Use of stochastic algorithms for distributed architectures in the optimization of electromagnetic devices

This paper presents a new method for electromagnetic device optimization by stochastic algorithms (e.g. simulated annealing and evolution strategies). In order to reduce the tremendous computational effort for such a problem, we propose to control the accuracy of the objective function evaluation during the computational process. The optimally increasing accuracy is determined using a test function which preserves the essential landscape characteristics of the real objective function. Moreover, based on the coarse-grain model, the distributed architecture of the used computer network is exploited