A Higher Rank Shifted Convolution Problem with Applications to L-functions

Basic Information

• Paper ID: 2511.03294
• Title: A higher rank shifted convolution problem with applications to L-functions
• Authors: Valentin Blomer, Junxian Li
• Classification: math.NT (Number Theory)
• Publication Date: November 5, 2025
• Paper Link: https://arxiv.org/abs/2511.03294v1

Abstract

This paper resolves a long-standing open case in the GL(3) × GL(2) shifted convolution problem: one factor is the classical divisor function while the other is a GL(3) Fourier coefficient. The proof employs an intertwined application of two distinct types of delta symbol methods. As an application, an asymptotic formula is established for the central values of L-functions of GL(3) automorphic forms twisted by Dirichlet characters (moduli q ≤ Q).

Research Background and Motivation

Research Problem

The shifted convolution problem investigates asymptotic formulas for products of two (typically multiplicative) arithmetic functions when their arguments differ by an additive shift. This is an important tool for measuring the correlation between two functions.

Problem Significance

1. Classical Background: The most classical case of divisor functions $\sum_{n≤x} τ(n)τ(n+1)$ has been studied for a century
2. Theoretical Importance: The shifted convolution problem is intimately connected to moments of L-functions, sieve theory, the Titchmarsh divisor problem, and other core problems in number theory
3. Technical Challenge: The shifted convolution problem for higher rank automorphic forms is extremely challenging, with nontrivial results remaining unknown for most cases

Limitations of Existing Methods

For the GL(3) × GL(2) case, the following progress has been made:

• Case (1.1): $\sum_{n≤x} τ_3(n)τ(n+1)$ - Hooley obtained the main term; Deshouillers gave a power-saving error term; Topacogullari achieved the current best $O(x^{5/6+θ/3+ε})$
• Case (1.2): $\sum_{n≤x} τ_3(n)λ(n+1)$ - Pitt established power-saving bounds; Tang achieved $O(x^{5/6+θ/3+ε})$
• Case (1.3): $\sum_{n≤x} A(n,1)λ(n+1)$ - Munshi obtained power-saving bounds; Xi achieved $O(x^{21/22+ε})$

However, Case (1.4): $\sum_{n≤x} A(n,1)τ(n+1)$ has remained open and could not be attacked by any method used for the preceding cases.

Research Motivation

1. Filling a Theoretical Gap: Case (1.4) is the last unresolved case in the GL(3) × GL(2) shifted convolution problem
2. Methodological Breakthrough: The exponential sum for the divisor function behaves poorly on the major arc (in the L² sense), preventing direct application of Jutila's circle method or the Kloosterman method
3. Application Value: Resolving this problem enables establishing an asymptotic formula for GL(3) L-function twisted moments, a breakthrough after 20+ years

Core Contributions

1. Resolving a Long-Standing Problem: First establishment of power-saving bounds for the shifted convolution of GL(3) Fourier coefficients with the divisor function
2. Innovative Methodology: First intertwined application of the Jutila method and the refined Kloosterman circle method, overcoming the limitations of each
3. Complete Uniformity: Achieves complete uniformity in all parameters h, λ₁, λ₂ in the bilinear shifted equation
4. Asymptotic Formula for L-function Moments: Establishes an asymptotic formula for central values of L-functions of GL(3) automorphic forms twisted by Dirichlet characters, with approximately Q² terms but conductor approximately Q³

Method Details

Core Theorems

Theorem 1: Let h, λ₁, λ₂ ∈ ℤ{0}, x ≥ 1. Let W, W₀ be smooth functions with compact support in 1,2. Let A(n,1) be the Hecke eigenvalue of a cusp form F on SL₃(ℤ). Then $\sum_{λ₁m-λ₂n=h} A(n,1)τ(m)W_0\left(\frac{|λ₁|m}{x}\right)W\left(\frac{|λ₂|n}{x}\right) \ll_{F,W,W_0,ε} x^{41/42+ε}$ holds uniformly for any ε > 0, uniformly in h, λ₁, λ₂.

Theorem 2: A more flexible variant replacing τ(m) with $τ_{A,B}(m) := \sum_{ab=m} v_1(a/A)v_2(b/B)$, achieving the same bound under the condition AB ≍ x/|λ₁|.

Technical Architecture

1. Dual Application of Delta Symbol Methods

The paper innovatively combines two delta symbol methods:

Jutila Method (Lemma 1):

• Provides flexible modulus selection, creating bilinear structure
• Approximates the delta function only in the L² sense
• Defines the periodic function: $χ(α) = \frac{1}{δL}\sum_q ω(q)\sum_{a\bmod q, (a,q)=1}\sum_{k∈ℤ} ψ\left(\frac{1}{δ}\left(α-\frac{a}{q}+k\right)\right)$
• Key property: $\int_0^1 (1-χ(α))^2 dα \ll ψ \frac{Q^2\|ω\|_∞|\log δ|^3}{L^2δ}$

Kloosterman Refined Method (Lemma 2):

• Sensitive to exponential sums of the divisor function on the major arc
• Uses Farey fraction decomposition of the 0,1 interval
• Provides square-root savings in the b/c estimate (for b modulo c)

2. Intertwined Application of Methods

Step 1: Initial Decomposition Write the sum as S₁ + S₂, where:

• S₁: contribution from χ(α)
• S₂: contribution from (1-χ(α))

Step 2: Estimation of S₂ (Minor Arc)

• Choose parameters C₀ < C with logC ≍ log x
• Partition the sum over c ≤ C into major arc (c ≤ C₀) and minor arc (C₀ < c ≤ C)
• Minor arc: use L² estimates and Voronoi summation (Lemmas 3, 4)
• Major arc: apply Kloosterman refinement (Lemma 2); key observation is that nontrivial (square-root saving) estimates can be obtained on b/c

Step 3: Special Modulus Selection Adopt Munshi's idea of choosing factorizable moduli: $ω(q) = \sum_{\frac{1}{2}Q_1≤p≤Q_1, p\text{ prime}, p∤hλ_1λ_2} \sum_{t∈ℕ, pt=q} ρ\left(\frac{t}{Q_2}\right)$ where Q = Q₁Q₂. This creates a bilinear structure.

Step 4: Estimation of S₁

• Apply Voronoi summation (Lemmas 3, 5) to dualize τ(m) and A(n,1)
• Key: maintain both m and p sums within absolute values (unlike Mu, Xi)
• Apply Cauchy-Schwarz inequality
• Use Poisson summation and character sum estimates (Lemmas 7, 8)

3. Refined Estimates for Character Sums

Lemma 7: Improved bounds for hyper-Kloosterman sums $Σ_{h,d,n_1,n_2}(c) \ll_ε c^{1+ε}c_2^{1/2}\frac{\sqrt{(n_1,c_1,d,h)}}{\sqrt{n_1}}$ where c = c₁c₂, c₁ is squarefree, c₂ is a perfect power, (c₁,c₂)=1.

Lemma 8: Character sum estimates for the twin prime case