A structure theorem for sets with doubling $4+\delta$

1. Problem Statement and Context

The Core Problem:
The paper addresses a central question in additive combinatorics regarding the structural classification of finite sets of integers $A$ with small "doubling." The doubling constant is defined as $\sigma[A] = |A+A|/|A|$.

• Freiman's Theorem: Establishes that sets with small doubling are contained in generalized arithmetic progressions (GAPs) of bounded dimension. However, the quantitative dependence of the dimension and size on the doubling constant $K$ is a major open area.
• Green's Question (Question 1.1): Green asked whether, for a set $A$ with $|A-A| \le K|A|$, one can find a structured subset $A' \subseteq A$ that obviously satisfies $|A'-A'| \le (K+\epsilon)|A'|$.
• The Gap at $K=4$:
  • For $K < 3$, Freiman's $3k-3$ theorem implies $A$ is contained in a 1-dimensional progression.
  • For $K < 4$, Eberhard, Green, and Manners (EGM) proved that $A$ contains a large subset with density $> 1/2$ inside a 1-dimensional progression.
  • The Challenge: When $K \ge 4$, two distinct structural types emerge:
    1. A dense subset of a 1-dimensional progression (doubling $\approx 4$).
    2. A dense subset of a 2-dimensional progression (doubling $\approx 4$).
  • The paper aims to prove that for $K = 4 + \delta$ (with $\delta$ sufficiently small), only these two structures exist.

2. Main Results

Theorem 1.2 (Main Structural Theorem):
Let $\epsilon', \delta > 0$ with $\delta$ sufficiently small. If $A \subset \mathbb{Z}$ is a finite non-empty set such that $|A+A| \le (4+\delta)|A|$, then one of the following holds:

1. 1D Case: There exists a 1-dimensional arithmetic progression $P$ such that $|P| \gg_\delta |A|$ and $|A \cap P|/|P| \ge 1/2 - \epsilon'$.
2. 2D Case: There exists a proper 2-dimensional arithmetic progression $Q$ such that $|Q| \gg_\delta |A|$ and $|A \cap Q|/|Q| \ge 1 - \epsilon'$.

Note: The same result holds for difference sets $A-A$.

Corollary 1.3 (Almost-All Version):
If a large proportion of pairs in $A \times A$ (specifically $(1-\delta')|A|^2$ pairs) have a restricted sumset size $\le (4+\delta)|A|$, then the same structural conclusions hold. This generalizes the Balog–Szemerédi–Gowers theorem.

Corollary 1.4 (Bogolyubov–Ruzsa Type Lemma):
Under the same hypotheses, the set $41A - 40A$ contains a proper $d$-dimensional progression with $d \le 2$ and size $\gg_\delta |A|$. This provides a sharp bound on the dimension ($d \le 2$) at the cost of a larger linear combination ($41A - 40A$).

3. Methodology and Proof Strategy

The proof is a sophisticated synthesis of additive combinatorics, harmonic analysis, and geometry of numbers. It proceeds through several stages:

A. Reduction to Dense Sets (The Dense Model Lemma)

The authors first reduce the general case to the case where $A$ is a dense subset of an interval $[N]$. They utilize:

• Freiman Isomorphisms: To map $A$ to a subset of a cyclic group or interval while preserving additive structure.
• Bogolyubov-type Lemmas: Using results by Lev and EGM to show that if $A$ has small doubling, specific linear combinations (like $9A-8A$ or $41A-40A$) contain large progressions.

B. Arithmetic Regularity Lemma

The core analytic tool is the Abelian Arithmetic Regularity Lemma (following EGM). This decomposes the characteristic function $1_A$ into three parts:

1. Structured ($f_{struct}$): A function depending on a nilsequence (specifically a Lipschitz function on a torus $T^d$).
2. Small ($f_{sml}$): A function with small $L^2$ norm.
3. Uniform ($f_{unf}$): A function with small $U^2$ Gowers norm (pseudorandom).

This allows the problem to be transferred from integers $\mathbb{Z}$ to a continuous setting: subsets of $\mathbb{Z}/q\mathbb{Z} \times [M] \times T^d$.

C. Fiber Analysis and Kneser-Type Inequalities

The set $A$ is partitioned into "fibers" over the torus $T^d$. The authors analyze the sumset of these fibers using:

• Inverse Kneser Theorem (Tao): If the sumset of two sets in $T^d$ is small, the sets must be close to the inverse image of intervals under a homomorphism.
• Weighted Kneser Inequalities: Used to handle the specific weights arising from the density of fibers.

The proof splits into three cases based on the density of these fibers:

1. 1-Dimensional Structured Case: A fiber has density $\ge 1/2$. This directly yields the 1D progression result.
2. 2-Dimensional Structured Case: Fibers are well-approximated by the inverse image of an interval under a homomorphism $\phi: T^d \to T$. This leads to the 2D progression result.
3. Expansion Case: If neither of the above holds, the authors prove that the sumset must expand significantly.

D. The Expansion Case (Brunn-Minkowski)

In the expansion case, the authors assume the set does not have the 1D or 2D structure. They apply a Planar Brunn-Minkowski inequality (via a lemma by EGM) to the densities of the fibers.

• They show that if the fibers are not structured, the sumset of the fibers must have a measure strictly larger than the sum of the measures.
• This leads to a contradiction with the assumption that $|A+A| \le (4+\delta)|A|$, proving that the "Expansion Case" cannot occur for sufficiently small $\delta$.

E. From Bohr Sets to Progressions

In the 2D structured case, the set $A$ is shown to have high density on an inhomogeneous Bohr set $B = \{n : n\theta \in I\}$.

• The authors use Geometry of Numbers (Minkowski's theorems on successive minima) and Continued Fractions to prove that such Bohr sets contain large proper 2-dimensional progressions.
• This step is crucial for translating the continuous torus structure back into a discrete integer progression.

4. Key Contributions

1. Resolution of the $K=4$ Threshold: The paper confirms that for doubling slightly above 4, the only possible structures are dense subsets of 1D or 2D progressions. This settles a specific variation of Green's question for $K=4+\delta$.
2. Optimality: The result is shown to be optimal up to the quantitative dependence between $\delta$ and $\epsilon'$. The examples provided (dense subsets of 1D and 2D progressions) are the only extremal cases.
3. Methodological Innovation:
   • Refinement of EGM: The authors extend the EGM framework (which worked for $K<4$) to handle the $K \ge 4$ regime by introducing a finer classification of fiber sizes (small, medium, large) and utilizing inverse Kneser theorems more aggressively.
   • Handling the "Gap": They successfully navigate the transition where the structure shifts from 1D to 2D, a regime where previous methods failed.
4. Sharp Dimension Bounds: The derived Bogolyubov-Ruzsa lemma provides a sharp bound on the dimension ($d \le 2$) of the resulting progression, a significant improvement over general Freiman-type results which often yield much larger dimensions.

5. Significance

• Additive Combinatorics: This work represents a major step forward in the quantitative understanding of Freiman's theorem. It bridges the gap between the well-understood $K < 3$ and $K < 4$ regimes and the more complex $K \ge 4$ regime.
• Applications: The structural results have implications for the Erdős sum-free set conjecture and other problems involving sum-free sets and additive energy.
• Technique Transfer: The combination of the Arithmetic Regularity Lemma with continuous geometric inequalities (Brunn-Minkowski) and discrete geometry of numbers demonstrates a powerful methodology for tackling inverse problems in additive number theory.
• Future Directions: The paper suggests that extending these results to arbitrary $K$ (answering Green's question for all $K$) is extremely difficult and likely requires new ideas beyond the current polynomial Freiman-Ruzsa conjecture framework.

In summary, Jing and Mudgal provide a definitive structural classification for sets with doubling just above 4, proving that such sets are essentially either "half-dense" in a line or "almost full" in a plane, and they do so by weaving together deep tools from analysis, combinatorics, and geometry.