A survey of the additive eigenvalue problem (with Appendix by M. Kapovich) (Q2257457)

This is a survey article on the classical Hermitian eigenvalue problem and its generalization to an arbitrary connected reductive group. The classical Hermitian eigenvalue problem starting with the work of \textit{H. Weyl} [Math. Ann. 71, 441--479 (1912; JFM 43.0436.01)] is the following: determines all triples of \(n\)-tuples of nonincreasing real numbers: \(\lambda=(\lambda_1,\dots,\lambda_n)\), \(\mu=(\mu_1,\dots,\mu_n)\) et \(\nu=(\nu_1,\dots,\nu_n)\) such that there exist Hermitian matrices \(A\), \(B\), \(C=A+B\) with eigenvalues \(\lambda\), \(\mu\) et \(\nu\). The solution was conjecture given by \textit{A. Horn} [Pac. J. Math. 12, 225--241 (1962; Zbl 0112.01501)] and proved by \textit{A. A. Klyachko} [Sel. Math., New Ser. 4, No. 3, 419--445 (1998; Zbl 0915.14010)] and \textit{A. Knutson} and \textit{T. Tao} [J. Am. Math. Soc. 12, No. 4, 1055--1090 (1999; Zbl 0944.05097)] The sets \(\lambda\), \(\mu\) et \(\nu\) are the solutions of the inequalities: \(\sum^{n}_{i=1}\nu_i =\sum^{n}_{i=1}\lambda_i+\sum^{n}_{i=1}\mu_i\) et \(\sum_{k\in K}\nu_i \leq\sum_{i\in I}\lambda_j+\sum_{j\in J}\mu_i\) for an appropriate family of sets \((I,J,K)\). It was proved by \textit{P. Belkale} [Compos. Math. 129, No. 1, 67--86 (2001; Zbl 1042.14031)] and \textit{A. Knutson} et al. [J. Am. Math. Soc. 17, No. 1, 19--48 (2004; Zbl 1043.05111)] that the above system of inequalities is overdetermined. The above Hermitian eigenvalue problem has been generalized to an arbitrary complex semisimple group. Let \(G\) be a connected, semisimple complex algebraic group. We fix a Borel subgroup, a maximal torus \(H\), and a maximal compact subgroup \(K\) and let \(\underline{h}\) and \(\underline{k}\) be their Lie algebras. Let \(\underline{h}_+\subset \underline{h}\) be the positive Weyl chamber. There is a natural homeomorphism \(f : \underline{k}/K \rightarrow \underline{h}_+\). For any positive integer \(s\), define the eigencone \[ \overline{\Gamma}_s(\underline{g})=\{(h_1,\dots,h_s)\in \underline{h}^s_+\mathrm{\;such \;that\;} \exists (k_1,\dots,k_s)\in \underline{k}^s \mathrm{\;with \;}\sum k_i=0\mathrm{\;et\;}f(k_i)=h_i\;\forall i\}. \] The aim of the general additive eigenvalue problem is to find the inequalities describing the eigencone \(\overline{\Gamma}(\underline{g})\) explicitly. The general additive eigenvalue problem was solved by \textit{A. Berenstein} and \textit{R. Sjamaar} [J. Am. Math. Soc. 13, No. 2, 433--466 (2000; Zbl 0979.53092)], \textit{M. Kapovich} et al. [Geom. Funct. Anal. 19, No. 4, 1081--1100 (2009; Zbl 1205.53037)] and \textit{P. Belkale} and \textit{S. Kumar} [Invent. Math. 166, No. 1, 185--228 (2006; Zbl 1106.14037)] using the deformed cup product (known as the Belkale-Kumar product) on singular homology with integral coefficients \(H^* (G/P , Z)\) of a flag variety \(G/P\). Then, one can determine the saturated tensor semigroup \(\overline{\Gamma}(G)\). Similar to the eigencone, one defines the saturated tensor semigroup \[ \overline{\Gamma}_s(G)=\{(\lambda_1,\dots,\lambda_s)\mathrm{\;dominant\;characters\;such\;that\;} \exists\; \] \[ [V(N(\lambda_1)\otimes\dots\otimes V(N(\lambda_s))]^G\neq0 \mathrm{\;for\;some \;}N\geq1\}. \] \(\overline{\Gamma}_s(G)\) is identified, via the Killing form, with the set of dominant characters in \(\overline{\Gamma}_s(\underline{g})\). The proof of previous result follows from the Hilbert-Mumford criterion for semistability, Kempf's maximally destabilizing one parameter subgroups, Kempf's parabolic subgroups associated to unstable points and the notion of Levi-movability. An ``explicit'' determination of the eigencone hinges upon understanding the Belkale-Kumar product in the Schubert basis, for all the maximal parabolic subgroups P. This product is easier to understand than the usual cup product since in general ``many more'' terms in the Belkale-Kumar product in the Schubert basis drop out. However, the Belkale-Kumar product has a drawback in that it is not functorial. Finally the author consider the saturation problem which aims at connecting the tensor product semigroup \[ T_s(G)= \{(\lambda_1,\dots,\lambda_s) \mathrm{\;dominant\;weights\;such \;that\;} [V (\lambda_1)\otimes \dots\otimes V(\lambda_s )]^G\neq0 \} \] with the saturated tensor product semigroup \(\Gamma_s (G)\). An integer \(d \geq 1\) is called a saturation factor for \(G\), if for any \((\lambda,\mu,\nu) \in \Gamma_3(G)\) such that \(\lambda+\mu+\nu\) belongs to the root lattice, then \((d\lambda,d\mu,d\nu)\in T_3 (G)\). Such a \(d\) always exists. If \(d = 1\) is a saturation factor for \(G\), we say that the saturation property holds for \(G\). The saturation theorem of \textit{A. Knutson} and \textit{T. Tao} [J. Am. Math. Soc. 12, No. 4, 1055--1090 (1999; Zbl 0944.05097)] asserts that the saturation property holds for \(G =\mathrm{SL}(n)\). A general result (though not optimal) on saturation factor is obtained by \textit{M. Kapovich} and \textit{J. J. Millson} [Groups Geom. Dyn. 2, No. 3, 405--480 (2008; Zbl 1147.22011)].