Research map / Image generation / Autoregressive and latent generation
CrystalDiT: Simple Diffusion Transformers for Crystal Generation
Xiaohan Yi, Guikun Xu, Zhong Zhang, Liu Liu, Yatao Bian, Xi Xiao, Peilin Zhao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use unified single-stream attention instead of dual-stream cross-attention architecture (ablation) | CrystalDiT · MP-20 | SUN rate (%): 6.36 -> 8.78 | Improved |
| use 2D periodic-table (period, group) atomic encoding instead of 1D atomic number (ablation) | CrystalDiT (Simple) · MP-20 | SUN rate (%): 6.28 -> 8.78 | Improved |
| 2D atomic encoding versus 1D encoding on UN rate (ablation) | CrystalDiT (Simple) · MP-20 | UN rate (%): 78.47 -> 63.28 | Hurt |
| 2D atomic encoding versus 1D encoding on stability (ablation) | CrystalDiT (Simple) · MP-20 | Stable in UN (%): 8.00 -> 13.87 | Improved |
| train longer (10k to 50k epochs) (ablation) | CrystalDiT (Simple) · MP-20 | UN rate (%): 80.72 -> 57.37 | Hurt |
| train longer (10k to 50k epochs) (ablation) | CrystalDiT (Simple) · MP-20 | chemical validity (%): 83.09 -> 89.40 | Improved |
| use dual-stream architecture on validity metrics (ablation) | CrystalDiT · MP-20 | structural validity (%): 97.79 -> 98.39 (higher validity) but UN 63.28 -> 40.28 | Mixed |
| CrystalDiT (Simple) versus FlowMM and MatterGen | crystal generation · MP-20 | SUN rate (%): 4.21 / 3.66 -> 8.78 | Improved |
| use two-stage latent autoencoder + DiT (ADiT) | ADiT · MP-20 | UN rate (%): 63.28 (CrystalDiT) vs 37.08 | Hurt |
| scale to larger structures (MPTS-52) | CrystalDiT (Simple) · MPTS-52 | SUN rate (%): 8.78 -> 6.73 | No clear effect |
Closest papers by meaning
- SymmCD: Symmetry-Preserving Crystal Generation with Diffusion Models (ICLR 2025 · similarity 0.91)
- Scalable Diffusion for Materials Generation (ICLR 2024 · similarity 0.90)
- Kinetic Langevin Diffusion for Crystalline Materials Generation (ICML 2025 · similarity 0.89)
- Generating Symmetric Materials using Latent Flow Matching (NeurIPS 2026 · similarity 0.89)
- WyckoffDiff -- A Generative Diffusion Model for Crystal Symmetry (ICML 2025 · similarity 0.89)
- Wyckoff Transformer: Generation of Symmetric Crystals (ICML 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
