2
$\begingroup$

Context

I have a linear time-invariant (single-input, single-output) system in state space representation (https://en.wikipedia.org/wiki/State-space_representation#Linear_systems):

$$ \mathbf{x'}(t) = \mathbf{A}*\mathbf{x}(t) + \mathbf{B}*{u(t)}$$

In which:

  • $\mathbf{x}(t)$ is a $n$-order vector of the variables of the system;

  • the output $\mathbf{y}(t)$ is the last variable of $\mathbf{x}(t)$, i.e., $\mathbf{x}_{n} = \mathbf{y}$;

  • $\mathbf{x'}(t)$ is a $n$-order vector of first order derivatives;

  • the $n$ x $n$ square matrix $\mathbf{A}$ is sparse; it's non-zero elements are in the: first column, first line and main diagonal;

$$ \mathbf{A} = \begin{bmatrix} {a}_{11} & {a}_{12} & {a}_{13} & \cdots & {a}_{1n} \\ {a}_{21} & {a}_{22} & 0 & \cdots & 0 \\ {a}_{31} & 0 & {a}_{33} & \cdots & 0 \\ \vdots & \vdots & \vdots & \ddots & \vdots \\ {a}_{n1} & 0 & 0 & \cdots & {a}_{nn} \\ \end{bmatrix} $$

  • the input $u(t)$ is a scalar;

  • $\mathbf{B}$ is a sparse $n$-order vector which the only non-zero element is the first.

To solve the state-space, I thought of doing integration, line by line: $$\mathbf{x}(t) = {e}^{\mathbf{A}t}\mathbf{x}(0) + \int_{0}^{t}{e}^{\mathbf{A}(t-\tau)}\mathbf{B}u(\tau)d\tau $$

That would be very simple if ${\mathbf{A}}$ was a diagonal matrix. Because if it is, then so will be ${e}^{\mathbf{A}t}$; and every element of it will be ${e}^{{a}_{ii}t}$.

Where ${a}_{ii}$ is an element of ${\mathbf{A}}$.


The question arises

So my question arised when I tried to diagonalize ${\mathbf{A}}$...

For ${\mathbf{A}}$ to be diagonalizable, it must be invertible, i.e., $$det({\mathbf{A}}) \neq 0$$

I went to check the determinant of ${\mathbf{A}}$, and I got: $${det(\mathbf{A})}=1.96*{10}^{16}$$

So, I am wondering: is ${\mathbf{A}}$ invertible? Can I diagonalize it?


Why do I ask?

Suppose that the determinant of ${\mathbf{A}}$ goes to infinity. Then the determinant of it's inverse will go to zero.

As if: $$det({\mathbf{A}}) = \infty $$ then $$ det({\mathbf{A}^{-1}}) = 0 $$

Therefore, the inverse of ${\mathbf{A}}$ is not invertible (This apparent contradiction is what is bugging me).

A note:

Because the determinant was so high, the inverse will have it's elements with near-zero values. This will certainly cause numerical errors.

$\endgroup$
  • 1
    $\begingroup$ You seem to be confusing two things here. Each individual matrix $A$ has a finite nonzero determinant, so it's invertible. If you're taking a limit of a sequence of matrices, then it totally depends on the sequence. $\endgroup$ – Greg Martin May 25 '16 at 0:20
  • 1
    $\begingroup$ It may be helpful to look into the theory of unbounded operators on a Hilbert space. en.wikipedia.org/wiki/Unbounded_operator $\endgroup$ – Nick Alger May 25 '16 at 0:28
  • $\begingroup$ The phrasing of the question is very confusing. However, it appears that -- despite talking about a single matrix $A$ -- the OP is dealing with a sequence of matrices of increasing size... Or maybe there's a square matrix of infinite order and the OP is approximating its determinant step-by-step... Clarifying the question would help to find an answer for it. @NickAlger: your comment is right on! Since that's probably what we're discussing here. $\endgroup$ – zipirovich May 25 '16 at 0:30
  • $\begingroup$ Remade the question. Now providing the context in which the question arised and why. $\endgroup$ – Pedro H. N. Vieira May 25 '16 at 2:17
4
$\begingroup$

I am interpreting the matrix $A$ to be a sequence of matrices, whose limit is taken entrywise, and $B$ likewise.

Consider $A_j=\left(\begin{smallmatrix}j&0\\0&1\end{smallmatrix}\right)$. We have $det(A_j)=j$, which approaches infinity as $j\to\infty$. However, $A_j^{-1}=\left(\begin{smallmatrix}\frac{1}{j}&0\\0&1\end{smallmatrix}\right)$, which approaches $\left(\begin{smallmatrix}0&0\\0&1\end{smallmatrix}\right)$. This limit exists, and indeed has determinant $0$.

This is not a contradiction; each individual matrix in the sequence has an inverse. The limiting matrix of the $A_j$ sequence doesn't exist, as its entries would no longer be real numbers. Hence the fact that the limiting $B_j$ matrix isn't invertible doesn't matter.

$\endgroup$
  • $\begingroup$ What if each individual matrix has a determinat which approaches infnity, right from the start (j = 1)? $\endgroup$ – Pedro H. N. Vieira May 25 '16 at 2:20
  • $\begingroup$ What does "approaches" mean in a single matrix? Any matrix with real entries will have a real determinant. $\endgroup$ – vadim123 May 25 '16 at 3:03
  • $\begingroup$ I mean that it's determinant has a very high numerical value, like 10^60. $\endgroup$ – Pedro H. N. Vieira May 25 '16 at 17:26
  • $\begingroup$ Although $10^{60}$ is indeed a large value, it is finite. It is not mathematically meaningful to say that a single number "approaches" infinity, no matter how large that number is. Approaches is used only for sequences of numbers. $\endgroup$ – vadim123 May 25 '16 at 17:29

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.