Electromagnetically large cylinders with duality symmetry by hybrid neural networks

Abstract

The exploration of electromagnetic duality symmetry is greatly constrained in natural optical materials due to the rare and weak intrinsic magnetism. Equal electric and magnetic responses can be obtained by using artificially engineered nonmagnetic particles with optically-induced magnetic responses, but this approach generally works only in the dipolar region. However, going beyond the dipolar approximation is currently a rather challenging task, as it requires equality between electric and magnetic responses of all multipolar orders. Here, we exploit artificial neural networks to explore electromagnetically large nonmagnetic cylinder systems exhibiting the electromagnetic duality symmetry beyond the dipolar approximation. A hybrid neural network that combines the forward neural network of one polarization with the inverse neural network of another polarization is developed. Such network enables us to construct dual-paired cylinders with matched angular scattering patterns for incident waves of cross polarizations. Our work demonstrates the great potential of artificial neural networks for exploring symmetries in fundamental physics.