Overview

• PXRDnet, a new AI algorithm developed by Columbia Engineering, reconstructs atomic structures of nanocrystals from degraded X-ray diffraction patterns.
• The method combines diffusion generative modeling, trained on 40,000 known atomic structures, with Rietveld refinement to polish predictions for accuracy.
• Published in *Nature Materials* on April 28, 2025, the breakthrough overcomes a century-old limitation requiring large, pure crystals for structural analysis.
• The tool can resolve features as small as 10 angstroms, enabling progress in next-generation batteries, drug discovery, and archaeological material identification.
• The project was led by Gabe Guo, Simon Billinge, and Hod Lipson, marking a significant leap for AI in solving long-standing scientific challenges.