Using machine learning to predict dinitrogen molecule density by analyzing transient-dependent nonlinear pulse propagation in an air-filled hollow core photonic crystal fiber

Abstract Pulse propagation in air-filled hollow core photonic crystal fibers has been well investigated within the last decade to generate nonlinear phenomena such as pulse compression, frequency conversion, supercontinuum (SC) generation, among others, in a highly reliable and reproducible manner. In this work, we extend the analysis to take into account the recently evidenced pulsewidth dependency of the nonlinear refraction index of air and the effects of its molecular composition, showing that this latter plays a drastic influence on the rotational Raman response as well as the nonlinear refractive index. Our study focuses on the dispersive wave and SC generation, presenting distinct effects on the spectrum due to the pulsewidths dependency along the propagation path, evidencing different spectral broadening features depending on the initial pulsewidth. Based on these results we propose a deep learning algorithm that can predict the dinitrogen molecule particle-density of an air sample from different atmospheric conditions with a relative error lower than 3%.