We compare the results of simulated and measured power efficiency and far-field beam pattern, for two reflective Fourier phase gratings, designed to generate 2 x 2 and 2 x 4 beams respectively from a single-beam, coherent source at 1.4 THz. The designed surface structures were manufactured on aluminum plates by a computer numerical control (CNC) micro-milling machine. Despite small differences between the designed and fabricated gratings, we measured power efficiencies of both gratings to be around 70%, which is in a good agreement with the simulated values. We also find a good agreement between the simulated and measured diffracted beam size and spatial distribution. We demonstrate the application of both gratings as multiple beam local oscillators to simultaneously pump (or operate) a 4-pixel array of superconducting heterodyne mixers.

In this paper we present a Krylov subspace model-order reduction technique for time- and frequency-domain electromagnetic wave fields in linear dispersive media. Starting point is a self-consistent first-order form of Maxwell's equations and the constitutive relation. This form is discretized on a standard staggered Yee grid, while the extension to infinity is modeled via a recently developed global complex scaling method. By applying this scaling method, the time- or frequency-domain electromagnetic wave field can be computed via a so-called stability-corrected wave function. Since this function cannot be computed directly due to the large order of the discretized Maxwell system matrix, Krylov subspace reduced-order models are constructed that approximate this wave function. We show that the system matrix exhibits a particular physics-based symmetry relation that allows us to efficiently construct the time- and frequency-domain reduced-order models via a Lanczos-type reduction algorithm. The frequency-domain models allow for frequency sweeps meaning that a single model provides field approximations for all frequencies of interest and dominant field modes can easily be determined as well. Numerical experiments for two- and three-dimensional configurations illustrate the performance of the proposed reduction method.