Fast frequency sweeps with many forcing vectors through adaptive interpolatory model order reduction

The goal of this work is to examine the efficacy of interpolatory model order reduction on frequency sweep problems with many forcing vectors. The adaptive method proposed relies on the implicit interpolatory properties of subspace projection with basis vectors spanning the forced response of the system and its derivatives. The algorithm is similar to a recently proposed adaptive scheme in that it determines both interpolation location and order within the frequency domain of interest. The bounds of efficiency of the approach as the number of forcing vectors grows are explored through the use of rough floating operation counts. This, in turn, guides choices made in the adaptive algorithm, including a new technique that prevents excessive subspace growth by capitalizing on subspace direction coupling. In order to demonstrate the algorithm's utility, a series of frequency sweep problems drawn from the field of structural acoustics is analyzed. It is demonstrated that increased order of interpolation can actually degrade the efficiency of the algorithm as the number of forcing vectors grows but that this can be limited by the subspace size throttling procedure developed here.