Extremal rays in the Hermitian eigenvalue problem for arbitrary types (Q2193979)

Generalizing the main results of the first author [Math. Ann. 373, No. 3--4, 1103--1133 (2019; Zbl 1416.15030)] for the case \(G = \mathrm{SL}_n\), this work studies the rays of the eigencone -- also called the tensor cone -- for an arbitrary connected complex semisimple groups \(G\). Upon fixing a maximal torus and Borel subgroup of \(G\), the finite-dimensional irreducible \(G\)-modules \(V(\lambda)\) are parametrized by dominant weights \(\lambda\). The tensor semigroup consists of all solutions \((\lambda_1,\lambda_2,\hdots,\lambda_s)\) to the problem: \(V(\lambda_1)\otimes \cdots\otimes V(\lambda_s)\) contains nonzero \(G\)-invariants; the closure of this semigroup is a convex polyhedral cone \(\Gamma(s,K)\) called the \textit{eigencone}. (The notation \(K\) stands for a maximal compact subgroup of \(G\) and hearkens to an original definition of the cone involving conjugacy classes in \(\mathfrak{k}\); if \(G = \mathrm{SL}_n\) then this eigencone parametrizes solutions to the Horn problem: possible sets of eigenvalues \(\lambda(A_1), \hdots, \lambda(A_s)\) of Hermitian matrices satisfying \(A_1+\hdots+A_s = 0\).) The inequalities determining \(\Gamma(s,K)\) have been well-studied and determined without redundancy by several contributions, in particular, due to Horn, Klyachko, Knutson-Tao, Belkale, Berenstein-Sjamaar, Kapovich-Leeb-Millson, Belkale-Kumar and Ressayre. Namely, to certain data in Schubert calculus of any \(G/P\), one associates a regular facet \(\mathcal{F}\) of \(\Gamma(s,K)\). In the present paper, the authors reduce the problem of producing extremal rays to finding all the extremal rays on such a face \(\mathcal{F}\). They first construct a list of ``type I'' rays \(\delta_1,\hdots,\delta_q\) from the Schubert calculus data of \(\mathcal{F}\), making use of a rigidity result in the cohomology of certain line bundles proved by Belkale-Kumar-Ressayre. These rays project to an orthogonal basis of a subspace of \(\mathbb Q\Gamma(s,K)\), hence are linearly independent, and moreover there is a linear decomposition \[ \prod_b \mathbb Z_{\ge 0} \delta_b \times \mathcal{F}_2 \simeq \mathcal{F}. \] They identify \(\mathcal{F}_2\) as the image of a smaller eigencone \(\Gamma(s,K(L^{ss}))\) -- here \(L\) is the standard Levi subgroup of \(P\) -- under an explicit linear map \(\mathrm{Ind}\). Therefore the ``type II'' extremal rays of \(\mathcal{F}_2\) are images of extremal rays of \(\Gamma(s,K(L^{ss}))\), and proceeding inductively, one can in principle find all the rays of \(\mathcal{F}\), hence of \(\Gamma(s,K)\). The paper concludes with a formula for the dimension of the kernel of \(\mathrm{Ind}\) and with several examples.