High moments of the Riemann zeta-function (Q1847831): Difference between revisions

The asymptotic formula for the integral \[ I_k(T) = \int_0^T|\zeta({1\over 2}+it)|^{2k} dt\qquad(k = 1,2,\ldots) \] represents a central problem in the theory of the Riemann zeta-function \(\zeta(s)\) (for an extensive account see the reviewer's monographs [John Wiley \& Sons (1985; Zbl 0556.10026)] and [Springer Verlag (1991; Zbl 0758.11036)]). Except for the classical cases \(k =1,2\) (op. cit.) no asymptotic formula of the form \[ I_k(T) \sim c_kT(\log T)^{k^2}\qquad(T\to\infty)\tag{1} \] is known when \(k > 2\). In fact, it is difficult even to guess what the value of the constant \(c_k\) should be, in case the above formula holds. \textit{J. B. Conrey} and \textit{A. Ghosh} [Int. Math. Res. Not. 1998, 775-780 (1998; Zbl 0920.11060)] conjectured that \[ c_3 = {42\over 9!}\prod_p {\left(1 - {1\over p}\right)}^4 \left(1 + {4\over p} + {1\over p^2}\right),\tag{2} \] where \(p\) denotes primes. In the present paper the authors propose that, when \(k=4\), (1) should hold with \[ c_4 = {24024\over 14!}\prod_p {\left(1 - {1\over p}\right)}^9 \left(1 + {9\over p} + {9\over p^2} + {1\over p^3}\right). \tag{3} \] This is achieved by considerations involving Dirichlet polynomials of the form \(D_{k,N}(s) = \sum_{n\leq N}d_k(n)n^{-s}\), where the divisor function \(d_k(n)\) denotes the number of ways \(n\) may be written as a product of \(k\) factors. Such polynomials appear in various approximate functional equations, used to approximate \(\zeta^k(s)\). After squaring and integration, the problem of the asymptotic evaluation of \(I_k(T)\) is reduced to problems involving the so-called general binary additive divisor problem, namely the study of the function \(D_k(x,h) = \sum_{n\leq x}d_k(n)d_k(n+h)\) where the `shift parameter' \(h\) may not be fixed. This condition makes the problem one of extra difficulty, and the authors make a conjecture involving the function \(m_k(x,h)\), the main term approximating \(D_k(x,h)\). This conjecture, too complicated to be reproduced here, is similar in nature to the one previously made by the reviewer [in New Trends Probab. Stat. 4, 69-89 (1997; Zbl 0924.11070)]. By using this conjecture the authors make a detailed analysis which leads to the conjectural value (3). It is worth remarking that recently \textit{J. P. Keating} and \textit{N. C. Snaith} [Commun. Math. Phys. 214, 57-89 (2000; Zbl 1051.11048)] made the general conjecture \[ c_k = {g_k a_k\over \Gamma(1+k^2)}, \] where \[ a_k = \prod_p \left\{{\left(1 - {1\over p}\right)}^{k^2} \sum_{j=0}^\infty d_k^2(p^j)p^{-j}\right\}, \] \[ g_k = \prod_{j=0}^{k-1}{j!\over(j+k)!}. \] It is remarkable that the values provided by the Keating-Snaith conjecture (an asymptotic formula should hold also when \(k \geq 0\) is not necessarily an integer, in which case they also produce the explicit value of the conjectural \(c_k\)) coincide with the known values when \(k=1,2\) and with the values furnished by (2) and (3), respectively. It is even more remarkable that the Keating-Snaith conjecture follows from considerations totally different from those employed by the authors, namely by modelling the behaviour of \(\zeta({1\over 2}+it)\) by analogies from random matrix theory. Further ramifications in this exciting field of research have been lately obtained by various authors, most notably by \textit{J. B. Conrey, D. W. Farmer, J. P. Keating, M. O. Rubinstein} and \textit{N. C. Snaith} [Integral moments of \(L\)-functions, preprint, 58 pp, arXiv:mat.NT/0206018].