Abstract

The interplay of nonlinear and thermal dynamics fundamentally constrains the power scalability of fiber lasers. Conventional development cycles rely on heuristic iterations between modeling and experiment, inevitably leading to extensive resource consumption and suboptimal architectures. In this work, we demonstrate a predictive in silico design paradigm for high-power fiber amplifiers by integrating a coupled multiphysics model with a physics-informed, surrogate-assisted optimization algorithm.

This framework enables the efficient co-optimization of macroscopic configurations and internal waveguide parameters, thereby directly identifying the system-level optimum under rigorous physical constraints. Computationally directed by this method, we designed an optimal gain fiber featuring a numerical aperture of 0.0585 and a centrally depressed refractive index profile. Upon physical fabrication and integration into a 0.5-nm-linewidth setup, the amplifier delivered an output power of 7.68 kW with high optical-to-optical efficiency (84%) and excellent beam quality (M2<1.5), setting a new power record for sub-nanometer-linewidth fiber lasers.

This generalizable framework paves the way to unlock previously inaccessible performance regimes in high-power lasers and accelerate the realization of fully functional digital twins for next-generation photonic systems.