Unnormalized single-set sum-difference exponent

Description of constant

For a finite non-empty set $A$ of integers, $C_{3e}$ is the least constant such that

\[|A - A| \le |A + A|^{C_{3e}}\]

for every such $A$; equivalently, $C_{3e} = \sup_A \log\lvert A-A\rvert / \log\lvert A+A\rvert$. This is Problem 6.43 of [GGSWT2025].

It measures how much larger the difference set can be than the sumset, without normalizing by $\lvert A\rvert$ — which is what distinguishes it from $C_{3d}$, the least $\theta$ with $\lvert A+A\rvert/\lvert A\rvert \le (\lvert A-A\rvert/\lvert A\rvert)^{\theta}$. The two normalizations have different classical bounds and different records, and conflating them has caused published confusion; see the caution below.

Known upper bounds

Bound Reference Comments
$\frac{3}{2}$ Elementary Ruzsa’s triangle inequality with $X = Y = Z = A$ gives $\lvert A-A\rvert \le \lvert A+A\rvert^2/\lvert A\rvert$, and $\lvert A+A\rvert \le \lvert A\rvert^2$ gives $\lvert A\rvert \ge \lvert A+A\rvert^{1/2}$; combining yields $\lvert A-A\rvert \le \lvert A+A\rvert^{3/2}$.
$\frac{4}{3} = 1.3333\dots$ [FrPi73] The classical Freiman–Pigaev inequality $\lvert A+A\rvert^{3/4} \le \lvert A-A\rvert \le \lvert A+A\rvert^{4/3}$. Quoted in this form by [GGSWT2025, Problem 6.43] and by [PeWe13, §4].

Known lower bounds

Bound Reference Comments
$1$ Trivial Any arithmetic progression has $\lvert A+A\rvert = \lvert A-A\rvert = 2\lvert A\rvert - 1$.
$\frac{\log(1+\sqrt{2})}{\log 2} = 1.2715\dots$ [HRY1999] The high-dimensional simplex $A = \{(x_1,\dots,x_N) \in \mathbb{Z}_{+}^{N} : x_1 + \dots + x_N \le N/2\}$, projected to $\mathbb{Z}$. See the note below for the asymptotics that produce this value.
$\approx 1.21$ [GGSWT2025] AlphaEvolve, given no human hints, did not recover the simplex construction within a few hours and reached only about $1.21$ — one of the cases [GGSWT2025] records as not matching the literature. Inferior to [HRY1999].

References

Contribution notes

Prepared with assistance from Claude Opus 5, working from §6.25 of [GGSWT2025] and the AlphaEvolve repository of problems, cross-checked against [GHR2007] and [PeWe13]. The elementary $3/2$ upper bound and the limit $\log(1+\sqrt{2})/\log 2$ were derived and checked here from the cardinality formulas recorded in [GHR2007]; the $4/3$ bound of [FrPi73] was not checked against the primary source, which is a 1973 Russian volume.