Space mapping-based optimization exploiting tolerant Cauchy approximations

We present a novel space mapping-based approach for the optimization of high frequency structures. A Cauchy rational function approximation with variable data tolerances is first constructed using the available inexpensive coarse model simulations. The parameters of the rational function are obtained through a fast linear program problem that is guaranteed to reach the global solution. The fine model data that become available during optimization are then added to the data set and the tolerances of the domain are adjusted to improve the overall model accuracy. The resulting new surrogate model is then utilized in optimization to predict a better design. Our approach is illustrated through two examples.