Wideband FEM computations via the adaptive BT-POD

This paper presents an adaptive sampling model order reduction (MOR) methodology for fast frequency sweep of wideband electromagnetic (EM) systems. The proposed algorithm is an automated, more reliable, robust and error-controllable version of the balanced truncation proper orthogonal decomposition (BT-POD) model order reduction [1], [2]. Although the framework in [1] and [2] works quite well, it is based on an ad-hoc uniform POD sampling strategy that is less robust and efficient and it has no stopping criteria. The adaptive BT-POD intelligently chooses POD sampling points according to goal-oriented error estimates and indicators [3], and provides reliable stopping criteria based on a global error indicator. In [4], an adaptive s-parameter rational interpolation is used to perform the fast sweep. In [5], an adaptive multipoint Krylov method using the s-parameter history to decide the next sample location was proposed. These are different than the present method because the error estimates are computed based on goal-oriented residual error indicators that are known to be more reliable. Numerical results on a large-scale waveguide filter model are used to illustrate the validity and performance of the proposed methodology.