A derivative free minimax optimizer for electromagnetic design

A derivative free minimax optimizer was developed. It targets electromagnetic (EM) design, when designed performance is available with numerically intensive solutions of partial differential equations, but can be also applied to other design environment - when design simulation is costly and inaccurate.
The core of the optimizer performs local searches using an original combination of adaptive polynomial approximation of responses, response caching and reliable trust region based search. Additionally, global searches can be performed using a variant of the Controlled Random Search algorithm.
Performance of the presented optimizer and other minimax optimizers is compared using both standard test problems as well as real EM design problems.