Physics-Informed Neural Networks for Fiber Laser Amplification and Its Associated Effects

Physics-informed neural networks (PINNs) offer a promising approach for solving partial differential equations by leveraging modern deep learning techniques and have demonstrated success across various disciplines. In this work, we develop a PINN-based model to explore multi-physics issues in high-power continuous-wave fiber lasers. This model effectively characterizes diverse effects through a unified natural-logarithmic gain equation with tailored treatments of gain coefficients. Specifically, learning the natural logarithm gain equation improves the convergence and stability, enabling accurate simulations of laser amplification in fiber amplifiers and fiber oscillators. Tailored treatments of gain coefficients help characterize amplified spontaneous emission (spectral domain), stimulated Brillouin scattering and stimulated Raman scattering (nonlinear domain), and transverse mode evolution (spatial domain) in three distinct physical domains. Moreover, this model can robustly estimate unknown parameters and efficiently predict threshold powers. The excellent agreement between all prediction results and reference values confirms the feasibility, extensibility, and generality of the model. This work establishes a new methodology for studying laser amplification and provides a promising avenue for advancing scientific machine learning in laser physics.