An $L^2$-bound for the Barban-Vehov weights

Olivier Ramaré,Sebastian Zuniga Alterman

2024-05-21

Abstract:Let $\lambda$ the Barban--Vehov weights, defined in $(1)$. Let $X\ge z_1\ge100$ and $z_2=z_1^\tau$ for some $\tau>1$. We prove that

Number Theory

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the L2 - mean estimation problem regarding Barban - Vehov weights. Specifically, the authors have studied the upper bounds of Barban - Vehov weights under different conditions and provided a more accurate and more efficient estimation method.

### Problem Background

In 1968, Barban and Vehov introduced a special weight function, called Barban - Vehov weights. These weights are defined as:

\[
\lambda_d=\lambda_d(z_2, z_1)=\frac{\mu(d)\left(\log^+\left(\frac{z_2}{d}\right)-\log^+\left(\frac{z_1}{d}\right)\right)}{\log\left(\frac{z_2}{z_1}\right)}
\]

where $\log^+x = \max(\log x, 0)$ and $z_2>z_1>0$. They found that the L2 - mean of these weights is very small, even when $n$ is less than $z_2^2$.

### Research Objectives

The main objective of this paper is to improve the estimation of the L2 - mean of Barban - Vehov weights. Specifically, the authors consider the following sum:

\[
\sum_{n\leq X}\frac{1}{n}\left(\sum_{d\mid n}\lambda_d\right)^2
\]

and attempt to find a tighter upper bound. In addition, they also study the more general weighted sum:

\[
\sum_{n\geq1}\frac{1}{n^{1 + \varepsilon}}\left(\sum_{d\mid n}\lambda_d\right)^2
\]

### Main Results

The authors have proved the following main theorems:

**Theorem 1.1**: Let $\varepsilon>0$ and $z_1\geq100$, and let $z_2 = z_1^{\tau}$ (where $\tau>1$), then:

\[
\sum_{n\geq1}\frac{1}{n^{1+\varepsilon}}\left(\sum_{d\mid n}\lambda_d\right)^2\leq\frac{e^{\gamma\varepsilon}}{\varepsilon}\frac{\log\left(\frac{z_2}{z_1}\right)}{\log z_1}\left[A(\tau + 1)+\left(B - C\frac{1}{z_1^{\varepsilon}}+(B - C\frac{1}{z_2^{\varepsilon}})\tau^2\right)\varepsilon\right]
\]

where $A = 1.084$, $B = 1.301$, $C = 0.318$.

**Corollary 1.2**: When $X\geq z_1\geq100$ and $z_2 = z_1^{\tau}$, we have:

\[
\sum_{n\leq X}\frac{1}{n}\left(\sum_{d\mid n}\lambda_d\right)^2\leq\frac{3.09\log X}{\log\left(\frac{z_2}{z_1}\right)}\left[1.084(\tau + 1)+1.301(1 + \tau^2)-0.116\frac{1}{\tau - 1}\right]
\]

In particular, when $\tau = 2$, the right - hand side of the above inequality can be simplified to $30\log X/\log z_1$.