A general framework for structure-preserving model reduction by Krylov
subspace projection methods is developed. It not only matches as many
moments as possible but also preserves substructures of importance in the
coeficient matrices L,G,C, and B that define a dynamical
system prescribed by the transfer function of the form H(s/)
= L*(G+sC)-1B. Many
existing structure-preserving model-order reduction methods for linear and
second-order dynamical systems can be derived under this general framework.
Furthermore, it also offers insights into the development of new
structure-preserving model reduction methods.