Abstract Let $S_2^*(q)$ be the set of primitive Hecke eigenforms of weight 2 and prime level $q$. For $p$ prime and $t\in \mathbb{R}$, we prove asymptotic formulas for the sums $$ \mathcal {A}(p^j,q,t)=\sum_{f\in S_2^*(q)} L\left(\frac{1}{2}+it,f\right)^2\lambda_f(p^j),\qquad j=1,2, $$ where $\lambda_f(p^j)$ is the $p^j$-th normalized Fourier coefficient of $f$ and $L(s,f)$ is the $L$-function associated to $f$.
Introduction
In the study of analytic number theory, the behavior of $L$-functions on the critical line ($Re(s) = 1/2$) remains one of the most significant areas of inquiry. These functions, which generalize the Riemann zeta function, provide deep insights into the distribution of prime numbers and the arithmetic of modular forms. A central technique for investigating these objects is the calculation of moments across a family of $L$-functions. This paper, authored by Wei Liu (2022), advances this field by establishing new asymptotic formulas for the second moments of automorphic $L$-functions associated with primitive Hecke eigenforms, specifically focusing on averages weighted by Fourier coefficients rather than the more common harmonic weights.
The Challenge of Non-Harmonic Weighting
Historically, the analysis of moments for $L$-functions has relied heavily on "harmonic" averages. In these cases, each term in the sum is weighted by the inverse of the Petersson inner product, $w_f = \frac{4\pi \langle f, f \rangle}{\Gamma(k-1)}$. This weighting is mathematically convenient because it allows researchers to employ the Petersson trace formula, which relates sums of Fourier coefficients to sums of Kloosterman sums.
However, many theoretical applications—such as studying the sub-convexity of $L$-functions or their non-vanishing properties—require "natural" or weighted averages where the weights are the Fourier coefficients $\lambda_f(p^j)$ themselves. These non-harmonic sums are significantly more difficult to compute because the orthogonality relations provided by the Petersson formula do not apply directly. The author addresses this difficulty by deriving asymptotic formulas for these sums when the level $q$ is a large prime.
Methodological Framework
The author utilizes a technical framework inspired by the work of VanderKam (1999), adapted for the second moment on the critical line $s = 1/2 + it$. The proof relies on several key pillars:
1. Approximate Functional Equation
To handle the $L$-function values $L(1/2 + it, f)$, the author employs an approximate functional equation. This tool represents the $L$-value as a finite sum of Fourier coefficients, effectively transforming a transcendental problem into a combinatorial one that can be manipulated using Hecke relations.
2. Hecke Multiplicativity
The interaction between the $L$-function and the weight $\lambda_f(p^j)$ is managed through the multiplicative properties of Hecke eigenvalues. Specifically, the relation: $$ \lambda_f(m)\lambda_f(n) = \sum_{d|(m,n)} \lambda_f\left(\frac{mn}{d^2}\right) $$ allows for the simplification of products of coefficients into linear combinations, which are more amenable to summation.
3. Burgess’s Bound and VanderKam’s Lemma
A critical component of the error estimation involves a lemma that bounds sums of Hecke eigenvalues. This estimation is deeply tied to the Burgess bound for Dirichlet $L$-functions, a fundamental result concerning the distribution of character sums. The precision of the final asymptotic formula is directly constrained by the strength of these bounds.
4. Contour Integration
The extraction of the main terms from the resulting sums is achieved via residue calculus. The author evaluates complex integrals involving the Riemann zeta function $\zeta(s)$ and its derivatives, accounting for both the central value ($t=0$) and off-center values ($t \neq 0$) on the critical line.
Key Results
The paper presents two primary theorems that provide explicit asymptotic formulas for the weighted sums $\mathcal{A}(p, q, t)$ and $\mathcal{A}(p^2, q, t)$.
- Theorem 1.1 ($j=1$): For the sum weighted by the first power of the Fourier coefficient, the author establishes an asymptotic formula valid for $p < q^{1/55 - \epsilon}$. The formula consists of a main term dominated by $\log q$ and an error term of $O_\epsilon(p^{39/32} q^{31/32+\epsilon})$.
- Theorem 1.2 ($j=2$): The result is extended to the square of the Fourier coefficient ($p^2$), though the valid range for $p$ is more restrictive, requiring $p < q^{1/110 - \epsilon}$.
These results demonstrate that while weighted averages behave similarly to their harmonic counterparts in terms of their main growth rates, the error terms are far more sensitive to the distribution of the eigenvalues.
Discussion and Future Directions
The limitations of the current study highlight the frontiers of analytic number theory. The restricted range of $p$ relative to $q$ is a direct consequence of our current knowledge of character sums. As the author notes, if the Lindelöf Hypothesis were assumed, the range for $p$ could be significantly extended, potentially allowing for $p$ to grow much faster relative to the level $q$.
Furthermore, while this work focuses on weight $k=2$, the methods suggest that similar results could be obtained for higher weights, albeit with increased complexity in the Hecke relations. Future research may also explore higher powers of the Fourier coefficients ($j > 2$) or higher moments of the $L$-functions, which would provide a more complete picture of the distribution of these values.
Conclusion
By moving beyond the "crutch" of harmonic weights, this paper provides a more natural perspective on the average behavior of automorphic $L$-functions. The explicit asymptotic formulas for $\mathcal{A}(p^j, q, t)$ serve as a valuable tool for researchers working on the sub-convexity problem and the broader distribution of values on the critical line.
Summary updated