5
$\begingroup$

$\newcommand{\g}{\mathfrak{g}}$Let $G$ be a compact Lie group and $\g$ its Lie algebra. I came across the the very important result that $G/T$ ($T$ a maximal torus of $G$) can be identified to a coadjoint orbit. However it is not at all clear to me how one can show this result. I guess we must somehow prove that there always exists an $F\in \g^*$ such that $\operatorname{Stab}(F)\cong T$. But this is not clear at all from the definition of $\operatorname{Stab}(F)$ so I guess the proof for that fact must be using more elegant ideas than that. So can anyone explain what ideas are used to show this or reference somewhere?

$\endgroup$

4 Answers 4

4
$\begingroup$

To be very explicit let's take a look at the case $G = U(n), T = U(1)^n$. As Allen says, by finding a suitable invariant form we can look at adjoint orbits rather than coadjoint orbits. Here $\mathfrak{g} = \mathfrak{u}(n)$ consists of the $n \times n$ Hermitian matrices. The generic element of $X \in \mathfrak{g}$ is semisimple with distinct eigenvalues, and for such an element it's a straightforward exercise to see that the stabilizer with respect to conjugation by $G$ is a conjugate of $T$ (namely the conjugate which is diagonal in a basis of eigenvectors for $X$).

$\endgroup$
1
  • 1
    $\begingroup$ Of course, in the spirit of being concrete, we might as well also be concrete about the invariant form; probably the trace form $(X, Y) \mapsto \operatorname{tr}(X Y)$ on $\mathfrak u(n)$ is most natural. $\endgroup$
    – LSpice
    Sep 13, 2022 at 0:57
3
$\begingroup$

Use the Haar measure on $G$ (compact!) to average a metric, obtaining a $G\times G$-invariant metric, and thus an identification $\mathfrak g \cong \mathfrak g^*$. Also, the geodesic spray $\mathfrak g \to G$ defined using this metric is $G$-equivariant. Hence a $G$-orbit in $\mathfrak g^*$ near enough $0$ is $G$-isomorphic to a conjugacy class in $G$.

Now take a topological generator $t$ of $T$, i.e. $\overline{\langle t\rangle} = T$. Then the conjugacy class $G\cdot t$ of $t$ is $G/Z_G(t) = G/Z_G(T)$. Now you need to know the standard fact that for $G$ connected, tori are self-centralizing. (Personally, I prove that by using the Bruhat decomposition of $G\cdot t$ to show it's simply-connected, and therefore the covering map $G/T \to G/Z_G(T)$ must be trivial.)

$\endgroup$
2
$\begingroup$

Fix a regular element $\lambda$ in $Lie(T)\subset Lie(G)$, then the coadjoint orbit $Ad(G) \lambda$ is isomorphic to $G/T.$ Best,

$\endgroup$
1
  • $\begingroup$ Strictly speaking, to satisfy the conditions of the question we must work (for example, as @AllenKnutson suggests, via a $G$-invariant bilinear form on $\operatorname{Lie}(G)$) on $\operatorname{Lie}^*(G)$ instead. $\endgroup$
    – LSpice
    Sep 13, 2022 at 0:56
0
$\begingroup$

Although there are three good answers already, perhaps you would like the explicit description $F = \sum \mathrm d\alpha$, where $\alpha$ runs over some system of simple roots.

$\endgroup$

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.