On supports of induced representations for \(p\)-adic special orthogonal and general spin groups (Q2068190)

Parabolic induction is a key tool in the representation theory of \(p\)-adic groups: it classifies representations of an ambient group \(G\) in terms of those of its Levi subgroups \(M\). An irreducible representation \(\pi\) can be classified in at least two ways by this theory. First, according to the Langlands classification, \(\pi\) is the unique maximal irreducible subrepresentation of an unramified twist of a tempered representation of a Levi. Second, \(\pi\) arises as the subquotient of the parabolic induction of a supercuspidal representation \(\sigma\) of some Levi \(M\). The pair \((M,\sigma)\) is called a cuspidal datum and its conjugacy class, the supercuspidal support of \(\pi\), is uniquely determined by \(\pi\). For the general linear group the situation is conceptually simpler: a cuspidal datum for \(\mathrm{GL}_n\) is just a set of supercuspidal representations of smaller general linear groups whose ranks sum to \(n\). In this setting Bernstein and Zelevinsky gave a third classification [\textit{I. N. Bernstein} and \textit{A. V. Zelevinsky}, Ann. Sci. Éc. Norm. Supér. (4) 10, 441--472 (1977; Zbl 0412.22015)]. They isolated a particular kind of cuspidal data, called segments, and classified irreducible representations \(\pi\) of \(\mathrm{GL}_n\), in the style of the Langlands classification, by parabolically inducing multisets of segments, called multisegments. Since parabolic induction is transitive, \(\pi\) can be constructed in two stages: parabolically induce the members of its multisegment individually, then parabolically induce the resulting representations. Stopping halfway, we produce a family \(\pi_1\), \(\dots\), \(\pi_k\) of irreducible representations of smaller \(\mathrm{GL}\)'s. This family carries much information about \(\pi\): for instance, \(\pi\) is essentially square-integrable if and only if the family is constant. The paper under review generalizes aspects of the construction of the previous paragraph to \(\mathrm{SO}_{2n}\), \(\mathrm{U}_n\), and \(\mathrm{GSpin}_n\), building on earlier work of the first author for \(\mathrm{SO}_{2n+1}\) and \(\mathrm{Sp}_{2n}\) [\textit{C. Jantzen}, Am. J. Math. 119, No. 6, 1213--1262 (1997; Zbl 0888.22013)], as well as \(\mathrm{O}_{2n}\) [\textit{C. Jantzen}, Can. J. Math. 57, No. 1, 159--179 (2005; Zbl 1065.22013)]. The correspondence so constructed respects temperedness, square-integrability, duality, and parabolic induction. There are various indeterminacies in the correspondence. The first, and most serious, is that it gives much less control over determinantal twists than the Bernstein-Zelevinsky theory. The other indeterminacies arise from structural properties of the classical groups under consideration: the sets of supercuspidal supports must be enlarged by dualizing in all types, using an outer automorphism in type \(D\), twisting by a central character in \(\mathrm{GSpin}\), and introducing some further indeterminacy in type \(B\). Relatedly, Moeglin and Tadić have classified the discrete series of classical groups in the style of Bernstein and Zelevinsky [\textit{C. Moeglin} and \textit{M. Tadic}, J. Am. Math. Soc. 15, No. 3, 715--786 (2002; Zbl 0992.22015)]. The first author generalized their work to tempered representations [\textit{C. Jantzen}, Manuscr. Math. 145, No. 3--4, 319--387 (2014; Zbl 1311.22025)]. The results of the paper under review are less explicit, but have the advantage of covering all irreducible representations.