In this article, we use the Balanced Truncation (BT)
method for constructing a reduced-order model for both stable and unstable finite-dimensional linear time-invariant (LTI) dynamical systems with non-homogeneous initial conditions. The L2-error bound for the stable system has been obtained by interpolating the non-zero initial conditions as an extra input by choosing Dirac’s delta function δ0 ∈ L2. This approach has been successfully extended to develop a framework for model reduction over a finite interval [0, T] for unstable (LTI) systems. The advantages and flexibility of this approach are demonstrated with a variety of numerical examples.