Predicting the soliton trapping in birefringence optical fibers via spectral neural operator enhanced by CNLSE-Layer

Soliton trapping is a crucial phenomenon among vector optical solitons in a birefringent fiber. To predict the generation of soliton trapping, we embed the coupled nonlinear Schrödinger equations (CNLSE) as one layer into the neural network. Here, our model employs the CNLSE-Layer to model the cross-phase modulation operator, with the spectral neural operator for modelling the group velocity dispersion operator. The proposed method has the potential to break the bottleneck of the high computational time for physics-informed deep learning methods by adding the physical information in the loss function. This work presents a novel, to the best of our knowledge, design paradigm for embedding the physics information in neural networks.