Data-driven prediction of spatial optical solitons in fractional diffraction

Abstract

A quasi-residual physics-informed neural network (QR_PINN) with efficient residual-like blocks, was investigated based on classical physics-informed neural network to solve nonlinear fractional Schrödinger equation and analyze the transmission of spatial optical solitons in saturable nonlinear media with fractional diffraction. A comprehensive verification of stable transmission of various solitons under PT-symmetric potential was carried out using the QR_PINN. In addition, the transmission of spatial optical solitons was studied under simple real potential (stable transmission) and complex Scarf-II potential (unstable transmission). The results show that the QR_PINN can accurately reconstruct the transmission of spatial optical solitons under fractional diffraction. Meanwhile, as the complexity of the potential function increases, the prediction accuracy of the QR_PINN slightly decreases. These results provide a new approach for the application of deep learning in the nonlinear fractional Schrödinger equation.