Zeros of linear combinations of functions of a special type that are connected with Selberg Dirichlet series (Q4349338)

The author proves that the interval \([T,2T]\) for \(T \geq T_0 > 0\) contains not less than \[ T\exp(\sqrt{\log\log\log T})\tag{1} \] zeros of odd order for the function \[ {\mathcal F}(t) = \Phi_1({\textstyle{1\over2}}+it), \qquad\Phi_1(s) = \sum_{j=1}^Na_j(\rho_j(s))^{-1/2}F_j(s),\tag{2} \] where the \(a_j\)'s are given real numbers (\(a_1 \not= 0\)), \(N\) is a given natural number, and the \(F_1(s),\ldots ,F_N(s)\) are primitive and distinct in the Selberg class of functions \(S\) of degree two, satisfying certain hypotheses. The Selberg class \(S\) was introduced by \textit{A. Selberg} [Proc. Amalfi conf. analytic number theory 1989 (eds. E. Bombieri etc.), Università di Salerno, Salerno, 367-385 (1992; Zbl 0787.11037)], and it is the class of functions generalizing the classical Riemann zeta-function \(\zeta(s)\). Namely \(S\) consists of the Dirichlet series \(F(s) = \sum_{n=1}^\infty a(n)n^{-s}\) (for \(\text{Re }s > 1\)) satisfying the following conditions: \((s-1)^mF(s)\) is an entire function for some integer \(m \geq 0\); \(a(n) \ll n^\varepsilon\); \(\Phi(s) = \bar{\Phi}(1-{\bar s})\), where \(\Phi(s) = \gamma_F(s)F(s)\), \(\gamma_F(s) = \varepsilon_1Q^s\prod_{\nu=1}^k\Gamma(\lambda_\nu s+\mu_n)\), \(|\varepsilon_1|= 1\), \(Q > 0\), \(\lambda_\nu > 0\), \(\text{Re } \mu_\nu \geq 0\); for \(\text{Re} s > 1\) \(F(s) = \prod_p(1 + a(p)p^{-s} + a(p^2)p^{-2s} + \ldots)\), \(\log F(s) = \sum_{n=1}^\infty b(n)n^{-s}\), where \(b(n) = 0\) unless \(n\) is a prime power, and \(b(n) \ll n^{\theta}\) for some \(0 \leq \theta \leq 1/2\). The degree \(d_F\) of \(F(s) \in S\) is the quantity \(d_F = 2\sum_{\nu=1}^k\lambda_\nu\), and \(F(s)\) is primitive if it cannot be factored as a product of elements of \(S\). The ``certain hypotheses'' mentioned in the enunciation of the author's result are of a technical nature. They are slightly stronger than some of the hypotheses originally made by Selberg and are satisfied by the standard examples of the class \(S\). There are three conditions which are needed, and for example the third one requires that, if \(F(s)\) is any of the functions \(F_1(s),\ldots ,F_N(s)\), then \[ \sum_{n\leq x}|a(n)|^2n^{-1} = A_F\log x + B_F + O(\log^{-4}x) \quad(A_F > 0). \] The functions \(\rho_j(s)\) appearing in the formulation of the theorem are defined as \(\rho(s) = \bar{\gamma}_F(1-\bar{s})/\gamma_F(s)\) if \(F(s) \in S\), so that by (2) \({\mathcal F}(t)\) is real for \(t\) real. The bound (1) falls short of the expected bound \(C_FT\log T, C_F > 0\), but nevertheless its proof is long and involved. The basic method is due to S. M. Voronin, who used it to obtain a non-trivial lower bound for the number of zeros of the so-called Davenport-Heilbronn zeta-function on the critical line \(\text{Re} s = 1/2\) (for which the Riemann hypothesis does not hold). In the course of the proof the author successfully displays a variety of different techniques from analytic number theory.