Learning Crystallographic Disorder: Bridging Prediction and Experiment in Materials Discovery.

Area of Science:

• Computational Materials Science
• Machine Learning in Materials Discovery
• Crystallography and Materials Informatics

Background:

• Computational materials discovery has generated vast numbers of predicted inorganic crystalline compounds.
• Current methods primarily focus on pristine crystalline materials, neglecting crucial factors like defects and disorder.
• This limitation hinders the experimental realization of computationally predicted materials.

Purpose of the Study:

• To incorporate crystallographic disorder into computational materials discovery workflows.
• To develop machine learning models capable of predicting the prevalence of disorder in predicted materials.
• To bridge the gap between computational predictions and experimental validation by accounting for disorder.

Main Methods:

• Development of machine learning (ML) based classification models.
• Training ML classifiers on data from the Inorganic Crystal Structure Database (ICSD).
• Estimating the prevalence of crystallographic disorder in large computational materials databases (e.g., Materials Project, GNoME).

Main Results:

• Successfully trained ML classifiers that capture chemical trends associated with crystallographic disorder.
• Demonstrated the ability to estimate disorder prevalence in large-scale computational materials databases.
• Established a foundation for disorder-aware computational materials discovery.

Conclusions:

• Machine learning offers a viable approach to introduce disorder into computational materials discovery.
• Disorder-aware workflows can significantly improve the accuracy and relevance of computational materials predictions.
• This work paves the way for more experimentally relevant materials discovery by accounting for inherent disorder.