Let $p$ be a prime number, and let $k=\mathbb{F}_p$ be the field of $p$ elements. Let $G=GL_n(k)$. We know that


so that the Sylow $p$-subgroups of $G$ have order $p^{\binom{n}{2}}$. One such subgroup is $U$, the upper-triangular unipotent subgroup consisting of all upper-triangular matrices with $1$'s on the diagonal. Let $A_{ij}=I_n+E_{ij}$ for $j>i$, where $E_{ij}$ is the matrix with a $1$ in the $ij$th entry and $0$'s elsewhere. If we find $i_1,\ldots,i_r,j_1,\ldots,j_r$ such that the $A_{i_kj_k}$ pairwise commute, then we have:

$$(\mathbb{Z}/p\mathbb{Z})^{\oplus r}\cong \langle A_{i_1j_1},\ldots,A_{i_rj_r}\rangle\subset U$$

Two questions:

  • Is every copy of $(\mathbb{Z}/p\mathbb{Z})^{\oplus r}$ inside of $U$ conjugate under $G$ to a subgroup of the form $\langle A_{i_1j_1},\ldots,A_{i_rj_r}\rangle$?
  • Do there exist distinct, conjugate subgroups of the form $\langle A_{i_1j_1},\ldots,A_{i_rj_r}\rangle$ and $\langle A_{i_1'j_1'},\ldots,A_{i_r'j_r'}\rangle$?

The largest value of $r$ for which such subgroups exist is $r=\lfloor n^2/4\rfloor$. In this case, I believe the answers to these questions are yes and no, respectively. I'd like to know if this also holds for smaller values of $r$.

  • $\begingroup$ let me know if you write up your notes. I'd be interested in reading them. These are pretty important ideas in a pretty important group. $\endgroup$ – Jack Schmidt Jul 12 '13 at 17:13

For $r=1$: No and yes as long as $n \geq 4$.

For $r$ maximal and $n$ odd, the second question has the answer “yes” as well.

Typically, I think the answer to the first question is very definitely “no”. Every subgroup has a “straightening” of the same order and roughly the same embedding that is of roughly the form you describe (they are called algebra subgroups, and may not be exactly products of root subgroups; see page 109 of the 3rd book of the GLS revised CFSG; it is called Malcev's striaghtening). However, for $r$ maximal it might be “yes” anyways.


For the first question: In $G=\operatorname{GL}(4,p)$, one has the following elementary abelian subgroup that is not conjugate to a root subgroup: $\langle A_{1,3} \cdot A_{1,4} \cdot A_{2,4}\rangle$. Every generator of every cyclic root subgroup has the property that $g-g^0$ has rank 1 which is invariant under GL conjugation, but this subgroup does not have that property.

For the second question: In $G=\operatorname{GL}(3,p)$, one has the following cyclic root subgroups are conjugate: $\langle A_{1,2} \rangle \cong_G \langle A_{2,3} \rangle$.

$r$ maximal

For $r=(n-1)(n+1)/4 = \lfloor n^2/4 \rfloor$, that is, for $n$ odd and $r$ maximal, the second question's answer is no: there are two elementary abelian root subgroups of this maximal rank that are conjugate under the inverse-transpose map, but not conjugate in GL. Each are formed of block matrices: $$P_{i,j} = \left\{ \begin{bmatrix} I & A \\ 0 & I \end{bmatrix} : A \in M_{i \times j}(\mathbb{F}_p) \right\}$$ The subgroups are $P_{(n-1)/2,(n+1)/2}$ and $P_{(n+1)/2,(n-1)/2}$.

  • $\begingroup$ Hi Jack, thanks again. In another question of mine that you answered, you used the term 'root subgroups'. Can you clarify what those are? Thanks. $\endgroup$ – Jared Jun 28 '13 at 23:03
  • $\begingroup$ Oh sorry, they are the subgroups generated by a $A_{i,j}$. $\endgroup$ – Jack Schmidt Jun 28 '13 at 23:05
  • $\begingroup$ If $i<j$ then they are called positive root subgroups, if $j=i+1$ then they are called simple root subgroups. If you work over fields other than $\mathbb{Z}/p\mathbb{Z}$, then the root subgroups also include all the matrices with the same zero-pattern (so $I_n +\alpha E_{i,j}$ for any $\alpha$ in the field). Many of the same ideas that work for GL work for other groups of lie type, though the explicit formula for root subgroups will change. $\endgroup$ – Jack Schmidt Jun 28 '13 at 23:08
  • $\begingroup$ An “algebra subgroup” is a subgroup of the form $\left\{ I_n + A : A \in \mathcal{A} \right\}$ where $\mathcal{A}$ is a set of matrices closed under addition, multiplication, and scalar multiplication. I think they are about as nice as products of root subgroups. You can use linear algebra to understand them, so that is good. $\endgroup$ – Jack Schmidt Jun 28 '13 at 23:11

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.