How to prove the covariance of error in a Kalman filter diverges when the system is not detectable?

93 Views Asked by At

Consider the following linear system

\begin{align} x_{t+1} = A x_t + w_t,\\ y_t = C x_t + v_t. \end{align}

If pair $(A, C)$ is not detectable, then how to prove $\mathrm{Cov}(x_t - \hat{x}_t)$ diverges?

Thanks!