Cuspidal discrete series for semisimple symmetric spaces (Q1760162)

Let \(G\) be a connected semisimple real Lie group with finite center and \(K\) a maximal compact subgroup of \(G\) associated with a Cartan involution \(\theta\) of \(G\). Write \(\mathfrak g\) and \(\mathfrak k\) for the Lie algebras of \(G\) and \(K\) respectively, and denote by \(\mathfrak g=\mathfrak k+\mathfrak p\) the corresponding Cartan decomposition of \(\mathfrak g\). The Schwartz space \(C(G)\) for \(G\) is dense in the Hilbert space \(L^{2}(G)\) of square integrable complex functions on \(G\). Harish-Chandra defined a cusp form on \(G\) as an element \(f\in C(G)\) satisfying \[ \int_{N}f(xny)dn=0 \] for all proper parabolic subgroups \(P\) in \(G\) with Langlands decomposition \(P=MAN\), and for all \(x,y\in G\). Moreover, Harish-Chandra proved that the space of cusp forms on \(G\) coincides with the intersection of \(C(G)\) and the sum of all discrete series representations in \(L^{2}(G)\); and that the space of cusp forms is nonzero if, and only if, the groups \(G\) and \(K\) have equal (complex) rank. Harish-Chandra used this notion of cusp forms to prove the Plancherel theorem for \(G\), i.e. the decomposition of \(L^{2}(G)\) into a finite sum of series of representations. These representations are generalized principal series representations induced from \(\theta\)-stable parabolic subgroups \(P=MAN\) with the property that \(M\) and \(M\cap K\) have equal (complex) rank. Such parabolic subgroups are called cuspidal. More generally, let \(G/H\) be a semisimple symmetric space for \(G\), i.e. \(H\) is a closed subgroup of \(G\) with the property that \(G_{e}^{\sigma}\subset H\subset G^{\sigma}\) where \(\sigma\) is an involution of \(G\), \(G^{\sigma}\) the group of fixed points of \(\sigma\) and \(G_{e}^{\sigma}\) its identity component. A Plancherel formula for \(G/H\), i.e. the decomposition of \(L^{2}(G/H)\) into a direct integral of representations induced from parabolic subgroups, is known. It is also known that discrete series representations occur in \(L^{2}(G/H)\) if, and only if, \(G/H\) and \(K/K\cap H\) have equal rank as symmetric spaces. (See the series of papers of \textit{E. P. van den Ban} and \textit{H. Schlichtkrull} in [Invent. Math. 161, No. 3, 453--566 (2005; Zbl 1078.22012); ibid., No. 3, 567--628 (2005; Zbl 1078.22013)] and the paper of \textit{P. Delorme} in [Ann. Math. (2) 147, No. 2, 417--452 (1998; Zbl 0906.22015)], as well as the references therein.) In the paper under review, the authors introduce a notion of cuspidal forms on \(G/H\) as follows. Let \(\mathfrak a\) be a \(\sigma\)-invariant maximal abelian subspace of \(\mathfrak p\) so that \(\mathfrak a_{1}=\mathfrak a\cap\mathfrak q\) is a maximal abelian subspace in \(\mathfrak p\cap\mathfrak q\). Fix a positive system \(\Sigma^{+}\) for the roots defined by nonzero weights of \(\mathfrak a_{1}\) in \(\mathfrak g\); and a positive system \(\Sigma_{\mathfrak g}^{+}\) for \(\mathfrak a\)-roots in \(\mathfrak g\) such that \(\Sigma^{+}\) consists of the nonzero restrictions to \(\mathfrak a_{\mathfrak q}\). Write \(\mathfrak a_{\mathfrak h}=\mathfrak a\cap\mathfrak h\) and consider the following subalgebras of \(\mathfrak g\): {\parindent=6mm \begin{itemize}\item{} \(\mathfrak n_{+}=\sum_{\beta\in\Sigma_{\mathfrak g}^{+}}\mathfrak g^{\beta}\) where \(\beta_{\mid_{\mathfrak a_{\mathfrak q}}}>0\) and \(\beta_{\mid_{\mathfrak a_{\mathfrak h}}}>0\); \item{} \(\mathfrak n_{-}=\sigma\theta(\mathfrak n_{+})\); \item{} \(\mathfrak n_{0}=\sum_{\beta\in\Sigma_{\mathfrak g}^{+}}\mathfrak g^{\beta}\) where \(\beta_{\mid_{\mathfrak a_{\mathfrak q}}}>0\) and \(\beta_{\mid_{\mathfrak a_{\mathfrak h}}}=0\); \item{} \(\mathfrak n^{\star}=\mathfrak n_{+}\oplus\mathfrak n_{0}\); \item{} \(\mathfrak n^{\star\star}=\mathfrak n_{-}\). \end{itemize}} Write \(N^{\star}=\exp(\mathfrak n^{\star})\) and \(N^{\star\star}=\exp(\mathfrak n^{\star\star})\). The authors define an element \(f\) in the Schwartz space for \(G/H\) to be a cusp form on \(G/H\) if \[ \int_{N^{\star}}f(gn^{\star}H)dn^{\star}=0 \] for all \(g\in G\) and for all proper \(\sigma\theta\)-stable proper parabolic subgroups of \(G\). Then they show that this notion agrees with Harish-Chandra's definition of a cusp form on \(G\) (viewing \(G\) as the symmetric space \(G\times G/\Delta(G\times G)\)), but that discrete series representations for \(G/H\) need not be cuspidal (as it is the case on \(G\)). This is done explicitly in the case of the real hyperbolic spaces \(SO_{e}(p,q+1)/SO_{e}(p,q)\). The authors prove that if \(p\geq q-1\) then all discrete series representations are cuspidal, while if \(p<q-1\) then there is a non-empty finite family of non cuspidal discrete series representations.