A generalization of the Bott-Weil theorem and Euler number multiplets of representations (Q1590157)

This paper significantly generalizes the Bott-Borel-Weil theorem by looking at Lie algebra cohomology and replacing the complex structure with a space of spinors. In more detail, let \({\mathfrak g}\) be a complex semisimple Lie algebra, \({\mathfrak p}\) a parabolic subalgebra with Levi decomposition \({\mathfrak p} = {\mathfrak g}_1 + {\mathfrak n}\) where \({\mathfrak n}\) is the nilradical, and \(V_\lambda\) a finite dimensional irreducible representation of \({\mathfrak g}\) with highest weight \(\lambda\). It is known that BBW is equivalent to knowing the \({\mathfrak g}_1\)-module structure of \(H^j({\mathfrak n}, V_\lambda)\). In this context, the analysis was done by \textit{B. Kostant} [Ann. Math. (2) 74, 329-387 (1961; Zbl 0134.03501)]. There a Laplacian \(L_\lambda\) was introduced on the cochain complex \(C({\mathfrak n}, V_\lambda)=V_\lambda \otimes \wedge {\mathfrak n}^*\) so that \(\text{Ker}(L_\lambda) \cong H^*({\mathfrak n}, V_\lambda)\) as \({\mathfrak g}_1\) modules. Then it was shown \(\text{Ker}(L_\lambda)\) is multiplicity free and decomposes into representations indexed by \(W^1\), a special set of coset representatives of the Weyl group of \({\mathfrak g}_1\) in the Weyl group of \({\mathfrak g}\). For \(\tau\in W^1\), let \(\tau*\lambda=\tau(\lambda+\rho)-\rho\) where \(\rho\) is the semisum of positive roots. Then \(\text{Ker}(L_\lambda)=\sum_{\tau\in W^1} \text{Ker}(L_\lambda)^{\tau*\lambda}\) where the superscript denotes the primary \(\sigma_{\tau*\lambda}\)-component and \(\sigma_{\tau*\lambda}\) is the irreducible \({\mathfrak g}_1\)-module with highest weight \(\tau*\lambda\). Even stronger, it was shown \(\text{Ker}(L_\lambda)=\sum_{\tau\in W^1}C({\mathfrak n}, V_\lambda)^{\tau*\lambda}\). To generalize the situation replace \({\mathfrak g}_1\) with any reductive Lie subalgebra, \({\mathfrak r}\), with the same rank as \({\mathfrak g}\). Let \(B_{{\mathfrak g}}\) be the Killing form on \({\mathfrak g}\) and orthogonally decompose \({\mathfrak g} = {\mathfrak r} + {\mathfrak q}\). Write \(\text{Cliff}({\mathfrak q})\) for the Clifford algebra over \({\mathfrak q}\) with respect to \(B_{{\mathfrak g}}\) restricted to \({\mathfrak q}\) and \(S\) for the unique irreducible spin module. By the action \({\mathfrak r}\) on \({\mathfrak q}\), view \(S\) as an \({\mathfrak r}\)-module and consider the \(V_\lambda\bigotimes S\). The main point is that in the case of \({\mathfrak r} = {\mathfrak g}_1\), \(V_\lambda\bigotimes S\) differs by \(C({\mathfrak n},V_\lambda)\) only by tensoring with a one-dimensional representation, \(\sigma_{\rho({\mathfrak n})}\), of \({\mathfrak g}_1\). Therefore in the case of \({\mathfrak r} = {\mathfrak g}_1\), it is easy to pass back and forth from \(C({\mathfrak n},V_\lambda)\) and \(V_\lambda\bigotimes S\) while in the general case, one only has \(V_\lambda\bigotimes S\). This paper establishes a BBW-type theorem for general equal rank \({\mathfrak r}\) and \(V_\lambda\bigotimes S\). In this general case, one does not have the Laplacian \(L_\lambda\). Rather, a certain cubic Dirac operator \(\square_\lambda\) is used instead. The kernel of this operator on \(V_\lambda\bigotimes S\), also the highest eigenvalue of an appropriate Casimir element, is shown to be a multiplicity free \({\mathfrak r}\)-module. The components are similarly indexed by the corresponding \(W^1\) with highest weights \(\tau \bullet \lambda = \rho({\mathfrak n}) + \tau * \lambda\).