Sag-compensated Structural Perturbation for Local-Minimum Escape in Aspheric Lens Design

Existing automated design methods can generate promising aspheric lenses from target specifications, yet further refinement of well-converged designs remains challenging. We propose sag-compensated structural perturbation (SCSP), which combines substantial structural changes with aspheric sag compensation to generate structurally distinct perturbed candidates with potential for further improvement through local reoptimization. Building on this operator, we develop an iterative refinement framework that performs parallel local reoptimization of diverse perturbed candidates and retains the best design for the next perturbation round. Controlled comparisons on two well-converged six-element lenses show that SCSP-assisted reoptimization consistently achieves lower final merit values than the compared perturbation strategies, with mean reductions of 51.2--56.6\% relative to the starting designs. Combining this refinement framework with diverse spherical starting-point generation and progressive aspherization enables automated design from target specifications. On two design tasks based on patent specifications, the resulting designs approach the imaging performance of the patent references and outperform those produced by three open-source lens-design frameworks. Applying the workflow across diverse specifications further yields a database of 500 high-quality aspheric lens designs. The database is publicly available at https://github.com/LensMind/Database.