Finding Zeros of System Function using Eigenvalues

AI Thread Summary
The discussion focuses on finding the zeros of a system function using eigenvalues, specifically relating to the state matrices A, b, c^t, and d. It highlights that the zeros of the system can be derived from the eigenvalues of the modified matrix (A - d^(-1)bc^t) when d is not zero. A participant expresses difficulty in simplifying the problem and seeks clarification on the linear algebra involved. The conversation reveals that using state space equations can lead to a clearer solution, and MATLAB is mentioned as a helpful tool for implementation. Ultimately, the key takeaway is the importance of understanding the relationship between the system's poles and zeros through eigenvalue analysis.
DSRadin
Messages
12
Reaction score
1

Hi all - working on this problem wanted to see if anyone had any advice - thanks!

As shown in section 4.4, the poles of the system H(z) with state matrices \mathbf{A, b, c^t, } d are given by the eigenvalues of \mathbf{A}.

Find: Show that, if d\neq0, the zeros of the system are given by the eigenvalues of the matrix \left (\mathbf{A}-d^{-1}\mathbf{b}\mathbf{c^t} \right ).

Hint: The poles of the inverse system H^{-1}(z) equal the zeros of H(z), and H^{-1}(z) has the output x(n) if its input is y(n).

2. H(z)=\mathbf{c^t}(\mathbf{zI-A})^{-1}\mathbf{b}+d
3. I understand why the poles of the system are eigenvalues of A. I have gone through this derivation in other work. I feel like there is something I am missing in the linear algebra that would simplify this problem. My attempt at a solution below stops short of solving for eigenvalues of the new matrix because i feel that proving this in generality must be cleaner than this brute force method.

H^{-1}(z)=\left [\mathbf{c^t}(\mathbf{zI-A})^{-1}\mathbf{b}+d \right ]^{-1}

\left ( \mathbf{A}-d^{-1}\mathbf{bc^t} \right ) = \left (\begin{bmatrix}<br /> a_{1 1} &amp; \cdots &amp; a_{1 N-1} \\<br /> \vdots &amp; \ddots &amp; \vdots \\<br /> a_{N-1 1} &amp; \cdots &amp; a_{N-1 N-1}<br /> \end{bmatrix} - \begin{bmatrix}<br /> \frac{b_1c_1}{d} &amp; \cdots &amp; \frac{b1c_{N-1}}{d} \\<br /> \vdots &amp; \ddots &amp; \vdots \\<br /> \frac{b_{N-1}c_1}{d} &amp; \cdots &amp; \frac{b_{N-1}c_{N-1}}{d}<br /> \end{bmatrix}\right ) = \begin{bmatrix}<br /> a_{1 1} - \frac{b_1c_1}{d} &amp; \cdots &amp; a_{1 N-1} - \frac{b1c_{N-1}}{d} \\<br /> \vdots &amp; \ddots &amp; \vdots \\<br /> a_{N-1 1} - \frac{b_{N-1}c_1}{d} &amp; \cdots &amp; a_{N-1 N-1} - \frac{b_{N-1}c_{N-1}}{d}<br /> \end{bmatrix} = \mathbf{A&#039;}

And then some eigen decomposition leads towards...

|\mathbf{A&#039;}-\lambda\mathbf{I}|=0 = det\begin{bmatrix}<br /> a_{1 1} - \frac{b_1c_1}{d}-\lambda &amp; \cdots &amp; a_{1 N-1} - \frac{b1c_{N-1}}{d}-\lambda \\<br /> \vdots &amp; \ddots &amp; \vdots \\<br /> a_{N-1 1} - \frac{b_{N-1}c_1}{d} -\lambda &amp; \cdots &amp; a_{N-1 N-1} - \frac{b_{N-1}c_{N-1}}{d}-\lambda<br /> \end{bmatrix}Is there something in the composition of \mathbf{A,b,c^t,} d that I am missing?

Thanks all.
 
Last edited:
Physics news on Phys.org
Just using the equation is the cleanest method I know of.
if you use MATLAB this method is extremely easy to implement.
I just want to note that you made an error in your step #3
 
Last edited:
Thank you for your response donpacino.

Unfortunately I'm not seeing it. (both the way to implement the equation and the error in 3)
I considered MATLAB but I was hesitant to take a deterministic approach and go with "if it works this once it will always work" kind of deal. Maybe i'l do that.
 
I= the identity matrix, which means the diagonal is equal to 1, and zero everywhere else

I=
[1 0 0
0 1 0
0 0 1]

so if A=
[1 1 1
1 1 1
1 1 1]

then |A-λI|=
[1-λ, 1, 1;
1, 1-λ , 1;
1, 1, 1-λ]

does that make sense?
 
Last edited:
doh! Yes absolutely it makes sense.
I got the answer now too - the key was to back it into state space equations and re-solve.

Thank you for the response!
-DR
 
Back
Top