3
$\begingroup$

The actual question: when $A$ is the companion matrix, why the general form of $M_i$ (the group of columns of the Jordan matrix $M$ that belongs to the block associated to $\lambda_i$) is:

$$ M_i^{h,j} = \binom {h-1}{j-1} \lambda_i^{h-j} $$

$A$ has order $n$, $h$ is the row index (thus $1 \le h \le n$), $j$ is the column index ($1 \le j \le \mu_i$ with $\mu_i$ the algebraic multiplicity of $\lambda_i$).

Explanation for the first column of $M_i$. Given the characteristic polynomial:

$$ p(z) = \frac 1a_n(a_nz^n + \dots + a_1z + a_0) $$

The corresponding companion matrix $A$ is:

$$ \begin{bmatrix} 0 & 1 &&& \\ & \ddots & 1 &&\\ &&\ddots&\ddots &\\ &&& 0 & 1\\ - \frac{a_0}{a_n} & - \frac{a_1}{a_n} & \cdots & -\frac{a_{n-2}}{a_n} & -\frac{a_{n-1}}{a_n} \end{bmatrix} $$

Because $A$ is the companion matrix, we know that there are $d$ distinct eigenvalues, each having a geometric multiplicity 1.

The Jordan matrix is made of $d$ blocks: $J = diag(J(\lambda_1), \dots, J(\lambda_d))$ .Each block $J(\lambda_i)$ associated with the eigenvalue $\lambda_i$ has the form:

$$ J(\lambda_i) = \begin{bmatrix} \lambda_i & 1 & & \\ & \lambda_i & \ddots &\\ & & \ddots & 1\\ & & & \lambda_i\\ \end{bmatrix} $$

The Jordan matrix $M$ can be partitioned in $d$ groups (a generic group would be $M_i$). The columns of $M_i$ forms the Jordan chain:

$$ M_i = \begin{bmatrix}v_{i,1} & v_{i, 2} & \dots & v_{i, \mu_i}\end{bmatrix} $$

The first vector $v_{i,1}$ is the eigenvector, others are the generalized eigenvectors. Last index is $\mu_i$, the algebraic multiplicity of the eigenvalue $\lambda_i$.

Let's calculate the first column of $M_i$. We know that $AM_i = M_iJ(\lambda_i)$. Computing the product on column at time and we get (for the first column):

$$ Av_{i,1} = \lambda_iv_{i,1} $$

The above is the exact definition of eigenvector. Solving $(A-\lambda_iI)v_{i,1} = 0$ is easy, because of the structure of $A$:

$$ A-\lambda_iI = \begin{bmatrix} -\lambda_i & 1 &&& \\ & \ddots & 1 &&\\ &&\ddots&\ddots &\\ &&& -\lambda_i & 1\\ - \frac{a_0}{a_n} & - \frac{a_1}{a_n} & \cdots & -\frac{a_{n-2}}{a_n} & -\frac{a_{n-1}} {a_n} -\lambda_i \end{bmatrix} $$

First $n-1$ are linearly independent, $A-\lambda_iI$ is singular so the last row can be ignored. First $n-1$ gives us (note that $v_{i,1}^{1} \dots v_{i,1}^{n}$ are the coordinates of the vector):

$$\begin{align} v_{i,1}^{2} &= \lambda_i v_{i,1}^{1}\\ v_{i,1}^{3} &= \lambda_i v_{i,1}^{2} = \lambda_i^2 v_{i,1}^{1} \\ v_{i,1}^{4} &= \lambda_i v_{i,1}^{3} = \lambda_i^3 v_{i,1}^{1} \\ \vdots\\ v_{i,1}^{n} &= \lambda_i^{n-1} v_{i,1}^{1} \end{align} $$

Done! The general form of the first row of $M_i$ is then:

$$ M_{i,1} = \begin{bmatrix} 1 \\ \lambda_i \\ \lambda_i^2 \\ \vdots \\ \lambda_i^{n-1} \end{bmatrix} $$

$\endgroup$
2
$\begingroup$

I cannot decipher your notation, notably of your actual question. However, you seem to be asking why for a companion matrix eigenvalues cannot have geometric multiplicity larger than $1$, or equivalently why there is only a single unbroken Jordan block for each eigenvalue $\lambda$ (if there would be several Jordan blocks for $\lambda$, each one would give an independent eigenvector for $\lambda$).

The clue to this is that companion matrices $A$ of size $n$ cannot by annihilated by a nonzero polynomial $P$ of degree less than $n$. To see that apply $P(A)$ on the right to the row vector $(1~0~0~\ldots~0)$, and see that it doesn't vanish.

Now assume that eigenvalue $\lambda$ has multiplicity $m$ as root of the characterisitic polynomial, but there is no Jordan block for $\lambda$ of size $m$ (there are multiple blocks, each of size less than $m$). That means that the corresponding block of $(A-\lambda I)^{m-1}$ already vanishes. It follows that the polynomial obtained by dividing the characterisitic polynomial by $X-\lambda$ already annihilates $A$. But we have seen that it cannot.

$\endgroup$

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.