Relative Kloosterman integrals for GL(3)

From MaRDI portal





Let \(E/F\) be a quadratic extension of number fields. Let \(G\) (resp. \(G'\)) denote the group \(GL(m,E)\) (resp. \(GL(m,F)\)), regarded as an algebraic group over \(F\). Let \(H\) denote the quasi-split unitary subgroup of \(G(F)\). An automorphic representation \(\pi\) of \(G(F_{\mathbf A})\) is said to be distinguished with respect to \(H\) if there is some \(\varphi\) in the space of \(\pi\) for which the integral \(\int_{H(F)\setminus H(F_{\mathbf A})} \varphi(h) dh\) is nonzero. The notion was first introduced by \textit{G. Harder}, \textit{R. P. Langlands} and \textit{M. Rapoport} [J. Reine Angew. Math. 366, 53-120 (1986; Zbl 0575.14004)]. An argument which is essentially given there shows that if \(\pi\) is \(H\)-distinguished then it is a base change from \(G'\). They conjecture the converse. The authors outline a possible proof of this conjecture, which is based upon a relative trace formula. This formula asserts the equality of the integrals \[ \iint K(h,n)\theta_ E(n) dh dn\qquad \text{and} \qquad \iint K'(n_ 1,n_ 2)\theta_ F (n_ 1 n_ 2) dn_ 1 dn_ 2, \] where \(K\) and \(K'\) are the kinds of geometric kernels found in the trace formula and \(\theta_ E\) and \(\theta_ F\) are (compatible) characters of the subgroup of upper triangular unipotent matrices \(N\). The kernels \(K\), \(K'\) are attached to functions on the groups \(G(F_{\mathbf A})\) and \(G'(F_{\mathbf A})\), respectively. The notion of relative trace formula was introduced in the fundamental paper of \textit{H. Jacquet} and \textit{K. F. Lai} [Compos. Math. 54, 243-310 (1985; Zbl 0587.12006)] and developed further in previous papers of \textit{H. Jacquet} [Proc. Indian Acad. Sci., Math. Sci. 97, 117-155 (1987; Zbl 0659.10031)], \textit{H. Jacquet} and \textit{Y. Ye} [C. R. Acad. Sci., Paris., Sér. I 311, No. 11, 671-676 (1990; Zbl 0715.11026)] and \textit{Y. Ye} [J. Reine Angew. Math. 400, 57-121 (1989; Zbl 0655.10020)], among others. It should be remarked that actually obtaining a theorem about automorphic representations from this conjectured equality requires a spectral analysis, which is based upon a truncation introduced by Jacquet and Lai [loc. cit.], but not discussed further in this article. Establishing the equality is a local question, requiring (at almost all places) the comparison of two particular local integrals. One is a Kloosterman integral on \(G'\); the other is a new kind of Kloosterman-like integral, a ``relative Kloosterman integral, on \(G\). The authors state a precise local conjecture concerning this comparison, including a precise transfer factor. They then establish this conjecture (the so- called fundamental lemma for the unit element of the Hecke algebra) when \(m=3\). To do so, they compute the Mellin transforms of the two local integrals, and show that they are equal. The matching of these Mellin transforms requires the Hasse-Davenport identity for Gauss sums. The equality when \(m=2\) was proved by the second author in his thesis [ibid.] (Ye also matched more general functions), with an improved proof given by Jacquet and Ye [ibid.] In the case \(m=2\), \textit{D. Zagier} proved the same identity in a classical language in [Invent. Math. 30, 1-46 (1975; Zbl 0308.10014)]. The authors conclude with some remarks on the matching of more general Hecke functions. One may also investigate the dual case: \(G'\)-distinguished and stable unitary base change from \(H\). This is discussed in a subsequent paper by the first author [Can. J. Math. 44, 1220-1240 (1992)].




Cited in
(30)








This page was built for publication: Relative Kloosterman integrals for GL(3)

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4030113)