Homological branching law for \((\mathrm{GL}_{n+1}(F), \mathrm{GL}_n(F))\): projectivity and indecomposability (Q2041860)

Let \(G_n = \mathrm{GL}_n(F)\) where \(F\) is a non-Archimedean local field, and let \(\mathrm{Alg}(G_n)\) denote the abelian category of smooth representations of \(G_n\) with coefficients in \(\mathbb{C}\). The paper investigates several homological properties of the restriction of representations from \(G_{n+1}\) to \(G_n\), where \(G_n\) is embedded as the upper-left block inside \(G_{n+1}\). Specifically, one studies \(\mathrm{Hom}_{G_n}(\pi|_{G_n}, \pi')\) or even their \(\mathrm{Ext}\)-analogs, where \(\pi \in \mathrm{Alg}(G_{n+1})\) and \(\pi' \in \mathrm{Alg}(G_n)\). As such, it is an instance of branching laws in representation theory. This provides a far-reaching generalization of the Gross-Prasad conjectures for \(\mathrm{GL}_n \times \mathrm{GL}_{n+1}\), which in its original form presuppose the genericity of \(\pi\) and \(\pi'\). Most of the results in this paper apply to smooth irreducible representations in general, and they are obtained by purely algebraic means. The author obtained numerous non-trivial results in this direction. For the sake of simplicity, let us confine ourselves to the main theorems presented in the Introduction. Theorem 1.1 asserts that for all irreducible \(\pi\), its restriction \(\pi|_{G_n}\) is projective if and only if \(\pi\) is relatively projective in the sense that either \(\pi\) is essentially square-integrable, or that \(n+1\) is even and \(\pi \simeq \rho_1 \times \rho_2\) for some cuspidal representations \(\rho_1, \rho_2\) of \(G_{(n+1)/2}\). Theorem 1.3 asserts that each nonzero Bernstein component \(\tau\) of \(\pi|_{G_n}\) is strongly indecomposable in the sense that any two nonzero \(G_n\)-submodules of \(\tau\) have nonzero intersection. In particular, every Bernstein component \(\pi|_{G_n}\) is indecomposable. Theorem 1.6 furnishes the branching law in the other direction, which takes a much simpler form: \(\mathrm{Hom}_{G_n}(\pi', \pi|_{G_n}) \neq \{0\}\) if and only if \(\pi\) and \(\pi'\) are both \(1\)-dimensional and \(\pi' = \pi|_{G_n}\). The main tool is Zelevinsky's theory, specifically the Bernstein-Zelevinsky induction from \(\mathrm{Alg}(G_{n-i})\) to \(\mathrm{Alg}(G_n)\), and their left adjoint functors, usually known as derivatives. This allows one to access \(\pi|_{G_n}\) by means of the Bernstein-Zelevinsky filtration, which implies that \(\pi|_{G_n}\) is glued from products of derivatives with Gelfand-Graev representations on smaller groups. The arguments are highly technical, and they seem to work only for general linear groups or their inner forms thus far. We only indicate the crucial and deep Theorem 1.5: for all irreducible \(\pi \in \mathrm{Alg}(G_n)\) and \(i \geq 0\) which is not the level \(i^*\) of \(\pi\) (i.e.\ \(\pi^{(i^*)} \neq 0\) and \(\pi^{(j)} = 0\) whenever \(j > i^*\)), the shifted derivatives \(\pi^{[i]}\) and its ``left'' or transposed variant \({}^{[i]}\pi\) share no isomorphic irreducible quotients, and ditto for the irreducible submodules. This exemplifies the critical observation that one should often work with two towers of Bernstein-Zelevinsky inductions simultaneously, which are related to each other by transpose, roughly speaking.