“…Model order reduction methods aim at significantly reducing the computational cost by approximating the quantity of interest starting from evaluations at only few wavenumbers. They rely on a two-step strategy: the offline stage consists in the computation of a finite dimensional basis-e.g., the basis of snapshots (see, e.g., [6,11,17,18,23,27,31,32,34,35]), or evaluations of the frequency response map and its derivatives at fixed centers (Padé method, see, e.g., [3,10,12,13,16]); the output of this phase, whose computational cost may be very high, is stored, to be used during the online phase, in which the approximation of the frequency response map corresponding to a given new value of the parameter is constructed. This stage does not involve the numerical solution of any large-scale PDE, and is expected to provide the output in real time.…”