Testing the functional equation of a high-degree Euler product (Q764075)

\(L\)-functions arise in at least two ways, as Dirichlet series \(\sum_{n=1}^\infty a_n n^{-s}\) satisfying Selberg's axioms (Selberg \(L\)-functions) and from modular forms via the Langlands program. Conjecturally these are broadly speaking the same thing, but it is very difficult to make the transition between them. In particular Langlands \(L\)-functions are expected to satisfy a functional equation, but this is known in only a few cases. The paper under review gives a computational method that provides evidence that a given Langlands \(L\)-function does satisfy the expected functional equation, on a case by case basis. The authors carry out their procedure in the first open case, the degree~\(10\) adjoint \(L\)-functional associated with the unique Siegel modular form \(F\) of level~\(1\) and degree~\(2\) that is a Hecke eigenform but not a Maaß\ form, which is of weight~\(20\). Thus \(F\) is not a Sato-Kurokawa lift, so the \(L\)-function should be primitive and thus, conjecturally, a Selberg \(L\)-function. In this particular case a lot has already been computed. The Hecke eigenvalues \(\lambda_F(n)\) for the Hecke operators \(T(n)\) are known for \(n=p\) or \(n=p^2\), for \(p\leq 79\). That is sufficient to compute arbitrarily many of the Dirichlet coefficients to any desired accuracy. As the Dirichlet coefficients also satisfy growth conditions (Ramanujan bounds) it is possible to give unconditional bounds on \(L(s)\) after computing finitely many of the coefficients. The functional equation is checked by computing some values depending a priori on an auxiliary function \(g\), as in [\textit{M. Rubinstein}, Lond. Math. Soc. Lect. Note Ser. 322, 425--506 (2005; Zbl 1168.11329)]. The values computed here for some different choices of \(g\) do indeed agree within the available precision. After that comes a discussion of how good this evidence for the validity of the functional equation really is. This is inevitably heuristic and involved treating the Dirichlet coefficients as if they were random variables, although their value is in fact known, but it does give a general idea that the computation is fairly convincing. The reviewer, at least, was not more convinced of the validity of the functional equation after reading the paper than before, but perhaps nothing short of a proof would have achieved that. The real use of such methods is perhaps to correct overenthusiastic conjecture rather than to confirm confident conjecture, but one should first check that they do give the ``correct'' answer in cases where one is confident, and that is what happens here.