Machine prediction of topological transitions in photonic crystals

TL;DR

This paper demonstrates that neural networks can accurately predict topological phases and transition boundaries in 1D photonic crystals, even outside the training data range, indicating an understanding of Maxwell's equations.

Contribution

The study introduces a neural network approach that effectively predicts topological phases and transition boundaries in photonic crystals beyond the training dataset.

Findings

Neural networks accurately predict topological phases.

Networks generalize beyond training parameter ranges.

Effective boundary prediction of topological transitions.

Abstract

We train artificial neural networks to distinguish the geometric phases of a set of bands in 1D photonic crystals. We find that the trained network yields remarkably accurate predictions of the topological phases for 1D photonic crystals, even for the geometric and material parameters that are outside of the range of the trained dataset. Another remarkable capability of the trained network is to predict the boundary of topological transition in the parameter space, where a large portion of trained data in the vicinity of that boundary is excluded purposely. The results indicate that the network indeed learns the very essence of the structure of Maxwell's equations, and has the ability of predicting the topological invariants beyond the range of the training dataset.