6
$\begingroup$

Let:

  • $G$ be a real semisimple Lie group;
  • $\rho$ be an irreducible representation of $G$ on a finite-dimensional real vector space;
  • $A$ be a "Cartan subspace" of $G$ (a Lie subalgebra which is abelian, composed of hyperbolic elements, and maximal);
  • $L$ be the centralizer of $A$ in $G$ (this group is often called "$MA$");
  • $W = N_G(A)/Z_G(A)$ be the restricted Weyl group of $G$, and $w_0$ its longest element.

I would like to classify the representations that satisfy the following two conditions:

  1. the space $V^L$ of points of $V$ fixed by every element of $L$ is nonzero;
  2. the action of $w_0$ on $V^L$ is nontrivial.

An important particular case (which it would already be nice to have) is when $G$ is split. In that case, $V^L$ is simply the zero weight space, which is nonzero whenever the highest weight is an integer combination of roots. So we just need to find the action of $w_0$ on the zero weight space.

This looks like it should have already been done somewhere, but I have been unable to find any references so far. Any pointers would be appreciated!

$\endgroup$
5
  • 1
    $\begingroup$ In your special case, which is equally about algebraic groups and their finite dimensional rational representations, there has been some scattered work done which might be relevant. I wrote up notes on this a couple of years ago at people.math.umass.edu/~jeh/pub/zero.pdf $\endgroup$ Jul 21, 2016 at 14:08
  • $\begingroup$ Yes, I have seen this paper at the question mathoverflow.net/questions/185797/… . It does not seem to contain the answer to my question but it is certainly useful to have in mind. $\endgroup$ Jul 21, 2016 at 14:35
  • 2
    $\begingroup$ Most of the time (not $A_{n>1}, D_{odd}, E_6$), $w_0$ acts as $-1$ on the weight lattice. In that case (and when $V$ is finite-dimensional) the trace of $w_0$ on $V^L$ is the trace of $w_0$ on $V$, i.e. computable from WCF. $\endgroup$ Jul 26, 2016 at 13:04
  • 2
    $\begingroup$ Yes, I have thought about this. I agree with your comment up to the last comma. Unfortunately however, the Weyl Character Formula yields $0/0$ (at least for $B_2$). $\endgroup$ Jul 27, 2016 at 12:48
  • $\begingroup$ Together with Bruno Le Floch, we solved this question for split $G$: arxiv.org/abs/1806.00347 . We are now working on the general case. $\endgroup$ Apr 15, 2019 at 10:15

1 Answer 1

3
$\begingroup$

The action of $w_0$ can be computed by Allen Knutson's suggestion (whether or not $-1\in W$). As Allen says, suppose $g\in N(T)$ is a representative of $w_0$. Then

$\text{trace}(\pi(w_0)) \text{ on } {V_0}=\text{trace}(\pi(g))\text{ on } V$

(since the only element of the root lattice fixed by $w_0$ is $0$).

It is known that $g$ is conjugate to $\exp(\pi i\rho^\vee)$. Note that $g^2=\exp(2\pi i\rho^\vee)\in Z(G)$, and $\lambda(g^2)=e^{2\pi i\langle\lambda,\rho^\vee\rangle}=1$. Write $\Delta(V)$ for the set of weights of $V$. These are in the root lattice; write $\text{ht}(\mu)$ for the height (writing $\mu$ as $\pm$ a sum of simple roots). Then

$$\text{trace}(\pi(w_0))=|\{\mu\in\Delta(V)\mid \text{ht}(\mu)\text{ even}\}|-|\{\mu\in \Delta(V)\mid ht(\mu)\text{ odd}\}|$$

Since $\pi(w_0)$ on $V_0$ is diagonalizable with $\pm1$ on the diagonal this computes the number of $1/-1$ eigenvalues.

$\endgroup$
2
  • $\begingroup$ One remark: the statement that "the only element of the root lattice fixed by $w_0$ is $0$" is not true in general! For example with $G = SL_3(\mathbb{R})$, the vector $e_1 - 2e_2 + e_3 = (e_1 - e_2) + (e_3 - e_2)$ is a nonzero $w_0$-invariant element of the root lattice. $\endgroup$ May 15, 2017 at 12:41
  • $\begingroup$ When we can ignore this issue (for example if we exclude from consideration $A_{n > 1}$, $D_{2n+1}$ and $E_6$), I agree that this gives an algorithm to compute the trace of $w_0$ for one given representation. (I already knew this algorithm, and a few others.) However, what I am looking for is a description of the set of all representations for which $w_0$ has maximal trace. If course it may be possible to use the former in order to get the latter, but I do not know how to do it, and I am afraid it is not so easy to do. $\endgroup$ May 15, 2017 at 12:41

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.