Crystal Structure Assignment for Unknown Compounds from X-ray Diffraction Patterns with Deep Learning.

Area of Science:

• Materials Science
• Computational Chemistry
• Crystallography

Background:

• Determining crystal structures from experimental data is vital for materials science.
• Current methods often require extensive domain expertise, hindering automation.
• Automating structure identification is crucial for high-throughput materials discovery.

Purpose of the Study:

• To develop a deep-learning model for automated identification of crystal structure types from experimental characterizations.
• To overcome the limitations of manual structure analysis and domain expertise dependency.
• To enable efficient analysis of materials from high-throughput experimentation.

Main Methods:

• Utilized a deep-learning model combining convolutional residual neural networks.
• Trained and validated the model on a dataset of over 60,000 compounds across 100 structure types.
• Analyzed the model's interpretability to understand how it quantifies structural resemblance.

Main Results:

• The deep-learning model accurately identifies crystal structure types from X-ray diffraction (XRD) patterns.
• The model can integrate additional structure types without retraining.
• The model quantifies resemblance based on both local and global features in XRD patterns.

Conclusions:

• A novel computational tool has been developed for automated materials structure analysis.
• This approach significantly reduces the reliance on expert knowledge for structure determination.
• The tool facilitates the rapid analysis of novel materials generated through high-throughput methods.