Isospectral commuting variety, the Harish-Chandra \(\mathcal D\)-module, and principal nilpotent pairs (Q447769)

Fix a connected complex reductive group \(G\) with Lie algebra \(\mathfrak g\), and with \(T\subset G\) a maximal torus. The corresponding Cartan subalgebra is denoted \(\mathfrak t\) and the rank of \(\mathfrak g\) by \(\mathbf{r}\). With \(N(t)\) the normalizer of \(T\) in \(G\), \(W=N(T)/T\) is the Weil group. \(W\) acts on \(\mathfrak t\) via the reflection representation, and on \(\bigwedge^{\mathbf{r}}\mathfrak t\) by the sign character \(w\mapsto\text{sign}(w)\). The notation \(\mathfrak D=\mathcal D_{\mathfrak g\times\mathfrak t}\) is used for the sheaf of differential operators on \(\mathfrak g\times\mathfrak t\). The subalgebras of \(\mathcal D(\mathfrak g)\) and \(\mathcal D(\mathfrak t)\) formed by the differential operators with constant coefficients may be identified with the corresponding symmetric algebras \(\text{Sym}\mathfrak g\) and \(\text{Sym}\mathfrak t\). Letting the group \(G\) act on \(\mathfrak g\) by the adjoint action, Harish-Chandra defined a \textit{radial part} map \(\mathsf{rad}:\mathcal D(\mathfrak g)^G\rightarrow\mathcal D(\mathfrak t)^W\). This is an algebra morphism, and its restriction to the subalgebra of \(G\)-invariant polynomials (\(G\)-invariant differential operators with constant coefficients) reduces to the Chevalley isomorphism, that is \(\mathbb C[\mathfrak g]^G\overset\sim\rightarrow\mathbb C[\mathfrak t]^W\), (\((\text{Sym}\mathfrak g)^G\overset\sim\rightarrow (\text{Sym}\mathfrak t)^W\)). The assignment \(a\mapsto\text{ad} a\) gives a linear map \(\text{ad}:\mathfrak g\rightarrow\mathcal D(\mathfrak g)\) with image \(\text{ad}\mathfrak g\) and one can form the left ideal \(\mathfrak D(\text{ad}\mathfrak g\otimes 1)\subset\mathfrak D\). The Harish-Chandra module is the left \(\mathfrak D\)-module defined as \[ {\mathcal M}=\mathfrak D/(\mathfrak D(\text{ad}\mathfrak g\otimes 1)+\mathfrak D\{u\otimes 1-1\otimes\mathsf{rad}(u),\;u\in\mathcal(\mathfrak g)^g\}). \] It is known that the Harish-Chandra module is a simple holonomic left \(\mathfrak D\)-module of geometric origin. Thus \(\mathcal M\) comes with a natural strucure of Hodge module; there is a canonical Hodge filtration on \(\mathcal M\). Taking an associated graded sheaf with respect to the Hodge filtration gives a coherent sheaf \(\widetilde{\text{gr}}^{\text{Hodge}}\mathcal M\) on \(T^\ast(\mathfrak g\times\mathfrak t).\) The author proves that the support of this is closely related to the commuting scheme of the Lie algebra \(\mathfrak g\). This is the core of this article: To exploit the powerful theory of Hodge modules to deduce new results concerning commuting schemes using information about the sheaf \(\widetilde{\text{gr}}^{\text{Hodge}}\mathcal M\). The main results are listed as follows: Put \(\mathfrak G=\mathfrak g\times\mathfrak g\) and consider the diagonal action of \(G\) on \(\mathfrak G\). Then the commuting scheme \(\mathfrak C\) is defined as the scheme-theoretic zero fiber of the commutator map \(\kappa:\mathfrak G:\rightarrow\mathfrak g\) sending \((x,y)\) to \([x,y]\). \(\mathfrak C\) is generically reduced and irreducible, but it is a standing problem if this is also reduced. \(\mathfrak T=\mathfrak t\times\mathfrak t\subset\mathfrak G\) is an \(N(T)\)-stable closed subscheme of \(\mathfrak C\), and the resulting action of \(N(T)\) factors through the diagonal action of the Weyl group \(W=N(T)/T\). The restriction of polynomial functions give algebra maps \(\mathsf{res}:\mathbb C[\mathfrak G]^G\twoheadrightarrow\mathbb C[\mathfrak C]^G\rightarrow\mathbb C[\mathfrak T]^W.\) The \textit{isospectral commuting variety} is defined to be the algebraic set \(\mathfrak X=\{(x_1,x_2,t_1,t_2)\in\mathfrak C\times\mathfrak T|P(x_1,x_2)=(\mathsf{res})(t_1,t_2),\;\forall P\in\mathbb C[\mathfrak C]^G.\) \(\mathfrak X\) is considered a reduced closed subscheme of \(\mathfrak G\times\mathfrak T\). Fix an invariant bilinear form \(\langle -,-\rangle :\mathfrak g\times\mathfrak g\rightarrow\mathbb C\). One gets an identification \(T^\ast(\mathfrak g\times\mathfrak t)=\mathfrak G\times\mathfrak T\), and can view \(\widetilde{\text{gr}}^{\text{Hodge}}\mathcal M\) as a coherent sheaf on \(\mathfrak G\times\mathfrak T\). For this part, this gives the main results that there is a natural \(\mathcal O_{\mathfrak G\times\mathfrak T}\)-module isomorphism \(\psi_\ast\mathcal O_{\mathfrak X_{\text{norm}}}\rightarrow\widetilde{\text{gr}}^{\text{Hodge}}\mathcal M\), and that \(X_{\text{norm}}\) is a Cohen-Macaulay and Gorenstein variety with trivial canonical sheaf. Nice corollaries on the geometry of \(X_{\text{norm}}\) follows from these main results. The second theme of the text is the group action: There is a natural \(G\times W\)-action on \(\mathfrak g\times\mathfrak t\) (resp. on \(\mathfrak G\times\mathfrak T\)). Also, \(\mathbb C^\ast\) acts by dilations. Thus \(\mathfrak g\times\mathfrak t\) is made a \((G\times W\times \mathbb C^\ast)\)-variety, and \(\mathfrak G\times\mathfrak T\) a \((G\times W\times\mathbb C^\times\times\mathbb C^\times)\)-variety. Then \(X_{\text{norm}}\) is \((G\times W\times\mathbb C^\times\times\mathbb C^\times)\)-stable, and the resulting \((G\times W\times\mathbb C^\times\times\mathbb C^\times)\)-action induces one on \(X_{\text{norm}}\). On the other hand, \(G\times W\times\mathbb C^\times\) acts on \(\mathfrak g\times\mathfrak t\), and the Harish-Chandra module \(\mathcal M\) has the natural structure of a \((G\times W\times\mathbb C^\times)\)-equivariant \(\mathfrak D\)-module, respecting the Hodge filtration on \(\mathcal M\). This means that \(G\times W\times\mathbb C^\times\) acts naturally on \(\widetilde{\text{gr}}^{\text{Hodge}}\mathcal M\). Combining with the action of \(\mathbb C^\times\) coming from the grading, \(\widetilde{\text{gr}}^{\text{Hodge}}\mathcal M\) is viewed as a \((G\times W\times\mathbb C^\times\times\mathbb C^\times)\)-equivariant coherent sheaf on \(\mathfrak G\times\mathfrak T\). The previous isomorphism of the first main theorem respects the \((G\times W\times\mathbb C^\times\times\mathbb C^\times)\)-equivariant structures, so that the equivariant structure on the sheaf \(\mathcal O_{\mathfrak X_{\text{norm}}}\) makes the coordinate ring \(\mathbb C[\mathfrak X_{\text{norm}}]\) a bigraded locally infinite \((G\times W)\)-module. Constructing a resolution and making use of the Atiyah-Bott-Lefschetz fixed point theorem, the author gives an explicit formula for the bigraded \(T\)-character of \(\mathbb C[\mathfrak X_{\text{norm}}]\). A coherent sheaf on the commuting scheme is constructed: The first projection \(\mathfrak G\times\mathfrak T\) restricts to a map \(p:\mathfrak X\rightarrow\mathfrak C\), and therefore the composition \(\mathfrak X_{\text{norm}}\rightarrow\mathfrak X\rightarrow\mathfrak C\) factors through a morphism \(p_{\text{norm}}:\mathfrak X_{\text{norm}}\rightarrow\mathfrak C_{\text{norm}}\). \(p_{\text{norm}}\) is a finite \((G\times\mathbb C^\times\times\mathbb C^\times)\)-equivariant morphism and \(W\) acts along its fibers. Then a coherent sheaf can be constructed from invariant subsheaves in \((p_{\text{norm}})_\ast\mathcal O_{\mathfrak X_{\text{norm}}}\). Properties of these sheaves are part of the main results giving geometric properties of the commuting schemes. Given the action of a group, also the small representations are studied. A representation \(L\) is called small if the set of weights is contained in the root lattice of \(\mathfrak g\) and if \(2\alpha\) is not a weight of \(L\) for any root \(\alpha\). Descriptions of the \(W\)-isotypic components of the sheaf \(\mathcal R\) is given, where \(\mathcal R\) is the sheaf corresponding to the \(W\)-representation in the zero weight space of a small \(G\)-representation. Let \(x=(x_1,x_2)\in\mathfrak C\) be a regular point, let \(\mathcal R_x\) be the fiber at \(x\). By definition, \(\mathcal R_x=\mathbb C[p^{-1}_{\text{norm}}(x)]\) where \(p^{-1}_{\text{norm}}(x)\) is the scheme theoretic fiber of the morphism \(p_{\text{norm}}\) which is a \(W\)-stable finite subscheme of \(\mathfrak X_{\text{norm}}\). So \(\mathcal R_x\) is a finite-dimensional algebra with a \(W\)-action. A regular pair \(e=(e_1,e_2)\in\mathfrak C^r\) is called a \textit{principal nilpotent pair} if there exists a rational homomorphism \(g:\mathbb C^\times\times\mathbb C^\times\rightarrow G\) such that \(\tau_i e_i=\text{Ad}g(\tau_1,\tau_2)(e_i),\;i=1,2,\;\forall (\tau_1,\tau_2)\in\mathbb C^\times\times\mathbb C^\times\). The article contains a thorough study of the fibers \(\mathcal R_x\) over principal nilpotent pairs. The author proves that there is an isomorphism \(\mathcal R_e=\text{gr}^F\mathbb C[W\cdot h]\) and that it is a Gorenstein algebra. Moreover, the line \((\text{gr}^F\mathbb C[W\cdot h])^{\text{sign}}\) is the socle of that algebra. For special nilpotent pairs called exceptional, a bihomogeneous polynomial \(\Delta_e\) is given, which class gives a base vector for the one-dimensional socle. Finally, the relation to results of M. Haiman is worked out. In Haiman's work on the \(n!\)-theorem, he introduced a certain \textit{isospectral Hilbert scheme}, a reduced finite scheme over \(\text{Hilb}^n(\mathbb C^2)\) called \(\widetilde{\text{Hilb}}^n(\mathbb C^2)\) which he proved is a normal Cohen-Macaulay and Gorenstein scheme. In the present article, the locally free sheaf \(\mathcal R|_{\mathfrak C^r}\) is used to construct a rank \(n!\) algebraic vector bundle \(\mathcal P\) on \(\text{Hilb}^n(\mathbb C^2)\) whose fibers have the regular representation of the symmetric group. Then Haimans result proves that \(\mathcal P\) is isomorphic to the Procesi bundle. The article is a great work with a lot of deep and important results. The proofs are long and demands hard work to understand. The reader is supposed to have a good knowledge in Lie theory. However, with such knowledge the article is stringent and with an impressing stucture making the hard theory consistent and possible to follow.