4
$\begingroup$

I need to answer (affirmatively, I hope) the following question:

In a Lie group $G$ whose Lie algebra $\mathfrak{g}$ is equipped with an $\mathrm{Ad}$-invariant scalar product, is the open subset $$\{g \in G \mid \operatorname{ker} (1 + \operatorname{Ad}_{g}) = \{0\}\}$$ dense in $G$?

In the case $G=\mathrm{GL}_n\mathbb{C}$, an affirmative answer goes as follows, which works for any complex reductive group as well. First we can restrict to the open dense subset $U$ of matrices with distinct eigenvalues. By conjugating any $g\in U$ to a diagonal matrix, we can see that $\mathrm{Ad}_g$ does not have eigenvalue $-1$ if and only if the proportion of any two eigenvalues is not $-1$, and this condition is dense.

However, since I'm writing an article in which everything else except this only relies on the property that $\mathfrak{g}$ is equipped with an $\mathrm{Ad}$-invariant scalar product, I feel like not to include the reductive assumption. But I fail to prove or find counterexamples to the above question in this case. In fact, I can't find any counterexample even for general Lie groups without any assumption.


Addendum As noticed by Peter McNamara, we should add the assumption that $G$ is connected, since for example in the case of the real orthogonal group $O(2n)$, the adjoint action of any elements not in the identity component has eigenvalue $-1$.

$\endgroup$
3
  • 6
    $\begingroup$ At least an assumption that G is connected should be included to rule out the orthogonal group O(2). $\endgroup$ Jul 13, 2012 at 2:13
  • 4
    $\begingroup$ I feel the proof should proceed by showing that the function f(g)=char poly of Ad(g) evaluated at -1 is an analytic function on G. Then on each connected component, it either is identically zero, or nonzero on a dense set. Since f doesn't vanish at identity, QED for connected G. $\endgroup$ Jul 13, 2012 at 5:57
  • $\begingroup$ This actually solves the problem... You should put it as an answer. $\endgroup$
    – Xin Nie
    Jul 13, 2012 at 8:54

2 Answers 2

7
$\begingroup$

[This is an embellished version of the answer I gave in the comment section]

The key to the solution lies in recalling that when one is playing with Lie groups, the functions that one encounters are not mere $C^\infty$ functions but are actually real-analytic. There is some discussion of this in the second edition of Knapp's book "Lie Groups Beyond an Introduction".

Now for $g\in G$, let $P_g(t)$ denote the charcteristic function of $\operatorname{Ad}(g)$. Then the function $g\mapsto P_g(-1)$ is an analtyic function on $G$. Since this function is nonzero at the identity, it is nonzero on a dense subset of the identity component of $G$. (so we have an affirmative answer for connected $G$.)

As is currenly noted in both the question and in the comments, for disconnected $G$, the example of the even orthogonal group limits what we can say away from the identity component.

I guess the case of $\pi_0(G)$ odd is still unresolved. Aber das ist eine andere Geschichte und soll ein andermal erzählt werden.

$\endgroup$
4
$\begingroup$

If $\pi_0(G)$ is odd, it's dense. For each nonzero connected component, for some $n$, $g \to g^{2^n}$ maps the connected component to itself. Take an element $g$ of the connected component. For some $m$, $g^{2^{nm}}$ has no $-1$ eigenvalues. (we can ignore the non-root of unity eigenvalues, and the roots of unity eventually become roots of unity of odd order, thus not $-1$.) So some element of the connected component has no $-1$ eigenvalues, so we can follow Peter McNamara's argument.

Note that these arguments work for any representation, not just the adjoint.

$\endgroup$
1
  • $\begingroup$ Although it's almost certainly clear, I did a double take: in "If $\pi_0(G)$ is odd, it's dense", 'it' is "the set satisfying the given condition" (no eigenvalue $-1$ for the adjoint action on the Lie algebra), not $\pi_0(G)$ (which wouldn't make sense anyway). (And of course "$\pi_0(G)$ is odd" really means "$\operatorname{card} $\pi_0(G)$ is odd".) $\endgroup$
    – LSpice
    May 30, 2019 at 0:07

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.