4
$\begingroup$

Let $G$ be a reductive group, say over an algebraically closed field of characteristic zero.

We have the following definitions for a pair of parabolic subgroups $P_1$ and $P_2$ with Levi quotients $M_1$ and $M_2$:

  • Recall that $P_1$ and $P_2$ are said to be associate if the projections $P_1 \cap P_2 \to M_1$ and $P_1 \cap P_2 \to M_2$ are surjective.
  • Furthermore, they are called opposite if the intersection $P_1 \cap P_2$ is a Levi subgroup.
  • I will say that $P_1$ and $P_2$ are somewhat opposite if there exists a parabolic $P$ containing both, with the property that the images of $P_1$ and $P_2$ in the Levi quotient of $P$ are opposite parabolics.

Is it true that any pair of associate parabolics are connected by a sequence of pairwise somewhat opposite parabolics?

$\endgroup$
4
  • $\begingroup$ Observation: upon choosing a maximal torus in $G$ that is contained in $P_1 \cap P_2$, this can be viewed as purely a question about root systems. But I don't know the answer for root systems. $\endgroup$
    – LSpice
    Commented Jun 9, 2023 at 18:35
  • $\begingroup$ In standard English, the question "Can any pair etc.etc.?" would normally be construed as "Is there any pair that can etc.etc.?", so that "any" becomes an existential quantifier. In other contexts "any" is a universal quantifier, as in "Any fool can do that." And "Any member can serve as chair of the committee" is not synonymous with "Every member can serve as chair of the committee" nor with "Some member can serve as chair of the committee". English-speaking mathematicians use the word "any" too much. This should be phrased as "Can every pair etc.etc." $\qquad$ $\endgroup$ Commented Jun 21, 2023 at 17:42
  • $\begingroup$ @Michael Hardy: Your pedantry is presumptuous and completely irrelevant to the question. $\endgroup$ Commented Jun 21, 2023 at 23:28
  • $\begingroup$ @Satan'sMinion : I disagree. There are contexts in which this makes a concrete practical difference in whether one is understood. $\endgroup$ Commented Jun 23, 2023 at 21:21

1 Answer 1

2
$\begingroup$

I have found a proof that any two associate parabolics can be connected by a sequence somewhat opposite parabolics, but a reference would still be appreciated.

First let me remark two parabolic subgroups $P_1$ and $P_2$ are associate if and only if there exists $M \hookrightarrow P_1 \cap P_2$ which is a Levi subgroup of both $P_1$ and $P_2$.

Next let us do some warm-ups about pairs $(P_1, P_2)$ of associate parabolics. We fix a common Levi subgroup $M \hookrightarrow P_1 \cap P_2$ for convenience.

  1. If $P_1$ and $P_2$ both contain a Borel subgroup $B$, then $P_1 = P_2$.
  2. If $P_1$ is submaximal (this means that $P_1$ is a maximal element in the poset of proper parabolics) then $P_2$ is either equal or opposite to $P_1$.
  3. Suppose that $Q$ is any parabolic subgroup containing $M$, and write $L$ for its Levi quotient. Also write $P_1^L$ and $P_2^L$ for the images of $P_1 \cap Q$ and $P_2 \cap Q$ in $L$, respectively. Then $P_1^L$ and $P_2^L$ are associate parabolics in $L$ containing the image of $M$ as a common Levi.

Proof of (1). This is clear because $P_1 \cap P_2$ is a parabolic subgroup.

Proof of (2). Choose a maximal torus $T \subset M$ and a Borel $T \subset B \subset P_1$. The assumption that $P_1$ is submaximal means that $P_1$ contains every positive simple root except for, say, $\alpha$. If $P_2$ contains $\alpha$ then $P_2 = P_1$; otherwise $P_2$ contains $-\alpha$ and is the opposite of $P_1$ with respect to our pinning.

Proof of (3). It suffices to show that the image of $M$ in $L$ is a Levi subgroup of $P_1^L$. Since the composition $P_1 \cap Q \hookrightarrow Q \to L$ has unipotent kernel, this is equivalent to the assertion that $M \hookrightarrow P_1 \cap Q$ is a Levi subgroup. But the composition $P_1 \cap Q \hookrightarrow P_1 \to M_1 = M$ has unipotent kernel by the same token and the inclusion $M \hookrightarrow P_1 \cap Q$ splits this map.

Now let me begin the proof of the original assertion with the same notation as above. That is, $P_1$ and $P_2$ are associate parabolics and we choose a common Levi $M$. For a parabolic $Q$ containing $P_1$ submaximally (by this I mean that the image of $P_1$ in the Levi quotient $Q \to L$ is a submaximal parabolic), I will describe a procedure that either does nothing to $P_1$ or creates a parabolic subgroup $P_1'$ with the following properties:

  1. It is somewhat opposite to $P_1$.
  2. It is associate to $P_2$.
  3. Its intersection with $P_2$ is strictly larger than $P_1 \cap P_2$.

Consider the parabolic subgroups $P_1^L$ and $P_2^L$ of $L$, which are either equal or opposite by (2) and (3) of the warm-up. In the first case we do nothing. Otherwise, define $P_1'$ to be the preimage of $P_2^L$ along $Q \to L$. It has the requisite properties: (1) holds by construction and (2) holds because $P_1'$ contains $M$ as a Levi subgroup. Finally (3) is verified by choosing a maximal torus in $M$ and comparing roots.

By virtue of this procedure, it remains to verify the following claim: if $P_1^L = P_2^L$ for every parabolic $Q$ containing $P_1$ submaximally, then $P_1 = P_2$.

Proof of the claim. Let's choose a pinning as in the proof of (1) and check that $P_2$ contains every positive simple root $\alpha$. This would imply that $P_2$ contains $B$ and hence equals $P_1$. Write $I$ for the subset of the Dynkin diagram corresponding to $P_1$.

  • If $\alpha \in I$ then $\alpha$ belongs to $M$ and hence $P_2$ also.
  • If $\alpha \notin I$ then we consider the parabolic $Q_\alpha$ corresponding to $I \cup \{\alpha\}$. Note that $Q_\alpha$ contains $P_1$ submaximally and that $\alpha$ belongs to the Levi of $Q_\alpha$.
$\endgroup$

You must log in to answer this question.

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