3

Consider the general linear group $G=GL(n,q)$. Let $d$ be an integer and suppose that there is an element in $G$ of order $d$. Is there any routine way to find this element?

Robin
  • 3,227
user33209
  • 113

3 Answers3

4

Short answer: Rational canonical form.

Suppose $g \in \newcommand{\GL}{\operatorname{GL}}\GL(n,q)$ has order $d$. Then the minimal polynomial of $g$ (1) has degree at most $n$, (2) divides $x^d-1$, and (3) does not divide $x^i-1$ for any $1 \leq i \leq d$. Conversely, given any such polynomial of degree $m \leq n$, its companion matrix is an element of order $d$ in $\GL(m,q)$ and there is a natural embedding of $\GL(m,q) \leq \GL(n,q)$.

For instance, there is an element of $\GL(5,3)$ of order $13$ with minimal polynomial $x^4+x+1$, so we form the companion matrix $M$ of $x^4+x+1$, and then embed it as $M_5$ in $\GL(5,3)$: $$M=\begin{bmatrix} 0 & 0 & 0 & -1 \\ 1 & 0 & 0 & -1 \\ 0 & 1 & 0 & 0 \\ 0 & 0 & 1 & 0 \\ \end{bmatrix}, \qquad M_5 = \begin{bmatrix} 0 & 0 & 0 & -1 & 0 \\ 1 & 0 & 0 & -1 & 0 \\ 0 & 1 & 0 & 0 & 0 \\ 0 & 0 & 1 & 0 & 0 \\ 0 & 0 & 0 & 0 & 1 \end{bmatrix} $$

Given a possible order $d$, one can factor $x^d-1$ over $\mathbb{F}_q$ and take enough irreducible factors so that they don't divide any $x^i-1$, but not so many as to make the total degree more than $n$. This is is possible to do by what I said above. There may be more than one solution, because there can be more than one conjugacy class of elements of order $d$.

We knew there was an element of order 13 since 13 divides $3^3-1$, and this particular element of order 13 chose the irreducible factors $x-1$ and $x^3+x^2+x-1$. If you just want a “simplest” element of order 13, we would take the companion matrix $N$ of $x^3+x^2+x-1$ and embed it as $N_5$ in $\GL(5,3)$: $$N=\begin{bmatrix} 0 & 0 & 1 \\ 1 & 0 & -1 \\ 0 & 1 & -1 \end{bmatrix}, \qquad N_5 = \begin{bmatrix} 0 & 0 & 1 & 0 & 0 \\ 1 & 0 & -1 & 0 & 0 \\ 0 & 1 & -1 & 0 & 0 \\ 0 & 0 & 0 & 1 & 0 \\ 0 & 0 & 0 & 0 & 1 \end{bmatrix}$$

Marc van Leeuwen gave an answer describing exactly which $d$ can occur using a similar technique.

I should say though that Don Antonio's answer is generally a better idea for specific orders. If $d$ divides $q^i-1$ for some $1 \leq i \leq n$, then one uses a block diagonal matrix with only one non-(zero or identity) block, which is a primitive $d$th root of unity in $\mathbb{F}_{q^i}^\times$. If $d$ divides $q$, then you use upper triangular matrices with 1s on the diagonal, but there is a maximum such order, and it is not related to $q$ exactly, only to $n$ and the prime divisor of $q$. If you have a number that can be factored into such numbers, then you can take an appropriate product of such matrices (being careful to arrange the blocks so that all of the factors commute). However, writing down the details of this in general is too difficult for me (though see Derek Holt's answer), and I would express it using a canonical form such as Jordan or Rational (which can be taken to coincide if you just want one element of each order).

Jack Schmidt
  • 55,589
  • Good explanation +1. I just have an issue with the last parenthesised remark: you need Rational here, not Jordan (which lives over an extension field of $\Bbb F_q$ only; I just can't see how it could be the same as the Rational canonical form which lives over $\Bbb F_q$). – Marc van Leeuwen Jun 26 '13 at 14:11
  • @MarcvanLeeuwen: I'll edit the last paragraph anyways after I've digested both your and Derek's answers. By “Jordan” I meant the block version as in your answer where eigenvalues are block matrices rather than single numbers. By “coincide” I meant that one could take the minimal polynomial to be square-free. Since what I've written is neither exactly true nor what I meant to say, it'll definitely be part of the rewrite. :-) – Jack Schmidt Jun 26 '13 at 14:57
2

I'll have a go at answering this, but I don't guarantee to get the details right.

Let $q$ be a power of the prime $p$ and write $d = d_{i_1}d_{i_2} \cdots d_{i_k}p^a$, where $1 \le i_1 < i_2 < \cdots i_k \le n$, $d_{i_j}|(q^{i_j}-1)$ and the $d_{i_j}$ are pairwise coprime.

If you can't do that then there is no element of order $d$ in ${\rm GL}(n,q)$. In fact, if $a=0$, then we need $i_1+i_2+\cdots i_k \le n$ and if $a>0$ we need $p^{a-1}+1 + i_1+i_2+\cdots i_k \le n$ (but we can manage with degree one less than that if $i_1=1$).

Assuming those conditions hold, then you can construct an element of order $d$ in ${\rm GL}(n,q)$ as a diagonal sum of elements of order $d_{i_j}$ in ${\rm GL}(i_j,q)$ and, if $a>0$, a unipotent element of order $p^a$ in ${\rm GL}(p^{a-1}+1,q)$. (But if $i_1=1$, then we can combine the unipotent element with the element in ${\rm GL}(1,q)$.)

Derek Holt
  • 90,008
  • I've corrected $d_1=1$ to $i_1=1$. According to a paper of Celler and Leedham-Green on computing the order of a matrix, for $g \in {\rm GL}(n,q)$ to have order divisible by $p^a$ with $a>0$, there must be an irreducible factor that occurs with multiplicity at least $p^{a-1}+1$ in the factorization of the minimal polynomial of $g$. I was using that to try and find the most efficient way of getting an element of order $d$. – Derek Holt Jun 26 '13 at 14:20
  • I think I've convinced myself that my suggestion is not as efficient as the simple way you describe. – Jack Schmidt Jun 26 '13 at 20:42
0

For any $\,d\in\Bbb N\;$ , The element

$$A_d:=\begin{pmatrix}\zeta&0&0&\ldots&0&0\\0&1&0&\ldots&0&0\\\ldots&\ldots&\ldots&\ldots&\ldots&\ldots\\0&0&\ldots&\ldots&0&1\end{pmatrix}\in\text{GL}_n(\Bbb C)\;,\;\;\zeta:=e^{\frac{2\pi i}{d}}$$

has order $\,d\,$ .

If GL$_(n,q)\;$ is meant to be the general linear group over a field with $\,q\,$ elements, $\,q\,$ a prime, then $\,d\;$ has to be a divisor of group's order...

DonAntonio
  • 211,718
  • 17
  • 136
  • 287
  • 2
    There are elements of other orders in $\operatorname{GL}(n,q)$. If $d$ is the prime divisor of $q$, or if $d$ divides $q^i-1$ for some $1 \leq i \leq n$. There are some other $d$ that exist as well (if $p$ divides $q$ and $k$ is smaller than some particular function of $p$ and $n$, then it has elements of order $p^k$ as well, but the function differs for small and large $p$; also there are some mixed elements). – Jack Schmidt Jun 26 '13 at 13:31
  • 2
    I've given a description of the orders that occur in $GL(n,\Bbb F_q)$ in this answer; it is somewhat messy. Certainly all divisors of $q^i-1$ with $1\leq i\leq n$ occur, from the multiplicative group of $\Bbb F_{q^i}$ realised as $\Bbb F_q$-linear operators. Also $p^k$ for the characteristic $p$ when $p^{k-1}<n$ (so @JackSchmidt the mystery function of $p$ and $n$ is just $\lceil\log_p n\rceil$). Some mixing occurs, subject to the restriction that it must fit inside dimension$~n$, which leads to the messy part. – Marc van Leeuwen Jun 26 '13 at 13:45
  • In fact in my question $d$ divides $q^i-1$ for some $1≤i≤n$. for example If $q=2$ and $d=2^i-1$ for some $i$, is there any routin element of order $d$? – user33209 Jun 26 '13 at 13:53