Two-sided Riemannian optimization model order reduction for linear systems with quadratic outputs

TL;DR

This paper develops a Riemannian optimization framework for structure-preserving model order reduction of linear systems with quadratic outputs, improving accuracy and efficiency through novel algorithms and stability constraints.

Contribution

It introduces a two-sided Riemannian optimization approach on Grassmann and Stiefel manifolds for $H_2$-optimal MOR, incorporating stability constraints and efficient Sylvester equation solvers.

Findings

Reduced models achieve high approximation accuracy.

Algorithms significantly improve computational efficiency.

Stability of reduced models is explicitly guaranteed.

Abstract

This paper investigates structure-preserving $H_2$-optimal model order reduction (MOR) for linear systems with quadratic outputs. Within a Petrov-Galerkin projection framework, the $H_2$-optimal MOR problem is first formulated as an optimization problem on the Grassmann manifold, for which a corresponding bivariable alternating optimization algorithm is proposed. Furthermore, to explicitly guarantee the asymptotic stability of the reduced-order model, a second approach is introduced by imposing specific constraints on the projection matrices. We reformulate the problem as a novel optimization task on the Stiefel manifold and construct a corresponding solution algorithm. The computational bottleneck in both iterative methods is addressed by developing an approximate solver for Sylvester equations based on orthogonal polynomial expansions, which significantly enhances the overall efficiency. Numerical experiments validate that the obtained reduced models provide significant advantages in approximation accuracy and computational efficiency.