Data-Driven Modeling of High-Q Cavity Fields Using Dynamic Mode Decomposition

Indranil Nayak, Fernando Teixeira

Abstract: We present a data-driven reduced-order model based on dynamic mode decomposition (DMD) for analysis and prediction of resonating electromagnetic fields inside a cavity. The full-order model employs a rectangular finite-difference time-domain (FDTD) scheme to generate the high-fidelity data which is used as input to the DMD algorithm. DMD extracts the dominant spatial patterns and respective frequencies which corresponds to the resonance modes of the cavity. The DMD based reduced-order model (ROM) is then used to extrapolate the fields in time for long-term predictions. We demonstrate the method for a closed rectangular cavity made of aluminium. Such time-domain ROMs not only facilitate analysis of the cavity resonances, but also help expedite the time consuming FDTD simulations by predicting late-time fields.