Study on machine learning–assisted optimal thermal design of ultra-stable laser cavity

The ultra-stable laser is a key component of the optical frequency standard system, where one of the primary limitations on its secondary stability index arises from thermal noise. This makes it essential to maintain the laser cavity within an extremely stable thermal environment. We develop and validate experimentally a thermal model for a typical Fabry–Pérot cavity for accuracy. To improve the thermal stability of the design, a machine learning–assisted optimization design method is proposed, which includes a deep neural network surrogate model, sensitivity analysis, and multi-objective optimization based on the Pareto front. The primary optimization targets are minimizing temperature fluctuations and system weight. A case study demonstrates the efficacy of this approach, showing that the thermal response time constant increased from 97 h before optimization to 190 h after optimization, with only a minimal weight cost. Furthermore, under an external temperature fluctuation of 1 mK, the silicon crystal’s temperature fluctuation is reduced to 0.46 μK—an improvement over the 6μK fluctuation reported in previous studies. We provide crucial guidance for the thermal design of ultra-stable laser cavities, enhancing their accuracy and performance in optical systems.