On the density of zeros of linear combinations of Euler products for \(\sigma>1\) (Q1686479)

Let \(F_1, \dotsc, F_N\) be distinct Dirichlet series, not identically zero, and let \(c=(c_1,\dotsc,c_N)\) be a vector of complex numbers. The author considers the linear combination \[ L_c = c_1 F_1 + \dotsc + c_N F_N, \] in particular the set of real parts of zeros of \(L_c\), that we denote here as \(B\): \[ B = \{\beta\in{\mathbb R}: \text{ there exists } \gamma\in{\mathbb R} \text{ such that } L_c (\beta+i\gamma)=0\}. \] Let \(\sigma^* = \sup B\) and let \(\sigma_u\) denote the abscissa of uniform convergence of \(L_c\). The paper contains a number of results, that we sketch below. \textit{E. Bombieri} and \textit{A. Ghosh} [Russ. Math. Surv. 66, No. 2, 221--270 (2011; Zbl 1247.11116)] conjectured that if \(N\geq 2\) and \(F_1, \dotsc, F_N\) are Dirichlet \(L\)-functions, then the set \(B\cap [1,\sigma^*]\) is dense in \([1,\sigma^*]\). The present paper disproves this conjecture by showing (Theorem 1.1) that for a large family of linear combinations \(L_c\) (that includes appropriate linear combinations of distinct Dirichlet \(L\) functions) there is a zero-free vertical strip \(\sigma_1 <\sigma<\sigma_2\) contained in \(1 <\sigma<\sigma^*\). If \(F_1, \dotsc, F_N\) are Dirichlet series with non-zero initial terms, absolutely convergent for \(\sigma>1\), and vectors of initial \(N\) coefficients linearly independent, the author shows (Theorem 1.2) the existence of \(L_c\) with at least \(N-1\) such zero-free vertical strips (with at least one zero between distinct strips). Theorems 1.3, 1.4 and 1.5 go in the other direction. Theorem 1.3 implies that if \(\sigma_u <\sigma^*\), then, for every \(\alpha>\sigma_u\), the set \(B\) is dense between \(\alpha\) and \(\sigma^*\) apart from a finite number of gaps and isolated points. Theorems 1.4 and 1.5 show, for a class of axiomatically defined linear combinations (all satisfying \(\sigma_u \leq 1\)) that there are no isolated points of \(B\) in \((1,\sigma^*]\). This implies that, for every \(\alpha>\sigma_u\), \(B\cap [\alpha,\sigma^*]\) is contained in a finite union of intervals and is dense in each of them. In particular, any zero \(\rho=\beta +i\gamma\) of \(L_c\) satisfying \(1 < \beta\) belongs to a family of zeros whose projection on the real line is a dense subset of a non-zero interval. Moreover, there is an \(\eta>0\) such that \(B\cap [\sigma^*-\eta,\sigma^*]\) is dense in \([\sigma^*-\eta,\sigma^*]\) (Corollary 1.6). These results do not violate the Riemann Hypothesis and its generalizations, e.g., the Grand Riemann Hypothesis (for the Selberg class), because, as the author notes, the functions \(L_c\) do not generally possess an Euler product, a suitable meromorphic continuation, and a functional equation. Theorem 1.8 shows the existence of linear combinations \(L_c\) of functions from the Selberg class such that \(L_c\) satisfies ``most'' conditions of the Selberg class (missing the Euler product), and still has \(\sigma^*>1\) and a zero-free strip between \(1\) and \(\sigma^*\).