Efficient finite element model order reduction of electromagnetic systems using fast converging Jacobi-Davidson Iteration

Finite element models of electromagnetic scattering/radiation problems involve absorbing boundaries for the computational domain truncation. Model order reduction based on eigenspace projection for such systems, invariably, has to deal with nonlinear eigenvalue problems. For small systems, the eigenspaces can be extracted using linearization techniques. However, the linearized system is at least double in size compared to the original system. Therefore, selective extraction of interior eigenvalues can take prohibitively long using traditional shift-invert techniques. In this paper, a Jacobi-Davidson based algorithm is used to extract the desired part of the spectrum. The reduced system is setup quickly and is shown to be accurate within a desired frequency band.