Columbia Engineering researchers have made a groundbreaking advancement in crystallography by developing a machine learning algorithm capable of determining the atomic structure of nanocrystals.

Traditional crystallography has relied on large, pure crystals for analysis; however, many samples are often only available as powders or in solution, complicating the process.

To tackle this challenge, the researchers utilized a machine learning technique known as diffusion generative modeling to interpret subpar X-ray diffraction patterns from nanocrystals.

The AI model was trained by jumbling atomic positions from known structures and linking them to their corresponding diffraction patterns, which led to successful crystal reconstruction.

This innovative algorithm achieves near-perfect reconstruction of atomic-scale structures from degraded diffraction patterns, a feat that was previously unattainable.

The study demonstrated that the algorithm could effectively analyze nanometer-sized crystals of various shapes that were previously difficult to characterize.

Hod Lipson highlighted the importance of this algorithm in solving a century-old problem in crystallography, suggesting that AI has the potential to address other scientific challenges as well.

Gabe Guo, the project leader, pointed out the evolution of AI capabilities, indicating its increasing significance in enhancing scientific research and innovation.

Professor Simon Billinge explained that the AI learned atomic arrangement patterns from a comprehensive database of 40,000 known structures, akin to how language models acquire patterns.

For over a century, crystallography has been a vital method for revealing atomic arrangements in materials through X-ray diffraction, yet it has faced difficulties with small crystal samples.