On algebraic integrability of Gelfand-Zeitlin fields (Q969616): Difference between revisions

Let \(\mathfrak{g}=\mathfrak{gl}(n,\mathbb{C})\). Denote by \(x_i\) the upper left \(i\times i\) corner of the matrix \(x\in\mathfrak{g}\) and let \(\mathfrak{g}_i\) denote all such matrices \(x_i\). Additionally let \(G_i\) be the corresponding Lie group \(\mathrm{GL}(i,\mathbb{C})\) acting by conjugation, and define \(f_{ij}(x)=\mathrm{tr}((x_i)^j)\) for \(1\leq i\leq n\) and \(1\leq j\leq i\). An element \(x\in \mathfrak{g}\) is called \textit{strongly regular} if the collection of all \(df_{ij}(x)\) are together linearly independent. The set of all strongly regular elements \(\mathfrak{g}_{\mathrm{sreg}}\) is open and Zariski dense in \(\mathfrak{g}\). The authors of this paper stratify \(\mathfrak{g}_{\mathrm{sreg}}\) using the following data: (1) let \(\mathfrak{l}_i\) be a Levi subalgebra of \(\mathfrak{g}_i\) with prescribed block sizes \(n_1,\dots,n_k\), (2) let \(\mathfrak{z}_{i,\mathrm{gen}}\) be the elements in the center of \(\mathfrak{l}_i\) whose centralizer in \(\mathfrak{g}_i\) is exactly \(\mathfrak{l}_i\), and (3) let \(e^i\) be the Jordan canonical form of a nilpotent element in \(\mathfrak{l}_i\). Then for each \(1\leq i\leq n\), define \(D_i=G_i\cdot(\mathfrak{z}_{i,\mathrm{gen}}+u_i)\subset \mathfrak{g}_i\) and let \(\mathcal{D}=(D_1,\dots,D_n)\). Each \(D_i\) consists of all regular elements of \(\mathfrak{g}_i\) whose Jordan form has blocks of sizes \(n_1,\dots,n_k\). Also, denote \(\mathfrak{z}_{\mathcal{D}}=\mathfrak{z}_{1,\mathrm{gen}}\times \cdots\times \mathfrak{z}_{n,\mathrm{gen}}\). With respect to \(\mathcal{D}\), define a stratum to be \(X_\mathcal{D}=\{x\in\mathfrak{g}_{\mathrm{sreg}}\;|\;x_i\in D_i\}\), and furthermore let \(\hat{\mathfrak{g}}_{\mathcal{D}}=\{(x,z_1,\dots,z_n)\in X_{\mathcal{D}}\times \mathfrak{z}_{\mathcal{D}}\;|\;x_i\in G_i\cdot(z_i+u_i)\}.\) Note that \(\mathfrak{g}_{\mathrm{sreg}}=\cup_{\mathcal{D}} X_{\mathcal{D}}\). Let \(\mu:\hat{\mathfrak{g}}_{\mathcal{D}}\to X_\mathcal{D}\) be projection onto the first factor. Let \(L_i\) be the Lie group corresponding to \(\mathfrak{l}_i\) and define \(W_i=N_{G_i}(\mathfrak{l}_i)/L_i\) and \(\Sigma_{\mathcal{D}}=W_1\times\cdots \times W_n\). Each \(W_i\) is a product of permutation groups acting by permuting blocks of the same size. The authors first main theorem is the following: the morphism \(\mu:\hat{\mathfrak{g}}_{\mathcal{D}}\to X_\mathcal{D}\) is an étale \(\Sigma_{\mathcal{D}}\)-covering. Moreover, each of \(\hat{\mathfrak{g}}_{\mathcal{D}}\) and \(X_\mathcal{D}\) are smooth, connected, irreducible varieties of dimension \(n(n-1)/2+\dim\mathfrak{z}_{\mathcal{D}}\). The Lie algebra \(\mathfrak{g}\) possesses a natural Poisson structure. Let \(\mathfrak{a}\) be the Lie algebra spanned by the Hamiltonian vector fields on \(\mathfrak{g}_{\mathrm{sreg}}\) associated to the functions \(f_{ij}\) defined above (these are called Gelfand-Zeitlin vector fields). The simply connected holomorphic group \(A\) associated to \(\mathfrak{a}\) is isomorphic to \(\mathbb{C}^{n(n-1)/2}\). The authors show that each \(X_{\mathcal{D}}\) is a union \(A\)-orbits in \(\mathfrak{g}_{\mathrm{sreg}}\), so the \(A\)-action restricts to \(X_{\mathcal{D}}\). Utilizing Poisson geometry, the authors prove their second main theorem: the Lie algebra \(\mathfrak{a}\) lifts via \(\mu\) to a Lie algebra \(\hat{\mathfrak{a}}\) associated to the connected abelian algebraic group \(Z_{\mathcal{D}}=Z_{G_1}(\mathfrak{z}_1+u_1)\times\cdots \times Z_{G_1}(\mathfrak{z}_1+u_1)\), \(Z_{\mathcal{D}}\) acts algebraically and freely on \(\hat{\mathfrak{g}}_{\mathcal{D}}\), and the tangent space to every \(Z_{\mathcal{D}}\)-orbit is \(\hat{\mathfrak{a}}\); i.e., the \(\hat{\mathfrak{a}}\)-action algebraically integrates to the action of \(Z_{\mathcal{D}}\). This work extends results obtained by \textit{B. Kostant} and \textit{N. Wallach} [Progr. Math. 243, 319--364 (2006; Zbl 1099.14037) and Progr. Math. 244, 387--420 (2006; Zbl 1099.14038)].