2
$\begingroup$

I have a reductive group $G$, sitting inside $GL_n$, everything over some algebraically closed field of characteristic 0. Let $\mu$ be a cocharacter (of a maximal torus in $G$) such that the induced filtration on the vector space is a 1-step filtration, with the filtered piece having dimension $n/2$ (I'm assuming that $n$ is even). Let $U$ be the subgroup of $G$ which acts trivially on the associated graded, and let $L$ be the centraliser of the cocharacter (in $G$). Is the adjoint action of $L$ on $Lie(U)$ irreducible?

And does the same result hold for any cocharacter (without the added conditions on the filtration)?

$\endgroup$

1 Answer 1

3
$\begingroup$

The adjoint action of $L$ on ${Lie}(U)$ is very rarely irreducible, but this does happen in your case provided that the derived subgroup of $G$ is simple. If $Lie(L)$ intersects properly with at least two simple ideals of $Lie(G)$ then intersecting ${Lie}(U)$ with those ideals we'll obtain some non-trivial proper $L$-submodules of $Lie(U)$.

So assume from now that the derived subgroup of $G$ is simple. One knows that the centraliser $L$ of $\mu$ in $G$ is a Levi subgroup of the parabolic subgroup $P_G(\mu)=L\cdot U$ of $G$. Now $G\subset GL_{n}$ by our general assumption and, moreover, we know that the unipotent radical, $\widetilde{U}$, of the parabolic subgroup $P_{GL_n}(\mu)=Z_{GL_n}(\mu)\cdot \widetilde{U}$ of $GL_n$ is abelian (this can be checked directly by looking at $n\times n$ matrices). As $U\subseteq \widetilde{U}$ we see that $U$ is abelian, too. But then it is well known that $L$ acts irreducibly on ${Lie}(U)$, although all proofs of this fact seem to rely on the classification of simple algebraic groups (there is a related discussion of irreducibility on this website prompted by a question of Jim Humphreys).

$\endgroup$

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.