1
$\begingroup$

I am given the two linear systems: \begin{eqnarray} \Sigma_1: \dot{x}&=&Ax+Bu\\ y&=&Cx+Du \end{eqnarray} and \begin{eqnarray} \Sigma_2: \dot{x}&=&\bar{A}x+\bar{B}u\\ y&=&\bar{C}x+\bar{D}u \end{eqnarray} with the same amount of inputs and outputs. Here $\bar{A} =SAS^{-1}$, $\bar{B} =SB$, $\bar{C}=CS^{-1}$ and $\bar{D}=D$ where $S$ is a non singular matrix. Thus, $\Sigma_1$ and $\Sigma_2$ are isomorphic.

Now let $R$ and $\bar{R}$ be the controllability matrices of $\Sigma_1$ and $\Sigma_2$ respectively and $W$ and $\bar{W}$ be their observability matrices. Then I've already shown that $\bar{R}=SR$ and $\bar{W}=WS^{-1}$. Now I need to show that system $\Sigma_1$ is observable iff $\Sigma_2$ is observable and similarly for controllable.

Now if $\Sigma_1$ is observable, the matrix $$R = (B\ \ AB\ \ A^2B\ \ \ldots\ \ A^{n-1}B)$$ has rank $n$, where $n$ is the dimension of the square matrix $A$. So I have to show that $$\bar{R} = (SB\ \ SAB\ \ SA^2B\ \ \ldots\ \ SA^{n-1}B)=SR$$ also has rank $n$. Similarly for the observability matrix. So basically what I need to show is that a matrix $R$ of full rank keeps full rank when multiplied by a non singular matrix $S$ of appropriate dimensions. This is probably due to the non singularity of $S$ but I am not sure how to show this.

Then finally I have to show that if $\Sigma_1$ and $\Sigma_2$ are observable then $S$ is unique.

Thanks for any help, I tagged this question as linear algebra as well because I think that methods from linear algebra may be used to fairly easily solve the problem.

$\endgroup$
2
$\begingroup$

Consider this:

Suppose $\text{rank} R = n$ but $\text{rank} SR < n$. Then there exists a non-zero vector $x$ such that $SRx = 0$ which implies that $Rx = 0$ also, because $S$ is non-singular (just multiply both sides with $S^{-1}$). But this is impossible since $R$ is non-singular. We obtained a contradiction, hence $\text{rank} SR = n$. The necessity part can also be shown similarly. Also, observability proof follows similarly.

However, we can find infinitely many non-singular matrices $S$, even if the system is not controllable and/or observable. What this theorem says is, controllability/observability is invariant under similarity transformations.

$\endgroup$
  • $\begingroup$ Thanks a lot for your answer that was very clear. So you are saying that $S$ will in fact not be unique. This makes sense, Thanks again! $\endgroup$ – Slugger Nov 27 '13 at 16:27

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.