Optimal H2 moment matching-based model reduction for linear systems by (non)convex optimization.

In this paper we compute families of reduced order models that match a prescribed set of moments of a highly dimensional linear time-invariant system. First, we fully parametrize the models in the interpolation points and in the free parameters, and then we fix the set of interpolation points and parametrize the models only in the free parameters. Based on these two parametrizations and using as objective function the H2-norm of the error approximation we derive non-convex optimization problems, i.e., we search for the optimal free parameters and even the interpolation points to determine the approximation model yielding the minimal H2-norm error. Further, we provide the necessary first-order optimality conditions for these optimization problems given explicitly in terms of the controllability and the observability Gramians of a minimal realization of the error system. Using the optimality conditions, we propose gradient type methods for solving the corresponding optimization problems, with mathematical guarantees on their convergence. We also derive convex SDP relaxations for these problems and analyze when the convex relaxations are exact. We illustrate numerically the efficiency of our results on several test examples.