3
$\begingroup$

Given an elementary abelian $p$-group $A$ of order $p^n$ for $n\geq 2$ and choose a subgroup $H$ of order $p$ from $Aut(A)\cong GL(n,p)$. We can use semi-direct product $A\rtimes B$ to construct a nonabelian $p$-group containing an elementary abelian maximal subgroup. Is there any other example and a complete characterization of $p$-groups satisfying this property?

$\endgroup$
1
  • 2
    $\begingroup$ The "modular $p$-group of order $p^3$" is the semidirect product $\mathbf{Z}/p^2\mathbf{Z} \rtimes \mathbf{Z}/p\mathbf{Z}$, where the right factor acts on the left factor via multiplication by $1 + p$. There is an elementary abelian subgroup of order $p^2$. Taking a direct product with an elementary abelian group gives you a non-split example of any order. I would guess that's basically it for odd $p$. $\endgroup$ Aug 3, 2021 at 14:32

1 Answer 1

6
$\begingroup$

Not every such group has this form, but they can be classified.

Let $G$ be a $p$-group containing an elementary abelian $p$-group $A$ as a maximal subgroup.

Being maximal, it is an index $p$ normal subgroup. Let $g$ be a generator of the quotient $G/A$. Then $g$ acts by conjugation on $A$ as a matrix $\sigma \in GL_n(\mathbb F_p)$

The unique eigenvalue of $\sigma$ is $1$ and thus it can be put in Jordan normal form, writing it as a block-diagonal matrix with $m_k$ unipotent blocks of size $k$ for some sequence $m_k$ satisfying $\sum_k k m_k =n$. Furthermore, since a unipotent block of size $k$ has order $>p$ if $k>p$, we must have $m_k =0 $ for $k>p$.

Let $\alpha \in A = g^p$. We can express any element of $G$ as $a g^r$ with $a\in A$ and $r \in \{0,\dots, p-1\}$. The multiplication table of $G$ is determined by the relations $ ga g^{-1} = \sigma(a)$ and $g^p = \alpha$. We obtain a group with $A$ as an index $p$ subgroup this way if and only if $\alpha$ is $\sigma$-invariant.

The space of $\sigma$-invariants in unipotent block of size $k$ for $\sigma$ is one-dimensional. So it might seem that the space of choices of $\alpha$ is $\mathbb F_p^{\sum_{k=1}^p m_k}$. However, up to automorphisms of $\mathbb F_p^n$ commuting with $\sigma$, if $\alpha$ has a nontrivial component in any block of size $k$ then we can cancel the component in every other block of size $k$ and every block of greater size. Thus, the only information that matters is the least $k$ such that $\alpha$ has a component in a block of size $k$, or equivalently, the least $k$ such that $\alpha$ is not in the image of $(\sigma-1)^k$. Call this $k_\alpha$, and if $\alpha=0$, set $k_\alpha=\infty$.

The groups with relation $g^p=\alpha_1$ and $g^p=\alpha_2$ are equivalent if we can write $ g^p =\alpha_1, (ag)^p=\alpha_2$, or in other words if $a + \sigma(a) + \sigma^2(a) + \dots + \sigma^{p-1}(a) = \alpha_2- \alpha_1 $. Restricting to any block of size $<p-1$, $a + \sigma(a) + \sigma^2(a) + \dots + \sigma^{p-1}(a)=0$, so $\alpha_1,\alpha_2$ become equivalent under this if and only if their difference is supported in blocks of size $p$. In other words, the cases $k_{\alpha}=p$ and $k_\alpha=\infty$ are equivalent.

The last source of equivalence of the pair $(G,A)$ is changing $g$ for a power of $g$, but this just has the effect of multiplying $a$ by a constant and thus is already accounted for.

So the classification of pairs $(G,A)$ where $G$ is a $p$-group and $A$ is a maximal elementary abelian subgroup is given by tuples $(m_1,\dots, m_p,k_{\alpha})$ of nonnegative integers where $\sum_{k=1}^p k m_k = n$, $1\leq k_{\alpha} \leq p-1$ or $k_{\alpha}=\infty$, and $m_{k_\alpha}>0$ unless $k_\alpha=\infty$.

$G$ is nonabelian if and only if $m_1 \neq n$.

Semidirect products only occur in the case $k_\alpha=\infty$, so there are lots of nonabelian groups that aren't semidirect products.

$\endgroup$
2
  • 2
    $\begingroup$ That's a really nice description! Just to be absolutely clear about your equivalence relation, I think you are regarding two pairs $(G_1,A_1)$ and $(G_2,A_2)$ as equivalent if there is a group isomorphism $\phi:G_1 \to G_2$ with $\phi(A_1)=A_2$. Is that right? (So it is conceivable that $G_1$ and $G_2$ could be isomorphic but not with an isomorphism mapping $A_1$ to $A_2$, and in that case $(G_1,A_1)$ and $(G_2,A_2)$ would not be equivalent.) $\endgroup$
    – Derek Holt
    Aug 3, 2021 at 19:00
  • 3
    $\begingroup$ @DerekHolt Yes, that's my intention. I think one can go a little further and consider isomorphisms just of $G$, by looking at the case where there exists another index $p$ elementary abelian subgroup. Such a subgroup, intersected with $A$, must necessarily be a codimension $1$ subspace invariant under $\sigma$, so we can only have $m_1 =n-2$, $m_2=1$, all other $m_k$ zero. Furthermore we need $(ag)^p$ trivial for some $a$, so $k_\alpha=\infty$, and so the group is entirely determined by this property and $n$, so we don't get any additional identifications. $\endgroup$
    – Will Sawin
    Aug 3, 2021 at 19:35

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.