Table of Contents
Fetching ...

Exponents in the local properties problem for difference sets have a gap at 2

Sanjana Das

TL;DR

The paper resolves a gap phenomenon at the quadratic threshold for the local properties problem in difference sets: for even k, at the threshold value ell = k^2/4, the minimum number of differences in an n-point set with the local property remains subquadratic, specifically g(n, k, k^2/4) = O(n^{2 - 2^{-29}}). It introduces a framework of k-configurations to encode equal-difference relations, classifies them as c-good or c-bad, and uses a Behrend-type random construction to build large sets with controlled difference counts while avoiding c-bad configurations. A core backbone argument—comprising a baseline bound, a stability step, and a huge-star case—shows that any c-good k-configuration certifies at most a fixed (quadratic-like) number of pairs, yielding the subquadratic upper bound with a constant gap to 2. The method extends to odd k with an analogous, though slightly more delicate, bound. Overall, the work demonstrates a constant-sized jump in the exponent at the quadratic threshold and highlights distinct behavior from the graph-case thresholds, with potential implications for related extremal arithmetic problems.

Abstract

We study the local properties problem for difference sets: If we have a large set of real numbers and know that every small subset has many distinct differences, to what extent must the entire set have many distinct differences? More precisely, we define $g(n, k, \ell)$ to be the minimum number of differences in an $n$-element set with the `local property' that every $k$-element subset has at least $\ell$ differences; we study the asymptotic behavior of $g(n, k, \ell)$ as $k$ and $\ell$ are fixed and $n \to \infty$. The quadratic threshold is the smallest $\ell$ (as a function of $k$) for which $g(n, k, \ell) = Ω(n^2)$; its value is known when $k$ is even. In this paper, we show that for $k$ even, when $\ell$ is one below the quadratic threshold, we have $g(n, k, \ell) = O(n^c)$ for an absolute constant $c < 2$ -- i.e., at the quadratic threshold, the `exponent of $n$ in $g(n, k, \ell)$' jumps by a constant independent of $k$.

Exponents in the local properties problem for difference sets have a gap at 2

TL;DR

The paper resolves a gap phenomenon at the quadratic threshold for the local properties problem in difference sets: for even k, at the threshold value ell = k^2/4, the minimum number of differences in an n-point set with the local property remains subquadratic, specifically g(n, k, k^2/4) = O(n^{2 - 2^{-29}}). It introduces a framework of k-configurations to encode equal-difference relations, classifies them as c-good or c-bad, and uses a Behrend-type random construction to build large sets with controlled difference counts while avoiding c-bad configurations. A core backbone argument—comprising a baseline bound, a stability step, and a huge-star case—shows that any c-good k-configuration certifies at most a fixed (quadratic-like) number of pairs, yielding the subquadratic upper bound with a constant gap to 2. The method extends to odd k with an analogous, though slightly more delicate, bound. Overall, the work demonstrates a constant-sized jump in the exponent at the quadratic threshold and highlights distinct behavior from the graph-case thresholds, with potential implications for related extremal arithmetic problems.

Abstract

We study the local properties problem for difference sets: If we have a large set of real numbers and know that every small subset has many distinct differences, to what extent must the entire set have many distinct differences? More precisely, we define to be the minimum number of differences in an -element set with the `local property' that every -element subset has at least differences; we study the asymptotic behavior of as and are fixed and . The quadratic threshold is the smallest (as a function of ) for which ; its value is known when is even. In this paper, we show that for even, when is one below the quadratic threshold, we have for an absolute constant -- i.e., at the quadratic threshold, the `exponent of in ' jumps by a constant independent of .
Paper Structure (26 sections, 23 theorems, 96 equations, 20 figures)

This paper contains 26 sections, 23 theorems, 96 equations, 20 figures.

Key Result

Theorem 1.1

There is an absolute constant $c < 2$ such that for all even $k$, we have

Figures (20)

  • Figure 1: A plot of the bounds we know on $g(n, k, \ell)$ in the regime where $\ell$ is quadratic in $k$, where the $x$-axis depicts the coefficient of $k^2$ in $\ell$ and the $y$-axis depicts the exponent of $n$ in the bound --- a point $(a, c)$ means that for $\ell \approx ak^2$ we have a bound of roughly $n^c$. The purple line represents lower bounds --- for $2/9 < a < 1/4$ the best lower bound comes from \ref{['eqn:fps-lower']} with $r = 3$, and for other values of $a$ the best lower bound comes from \ref{['eqn:das-interm-lower']}. The blue line represents upper bounds, which come from \ref{['eqn:das-interm-upper']}.
  • Figure 2: A version of Figure \ref{['fig:previous-bounds']} incorporating Theorem \ref{['thm:main']} (the value of $c$ is not to scale). There is a 'gap' on the $y$-axis between $c$ and $2$ --- the exponent of $n$ in $g(n, k, \ell)$ can never lie in this range when $k$ is even.
  • Figure 3: A star of size $8$, where we depict a difference equality $x_{i_1} - x_{i_2} = x_{i_3} - x_{i_4}$ by placing $x_{i_1}$, $x_{i_2}$, $x_{i_4}$, $x_{i_3}$ in a parallelogram. A star of size $2p$ then consists of $p$ pairs with the same midpoint, which is the reason for the name. The shaded parallelogram shows that this star certifies $(8, 3)$ and $(8, 4)$.
  • Figure 6: An example of a minimal implication (we have ${*} = -{*_1} - {*_2} + {*_3} + {*_4}$) and the corresponding graph used to prove \ref{['item:signs-pm1']}, where every edge is labelled with the ordinary variable it corresponds to.
  • Figure 7: Two overlapping $2$-full sets of difference equalities as in Lemma \ref{['lem:2-full-intersection']}, with $\mathcal{T}_1$ in purple and $\mathcal{T}_2$ in blue. Here $\mathcal{T}_1$ has $4$ equations on $9$ variables and $\mathcal{T}_2$ has $5$ equations on $11$ variables; their intersection has $3$ equations on $7$ variables, and their union has $6$ equations on $13$ variables.
  • ...and 15 more figures

Theorems & Definitions (60)

  • Theorem 1.1
  • Proposition 1.2
  • Example 2.1
  • Example 2.2
  • Lemma 2.3
  • Lemma 2.4
  • Lemma 3.1
  • proof : Proof of Lemma \ref{['lem:random-constr']}
  • Lemma 4.1
  • Lemma 4.2
  • ...and 50 more