Convolutional neural network analysis of optical texture patterns in liquid-crystal skyrmions

Area of Science:

• Materials Science
• Condensed Matter Physics
• Data Science

Background:

• Liquid crystals exhibit optical birefringence, creating intricate patterns under polarized light microscopy.
• These patterns serve as fingerprints for liquid crystal properties, including elastic constants and director orientation.
• Liquid crystals host topological defects, including singular disclinations and non-singular solitons like skyrmions.

Purpose of the Study:

• To demonstrate the effectiveness of machine learning in predicting liquid crystal system parameters using skyrmion optical signatures.
• To reduce computational cost by focusing analysis on skyrmion-localized regions.
• To explore the potential of data science methods for materials characterization.

Main Methods:

• Utilizing simulated polarized optical microscopy images of liquid crystal skyrmions.
• Training convolutional neural networks (CNNs) on these images.
• Focusing the machine learning analysis on skyrmion-localized regions.

Main Results:

• Trained CNNs accurately predict key system parameters such as free energy, cholesteric pitch, and electric field strength.
• The method demonstrates high accuracy in parameter prediction from skyrmion optical signatures.
• Significant reduction in computational cost achieved by analyzing localized regions.

Conclusions:

• Optical signatures of liquid crystal skyrmions are valuable for machine learning-based materials characterization.
• Machine learning, particularly CNNs, can efficiently extract critical information from complex liquid crystal systems.
• This research paves the way for advanced applications utilizing skyrmions and data science.