no. 2 · Monday 26 October 2026
An Optimal Projection Framework for Structure-Preserving Model Reduction of Linear Systems
Abstract
This paper presents a model order reduction framework for linear systems where the $\mathcal{H}_2$ optimization is combined with the Petrov-Galerkin projection to preserve a wide range of system properties, including stability, bounded realness, sector boundedness, and dissipativity. The model reduction problem is formulated as a nonconvex optimization problem on a generalized Stiefel manifold, aiming to minimize the $\mathcal{H}_2$ norm of the pproximation error between the full-order and reduced-order models. An explicit expression for the Euclidean gradient of the objective function is derived, and a Riemannian gradient procedure with an efficient generalized Cayley update is applied to seek a (locally) optimal projection, followed by a theoretical analysis of the algorithm. Finally, the performance of the proposed method is demonstrated by a numerical example.
- comments
- accepted by IEEE Transactions on Automatic Control
- fields
- Optimization and Control