Predicting Chern numbers in photonic crystals using generative adversarial network-based data augmentation.

Area of Science:

• Topological photonics
• Condensed matter physics
• Materials science
• Computational physics

Background:

• The Chern number is crucial for understanding topological properties in photonic crystals and optical systems.
• Calculating Chern numbers is computationally intensive and time-consuming, hindering practical applications.
• Limited numerical data often poses a challenge for deep learning models in scientific computing.

Purpose of the Study:

• To develop a deep learning approach for predicting the Chern number of two-dimensional photonic crystals.
• To address the challenge of insufficient training data in deep learning for complex physical quantity calculations.
• To improve the efficiency and accuracy of Chern number determination.

Main Methods:

• Integration of deep learning models with Maxwell's equations for Chern number prediction.
• Implementation of a numerical-to-image generative adversarial networks (GANs) data augmentation strategy.
• Application to a two-dimensional photonic crystal with a square lattice structure.

Main Results:

• Achieved an average prediction accuracy of 92.25% on the test dataset.
• The proposed data augmentation method improved Chern number prediction accuracy by 7.95%.
• Demonstrated excellent predictive performance, validating the deep learning approach.

Conclusions:

• The developed deep learning method provides an accurate and efficient way to predict Chern numbers.
• The novel GANs-based data augmentation effectively overcomes data scarcity in scientific machine learning.
• This approach offers a promising solution for deep learning applications involving complex physical calculations and has potential in other scientific domains.