This paper proposes a new balanced realization and model reduction method f
or possibly unstable systems by introducing some new controllability and ob
servability Gramians, These Gramians can be related to minimum control ener
gy and minimum estimation error. In contrast to Gramians defined in the lit
erature for unstable systems, these Gramians can always be computed for sys
tems without imaginary axis poles and they reduce to the standard controlla
bility and observability Gramians when the systems are stable. The proposed
balanced model reduction method enjoys the similar error bounds as does fo
r the standard balanced model reduction, Furthermore, the new error bounds
and the actual approximation errors seem to be much smaller than the ones u
sing the methods given in the literature for unstable systems. Copyright (C
) 1999 John Wiley & Sons, Ltd.