Abstract

Generalized Simulated Annealing (GSA) is emerging as an interesting stochastic procedure for global optimization in lens design problems. However, except for the trivial cases, the 'curse of dimensionality' associated with the large number of variables in practical lens design problems often forbids an effective search for true global optimum from an arbitrary starting point in a reasonable time frame. We propose to overcome this impasse by making use of the traditional strategies for lens design, modifying them as and when necessary. It is important to note that these strategies were developed over the years primarily for the sake of tackling the problem of large dimensionality. Essentially, they involve formulation of approximate models with varying degrees of complexity, e.g.,thin lens design, third order design, thick lens design etc., and using them in a bottom up approach. The reduced number of degrees of freedom at the subproblem level facilitates practical utilization of global optimization procedures. These optima can be used as convenient starting points for further optimization in the final configuration space. Although there is no guarantee that a globally optimum solution at the subproblem level would ultimately lead to the global optimum in the final configuration space, it is obvious that the former, in most cases, is more likely to lead to practically useful solutions.

© 1998 Optical Society of America