Predicting emergence of crystals from amorphous precursors with deep learning potentials

Area of Science:

• Materials Science
• Crystallography
• Computational Chemistry

Background:

• Crystallization of amorphous precursors is crucial for natural and synthetic matter formation.
• Predicting crystallization outcomes is vital for materials development but challenging for current models.

Purpose of the Study:

• To develop a predictive method for identifying crystallization products from amorphous precursors.
• To enable new research directions in materials synthesis and geological/biological processes.

Main Methods:

• Utilizing universal deep learning interatomic potentials.
• Sampling local structural motifs at the atomistic level.
• Applying the method to diverse inorganic material systems.

Main Results:

• Accurate prediction of crystallization products from amorphous precursors.
• Identification of the most likely initial crystal polymorphs.
• Successful application across oxides, nitrides, carbides, halides, chalcogenides, and alloys.

Conclusions:

• Deep learning interatomic potentials offer a powerful tool for predicting amorphous crystallization.
• This approach overcomes limitations of traditional molecular modeling and ab initio methods.
• The method has broad applicability in materials science and related fields.