27
$\begingroup$

$\DeclareMathOperator\SL{SL}\DeclareMathOperator\Sp{Sp}$In a group $G$, an element $g$ is said to be primitive if there is no $h \in G$ and integer $n >1$ such that $g = h^n$. (For clarification, I consider finite order elements to be not primitive)

I was wondering, in the case $G$ is $\SL_n(\mathbb{Z})$ or $\Sp_{2n}(\mathbb{Z})$, if there exists a criterion for primitivity of matrices. I actually even struggle to find examples of primitives matrices in these groups.

In $\Sp_{2}(\mathbb{Z})=\SL_{2}(\mathbb{Z})$, the matrix $\pmatrix{1 & 1 \\ 0 & 1}$ is primitive. (This can be shown by considering its action on $\mathbb{H^2}$, for example.)

But this does not generalize (easily at least) to higher dimension. For example, and quite surprisingly maybe $$ \pmatrix {1 & 0 & 0& 1\\ 0 & 1 & 0 & 0\\ 0 & 1 & -1 & 1\\ 0 & 1& -1 & 0 }^3 = \pmatrix{1 & 1 & 0& 0\\ 0 & 1 & 0 & 0\\ 0 & 0 & 1 & 0\\ 0 & 0& 0 & 1} $$

Anyway, it seems like some things should be known, but it is very hard to find anything on google since primitive matrix usually means something else …. I would appreciate any input.

$\endgroup$
9
  • 10
    $\begingroup$ The MO post "Condition for a matrix to be a perfect power of an integer matrix" seems relevant, although it doesn't have a complete answer to your question: mathoverflow.net/questions/375584/… $\endgroup$ Jan 21, 2021 at 16:35
  • 1
    $\begingroup$ @GeoffRobinson wait, if $a^2=1$ then $a=a^3$? $\endgroup$ Jan 21, 2021 at 17:15
  • 2
    $\begingroup$ @DenisT I follow his definition in the first paragraph of the post. $\endgroup$ Jan 21, 2021 at 17:24
  • 3
    $\begingroup$ The problem in $\mathrm{GL}_2(\mathbf{Z})$ or $\mathrm{SL}_2(\mathbf{Z})$ is easily solvable using the amalgam decomposition. For $\mathrm{SL}_d(\mathbf{Z})$, $d\ge 3$ I don't even see in an obvious way that it's algorithmically solvable, but I'm pretty sure it is; how efficiently, I'm less sure. $\endgroup$
    – YCor
    Jan 21, 2021 at 19:17
  • 3
    $\begingroup$ About primitive: it can be shown that every element of infinite order in $\mathrm{GL}_n(\mathbf{Z})$ (or in any of its subgroups) of infinite order is a power of a primitive element: indeed every abelian subgroup in such a group is finitely generated (use that it's a discrete subgroup of its Zariski closure, which is a virtually connected abelian Lie group). $\endgroup$
    – YCor
    Jan 21, 2021 at 23:47

1 Answer 1

26
$\begingroup$

I actually even struggle to find examples of primitives matrices in these groups.

Here is a relatively easy sufficient condition. If $M \in SL_n(\mathbb{Z})$ is the $k^{th}$ power of some other matrix $N$ then every eigenvalue $\lambda$ of $N$ has the property that $\lambda^k$ is an eigenvalue of $M$, and conversely. If we can choose $M$ and an eigenvalue $\mu$ of $M$ such that every $k^{th}$ root of $\mu$ has the property that its minimal polynomial has degree larger than $n$, then $N$ cannot exist.

For this to work $\mu$ can't be a root of unity. As an explicit example we'll take $M = \left[ \begin{array}{ccc} 1 & 1 & 0 \\ 1 & 0 & 0 \\ 0 & 0 & -1 \end{array} \right]$, the Fibonacci matrix augmented by a $-1$ to live in $SL_3(\mathbb{Z})$, whose eigenvalues are $-1$ and the golden ratios $\phi, \varphi = \frac{1 \pm \sqrt{5}}{2}$. Any $\lambda$ such that $\lambda^k = \phi, \varphi$ for $k \ge 2$ generates an extension of $\mathbb{Q}$ which contains $\mathbb{Q}(\sqrt{5})$ and hence is either equal to $\mathbb{Q}(\sqrt{5})$ or else has even degree greater than $2$ and so at least $4$. Moreover $\lambda$ is an algebraic integer and a unit.

But $\phi, \varphi$ each generate the unit group of $\mathcal{O}_{\mathbb{Q}(\sqrt{5})} = \mathbb{Z}[\phi]$ (up to signs). So if $\mathbb{Q}(\lambda) = \mathbb{Q}(\sqrt{5})$ then $\lambda$ must be $\pm \phi$ or $\pm \varphi$, but this is ruled out by taking the absolute value. So the minimal polynomial of $\lambda$ must have degree at least $4$, which means $\lambda$ can't be an eigenvalue of a matrix in $SL_3(\mathbb{Z})$.

$\endgroup$
0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.