Physics-informed hybrid learning for predicting output behavior of 2 µm cross-level solid-state lasers

Mid-infrared solid-state lasers hold significant importance in applications such as medical therapy, lidar, infrared countermeasures, and precision manufacturing. However, designing high-performance systems requires precise modeling of the complex multiphysics coupling among gain, thermal, and optical effects, posing substantial challenges. Conventional analytical models, which rely on simplifying assumptions, are often inadequate to accurately describe the strong nonlinear behaviors exhibited by crystal systems subject to multi-structural and multi-mechanism interactions. We propose a physics-informed, multi-model fusion machine learning framework that systematically extracts physical, chemical, and thermal properties of crystals, significantly improving the modeling of dynamic processes in cross-energy-level and cross-structure lasers. The framework is validated on three representative systems-single-doped Tm: YAP, multi-segment bonded Tm: YAG, and integrated Tm/Ho: YLF-achieving high-precision output predictions across broad pump ranges and thermal stability limits. Near thermal instability, the model remains robust, reaching R² = 0.952 with RMSE and MAE of 0.276 and 0.183, respectively, surpassing traditional methods. In the integrated Tm/Ho: YLF laser, the method maintains excellent generalization under matrix variations and cross-level energy transfer, with a minimum relative error of only 0.102%. This framework not only accurately identifies the dominant performance factors but also provides a novel and efficient modeling tool for the physical understanding and structural optimization of complex laser systems.