[Paper Review] On sets of large exponential sums

Ilya D. Shkredov|ArXiv.org|May 26, 2006

Limits and Structures in Graph Theory8 references4 citations

TL;DR

This paper establishes that the set of large Fourier coefficients $\mathcal{R}_{\alpha}$ of a subset $A \subseteq \mathbb{Z}/N\mathbb{Z}$ with density $\delta$ exhibits strong additive structure: the number of quadruples $(r_1,r_2,r_3,r_4) \in \mathcal{R}_{\alpha}^4$ satisfying $r_1 + r_2 = r_3 + r_4$ is at least $|\mathcal{R}_{\alpha}|^{2+\epsilon}$ for some $\epsilon > 0$, indicating high internal additive coherence. The result is derived via Fourier analysis and convolution estimates, improving bounds on Bohr sets in $2A - 2A$ and refining prior work on Freiman-type theorems.

ABSTRACT

Let A be a subset of Z / NZ, and let R be the set of large Fourier coefficients of A. Properties of R have been studied in works of M.-C. Chang and B. Green. Our result is the following : the number of quadruples (r_1, r_2, r_3, r_4) \in R^4 such that r_1 + r_2 = r_3 + r_4 is at least |R|^{2+ε}, ε>0. This statement shows that the set R is highly structured. We also discuss some of the generalizations and applications of our result.

Motivation & Objective

• To understand the additive structure of the set $\mathcal{R}_{\alpha}$ of large Fourier coefficients of a subset $A \subseteq \mathbb{Z}/N\mathbb{Z}$ with density $\delta$.
• To establish quantitative lower bounds on the number of additive quadruples in $\mathcal{R}_{\alpha}$, showing it exceeds $|\mathcal{R}_{\alpha}|^2$ by a superpolynomial factor.
• To improve existing bounds on Bohr sets contained in $2A - 2A$, refining results relevant to Freiman's theorem and additive combinatorics.
• To demonstrate that $\mathcal{R}_{\alpha}$ is not just sparse but highly structured, even when $\delta \to 0$ as $N \to \infty$.

Proposed method

• Use of Parseval's identity to relate $L^2$-norms of the Fourier transform to the additive energy of $A$.
• Application of convolution identities: $\widehat{A*A*A*A}(x) = \sum_r |\widehat{A}(r)|^4 e(rx)$, linking additive energy to exponential sums.
• Construction of a Bohr set $B_1 = B(\mathcal{R}_{\alpha}^*, 1/20)$ on which $|\widehat{A}(r)|^4$ remains large, ensuring positivity of the convolution.
• Use of a structured generating set $\Lambda^*$ of size $O(\delta^{-1} \log(1/\delta))$ to represent elements of $\mathcal{R}_{\alpha}^*$ via signed sums, enabling the construction of a smaller Bohr set $B_2 \subseteq B_1$.
• Establishing $B_2 \subseteq 2A - 2A$ by showing that for $x \in B_2$, the sum $\sum_r |\widehat{A}(r)|^4 e(rx) > 0$, implying $x \in 2A - 2A$.
• Use of the triangle inequality and bounds on $\|rx/N\|$ to show that $B_2 \subseteq B_1$, leveraging the representation of $r \in \mathcal{R}_{\alpha}^*$ via $\Lambda^*$.

Experimental results

Research questions

• RQ1 What is the minimal number of additive quadruples $(r_1,r_2,r_3,r_4)$ in $\mathcal{R}_{\alpha}^4$ satisfying $r_1 + r_2 = r_3 + r_4$?
• RQ2 Can the set $\mathcal{R}_{\alpha}$ be shown to possess nontrivial additive structure, even when $\delta$ is small?
• RQ3 How can the structure of $\mathcal{R}_{\alpha}$ be leveraged to find large Bohr sets inside $2A - 2A$?
• RQ4 To what extent does the number of solutions to $r_1 + r_2 = r_3 + r_4$ in $\mathcal{R}_{\alpha}$ exceed the trivial count?

Key findings

• The number of solutions to $r_1 + r_2 = r_3 + r_4$ with $r_i \in \mathcal{R}_{\alpha} \setminus \{0\}$ is at least $|\mathcal{R}_{\alpha}|^{2+\epsilon}$ for some $\epsilon > 0$, indicating strong additive structure.
• For $k=2$, the number of such quadruples is $\Theta(\delta / \alpha^4)$, which exceeds the trivial bound of $3|\mathcal{R}_{\alpha}|^2$ when $\delta / \alpha^2$ is large.
• The set $2A - 2A$ contains a Bohr set $B(K, \varepsilon)$ with $|K| \leq 2^{33} \delta^{-1} \log(1/\delta)$ and $\varepsilon = 1/(2^8 \log(1/\delta))$, improving prior bounds.
• The cardinality of this Bohr set is at least $\frac{1}{2} \cdot 2^{-2^{35} \delta^{-1} \log(1/\delta) \log \log(1/\delta)} N$, showing exponential decay in the exponent.
• The result holds under the condition $(N,6) = 1$ and for $\delta \leq 2^{-256}$, ensuring the applicability of the method in the small density regime.
• The proof establishes that $B_2 \subseteq B_1 \subseteq 2A - 2A$, where $B_2$ is a Bohr set defined over a generating set $\Lambda^*$ of size $O(\delta^{-1} \log(1/\delta))$.