Spectral reciprocity via integral representations (Q6093184): Difference between revisions

Let \(F\) be a number field. We fix a cuspidal automorphic representation \(\Pi\) of \(\mathrm{GL}(3)\) over \(F\). For each automorphic representation \(\pi\) of \(\mathrm{GL}(2)\) over \(F\), we have respectively, the Hecke \(L\)-function and the adjoint \(L\)-function, and the Rankin-Selberg \(L\)-function: \[ \Lambda(s, \pi),\quad \Lambda(s, \mathrm{Ad}, \pi),\quad \Lambda(s, \Pi\times \pi). \] Let \[ \mathcal{C}_{s,w}(\Phi):= \sum_{\pi\in C(S)}\frac{\Lambda(s, \Pi\times \pi) \Lambda(w, \pi)}{\Lambda(1, \mathrm{Ad}, \pi)}H(\pi) \] and \[ \mathcal{E}_{s,w}(\Phi):= \sum_{\omega\in \Xi(S) }\int_{-\infty}^{+\infty} \frac{\Lambda(s, \Pi\times \pi (\omega, it)) \Lambda(w, \pi(\omega, it))}{\Lambda^\ast(1, \mathrm{Ad}, \pi(\omega,it))} H(\pi(\omega,it)) \frac{\mathrm{d}t}{2\pi}. \] The notations must be explained. In the above, \(\Phi\) is a decomposable vector in the representation space of \(\Pi\) and \(H(\pi)\) is a weight factor defined in the article involving \(\Phi\) and local \(L\)-factors. The set \(S\) is any finite set of places containing all the Archimedean ones and those for which \(\Pi_v\) is ramified. The set \(C(S)\) (resp. \(\Xi(S)\)) denotes the collection of cuspidal automorphic representations of \(\mathrm{GL}(2)\) with trivial central character (resp. unitary normalized idele characters) over \(F\) that are unramified everywhere outside \(S\). Finally, \(\pi(\omega, it)=\omega |\cdot|^{it} \boxplus \omega^{-1}|\cdot|^{-it} \), the isobaric sum and the factor \(\Lambda^\ast(1, \mathrm{Ad}, \pi(\omega,it))\) denotes the first nonzero Laurent coefficient of \(\Lambda(s, \mathrm{Ad}, \pi(\omega,it))\) at \(s=1\). Let \((s', w')=(\frac{1}{2}(1+w-s), \frac{1}{2}(3s+w-1))\). To each \(\Phi\), the article associates to it another vector \(\Phi^\vee\). If the real parts of \(s, w, s', w'\) are large enough, the main theorem of the article is the following identity: \[ \mathcal{M}_{s,w}(\Phi)+ \mathcal{D}_{s,w}(\Phi)= \mathcal{M}_{s',w'}(\Phi^\vee)+ \mathcal{D}_{s',w'}(\Phi^\vee), \] where \(\mathcal{M}_{s,w}(\Phi)=\mathcal{C}_{s,w}(\Phi)+\mathcal{E}_{s,w}(\Phi)\) and \(\mathcal{D}_{s,w}(\Phi)\) is defined using an explicit integral. Then the article tries to do analytic continuation for the term \(\mathcal{E}_{s,w}(\Phi)\) to obtain a formula that is valid at the central point \(s=s'=w=w'=\frac{1}{2}\). This is done under a technical condition. As an application, the article provides the following corollary. Let \(\Pi\) be an unramified cuspidal automorphic representation of \(\mathrm{GL}(3)\) over \(F\), and let \(\mathfrak{p}\) be a prime ideal of \(F\) with absolute norm \(q\). Then, for every \(\varepsilon> 0\), we have \[ \frac{\varphi(q)}{q}\sum_{\pi} \frac{\Lambda(\frac{1}{2}, \Pi\times \pi) \Lambda(\frac{1}{2}, \pi)}{\Lambda(1, \mathrm{Ad}, \pi)} = \frac{4\Lambda(1,\Pi)\Lambda(0, \Pi) }{\xi_F(2)}+ O_{\varepsilon}(q^{\vartheta-\frac{1}{2} +\varepsilon }), \] where the sum over \(\pi\) is taken over a collection of cuspidal automorphic representations whose conductor is \(\mathfrak{p}\), \(\vartheta\) is an admissible exponent towards the Ramanujan conjecture, \(\xi_F\) is the Dedekind zeta function of \(F\) and \(\varphi\) is the Euler function. In particular, for \(q\) sufficiently large, there is at least one automorphic representation \(\pi\) of conductor \(\mathfrak{p}\), unramified for every Archimedean place and such that \(\Lambda(\frac{1}{2}, \Pi\times \pi)\) and \(\Lambda(\frac{1}{2}, \pi)\) are both nonzero.