On the global Gan-Gross-Prasad conjecture for general spin groups (Q2189533)

This work is an amplification of the refined Gan-Gross-Prasad conjecture by Ichino-Ikeda to \(\mathrm{GSpin}\) groups, which predicts a remarkable identity between period integrals of automorphic forms and \(L\)-values. Let \(n \geq 2\) be an integer and consider quadratic spaces \((V_n, q_n) \subset (V_{n+1}, q_{n+1})\) over a number field \(F\), which are of dimension \(n\) and \(n+1\) respectively, such that \(V_n\) is not the hyperbolic plane. From these we form the group \(\mathrm{GSpin}(V_n)\). It has a central subgroup \(Z_n^\circ \simeq \mathbb{G}_m\) such that \(\mathrm{GSpin}(V_n)/Z_n^\circ\) is isomorphic to \(\mathrm{SO}(V_n)\). The same construction applies to \(V_{n+1}\) as well. Now consider a cuspidal tempered representation \(\pi_n\) of \(\mathrm{GSpin}(V_n)\), with central character \(\omega_{\pi_n}\) restricted to \(Z_n^\circ(\mathbb{A})\). Ditto for \(\pi_{n+1}\). Assume in what follows that \(\omega_{\pi_{n+1}} \omega_{\pi_n} = \omega^2\) where \(\omega\) is a chosen automorphic character of \(\mathbb{G}_m\). Let \(\phi \in \pi_{n+1}\) (resp. \(f \in \pi_n\)). The global protagonist of this work is the period integral \[ \mathcal{P}_\omega(\phi, f) = \int_{Z_n^\circ(\mathbb{A}) \backslash [\mathrm{GSpin}(V_n)]} \phi f \cdot (\omega \circ N)^{-1} \] where \(N\) denotes the spinor norm. On the local side, one defines at each place \(v\) of \(F\) the local pairing \[ \alpha^\natural_{\omega_v}: (\pi_{n+1, v} \boxtimes \overline{\pi_{n+1, v}}) \times (\pi_{n, v} \boxtimes \overline{\pi_{n, v}}) \to \mathbb{C} \] as \[ \int_{Z_n^\circ(F_v) \backslash \mathrm{GSpin}(V_n, F_v)} \mathcal{B}_{\pi_{n+1}, v} \mathcal{B}_{\pi_n, v} \cdot (\omega_v \circ N)^{-1} \] where \(\mathcal{B}_{\pi_n, v}\) and \(\mathcal{B}_{\pi_{n+1}, v}\) are the local components of the global Petersson pairings \(\mathcal{B}_{\pi_n}\) and \(\mathcal{B}_{\pi_{n+1}}\) relative to Tamagawa measures. These integrals are shown to be convergent under the temperedness assumption. The unramified computations also yield the expected \(L\)-values (see below). The refined Gan-Gross-Prasad conjecture proposed in this work reads as \[ |\mathcal{P}_\omega(\phi, f)|^2 = \frac{\Delta_{\mathrm{SO}(V_{n+1})}}{2^\beta} \cdot \frac{L(1/2, \pi_{n+1} \times \pi_n \otimes \omega^{-1})}{L(1, \pi_{n+1}, \mathrm{Ad}) L(1, \pi_n, \mathrm{Ad}) } \prod_v \alpha_{\omega_v}(\phi_v, f_v) \] for factorizable \(\phi\) and \(f\), where \(\alpha_{\omega_v}\) are certain renormalized (by \(L\)-values) version of \(\alpha_{\omega_v}^\natural\) so that the infinite product makes sense, \(\Delta\) is a quantity related to the motive of the group, and \(2^\beta = 4 |\mathcal{S}_{\phi_{n+1}}| |\mathcal{S}_{\phi_n}|\) if one accepts the Langlands parameterization. All these are \(\mathrm{GSpin}\) analogues of the Ichino-Ikeda conjecture for classical groups. It is also remarked that the \(L\)-value in the numerator (resp. denominator) is expected to be finite (resp. nonzero). As a reality check, in Section 4, these conjectures are verified when \(n = 2, 3\), and also for \(n=4\) under extra assumptions. Specifically, the case \(n=2\) reduces to the study of toric periods, which boils down to Waldspurger's formula (non-split case) and the Jacquet-Langlands or Hecke theory (split case). The case \(n=3\) reduces to the triple product formula. These reductions are based accidental isomorphisms in low ranks. The treatment of the case \(n=4\) requires assumptions on the groups, namely quasi-splitness, as well as on the representations under consideration. The conjectures are then reduced to the work [\textit{W. T. Gan} and \textit{A. Ichino}, J. Inst. Math. Jussieu 10, No. 2, 235--324 (2011; Zbl 1241.11058)] on \(\mathrm{SO}_4 \subset \mathrm{SO}_5\) and its version for similitude groups. One should remark that the reduction procedures above are not always trivial. It may involve some restriction procedure, or enhancements of known results such as Waldspurger's formula.