In this contribution, we present a model order reduction algorithm for linear systems with multiple inputs and multiple outputs that aims at finding the global optimal reduced model of prescribed order n, with respect to the H2 norm. Our approach is based on globalized local optimization, which requires a global sampling of the search space and subsequent local H2 optimization. The increased cost resulting from repeated H2 optimization will be mitigated by exploiting the Model Function framework for H2-optimal model reduction, making the optimization cost negligible compared to the cost of reduction. Numerical investigations motivate the need for globalized approaches in H2-optimal reduction and demonstrate how our method is capable of finding global optima, at a far lower cost than running conventional H2-optimal reduction for different initial samples.     «

In this contribution, we present a model order reduction algorithm for linear systems with multiple inputs and multiple outputs that aims at finding the global optimal reduced model of prescribed order n, with respect to the H2 norm. Our approach is based on globalized local optimization, which requires a global sampling of the search space and subsequent local H2 optimization. The increased cost resulting from repeated H2 optimization will be mitigated by exploiting the Model Function framework...     »