Near optimal three-fold additive energy bound for points on convex curves

Technical Summary: Near Optimal Three-Fold Additive Energy Bound for Points on Convex Curves

Problem Statement
The paper addresses the problem of bounding the three-fold additive energy, denoted $J_3(\gamma(X))$, for a finite set of points $X \subset \mathbb{R}$ mapped onto a strictly convex curve $\gamma(t) = (t, f(t))$. Specifically, $J_3(\gamma(X))$ counts the number of sextuples $(x_1, \dots, x_6) \in X^6$ such that the sum of the first three points on the curve equals the sum of the last three:
$$\sum_{i=1}^3 \gamma(x_i) = \sum_{i=4}^6 \gamma(x_i).$$
While diagonal solutions (where the sets of indices are identical) contribute roughly $|X|^3$ to this count, the natural target for a non-trivial bound is $O(|X|^{3+\varepsilon})$. Previous results for specific curves (like the circle or parabola) achieved bounds such as $|X|^{7/2}$ or $|X|^{4-o(1)}$, often relying on deep number theory or requiring the points to be "well-spaced" (i.e., the minimal distance between consecutive points is comparable to $|X|^{-1}$). The authors aim to establish the near-optimal bound $J_3(\gamma(X)) \ll_\varepsilon |X|^{3+\varepsilon}$ for any strictly convex curve and any finite set $X$, removing the well-spaced restriction.

Methodology and Key Ideas
The proof relies on a novel combinatorial-geometric argument that transforms the problem of counting equal triple sums into a recurrence relation, iterated to achieve the near-cubic exponent. The core methodology involves:

1. Interlacing Geometry via Karamata's Inequality:
   The central geometric insight is that for strictly convex functions, the solutions to $\gamma(x) + \gamma(y) + \gamma(z) = \gamma(x') + \gamma(y') + \gamma(z')$ with ordered triples $x<y<z$ and $x'<y'<z'$ exhibit a strict "interlacing" property. If $x < x'$, then the variables must satisfy:
   $$x < x' < y' < y < z < z'.$$
   This ordering is derived from Karamata's inequality (Lemma 2.5) and significantly reduces the complexity of the solution set.

2. Recurrence via Gap Parameters:
   The authors define an extremal quantity $T(N, D)$, which counts six-tuples satisfying the sum equality where the "gaps" (rank differences) between paired elements ($x$ vs $x'$, $y$ vs $y'$, $z$ vs $z'$) are bounded by a parameter $D$.

   • First Estimate (Combinatorial): Using the interlacing property and a Szemerédi–Trotter type theorem for translations of convex curves, they derive a bound relating the total energy $J(N)$ to $T(N, D)$:
     $$J(N) \ll \frac{N^4}{D} + \frac{N}{D} T(N, D).$$
   • Second Estimate (Harmonic Analysis): They establish a bound for $T(N, D)$ in terms of the energy at the smaller scale $D$:
     $$T(N, D) \ll_\eta \left(\frac{N}{D}\right)^{2+\eta} J(D).$$
     This step utilizes $L^2$ orthogonality, the boundedness of the Hilbert transform, and vector-valued interpolation (specifically, disjoint-cone $l_3(L_3)$ flat decoupling). The strict convexity ensures that chord vectors from disjoint rank blocks lie in disjoint conic regions in the frequency plane, allowing for orthogonality.

3. Iteration:
   By combining these two estimates and choosing an appropriate scale $D$ (e.g., $D = N^{1-\varepsilon/4}$), the authors set up a recurrence. Assuming polynomial growth $J(N) \sim N^\alpha$, the iteration forces $\alpha \le 3$, yielding the near-optimal bound.

Key Contributions and Results

• Theorem 1.1: For every $\varepsilon > 0$, there exists a constant $C_\varepsilon$ such that for any finite $X \subset \mathbb{R}$ and strictly convex $\gamma$:
  $$J_3(\gamma(X)) \le C_\varepsilon |X|^{3+\varepsilon}.$$
  This bound is sharp up to the $|X|^\varepsilon$ factor. The authors note that for $k \ge 4$, the bound $J_k(\gamma(X)) \ll_\varepsilon |X|^{2k-3+\varepsilon}$ is the best possible uniform estimate, which is a direct consequence of the $k=3$ case via interpolation.
• Removal of Well-Spaced Restriction: Unlike previous decoupling approaches (e.g., Bourgain-Demeter) which required points to be well-spaced (minimal distance $\delta \sim |X|^{-1}$), this result holds for arbitrary finite sets, including rescaled lattice points on the unit circle which are not well-spaced.
• Applications to Convex Sequences: The authors apply their result to convex sequences $A \subset \mathbb{R}$ (where $a_{k+1} - a_k > a_k - a_{k-1}$). They prove improved lower bounds for the size of the sumset and difference set:
  $$|A - A| \gg_\varepsilon |A|^{5/3-\varepsilon}, \quad |A + A| \gg_\varepsilon |A|^{8/5-\varepsilon}.$$
  These results improve upon previous bounds obtained by the first author and others.

Significance and Claims
The paper claims that the interlacing property of equal triple sums on convex curves is the "key input" that allows the problem to be reduced to a recurrence, bypassing the need for the heavy number-theoretic inputs previously thought necessary for lattice point problems. The authors emphasize that their argument is "purely combinatorial" in establishing the first estimate and uses "harmonic analysis" (specifically $L^2$ orthogonality and interpolation) for the second, without relying on the specific arithmetic structure of the set $X$.

The work also highlights a unique aspect of its development: the authors explicitly state that communication with AI (GPT-5.6) was central to the work. The AI helped identify the correspondence between triple sums and roots of cubic polynomials (leading to the interlacing observation) and assisted in verifying the complex iteration formulas required to close the argument.

The paper concludes that the result is "near-optimal" and resolves the question of the three-fold additive energy bound for general strictly convex curves, providing a unified framework that applies to both well-spaced and non-well-spaced configurations.