11
$\begingroup$

Given a finite group $G$, I am interested in finding a non-trivial proper subgroup $H$ of $G$ such that $\mathrm{Ind}_H^G\mathbf{1}$ contains all the irreducible representations of $G$, that is, permutation representation of $G$ on $G/H$ contains all the representations of $G$. I can see that if $G$ is abelian then this is not possible. Because, if $G$ is abelian and $H$ is such a subgroup then $\mathrm{Ind}_H^G\mathbf{1}$ will contain all $|G|$ many representations but $\mathrm{dim}(\mathrm{Ind}_H^G\mathbf{1})=[G: H]$ which is a contradiction.

Now, I am thinking about nilpotent groups. Given a finite nilpotent group $G$, is it possible to have such a subgroup? What about solvable groups?

I was also looking at this problem from another point of view. If $\mathrm{Ind}_H^G\mathbf{1}$ contains all the irreducible representations of $G$, then by Frobenius reciprocity we can say that for every irreducible representation $(\rho, V)$ of $G$, $V^{H}\neq {0}$. Hence, there is a one-one correspondence between the simple modules for the group algebra $\mathbb{C}[G]$ and the simple modules for the Hecke algebra $e_H\mathbb{C}[G]e_H$ (where $e_H=\frac{1}{|H|}\underset{h\in H}{\sum}h$). And I want to ask if there is some sort of classification of groups having subgroup with the property discussed above.

Also, given a compact connected Lie group $G$, when does it possess a closed subgroup $H$ such that $L^2(G/H)$ contains all the representations of $G$?

I posted this in MSE. Got no answer. That is why I am posting it here.

$\endgroup$
12
  • 2
    $\begingroup$ If such an $H$ exists, we may assume that $H$ has prime order. Also, $H$ is not normal. So for the quaternion group of order $8$, no such $H$ exists. I haven't checked whether the necessary condition, that $G$ has a non-normal subgroup of prime order, is sufficient. $\endgroup$ Commented Sep 27 at 8:18
  • 2
    $\begingroup$ If $G$ is nilpotent, then every irreducible character $\chi$ of $G$ has the form ${\rm Ind}_{X}(G)(\lambda),$ where $X$ is a subgroup of $G$ necessarily containing $Z(G)$, and $\lambda$ is a linear character of $X$. Mackey's formula then gives a criterion to check whether $H$ has fixed points on the induced representation: something like, there are fixed points if and only if $X \cap gHg^{-1} \leq {\rm ker} \lambda$ for some $g \in G.$ Not sure how helpful that is. $\endgroup$ Commented Sep 27 at 10:15
  • 3
    $\begingroup$ I do not have a good answer, but with Michael Magee, we have an example that we have found meaningful, for operator algebraic considerations: the pairs $(G,H)$ equal to $(\mathrm{SL}_4(\mathbf Z/N\mathbf Z),\mathrm{SL}_2(\mathbf Z/N\mathbf Z))$ for integers $N$ (or equivalently $\mathrm{SL}_4(\mathbf Z_p),\mathrm{SL}_2(\mathbf Z_p))$ for primes $p$) have the property that all nonzero representation over $\mathbf C$ of $G$ have a nonzero $H$-invariant vector. This is false for $4$ replaced by $3$, see arxiv.org/abs/2312.03220 $\endgroup$ Commented Sep 27 at 14:06
  • 1
    $\begingroup$ @LSpice in a corner, say the upper-left. $\endgroup$ Commented Sep 27 at 15:24
  • 1
    $\begingroup$ If $\sigma\in G$ has order $2$ and $H=\langle\sigma\rangle$ does not work, then there exists an irrep $\rho$ such that $\rho(\sigma)=-{\rm Id}$, so that $\sigma$ is central of order $2$ in $G/\ker\rho$. In particular, in every nonabelian simple group, such an $H$ always works. $\endgroup$
    – Aurel
    Commented Oct 1 at 21:50

1 Answer 1

11
$\begingroup$

Let us write ($*$) for the property that a subgroup $H$ of a finite group $G$ has nonzero fixed vectors in every representation of $G$.

Proposition 1: if $G$ is nilpotent, then no nontrivial subgroup $H$ of $G$ satisfies ($*$).

Proof: Assume such an $H$ exists. We may assume that $H$ is cyclic, generated by some nontrivial $h\in G$. Let $1=G_0 \triangleleft G_1\triangleleft \dots \triangleleft G_n = G$ be a central series. For some $i$, the element $h$ belongs to $G_{i+1}$ but not to $G_i$. Therefore, the image of $h$ in $G/G_i$ is nontrivial, but central. For every irreducible representation of $G/G_i$, the element $h$ has fixed points, but since it is central it acts trivially (Schur's lemma); therefore $h$ is trivial in $G/G_i$, a contradiction.

More generally, let $G$ be a finite group. We have the following simple obstructions:

  • If $N\triangleleft G$ is a normal subgroup and $H$ satisfies ($*$), then so does the image of $H$ in $G/N$.
  • A nontrivial central subgroup $H$ of $G$ never satisfies ($*$).
  • Combining these two, a subgroup $H$ whose image is nontrivial and central in some quotient $G/N$ never satisfies ($*$).

There is a converse for subgroups of order $2$.

Proposition 2: Let $H$ be generated by an element $h$ of order $2$. Assume that the image of $h$ in every quotient $G/N$ is either trivial or noncentral. Then $H$ satisfies ($*$).

Proof: Let $(V,\rho)$ be an irreducible representation of $G$ and let $N=\ker\rho$. If $h$ is trivial in $G/N$, then $h$ acts trivially on $V$ and therefore $H$ has nonzero fixed points in $V$. If $h$ is noncentral in $G/N$, then $\rho(h)$ has at least two distinct eigenvalues, so it has $1$ as an eigenvalue, i.e. $H$ has fixed points in $V$.

Corollary 3: Let $G$ be a nonabelian finite simple group, and let $H$ be a subgroup of order $2$. Then $H$ satisfies ($*$).

Proof: there is no nontrivial quotient, and $Z(G)$ is trivial. So Proposition 2 applies.

Corollary 4: Let $G = S_n$ for $n\ge 4$, and let $H$ be generated by a double transposition. Then $H$ satisfies ($*$).

Proof: $H$ is not central, and belongs to all the proper normal subgroups. So Proposition 2 applies.

$\endgroup$
4
  • 3
    $\begingroup$ This is a beautiful answer! I would just add a small comment regarding Corollary 3 for the non-group theorists among us: every nonabelian finite simple group has even order (this is the Feit-Thompson Theorem), hence has an element of order 2. $\endgroup$ Commented Oct 2 at 21:10
  • $\begingroup$ @Aurel In the proof of Proposition $2$, "If $h$ is noncentral in $G/N$, then $\rho(h)$ has at least two distinct eigenvalues". Why is that so? $\endgroup$ Commented Oct 3 at 6:37
  • 1
    $\begingroup$ @SoumyadipSarkar First, $h$ is noncentral in $G/N$ iff $\rho(h)$ is noncentral in $\rho(G)$. Now, if $\rho(h)$ had only one eigenvalue, then if would be a homothety (having finite order, it is diagonalisable), so it would commute with $\rho(h)$. $\endgroup$
    – Aurel
    Commented Oct 3 at 7:32
  • $\begingroup$ I meant to write that it would commute with $\rho(G)$. $\endgroup$
    – Aurel
    Commented Oct 3 at 8:21

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .