the heavyweight equivariant GNNs for crystal modeling: a lightweight Transformer with clever tokenization and geometric attention biases can achieve state-of-the-art results with significantly faster sampling.
Generative models for crystalline materials often rely on equivariant graph neural networks, which capture geometric structure well but are costly to train and slow to sample. We present Crystalite, a lightweight diffusion Transformer for crystal modeling built around two simple inductive biases. The first is Subatomic Tokenization, a compact chemically structured atom representation that replaces high-dimensional one-hot encodings and is better suited to continuous diffusion. The second is the Geometry Enhancement Module (GEM), which injects periodic minimum-image pair geometry directly into attention through additive geometric biases. Together, these components preserve the simplicity and efficiency of a standard Transformer while making it better matched to the structure of crystalline materials. Crystalite achieves state-of-the-art results on crystal structure prediction benchmarks, and de novo generation performance, attaining the best S.U.N. discovery score among the evaluated baselines while sampling substantially faster than geometry-heavy alternatives.