Elsevier

Journal of Number Theory

Volume 223, June 2021, Pages 12-34
Journal of Number Theory

Computational Section
Numerical verification of Littlewood's bounds for |L(1,χ)|
Image 1

https://doi.org/10.1016/j.jnt.2020.12.017Get rights and content

Abstract

Let L(s,χ) be the Dirichlet L-function associated to a non-principal primitive Dirichlet character χ defined modq, where q is an odd prime. In this paper we introduce a fast method to compute |L(1,χ)| using the values of Euler's Γ function. We also introduce an alternative way of computing logΓ(x) and ψ(x)=Γ/Γ(x), x(0,1). Using such algorithms we numerically verify the classical Littlewood bounds and the recent Lamzouri-Li-Soundararajan estimates on |L(1,χ)|, where χ runs over the non-principal primitive Dirichlet characters modq, for every odd prime q up to 107. The programs used and the results here described are collected at the following address http://www.math.unipd.it/~languasc/Littlewood_ineq.html.

Introduction

Let q be an odd prime, χ be a Dirichlet character modq and L(s,χ) be the associated Dirichlet L-function. The goal of this paper is to introduce a fast algorithm to compute the values of |L(1,χ)| for every non-principal primitive Dirichlet character χ defined modq and, using such a new method, to numerically study a generalisation of the classical bounds of Littlewood [12] for |L(1,χd)|, where χd is a quadratic Dirichlet character. Assuming the Riemann Hypothesis for L(s,χd) holds, in 1928 Littlewood proved, for dm2, that(12eγπ2(1+o(1))loglogd)1<L(1,χd)<2eγ(1+o(1))loglogd as d tends to infinity, where γ is the Euler-Mascheroni constant. In 1973 Shanks [16] numerically studied the behaviour of the upper and lower Littlewood indices defined asULI(d,χd):=L(1,χd)2eγloglog|d|andLLI(d,χd):=L(1,χd)12eγπ2loglog|d| for several small discriminants d. Such computations were extended by Williams-Broere [17] in 1976 and by Jacobson-Ramachandran-Williams [5] in 2006.

Recently Lamzouri-Li-Soundararajan [8, Theorem 1.5] proved an effective form of Littlewood's inequalities: assuming the Generalised Riemann Hypothesis holds, for every integer q1010 and for every non-principal primitive character χmodq, they obtained that|L(1,χ)|2eγ(loglogqlog2+12+1loglogq) and1|L(1,χ)|12eγπ2(loglogqlog2+12+1loglogq+14loglogqlogq).

Using our method we will compute the values of |L(1,χ)| for every non-principal primitive Dirichlet character χ defined modq, for every odd prime q up to 107. This largely extends previous results. Moreover, lettingMq:=maxχχ0|L(1,χ)|,mq:=minχχ0|L(1,χ)|,f(q):=loglogqlog2+1/2+1/loglogq,g(q):=f(q)+14(loglogq)/logq, we obtain the following

Theorem 1

Let 3q107, q be a prime number and Mq be defined in (4). Then we have 0.604599=M3MqM4305479=6.399873. Moreover, we also have0.3252eγf(q)<Mq<0.622eγf(q), where the lower bound holds just for q79, and0.4<maxχχ0ULI(q,χ)<0.66, where the upper bound holds just for q5.

We also have an analogous result on mq.

Theorem 2

Let 3q107, q be a prime number and mq be defined in (4). Then we have 0.198814=m991027mqm11=0.618351. Moreover, we also haveπ212eγ2.35g(q)<mq<π212eγ5g(q), where the upper bound holds just for q953, and1.13<minχχ0LLI(q,χ)<2, where the lower bound holds just for q373.

Theorem 1, Theorem 2 are in agreement with Littlewood's bounds in (1) and the Lamzouri-Li-Soundararajan estimates in (2)-(3).

The paper is organised as follows: in Section 2 we will see how to compute |L(1,χ)| using the values of Euler's Γ function and the Fast Fourier Transform algorithm; we will also describe the actual computation we performed and how Theorem 1, Theorem 2 are obtained. In Sections 3-4 we will see how to efficiently evaluate logΓ(x), and ψ(x)=Γ/Γ(x), for x(0,1) using precomputed values of the Riemann zeta-function at positive integers. We will also insert some tables and figures (the scatter plots were obtained using GNUPLOT, v.5.2.8).

Section snippets

Acknowledgments

I wish to thank Luca Righi (University of Padova) for his help in developing the C language implementation of the algorithms described in Sections 3-4. I would also like to thank the referee for his/her remarks.

Computation of |L(1,χ)| and proofs of Theorems 1-2

Recall that q is an odd prime and let χ be a primitive non-principal Dirichlet character mod q. The values of |L(1,χ)| can be computed in two different ways. Recalling eq. (3.1) of [2], we have L(1,χ)=q1a=1q1χ(a)ψ(a/q), so that|L(1,χ)|=1q|a=1q1χ(a)ψ(aq)|, where ψ(x)=Γ/Γ(x) is the digamma function and Γ is Euler's function. As we will see later, for computational purposes it is in fact more efficient to distinguish between the parity of the Dirichlet characters. If χ is an even character

An alternative algorithm to compute logΓ(x), x(0,1)

We describe here an alternative way of computing logΓ(x), x(0,1), which is based on the well-known Euler formula (see, e.g., Lagarias [7, eq. (2.3.3)]):logΓ(x)=γ(1x)+k=2+ζ(k)k(1x)k, where ζ(s) is the Riemann zeta-function.

We follow the argument used in Languasco-Righi [11] to study the Ramanujan-Deninger Gamma function Γ1(x). We immediately remark that the series in (12) absolutely converges for x(0,2); this fact and the well-known relationlogΓ(1+x)=logΓ(x)+logx,x>0, let us obtain log

An alternative algorithm to compute ψ(x), x(0,1)

Here we apply to the digamma function ψ(x)=Γ/Γ(x), x(0,1), the same argument used in Section 3. The starting point is the well-known Euler formula (see, e.g., Lagarias [7, Theorem 3.1.2]):ψ(x)=γk=2+ζ(k)(1x)k1.

We immediately remark that the series in (23) absolutely converges for x(0,2); this fact and the well-known relationψ(1+x)=ψ(x)+1x let us obtain ψ(x), x(0,1), in two different ways. Recalling ψ(1)=γ and ψ(1/2)=2log2γ, we also remark that, letting nN, n2, for every x(0,2)

References (17)

  • H. Cohen

    Number Theory, vol. II: Analytic and Modern Tools

    (2007)
  • K. Ford et al.

    Values of the Euler ϕ-function not divisible by a given odd prime, and the distribution of Euler-Kronecker constants for cyclotomic fields

    Math. Comput.

    (2014)
  • M. Frigo et al.

    The design and implementation of FFTW3

    Proc. IEEE

    (2005)
  • N.J. Higham

    The accuracy of floating point summation

    SIAM J. Sci. Comput.

    (1993)
  • M.J. Jacobson et al.

    Numerical results on class groups of imaginary quadratic fields

  • W. Kahan

    Further remarks on reducing truncation errors

    Commun. ACM

    (1965)
  • J.C. Lagarias

    Euler's constant: Euler's work and modern developments

    Bull. Am. Math. Soc.

    (2013)
  • Y. Lamzouri et al.

    Conditional bounds for the least quadratic non-residue and related problems

    Math. Comput.

    (2015)

    Math. Comput.

    (2017)
There are more references available in the full text version of this article.

Cited by (10)

  • The Kummer ratio of the relative class number for prime cyclotomic fields

    2024, Journal of Mathematical Analysis and Applications
  • Uniform effective estimates for |L(1,χ)|

    2022, Journal of Number Theory
View all citing articles on Scopus
View full text