Feed-forward neural network as nonlinear dynamics integrator for supercontinuum generation.

Area of Science:

• Nonlinear optics
• Computational physics

Background:

• Ultrashort pulse propagation in optical fibers is governed by the generalized nonlinear Schrödinger equation (GNLSE).
• Simulating these dynamics, especially for applications like supercontinuum generation, is computationally intensive and time-consuming.

Purpose of the Study:

• To develop a faster and more efficient method for simulating the nonlinear propagation of ultrashort pulses in optical fibers.
• To emulate the numerical integration of the GNLSE using a machine learning approach.

Main Methods:

• Training a feed-forward neural network (FFNN) to learn the differential propagation dynamics described by the GNLSE.
• Comparing the FFNN approach with a recurrent neural network (RNN) for accuracy, speed, and memory efficiency.

Main Results:

• The FFNN successfully emulates direct numerical integration of the GNLSE, accurately predicting pulse propagation and supercontinuum generation.
• The FFNN demonstrates faster training and computation times compared to the RNN.
• The FFNN exhibits reduced memory requirements relative to the RNN.

Conclusions:

• Feed-forward neural networks offer a computationally efficient and accurate alternative to traditional numerical methods for simulating complex nonlinear fiber optics phenomena.
• This generic machine learning approach has the potential for broad application in various physical systems requiring differential equation solving.