DiffractGPT: Atomic Structure Determination from X-ray Diffraction Patterns Using a Generative Pretrained...

Area of Science:

• Materials Science
• Crystallography
• Artificial Intelligence

Background:

• Crystal structure determination from powder diffraction is challenging.
• Requires significant expertise and computational power.
• Current methods are often slow and resource-intensive.

Purpose of the Study:

• Introduce DiffractGPT, a generative pretrained transformer model.
• Enable fast and accurate prediction of atomic structures from X-ray diffraction (XRD) patterns.
• Facilitate inverse design of materials.

Main Methods:

• Trained DiffractGPT on thousands of atomic structures and simulated XRD patterns from the JARVIS-DFT dataset.
• Evaluated model performance with and without chemical information (element list, chemical formula).
• Utilized generative pretrained transformer architecture.

Main Results:

• DiffractGPT accurately predicts crystal structures from XRD patterns.
• Incorporating chemical information significantly enhances prediction accuracy.
• The model demonstrates fast and straightforward training.

Conclusions:

• DiffractGPT offers a significant advancement in automating crystal structure determination.
• Provides a robust tool for data-driven materials discovery and design.
• Bridges computational, data science, and experimental communities.