Differentiable Optimization Workflow for Large-Aperture Reflective Optical Systems Inspired by Curriculum Learning

We present a differentiable, curriculum-based optimization workflow for the engineering-oriented design of large-aperture reflective optical systems. The method integrates physics-informed differentiable ray tracing with a progressive, two-stage optimization strategy that evolves from simple Ritchey–Chrétien (RC) foundations to complex four-mirror architectures. Without relying on pretrained models or large datasets, the workflow optimizes geometric and physical parameters under field-weighted RMS, focal length, and dynamic obscuration constraints while maintaining minimal perturbation to primary and secondary mirrors. Validated through Zemax-based analysis, the optimized systems achieve high imaging quality with improved RMS uniformity and stable convergence across varying aperture scales. This approach provides a practical and scalable pathway for the design and optimization of reflective optical instruments, offering strong robustness and adaptability for diverse imaging applications.