Lens Optimization in a Multi-objective Framework

This paper demonstrates the computational advantages of a multi-objective framework that can overcome the generic and domain-related challenges in optical system design and optimisation. Non-dominated sorting genetic algorithms-II (Deb, 2003) is employed in this study. The optical systems studied in this paper are Cooke triplets, Petzval lens systems and achromatic doublets. We report the results of four studies. In the first study, we optimise the optical systems using computationally efficient image quality objective functions. Our approach uses only two paraxial rays to estimate the objective functions and thus improves the computational efficiency. This timesaving measure can partially compensate for the typically enormous number of fitness function evaluations required in evolutionary algorithms. The reduction in reliability due to the computations from a single ray pair is compensated by the availability of multiple objective functions that help us to navigate to the optima. In the second study, hybridisation of evolutionary and gradient-based approaches and scaling techniques are employed to speed up convergence and enforce the constraints. The third study shows how recent developments in optical system design research can be better integrated in a multi-objective framework. The fourth study optimises an achromatic doublet with suitable constraints applied to the thicknesses and image distance.