On the structure of the Selberg class, VI: non-linear twists (Q4677362)

This is the sixth in a series of papers in which the authors investigate the structure of the Selberg class and prove some very interesting results about the analytic properties of Dirichlet series associated with the Dirichlet series of the extended Selberg class. There is a very good survey of the subject in the authors' paper \textit{J. Kaczorowski} and \textit{A. Perelli} [The Selberg class: a survey (in Number theory in Progress, Proc. Conference in Honor of A. Schinzel, K. Györy (ed.) et al., Berlin: de Gruyter, 953--992 (1999; Zbl 0929.11028)]. This paper offers an attractive and well-referenced account of the subject and extends the results of the authors' original paper [\textit{J. Kaczorowski} and \textit{A. Perelli}, On the structure of the Selberg class. I: \(0 \leq d\leq 1\), Acta Math. 182, No. 2, 207--241 (1999; Zbl 1126.11335)]. We begin our review by recalling the basic ideas concerning the Selberg class, \({\mathcal S}\), and the extended class \({\mathcal S}^\#\). The class \({\mathcal S}\) denotes the class of Dirichlet series, \(F(s)\), satisfying the following five conditions: (i) the abscissa of convergence, \(\sigma_a(s)\), satisfies \(\sigma_a(F)\leq 1\); (ii) \((s-1)^mF(s)\) is an entire function of finite order, for some integer \(m\geq 0\); (iii) \(F(s)\) satisfies a functional equation of the form \(\Phi(s)=\omega \overline\Phi(1-s)\) where \(|\omega |=1,\;\overline f(s)=\overline{f(\overline s)}\) and \(\Phi(s)=Q^s \prod^r_{j=1}\Gamma(\lambda_js+\mu_j)F(s)=\gamma(s)F(s)\), with \(Q<0\), \(\lambda_j>0\) and \(\Re\mu_j\geq 0\); (iv) the Dirichlet coefficients, \(a(n)\), of \(F(s)\) satisfy the Ramanujan conjecture \(a(n)\ll n^\varepsilon\) for every \(\varepsilon >0\); (v) \(\log F(s)\) is a Dirichlet series with coefficients \(b(n)\) such that \(b(n)=0\) unless \(n= p^m\), where \(p\) denotes a prime, \(m\geq 1\) and \(b(n)\ll n^\vartheta\) for some \(\vartheta<\frac 12\). The extended Selberg class \({\mathcal S}^\#\) consists of the non-zero functions satisfying only the conditions (i), (ii) and (iii) and \(m_F\geq 0\) denotes the order of the pole at \(s=1\) of \(F(s)\in{\mathcal S}^\#\). The degree \(d_F\), \(\xi\)-invariant \(\xi_F\), conductor \(q_F\) and root number \(\omega^*_F\) of \(F(s)\in{\mathcal S}^\#\) are defined in the usual way and in particular \[ d_F=2 \sum^r_{j=1} \lambda_j,\;\xi_F=2 \sum^r_{j=1}(\mu_j-\tfrac 12)= \eta_F+i\theta_F. \] \[ q_F=(2 \pi)^{d_F}Q^2 \prod^r_{j=1} \lambda_j^{2\lambda_j},\;\omega^*_F=\omega e^{-i\frac {\pi}{2} (\eta_F+1)}\left(\frac{q_F}{(2\pi)^{d_F}} \right)^{i\frac {\theta_F} {d_F}} \prod^r_{j=1}\lambda_j^{-2i{\mathfrak I}\mu_j}. \] In the first of this series of papers (op. cit.) the authors considered the linear twist \[ F(s,\alpha)= \sum^\infty_{n=1}\frac{a(n)}{n^s}e(-n\alpha),e(x)= e^{2\pi ix}, \] where \(\alpha \in\mathbb R\), \(\sigma>1\), of a function \(F(s)\) of degree 1 in the extended Selberg class \({\mathcal S}^\#\), and showed, in particular, that \(F(s,\alpha)\) has a meromorphic continuation to \(\sigma >0\), with a simple pole at \(s=1-i\theta_F\), under certain conditions. In this paper, they consider the nonlinear twist \[ F(s,\alpha)= \sum^\infty_{n=1}\frac{a(n)}{n^s}e(-n^{1/d}\alpha),\;\sigma>1, \] where \(d>0\), (which reduces to the linear twist when \(d=1)\) and they obtain the basic analytic properties of \(F(s,\alpha)\), together with some interesting applications. Their fundamental result is that if \({\mathcal S}_d^\#\) denotes the Selberg class of functions of degree \(d\) and if \(d<0\) and \(\alpha>0\), then \(F(s,\alpha)\) has a meromorphic continuation to \(\mathbb C\). The function \(F(s, \alpha)\) is entire if \(a (n_\alpha)=0\), where \(a(n_\alpha)\) is defined in terms of the conductor, namely \[ n_\alpha=q_Fd^{-d}\alpha^d,\;a(n_\alpha)=\begin{cases} 0, &\text{ if }n_\alpha\not\in\mathbb N\\ a(n_\alpha), &\text{ if }n_\alpha \in \mathbb N\end{cases} \] and if \(a(n_\alpha)\neq 0\), then \(F(s,\alpha)\) has at most simple poles at the points \[ s_k=\frac{d+1}{2d}-\frac kd-i\frac {\theta_F}{d},\;k=0,1,2,\dots \] with non-zero residue at \(s_0\). The proof uses the incomplete Fox hypergeometric function \[ H_K(z,s)= \frac{1}{2\pi i}\int_{(-K-\tfrac 12)} h(w,s)z^w\,dw,\quad\text{where }h (w,s)=h^*(w,s)\Gamma(w), \] \[ h^*(w,s)= \prod^r_{j=1}\frac{\left( \Gamma\left(\lambda_j(1-s)+\overline\mu_j-\frac {\lambda_j} {d}w \right)\right)}{\left(\Gamma\left(\lambda_js+\mu_j+\frac {\lambda_j} {d}w\right)\right)}. \] The properties of the hypergeometric functions required are established in the paper. The paper includes interesting applications; for example if \(F\in{\mathcal S}_d\), \(d\geq 1\) and if the series \(\sum^\infty_{n=1}\frac{a(n)-1}{n^s}\) converges for \(\sigma>\tfrac 15-\delta\), \(\delta>0\), then \(F(s)=\zeta(s)\), the Riemann zeta function, and there is a similar result concerning the Dirichlet series \(F(s)=L(s+i \theta,\chi)\), \(\theta\in\mathbb R\) and \(\chi\) a primitive Dirichlet character. In connection with that example, the authors draw attention to the fact that their proof is indirect and they propose the problem of finding a more direct proof.