Bayesian trained rational functions for electromagnetic design optimization

This paper presented an interpolation scheme based on Bayesian trained quadratic rational functions for approximating frequency based electromagnetic return loss responses. Initial results indicate that this scheme is an efficient tool in catching nulls and characterizing resonance behavior. With the implementation of the adjoint variable method for effective gradient evaluations, this may be an alternative tool to predict the nulls and corresponding BW values for practical heuristic design optimization studies. Future work includes elaborating on coef and adaptive selection of sample points and finally applying it to a global design optimization example.