The Gan-Gross-Prasad conjecture for \(\mathrm{U}(n) \times \mathrm{U}(n)\) (Q2251910)

The global, coarse Gan-Gross-Prasad conjectures predict a connection between central \(L\)-values of Rankin-Selberg \(L\)-functions and certain periods of automorphic forms on classical groups. For the unitary groups, one considers periods of automorphic forms on \(\mathrm{U}(n) \times \mathrm{U}(m)\) for various \(n,m\). In the case \(\mathrm{U}(n) \times \mathrm{U}(n+1)\), \textit{H. Jacquet} and \textit{S. Rallis} proposed in [Clay Math. Proc. 13, 205--264 (2011; Zbl 1222.22018)] an approach via the relative trace formulae, which is then successfully applied by \textit{W. Zhang} [Ann. Math. (2) 180, No. 3, 971--1049 (2014; Zbl 1322.11048)]. The paper under review considers the case \(n=m\). To be precise, we fix a quadratic extension \(k|k'\) of number fields, an Hermitian space \(V\) of dimension \(n\) over \(k\), together with an irreducible cuspidal automorphic representation \(\pi = \pi_1 \otimes \pi_2\) on \(G(\mathbb{A}')\) where \(G := \mathrm{U}(V) \times \mathrm{U}(V)\). Let \(\iota: \mathrm{U}(V) \to G\) stand for the diagonal embedding. One considers the Fourier-Jacobi periods \[ \mathcal{FJ}_{\psi,\mu}(\varphi, \phi) = \int_{[\mathrm{U}(V)]} \varphi(\iota(g)) \theta_{\psi^{-1}, \mu^{-1}}(g,\phi)\,dg, \] where \([\mathrm{U}(V)] := \mathrm{U}(V)(k') \backslash \mathrm{U}(V)(\mathbb{A}')\), \((\psi,\mu)\) is the datum used to define the Weil representation of \(\mathrm{U}(V)(\mathbb{A}')\) realized on the Schrödinger model \(\mathcal{S}(L(\mathbb{A}'))\), and various \(\varphi \in \pi\), \(\phi \in \mathcal{S}(L(\mathbb{A}'))\). Let \(\mathrm{BC}(\pi_i)\) denote the base change \(k|k'\) (\(i=1,2\)). The global Gan-Gross-Prasad conjecture predicts that \(L(\frac{1}{2}, \mathrm{BC}(\pi_1) \times \mathrm{BC}(\pi_2)) \neq 0\) if and only if there exists an hermitian space \(V'\) of the same dimension as \(V\), an automorphic representation \(\pi'\) on \(G'(\mathbb{A}')\) (the group attached to \(V'\)) that is nearly equivalent to \(\pi\), such that \(\mathcal{FJ}\) is not identically zero for \(\pi'\). Note that in the split case \(k' = k \times k\), this has been completely answered by the fundamental work of \textit{H. Jacquet} et al. [Am. J. Math. 105, 367--464 (1983; Zbl 0525.22018)]. In this paper, the foregoing result is established under the assumptions that (1) there exist non-archimedean places \(v_1\), \(v_2\) of \(k'\) that split in \(k\) such that \(\pi_{v_1}\), \(\pi_{v_2}\) are supercuspidal; (2) all archimedean places of \(k'\) split in \(k\). The result is then applied to the case \(\mathrm{U}(n) \times \mathrm{U}(n+1)\) (Theorem 8.2.1), via the \(\theta\)-correspondence. The method is based on a comparison of relative trace formulae, generalized by \textit{Y. Liu} [Manuscr. Math. 145, No. 1--2, 1--69 (2014; Zbl 1301.11050)] to all \(n,m\). The relevant fundamental lemma is proven Liu's paper whereas the smooth transfer is done in the present paper. Both are based on a reduction to the case \(\mathrm{U}(n) \times \mathrm{U}(n+1)\) treated by Zhang [loc. cit.].