Supercuspidal representations of \(\mathrm{GL}_n(\mathrm{F})\) distinguished by a Galois involution (Q2325555)

Let \(F/F_0\) be a quadratic extension of non-Archimedean local fields of residual characteristic \(p\neq 2\), and denote the Galois involution by \(\sigma\). Let \(\omega = \omega_{F/F_0}\) be the quadratic character of \(F_0^\times\) trivial on the image of the norm \(N_{F/F_0}\). Denote \(G=GL_n(F)\) and \(G^\sigma = GL_n(F_0)\). Let \(R\) be an algebraically closed field of characteristic \(\ell \neq p\). One may ask under what conditions is an irreducible \(R\)-representation \(\pi\) of \(G\) distinguished resp. \(\omega\)-distinguished, i.e., satisfying \(\mathrm{Hom}_{G^\sigma}(\pi,1)\neq 0\) resp. \(\mathrm{Hom}_{G^\sigma}(\pi,\omega\circ \det)\neq 0\). Classically, when \(R=\mathbb{C}\), it is known that a distinguished irreducible representation is necessarily \(\sigma\)-selfdual, which means that the contragredient \(\pi^\vee\) of \(\pi\) is isomorphic to \(\pi\circ \sigma\). Some converse statements are also known, e.g., when \(R=\mathbb{C}\) and char\((F)=0\), if a discrete series representation is \(\sigma\)-selfdual, then it is either distinguished or \(\omega\)-distinguished (the Dichotomy Theorem) but not both (the Disjunction Theorem). In the paper under review, the author provides certain distinction criteria for supercuspidal representations, those not appearing as sub-quotients of proper parabolically induced representations, under mild conditions on \(F\) and \(R\). His criteria involve two structural data of a given supercuspidal: a tamely ramified extension \(D/F\) equipped with an involution extending \(\sigma\), and a character \(\delta_0\) of \((D^\sigma)^\times \), either trivial or equal to \(\omega_{D/D^\sigma}\). He asserts that when \(p\neq 2\), a \(\sigma\)-selfdual supercuspidal \(R\)-representation of \(G\) is distinguished if and only if either \(\ell=2\), or \(\ell\neq 2\) and certain conditions on the data \((F/F_0,D/D^\sigma,\delta_0)\) are satisfied. The precise conditions are stated in Theorem 1.2. The method he uses is purely local. We first view an irreducible supercuspidal \(\pi\) as a cuspidal representation, which means that its proper Jacquet modules are all zero. It is known that \(\pi\) is compactly induced from an extended maximal simple type, or a type for short, which is a finite dimensional irreducible representation \(\boldsymbol{\lambda}\) of a compact-mod-center open subgroup \(\boldsymbol{J}\). The pair \((\boldsymbol{J},\boldsymbol{\lambda})\) is uniquely determined by \(\pi\) up to \(G\)-conjugacy. The type \(\boldsymbol{\lambda}\) can be decomposed into a tensor product \(\boldsymbol{\boldsymbol{\kappa}}\otimes \boldsymbol{\rho}\). Here \(\boldsymbol{\boldsymbol{\kappa}}\) is an extension of the Heisenberg representation associated to the simple character attached to \(\boldsymbol{\lambda}\) (see Section 5D for the meanings of these terminologies), and \(\boldsymbol{\rho}\) is a type whose restriction to the unique maximal compact subgroup \({J}\) is of level zero. By definition, \(\boldsymbol{\rho}\) is trivial on the normal pro-p-subgroup \(J^1\) of \(J\), and \(\boldsymbol{\rho}|_J\) is inflated from a cuspidal representation \(\rho\) of the finite reductive quotient \(J/J^1\cong GL_m(\boldsymbol{l})\), for some \(m\in \mathbb{N}\) and some finite field extension \(\boldsymbol{l}\) over the residue field \(\boldsymbol{k}\) of \(F\). It is the residue field of the tame parameter field \(T\) associated to \(\boldsymbol{\lambda}\), a tamely ramified extension of \(F\) equipped with an involution extending \(\sigma\). The field \(D\) is unramified over \(T\) of degree \(m\). He shows that if \(\pi\) is cuspidal and \(\sigma\)-selfdual, then it contains a \(\sigma\)-selfdual type, which means that \(\boldsymbol{J}\) is \(\sigma\)-invariant and \(\boldsymbol{\lambda}^\vee\cong \boldsymbol{\lambda}\circ \sigma\). In this case, \(\pi\) is distinguished if and only if it contains a distinguished \(\sigma\)-selfdual type. He then separates into two cases, depending on whether the tame parameter field \(T\) is unramified over \(T^\sigma\) or not. If \(T/T^\sigma\) is unramified, there is a unique \(G^\sigma\)-conjugacy class of \(\sigma\)-selfdual types in \(\pi\), while if \(T/T^\sigma\) is ramified, we choose the \(G^\sigma\)-conjugacy class of \(\sigma\)-selfdual types with `maximal index'. He shows that \(\pi\) is distinguished if and only if the types in the above chosen class are distinguished. If \( \boldsymbol{\lambda} = \boldsymbol{\boldsymbol{\kappa}}\otimes \boldsymbol{\rho}\), to assure that \(\boldsymbol{\kappa}\) can be chosen to be distinguished, we require \(\pi\) to be supercuspidal, or equivalently \(\rho\) to be supercuspidal. The distinction problem is now reduced to the level zero case, and is then handled by the author by extending the results of \textit{C. Coniglio-Guilloton} [Bull. Soc. Math. Fr. 144, No. 2, 163--216 (2016; Zbl 1348.22023)].