Conjectures and Experiments Concerning the Moments of L(1/2, χ d )

Matthew W. Alderson, Michael RubinsteinExp. Math 2012 · DOI

TL;DR

This paper presents a large-scale computational study of the moments of quadratic Dirichlet L-functions at the critical point $s=1/2$. By performing extensive numerical experiments, the authors provide strong evidence supporting existing conjectures about the asymptotic behavior of these moments and offer insights into the presence of lower-order terms, which have been a subject of theoretical debate.

What Problem Does This Paper Address?

The behavior of L-functions, particularly at their critical points, is a central theme in number theory. For quadratic Dirichlet L-functions, $L(s, \chi_d)$, understanding their moments, i.e., averages of $L(1/2, \chi_d)^k$ over families of discriminants $d$, is crucial for testing deep conjectures related to their distribution and analytic properties. While leading-order asymptotic formulas have been proposed, the existence and form of lower-order terms have been less clear, with different theoretical approaches yielding varying predictions. This paper aims to provide numerical evidence to resolve these discrepancies and validate the proposed asymptotic formulas.

Methods

The authors employ a two-pronged computational strategy, tailored to the sign of the discriminant $d$:

  • For $d < 0$: They utilize a method based on the theory of binary quadratic forms and the K-Bessel expansions of Epstein zeta functions, as developed by Chowla and Selberg. This approach involves computing sums over primitive quadratic forms.
  • For $d > 0$: A smooth approximate functional equation for $L(s, \chi_d)$ is used, which expresses the L-function in terms of the incomplete gamma function.

Both methods are designed for efficient computation of $L(1/2, \chi_d)$ values. The paper also details significant algorithmic optimizations, including:

  • Parallelization: Distributing computations across multiple processors.
  • Efficient GCD computation: Optimizing primitivity checks.
  • Prefetching: Anticipating data needs to reduce memory access latency.
  • Taylor series expansions: Precomputing and using Taylor series for the K-Bessel function.
  • Precomputed tables: Storing values of $\chi_d(n)$ and $\log|d|$ to speed up calculations.

Datasets

The paper does not introduce new benchmark datasets. Instead, it utilizes a large dataset of fundamental discriminants $d$ ranging from $-5 \times 10^{10}$ to $1.3 \times 10^{10}$. This extensive range allows for robust statistical analysis and testing of asymptotic formulas.

Key Insights and Ideas

  • Strong Support for CFKRS Conjectures: The numerical results provide compelling evidence for the full asymptotic expansion of the moments of quadratic Dirichlet L-functions conjectured by Conrey, Farmer, Keating, Rubinstein, and Snaith (CFKRS). The agreement is remarkably good, especially for lower moments.
  • Evidence for Lower-Order Terms: The analysis of the remainder terms suggests the presence of additional lower-order terms, as predicted by Diaconu, Goldfeld, and Hoffstein (DGH) and Zhang. This is particularly evident for the third moment ($k=3$), where a bias consistent with the predicted $X^{3/4}$ term is observed.
  • Differences Between Positive and Negative Discriminants: The paper highlights subtle differences in the behavior of the remainder terms for positive and negative discriminants, potentially due to the magnitude of "noise" and the size of predicted constant factors.
  • Algorithmic Efficiency is Crucial: The detailed discussion of computational "hacks" underscores the significant impact of algorithmic optimizations on the feasibility of such large-scale number theoretic computations.

Key Results

  • The computed moments for $k=1, \dots, 8$ show excellent agreement with the CFKRS predictions, typically within 7-8 decimal places for $k=1$ and 4-5 decimal places for $k=8$.
  • For $k=3$, the numerical data supports the existence of a lower-order term of size $X^{3/4}$, with the observed bias aligning with Zhang's predictions for both positive and negative discriminants, though the fit is more conclusive for $d>0$.
  • The computational methods developed allowed for the evaluation of $L(1/2, \chi_d)$ for a vast range of discriminants, $-5 \times 10^{10} < d < 1.3 \times 10^{10}$.
  • The computational approach for $d<0$ was found to be significantly faster (approximately 20 times) than for $d>0$, consistent with theoretical estimates of the number of operations.

Limitations and Future Work

  • Noise in Remainder Terms: The authors acknowledge that the "noise" in the remainder terms can make it challenging to definitively detect lower-order terms, especially when their constant factors are small.
  • Unproven Hypotheses: Some of the theoretical predictions regarding lower-order terms rely on unproven hypotheses about the meromorphic continuation and growth of certain zeta functions.
  • Experimental Nature: The study is primarily experimental, aiming to test conjectures rather than provide rigorous proofs. The authors suggest that interval arithmetic packages would be needed for rigorous results.
  • Future Work: The paper implicitly suggests further investigation into the precise form and behavior of lower-order terms for higher moments and a deeper theoretical understanding of the observed biases. The authors also mention that their computations were repeated with higher precision to mitigate concerns about numerical errors, reinforcing the validity of their findings.

Summary updated

BibTeX
@article{alderson2012conjectures,
  author = {Matthew W. Alderson and Michael Rubinstein},
  title = {{Conjectures and Experiments Concerning the Moments of L(1/2, χ d )}},
  journal = {Exp. Math.},
  year = {2012},
  doi = {10.1080/10586458.2012.687238},
}
Figure