Physics-Guided Descriptors for Prediction of Structural Polymorphs.

Area of Science:

• Materials Science
• Computational Chemistry
• Solid-State Physics

Background:

• Predicting stable crystal structures (polymorphs) is crucial for material properties.
• Traditional methods are computationally expensive and struggle with complex materials.

Purpose of the Study:

• To develop an efficient computational method for discovering low-energy polymorphs.
• To accelerate the exploration of material phase spaces.

Main Methods:

• Combining machine learning (ML) with density functional theory (DFT) calculations.
• Utilizing physics-guided descriptors based on structural distortion modes.
• Generating and evaluating crystal structures using DFT and training ML models.

Main Results:

• Successfully rediscovered known metastable phases of Bismuth Ferrite (BiFeO3).
• Identified 21 novel low-energy polymorphs for BiFeO3.
• Demonstrated the efficiency of the ML-DFT approach in exploring material phase space.

Conclusions:

• The developed method accelerates the prediction of low-energy polymorphs.
• This approach offers a new pathway for discovering novel solid-state materials.
• Physics-guided ML-DFT is a powerful tool for materials discovery.