XRDiff: Crystal Structure Prediction from Powder X-Ray Diffraction Data Using Diffusion Models
Researchers have developed XRDiff, a novel diffusion model capable of predicting crystal structures from powder X-ray diffraction (PXRD) data. This model can infer structures based on known stoichiometry or, more challenging, elemental composition and unit cell atom count. XRDiff demonstrates strong performance in differentiating between polymorphs using simulated data and shows promise for generalization to experimental data through a peak-based encoding that is robust to noise and artifacts. AI
IMPACT This research could accelerate materials discovery by enabling faster and more accurate crystal structure determination from experimental data.