4
$\begingroup$

Consider the Verma module $M(\lambda)=\oplus_{\mu \in \mathfrak{h}^{\star}} M(\lambda)_{\mu}$. Denote its BGG dual by $M(\lambda)^{\vee}=\oplus_{\mu \in \mathfrak{h}^{\star}} M(\lambda)^{\star}_{\mu}$ as in section $3.2$ of Humphreys' book.

When $\lambda$ is anti-dominant, $M(\lambda)$ is simple and $M(\lambda) \cong M(\lambda)^{\vee}$. My question is whether $$\dim M(\lambda)^{\vee}_{\mu}=\dim M(\lambda)_{-\mu}?$$ I have doubts because on one hand, I feel that since $M(\lambda) \cong M(\lambda)^{\vee}$ we should have $\dim M(\lambda)^{\vee}_{\mu}=\dim M(\lambda)_{\mu}$. On the other hand, I feel that the weights of $M(\lambda)^{\vee}$ should be negative of the weights of $M(\lambda)$. Hence, it might be true that $\dim M(\lambda)^{\vee}_{\mu}=\dim M(\lambda)_{-\mu}$.

Thank you for explaining.

$\endgroup$
4
  • $\begingroup$ It's useful to have some examples of Verma modules and their category $\mathcal{O}$ duals in mind (starting with rank $1$): the weight diagram is far from symmetric. Even when $\mu$ occurs as a weight of $M(\lambda)$ (with some multiplicity), it's usually not the case that $-\mu$ also occurs as a weight. (By the way, the possessive in English is awkward for names like mine ending in 's', but the simplest convention then is just to add an apostrophe.) $\endgroup$ Jan 6, 2018 at 18:51
  • $\begingroup$ I corrected the English. If $\mu $ is a weight of $M(\lambda)$, then can $-\mu$ be a weight of BGG dual of $M(\lambda)$? Also, is my question on the dimension true? It will be very helpful if you explain a little bit. $\endgroup$ Jan 6, 2018 at 19:41
  • $\begingroup$ To be more explicit, the weights and multiplicities are the same in any Verma module and its BGG dual; only the composition factor picture is turned upside down. So your question has a negative answer. In rank 1 for example, where weights can be identified with integers, the antidominant Verma module $M(-1)$ equals the simple module $L(-1)$ and is then isomorphic to its BGG dual $M(-1)^\vee$; the weights (with multiplicity 1) are $-1,-3,-5, \dots$ in both cases. $\endgroup$ Jan 6, 2018 at 21:33
  • $\begingroup$ To answer your newer question, it's quite possible (for weights which are not antidominant) that some weights of a Verma module have negatives which are also weights: again this is seen in rank 1, when you start with a nonnegative highest weight and get two composition factors. The top one is finite dimensional, with positive weights occurring along with their negatives as in the classical case. (Of course, $M(\mu)$ with $\mu = 0$ is an extreme example of this.) Dual Verma modules have the same weights and multiplicities. $\endgroup$ Jan 6, 2018 at 21:44

1 Answer 1

4
$\begingroup$

If you compute the dual of a weight module $M$ using the usual notion of dual of a module over a Lie algebra, then you get a module $M^{\ast}$ such that $M^{\ast}_{-\mu}$ is the vector space dual of $M_{\mu}$ (and thus has the same dimension). This why you think the displayed equation is correct.

But that's not the notion of dual that anyone uses in category $\mathcal{O}$, because as you can see it doesn't send objects in category $\mathcal{O}$ to objects in category $\mathcal{O}$. Thus, $M^\vee$ is something different: it's the usual vector space dual (EDIT: not actually the full one; as the OP writes, it's the sum of the vector space duals of the individual weight spaces, which is a version of graded dual), with the action twisted by the Cartan involution $E_i\mapsto -F_i, F_i\mapsto -E_i, H_i\mapsto -H_i$ (for the classical groups, this is transpose and negate). Because of the negative sign in front of $H_i$, this flips the sign of weights, and $M^\vee_\mu$ is the vector space dual of $M_\mu$ (and thus the same dimension).

$\endgroup$
4
  • $\begingroup$ @Ben: Your quick sketch of the BGG construction is partly out of focus, especially the usual vector space dual (too large when $M$ is infinite dimensional); their construction didn't use the Cartan involution, but rather a related anti-involution. See the remarks at the end of $\S4$ in mathnet.ru/php/… (the English translation is OK but has "complete" rather than "full subcategory" etc.). See also Gaitsgory's lectures math.harvard.edu/~gaitsgde/267y/index.html or my 2008 book. $\endgroup$ Jan 7, 2018 at 14:21
  • 1
    $\begingroup$ @JimHumphreys Of course I'm well aware that the full dual is too large, but I didn't want to water down my point with topological issues (especially since the OP seemed to be straight on this point). I also am fairly sure that I'm doing exactly what BGG did (my Russian is a little rusty), but I broke it up into two steps of taking usual dual (which implicitly chooses the anti-involution of taking negative) and then twisting by Cartan, whereas BGG does it in one step composing these (to get an anti-involution which is the identity on the Cartan). $\endgroup$
    – Ben Webster
    Jan 7, 2018 at 18:57
  • $\begingroup$ I think this is a good description of what is going on, except I agree with Jim that calling it the usual vector space dual might lead to confusion (for someone other than the OP potentially reading this answer without thoroughly reading the question). I suppose a short term for what it really is would be "graded dual" (with the grading by the dual of the Cartan). $\endgroup$ Jan 17, 2018 at 8:39
  • $\begingroup$ @TobiasKildetoft Since my not saying "graded dual" is apparently upsetting for people, I added a note. $\endgroup$
    – Ben Webster
    Jan 17, 2018 at 13:58

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.