I'm trying to determine the rational canonical form of a diagonal matrix $$ A=\begin{pmatrix} a_1 & 0 & \cdots & 0\\ 0 & a_2 & \cdots & 0\\ \vdots & \vdots & & \vdots\\ 0 &0 & \cdots & a_n \end{pmatrix} $$ where the $a_i$'s are all different. If my intuition is correct, since the characteristic polynomial (in this case also the minimal polynomial) of $A$ is just the product $(x-a_1)\cdots(x-a_n)$ and all the factors are different we have that the $(x-a_i)$'s are the invariant factors of $A$. Then the rational canonical form of $A$ is again $A$.

Is this correct? Is there a more formal way to work this problem? I'd appreciate any suggestions. Thanks in advance.

  • 1
    $\begingroup$ Yes, $A$ in this case is the rational canonical form of $A$. You might consult (or refer to) the definition. $\endgroup$ – hardmath Dec 16 '17 at 1:00

Yes, the rational canonical form is just $A$. Your intuition is good, but can be expounded futher:

Let $V$ be a vector space where the matrix of some linear operator $T$ is represented by $A$ in some basis. Since $A$ is diagonal, $T$ is a diagonalizable operator, so $V$ has a basis where each vector is an eigenvector for $T$. The Cyclic Decomposition Theorem lets us decompose $V$ into a direct sum of $T$-cyclic subspaces for some vectors $\alpha_1,...,\alpha_k$. Namely, we can decompose $V$ as:

$$V = Z(\alpha_1; T) \oplus ... \oplus Z(\alpha_n; T)$$

Where $Z(\alpha_i; T) = \{ v \in V: v = g(T) \alpha_i \ \text{ for some polynomial } \ g(x) \}$, and $\alpha_1,...,\alpha_n$ are the distinct eigenvectors of $A$.

These subspaces are invariant under $T$, so let $T_i$ be the operator induced by $T$ on $Z(\alpha_i; T)$. Then $T_i$ has minimal polynomial $p_i(x) = x - a_i$, where $a_i$ is the eigenvalue corresponding to the eigenvector $\alpha_i$. Since $\dim(Z(\alpha_i; T)) = 1$ and $T_i$ has a cyclic vector, we get that the rational form of $T_i$ is just its $1 \times 1$ companion matrix:

$A_i = \begin{bmatrix} a_i \end{bmatrix}$.

Thus $$A = A_1 \oplus ... \oplus A_n = \begin{bmatrix} a_1 & 0 &... & 0 \\ 0 & a_2 & ... & 0 \\ . \\ . \\ 0 & 0 & ... & a_n \end{bmatrix}$$

Is the rational form of $A$


I totally disagree with the answers given previously. The authors confuse the Frobenius normal form with the primary rational canonical form.

The Frobenius decomposition has the following form

$F:=diag(C_{p_1},\cdots,C_{p_k})$ where the $C_{p_i}$ are the companion matrices of the polynomial $p_i$, and overall, $p_i$ is a divisor of $p_{i+1}$. In particular, $p_k$ is the minimal polynomial and $p_1\cdots p_k$ is the characteristic polynomial of $A$.

When the eigenvalues of $A$ are distinct, then the vector $(1,\cdots,1)$ is cyclic over whole vector space $K^n$.

Then the Frobenius form of $A$ is $F=C_p$ where $p$ is the characteristic polynomial of $A$.

It suffices to convince oneself to test in Maple

"FrobeniusForm (DiagonalMatrix ([1,2,3]);"


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.