Disentangling autoencoders and spherical harmonics for efficient shape classification in crystal growth simulations

Area of Science:

• Materials Science
• Crystallography
• Computational Chemistry

Background:

• Controlling crystal growth is crucial for material properties but current methods are costly and time-consuming.
• Existing machine learning applications in crystal growth primarily focus on structure-property relationships, not morphological control.

Purpose of the Study:

• To develop an efficient computational framework for controlling crystal morphology during growth.
• To reduce the analytical and computational burden associated with crystal growth simulations.

Main Methods:

• Utilized disentangling autoencoders combined with particle aspect ratio and spherical harmonics descriptors.
• Developed a machine learning approach to analyze and predict crystal growth pathways.

Main Results:

• Revealed continuous transformation pathways between different crystal morphologies.
• Preserved underlying crystallographic principles during morphological transformations.
• Significantly reduced data analytics burdens and design study timelines.

Conclusions:

• The developed framework enhances simulation workflows for crystal growth.
• Enables efficient exploration of crystal morphologies for targeted material design.
• Facilitates the development of crystalline materials with specific functional properties.