Smooth transfer of Kloosterman integrals (Q5907050)

Let \(E/F\) be a quadratic extension of local non-Archimedean fields. Let \(G= \text{GL}(m)\), \(A\) be the diagonal torus of \(G\) and \(N\) be the unipotent radical of the standard Borel subgroup. Let \(\psi : \;F \to C^\times\) be a non trivial character. For \(n\in N(F)\), define \(\theta(n)=\psi(\sum_{i} n_{i,i+1})\). For any smooth function \(\Phi\) of compact support on \(G(F)\), the (diagonal) Kloosterman orbital integral is defined by \[ \Omega(\Phi,a)=\int_{N(F)\times N(F)} \Phi(\;^tn_1 an_2)\theta(n_1n_2)\,dn_1\,dn_2 \] for \(a\in A(F)\). Let \(S(F)\) be the subset of Hermitian matrices in \(\text{GL}(m,E)\). For \(n\in N(E)\), define a character \(\theta\) by \(\theta(n\bar{n})= \psi(\sum_{i} n_{i,i+1}+\bar{n}_{i,i+1})\). The non-trivial element of \(\text{Gal}(E/F)\) is denoted by a bar. For any smooth function \(\Psi\) of compact support on \(S(F)\), the (diagonal) relative Kloosterman orbital integral is defined by \[ \Omega(\Psi,a)=\int_{N(E)} \Psi( ^ t \bar{n}an)\theta(n\bar{n})\, dn \] for \(a\in A(F)\). The functions \(\Phi\) et \(\Psi\) match if for every \( a\in A(F)\) \[ \Omega(\Phi,a)=\gamma(a) \Omega(\Psi,a). \] Here \(\gamma\) is a transfer factor defined explicity by \(\gamma(a)=\eta(a_1) \ldots \eta(a_1a_2\ldots a_{m-1})\) where \(\eta\) is the quadratic character of \(F^\times\). The main result of this paper is that for all \(\Phi\) there is a \(\Psi\) (and vice-versa) such that the functions \(\Phi\) and \(\Psi\) match. The method of the author is to linearize the problem by replacing \(G(F)\) (resp. \(S(F)\)) by the vector space of (resp. hermitian) matrices. Then he proves an inversion formula between the orbital integrals of a function and its Fourier transform. As a consequence, if the functions \(\Phi\) and \(\Psi\) match then their Fourier transforms also match. This provides many matching functions and is used to prove the existence of the transfer. This is in strong analogy with the work of \textit{J.-L. Waldspurger} on the Langlands-Shelstad transfer [Compos. Math. 105, No. 2, 153--236 (1997; Zbl 0871.22005)]. The matching here is relevant to a relative trace formula which may be used to study the quadratic base change (cf. for GL(3) [\textit{H. Jacquet} and \textit{Y. Ye}, Trans. Am. Math. Soc. 348, No. 3, 913--939 (1996; Zbl 0861.11033)]). In this context, there is also a fundamental lemma. \textit{B. C. Ngô} has established it for function fields [Duke Math. J. 96, No. 3, 473--520 (1999; Zbl 1047.11517)]. The author uses the previous matching to prove the fundamental lemma for number fields (in this article for \(m=2,3\) and for all \(m\) in [Kloosterman identities over a quadratic extension, Ann. Math. (2) 160, No. 2, 755--779 (2004; Zbl 1071.11026)].