Paper detail

On the stability of projection-based linear reduced-order models: Descriptor vs non-descriptor forms

Two comprehensive approaches are considered for constructing projection-based reduced-order computational models for linear dynamical systems. The first one reduces the governing equations written in the descriptor form, using a Galerkin or Petrov-Galerkin projection onto a reduced-order basis or pair of them, respectively. These bases can be constructed by any preferred method such as the Proper Orthogonal Decomposition, Balanced Proper Orthogonal Decomposition, or Moment Matching method. The second approach transforms first the governing equations into their non-descriptor form before applying the same projection-based model order reduction method. For several structural and coupled fluid-structure dynamical systems, it is observed that the first reduction approach leads to reduced-order models that exhibit significantly better numerical stability and accuracy properties than those resulting from the second one. These observations are reinforced by theoretical results that anticipate and confirm the better stability properties obtained in general when reducing the descriptor rather than non-descriptor form of the equations governing a linear dynamical system.