On the asymptotic accuracy of reduced-order models.

Popular model reduction methods can easily be adapted to retain the asymptotic response to inputs with rational transform. To this purpose, the forced response of the high-order system is decomposed into a transient and a steady-state component. Then, the reduced-order model is obtained by combining the unaltered steady-state component with an approximation of the transient component. Examples show that forcing the reduced-order model to retain the steady-state component does not compromise the transient accuracy. [ABSTRACT FROM AUTHOR]