Representations of automorphism groups of finite \(\mathfrak o\)-modules of rank two. (Q952434)

Let \(F\) be a non-Archimedean field with ring of integers \(\mathfrak o\) and maximal ideal \(\mathfrak p\). The article describes all the irreducible representations of the group \(\Aut(M)\) for any finite \(\mathfrak o\)-module \(M\) of rank two. For each partition \(\lambda=(\ell_1,\ell_2)\) of length \(2\), let \(\mathfrak o_{\ell_i}:=\mathfrak o/\mathfrak p^{\ell_i}\) (\(i=1,2\)), let \(M_\lambda\) denote the \(\mathfrak o\)-module \(\mathfrak o_{\ell_1}\oplus\mathfrak o_{\ell_2}\) of type \(\lambda\), and let \(G_\lambda:=\Aut_{\mathfrak o}(M_\lambda)\). The complex irreducible representations of the groups \(G_\lambda\) are described in detail in the paper. Write \(\mu\leq\lambda\) if a module of type \(\mu\) can be embedded in a module of type \(\lambda\). Then representations of \(G_\lambda\) are built from representations of groups \(G_\mu\) where \(\mu<\lambda\) using induction functors from two families of subgroups. The first family generalizes the family of maximal parabolic subgroups in the case of \(G_{1^2}=\Aut_{\mathbf F_q}(\mathbf F_q^2)\). Given a direct sum decomposition \(M_\lambda=M_\mu\oplus M_{\lambda/\mu}\) (where \(\lambda/\mu\) denotes the type of the quotient \(M_\lambda/M_\mu\)), let \(P_{\mu,\lambda}\) be the stabilizer of \(M_\mu\) in \(M_\lambda\). Then \(P_{\mu,\lambda}\) surjects on \(G_\mu\times G_{\lambda/\mu}\), and (by imitating the Harish-Chandra induction functor) one can construct representations of \(G_\lambda\) by pulling back to \(P_{\mu,\lambda}\) representations of its ``Levi factor'' \(G_\mu\times G_{\lambda/\mu}\) and inducing the latter to \(G_\lambda\). The resulting induction functor from \(G_\mu\times G_{\lambda/\mu}\) to \(G_\lambda\) is called `geometrical induction'. The second family has no analogue in the case of \(G_{1^2}\). Given types \(\mu\leq\lambda\), call \(\mu\) `symmetric' in \(\lambda\) if the embedding of \(M_\mu\) in \(M_\lambda\) is unique up to automorphism. In such case let \(\lambda/\mu\) denoted the well defined type of the quotient \(M_\lambda/M_\mu\). Then there is a unique embedding (up to automorphism) of a module \(M_{\lambda/\mu}\) of type \(\lambda/\mu\) such that \(M_\lambda/M_{\lambda/\mu}\) is of type \(\mu\). Now let \(\mu\) be a symmetric type in \(\lambda\). In order to guarantee that there is no overlap with the first family, assume also that the partitions \(\mu\) and \(\lambda\) have same length (that is, \(M_\mu\) and \(M_\lambda\) have same rank). Then the author considers the following subgroups: \[ P_{\mu\hookrightarrow\lambda}:=\text{Stab}_{G_\lambda}(M_\mu\subset M_\lambda)\quad\text{and}\quad P_{\mu\twoheadrightarrow\lambda}:=\text{Stab}_{G_\lambda}(M_{\lambda/\mu}\subset M_\lambda). \] The fact that \(\mu\) is symmetric in \(\lambda\) translates into canonical epimorphisms from \(P_{\mu\hookrightarrow\lambda}\) and \(P_{\mu\twoheadrightarrow\lambda}\) to \(G_\mu\), the first defined by restricting the elements of \(P_{\mu\hookrightarrow\lambda}\) to \(M_\mu\) and the second defined by identifying the elements of \(P_{\mu\twoheadrightarrow\lambda}\) as automorphisms of \(M_\lambda/M_{\lambda/\mu}\). Again representations of \(G_\lambda\) are constructed by pulling back representations from \(G_\mu\) to either \(P_{\mu\hookrightarrow\lambda}\) or \(P_{\mu\twoheadrightarrow\lambda}\), and then inducing them to \(G_\lambda\). The resulting induction functor from \(G_\mu\) to \(G_\lambda\) is called `infinitesimal induction'. The types which eventually contribute to the infinitesimal induction are types \(\mu\) which are symmetric in \(\lambda\) and of same length and same height (where the height of a partition stands for its largest part). The set of these types will be denoted by \(\mathcal I_\lambda\). Let a `twist' of representation be its tensor product with a one-dimensional representation. For a partition \(\lambda=(\ell_1,\ell_2)\) let \(\lambda_-:=(\ell_1-1,\ell_2-1)\). There is a natural reduction map \(G_\lambda\twoheadrightarrow G_{\lambda_-}\). Call a representation of \(G_\lambda\) `primitive' if none of its twists is a pull-back from \(G_{\lambda_-}\). By using an inductive approach one can focus only on primitive representations as the non-primitive representations are twists of representations of groups of lower rank. A primitive irreducible representation of \(G_\lambda\) will be called `cuspidal' if none of its twists is contained in some geometrically or infinitesimally induced representation. Let \(\widehat C_\lambda\) denote the set of cuspidal irreducible representations of \(G_\lambda\). It is proved in the paper (Theorem~1.2) that the primitive irreducible representations of \(G_\lambda\) fall precisely into one of the following classes: (1) Contained in a geometrically induced representation. (2) Infinitesimally induced from a unique cuspidal representation \(\rho\in\widehat C_\mu\) (\(\mu\in\mathcal I_\lambda\)) up to twist. (3) Cuspidal. The cuspidal representations are explicitly constructed in Section 6 of the paper. The dimensions of the primitive irreducible representations of \(G_\lambda\) are computed as well as the number of irreducible representations of a given dimension (number which occurs to be a polynomial in \(q\) with rational coefficients). Interesting conjectures in greater ranks are also provided.