Automating physical intuition in nonlinear fiber optics with unsupervised dominant balance search

Andrei V. Ermolaev,Christophe Finot,Goery Genty,John M. Dudley

2024-04-26

Abstract:Identifying the underlying processes that locally dominate physical interactions is the key to understanding nonlinear dynamics. Machine-learning techniques have recently been shown to be highly promising in automating the search for dominant physics, adding important insights that complement analytical methods and empirical intuition. Here we apply a fully unsupervised approach to the search for dominant balance during nonlinear and dispersive propagation in optical fiber, and show that we can algorithmically identify dominant interactions in cases of optical wavebreaking, soliton fission, dispersive wave generation, and Raman soliton emergence. We discuss how dominant balance manifests both in the temporal and spectral domains as a function of propagation distance.

Pattern Formation and Solitons,Optics