Crystal Structure Determination from Powder Diffraction Patterns with Generative Machine Learning.

Area of Science:

• Materials Science
• Crystallography
• Machine Learning

Background:

• Powder X-ray diffraction (PXRD) is crucial for materials characterization, but complete structure determination is challenging and time-consuming.
• Existing machine learning (ML) methods for PXRD analysis only predict partial structural information.

Purpose of the Study:

• To develop a pioneering generative ML model for complete crystal structure determination directly from experimental PXRD data.
• To overcome limitations of current ML approaches in PXRD analysis.

Main Methods:

• Developed a generative machine learning model capable of analyzing PXRD patterns.
• Validated the model on simulated and experimental diffraction data from databases like RRUFF and Materials Project.
• Applied the model to determine previously unreported structures from the Powder Diffraction File and newly synthesized high-pressure materials.

Main Results:

• The ML model achieved state-of-the-art performance on simulated and experimental PXRD data.
• Successfully predicted crystal structures for 134 experimental patterns and thousands of simulated patterns.
• Determined unreported structures for materials including NaCu2P2, Ca2MnTeO6, and high-pressure compounds like Rh3Bi.

Conclusions:

• The developed generative ML model enables full crystal structure solutions from PXRD data.
• This approach significantly accelerates materials discovery, especially for novel materials and under conditions preventing single-crystal growth.
• The model is poised to advance automated materials discovery pipelines and explore new chemical domains.