0

Let the matrix

\begin{equation} A=\begin{bmatrix} 2 & 1 & 2 \\ -2 & -1 & -4 \\\ 1 & 1 & 3 \end{bmatrix}. \end{equation}

So far I found the characteristic polynomial $C_A(x)=(2-x)(x-1)^2$, the minimal polynomial $m_A(x)=(2-x)(x-1)$ and the Jordan matrix form is

\begin{equation} J=\begin{bmatrix} 2 & * & *\\ * & 1 & * \\\ * & * & 1 \end{bmatrix}. \end{equation}

Does someone could tell me from what I did how it possible to find the rational canonical form of $A$?

user26857
  • 53,190

1 Answers1

7

There are two "rational canonical forms", one obtained using elementary divisors and another obtained by invariant factors.

The invariant factors form a list where each term divides the next, their product is the characteristic polynomial, and the minimal polynomial is the largest one.

Therefore in this problem, $(x-2)(x-1)$ is the largest invariant factor. The other invariant factor is $(x-1)$ simply because there is nothing left over.

The rational canonical form is obtained by putting the companion matrices corresponding to the invariant factors on the main diagonal.

So the rational canonical form of $A$ given by the invariant factors is: $$\begin{pmatrix} 0 & -2 & 0\\ 1 & 3 & 0\\ 0 & 0 & 1 \end{pmatrix}_.$$

The matrix $$\begin{pmatrix} 0 & -2 \\ 1 & 3 \end{pmatrix}_.$$

Is the companion matrix of $(x-2)(x-1)$.

https://en.wikipedia.org/wiki/Companion_matrix

Ken Duna
  • 5,906
  • 1
    It is easy to obtain the invariant factors from the elementary divisors, and there is a great algorithm for computing the elementary divisors of a matrix in the book "Linear Algebra" by Friedberg, Insel, and Spence. – Ken Duna Apr 29 '16 at 14:38
  • Is it necessary to use this algorithm to obtain the elementary divisor? –  Apr 29 '16 at 14:45
  • No. This is just a convenient way, and the way I like to do it. There are methods to find the invariant factors, such as the Smith-Normal Form. However, I find that very cumbersome. http://math.stackexchange.com/questions/340352/invariant-factors-vs-elementary-divisors Here is a link to a question about computing rational canonical forms that has an answer (by the guy who wrote the book the questioner was studying from) on how to computing the invariant factors. – Ken Duna Apr 29 '16 at 14:48
  • What I like about Friedberg's Algorithm is that the computations involved are just matrix multiplication and computing ranks. Also, by computing the elementary divisors, you have also computed the invariant factors (it is really easy to convert elementary to invariant). But as Bruce Cooperstein notes in his answer in the link above, there are no efficient ways to convert invariant factors to elementary divisors. – Ken Duna Apr 29 '16 at 14:52
  • @SpamIAm That is not correct. The rational canonical form whether computed using invariant factors or elementary divisors is comprised of companion matrices. The Jordan form does not use companion matrices.

    Also, elementary divisors are defined for finite dimensional vector spaces over $\mathbb{Q}$. In fact, they exist for any finitely generated module over a PID.

    – Ken Duna Apr 29 '16 at 15:53
  • @SpamIAm The only incorrect thing you said was that there is only one notion of rational canonical form. You can form a matrix of companion matrices associated to the elementary divisors. There are many books and sources that also refer to this as a rational canonical form. In this example, it is easy to find the elementary divisors based on the minimal polynomial and characteristic polynomial. But as the invariant factors were even more clear here, I used those. – Ken Duna Apr 29 '16 at 21:01
  • @SpamIAm I think I have been unclear. The matrices you get when using the elementary divisors and invariant factors are different. But both are canonical in their own way. There are two theorems that are both commonly referred to as the Fundamental Theorem of Finitely generated modules over a PID. The rational canonical form using invariant factors is a special case of one, while the rational canonical form using elementary divisors is a special case of the other. – Ken Duna Apr 29 '16 at 21:07
  • @SpamIAm http://cseweb.ucsd.edu/~gill/CILASite/Resources/12Chap8.pdf

    Page 416. I also know many professors who refer to the elementary divisors decomp using companion matrices as a rational canonical form.

    The rational canonical form obtained from elementary divisors is NOT the Jordan form. They are both obtained from the elementary divisors, but one uses companion matrices and the other uses Jordan blocks. I am not fighting you on the fact that the Jordan form comes from elementary divisors, that is a completely different idea from the rational canonical form via elementary divisors.

    – Ken Duna Apr 29 '16 at 21:46
  • 1
    So far, no one has said anything mathematically incorrect. We just have a disagreement on terminology. If you do not want to refer to the companion matrix of elementary divisors decomposition as a rational canonical form, that is fine and I don't think I will be able to change your mind. Let's just move on and be friends :) – Ken Duna Apr 29 '16 at 21:59
  • 3
    Fair enough, I had never seen such a definition before. Thanks for the reference! – Viktor Vaughn Apr 30 '16 at 02:06