On a strange invariant bilinear form on the space of automorphic forms (Q730317)

Let \(F\) be a global field, that is, either an algebraic number field, or a field finitely generated over the finite field \(\mathbb{F}_p\) of transcendence degree one. Let \(\mathbb{A}\) denote the ring of adèles of \(F\). Let \(G=\mathrm{SL}(2)\) be the special linear group defined over \(F\) of rank one. Fix the standard maximal compact subgroup \(K\) of \(G(\mathbb{A})\). Let \(\mathcal{A}\) denote the space of \(K\)-finite smooth functions on \(G(\mathbb{A})\), and \(\mathcal{A}_c\) the subspace of compactly supported functions. Then, there is an obvious bilinear form \(\mathcal{B}_{\mathrm{naive}}\) on \(\mathcal{A}\), which makes sense if one of the functions is compactly supported. More precisely, \[ \mathcal{B}_{\mathrm{naive}}(f_1,f_2)=\int_{G(\mathbb{A})/G(F)}f_1(x)f_2(x)\,dx, \] where \(f_1,f_2\in\mathcal{A}\) with at least one of them in \(\mathcal{A}_c\), and \(dx\) is a fixed Haar measure on \(G(\mathbb{A})\). Motivated by the geometric Langlands program, in particular by the results of the first author and \textit{D. Gaitsgory} [Camb. J. Math. 3, No. 1--2, 19--125 (2015; Zbl 1342.14041)] and of \textit{D. Gaitsgory} [Ann. Sci. Éc. Norm. Supér. (4) 50, No. 5, 1123--1162 (2017; Zbl 1423.11118)], the authors consider a bilinear form \(\mathcal{B}\) on \(\mathcal{A}_c\), slightly different from \(\mathcal{B}_{\mathrm{naive}}\), which has natural interpretation in terms of the geometric Langlands program. They refer to this form as a strange invariant bilinear form, as it is related to a ``strange'' functional equation for the Eisenstein series and miraculous duality on the moduli stack of bundles as in [Zbl 1423.11118]. Although the main body of the paper is entirely concerned with the classical theory of automorphic forms, the appendices provide a dictionary between classical and geometric theory, which explains why the bilinear form \(\mathcal{B}\) is natural from the geometric point of view. The bilinear form \(\mathcal{B}\) can be defined as \[ \mathcal{B}(f_1,f_2)= \mathcal{B}_{\mathrm{naive}}(f_1,f_2)-\langle M^{-1}\mathrm{CT}(f_1),\mathrm{CT}(f_2)\rangle, \] where \(\langle\,,\,\rangle\) denotes the scalar product, \(\mathrm{CT}\) stands for the constant term, and \(M\) is the standard intertwining operator. The form \(\mathcal{B}\) is \(G(\mathbb{A})\)-invariant, where in the number field case this must be interpreted in a different, but standard way, because the action of \(G(\mathbb{A})\) does not preserve the \(K\)-finiteness, thus, does not act on \(\mathcal{A}\). It is also proved that \(\mathcal{B}\) is not positive definite in the function field case. Note that \(G\) is of \(F\)-rank one, so that \(\mathrm{CT}\) and \(M\) are uniquely determined. The existence of the inverse \(M^{-1}\) of the intertwining operator \(M\) is proved in the paper as a key technical result. It turns out that, there exists a linear operator \(L:\mathcal{A}_c\to\mathcal{A}\), such that \[ \mathcal{B}(f_1,f_2)=\mathcal{B}_{\mathrm{naive}}(Lf_1,f_2), \] where \(L\) acts as identity on cusp forms in \(\mathcal{A}_c\), and its action on the Eisenstein series can be explicitly described. Motivated again by the geometric Langlands program, the authors introduce a subspace \(\mathcal{A}_{ps-c}\) of \(\mathcal{A}\). It is proved that, in the function field case, the operator \(L\) induces an isomorphism \(\mathcal{A}_c\to\mathcal{A}_{ps-c}\). The inverse of \(L\) is then explicitly computed, and the bilinear form \(\mathcal{B}\) is lifted to \(\mathcal{A}_{ps-c}\). The formula for the inverse of \(L\) serves as the main heuristics in a conjectural analog for the \(D\)-module setting, stated at the end of the paper in Appendix C. However, the analogous result in the number field case does not hold. Finally, the authors turn to the case of \(K\)-invariant functions. As a technical tool, the action of the intertwining operator \(M\), and its inverse \(M^{-1}\), on \(K\)-invariant functions is described. This allows a description of the bilinear form \(\mathcal{B}\) restricted to the subspace of \(K\)-invariant functions. In the function field case, the description is given in terms of Picard-Lefschetz oscillators of \textit{S. Schieder} [Duke Math. J. 167, No. 5, 835--921 (2018)]. In the number field case, the analogous description is in terms of Arakelov \(G\)-bundles. The techniques used in the paper are carefully chosen to be compatible with the ideas coming from the geometric Langlands program. In particular, the manipulations with functions are only those that can be also applied to \(l\)-adic sheaves and \(D\)-modules. Hence, for instance, the spectral decomposition is avoided, because in the setting of \(l\)-adic sheaves the Mellin transform on a torus [\textit{O. Gabber} and \textit{F. Loeser}, Duke Math. J. 83, No. 3, 501--606 (1996; Zbl 0896.14009)] is not an invertible functor. In the number field case, the considered spaces of functions on \(G(\mathbb{A})\) are not natural. For example, the space of cusp forms is not contained in \(\mathcal{A}_c\). One would prefer to impose on functions in \(\mathcal{A}\) certain growth conditions, such as uniform moderate growth, instead of compact support. Then, the bilinear form \(\mathcal{B}\) could be extended to these spaces by continuity, and a similar theory could be developed. However, this is out of the scope of the present paper.