Data-Driven Crystal Structure Prediction for Ternary Metal Chalcogenides.

Area of Science:

• Materials Science
• Computational Chemistry
• Crystallography

Background:

• Efficient discovery of stable inorganic crystal structures is crucial for materials innovation.
• Data-driven methods are increasingly used for accelerated crystal structure prediction.

Purpose of the Study:

• To compare data-driven approaches for accelerated crystal structure prediction.
• To evaluate generative artificial intelligence (genAI) against traditional methods for identifying novel inorganic materials.

Main Methods:

• Crystal structure prediction using substitution, generative AI (Chemeleon), and evolutionary global optimization.
• Optimization of candidate structures with machine-learned interatomic potentials for energy estimation and uncertainty quantification.
• Application to ternary metal chalcogenide compositions, including sulfides.

Main Results:

• The genAI approach matched and surpassed traditional methods in identifying diverse, low-energy structures.
• Machine-learned potentials provided reliable energy estimates and uncertainty quantification for candidate structures.
• Successful application to technologically relevant materials like Na2SiS3, RbPS3, and KMo2S4.

Conclusions:

• Generative models show significant promise for scalable structural exploration of inorganic materials.
• AI-driven approaches offer a powerful toolkit for accelerating materials discovery.
• This study validates genAI as a competitive and effective method in materials science.