Spectacle lens design with double aspheric surfaces using differentiable ray tracing

Traditional spectacle lens design methodologies have been hindered by their high complexity and low efficiency, primarily due to their reliance on Coddington equations or classical optimization algorithms. We propose an efficient spectacle lens design method based on differentiable ray tracing (DRT), where the partial derivatives of the merit function with respect to the lens surface parameters are computed through automatic differentiation. A -12D lens design demonstrates that the proposed method outperforms traditional optimization approaches, including simulated annealing (SA), particle swarm optimization (PSO), and genetic algorithm (GA), in terms of optimization efficiency, edge thickness, and mean power error, while achieving lower distortion compared to SA and PSO. Additionally, we explored the impact of vertex distance variation on the design results, as well as the spatial distribution of optimized lenses in relation to distortion and mean power error.