Ramsey numbers of semi-algebraic and semi-linear hypergraphs

TL;DR

This paper investigates the Ramsey numbers of semi-algebraic hypergraphs, providing new bounds and refuting a conjecture about their growth, especially in the case of linear polynomials.

Contribution

It establishes new lower bounds for semi-algebraic Ramsey numbers and disproves a conjecture by showing super-polynomial growth in certain cases.

Findings

Refutes the conjecture that $R_{3}^{ extbf{t}}(s,n)$ is polynomial in $n$ for fixed $s$.

Provides upper bounds for linear semi-algebraic hypergraph Ramsey numbers.

Establishes lower bounds demonstrating super-polynomial growth in specific cases.

Abstract

An $r$-uniform hypergraph $H$ is semi-algebraic of complexity $\mathbf{t}=(d,D,m)$ if the vertices of $H$ correspond to points in $\mathbb{R}^{d}$, and the edges of $H$ are determined by the sign-pattern of $m$ degree-$D$ polynomials. Semi-algebraic hypergraphs of bounded complexity provide a general framework for studying geometrically defined hypergraphs. The much-studied semi-algebraic Ramsey number $R_{r}^{\mathbf{t}}(s,n)$ denotes the smallest $N$ such that every $r$-uniform semi-algebraic hypergraph of complexity $\mathbf{t}$ on $N$ vertices contains either a clique of size $s$, or an independent set of size $n$. Conlon, Fox, Pach, Sudakov, and Suk proved that $R_{r}^{\mathbf{t}}(n,n)<\mbox{tw}_{r-1}(n^{O(1)})$, where $\mbox{tw}_{k}(x)$ is a tower of 2's of height $k$ with an $x$ on the top. This bound is also the best possible if $\min\{d,D,m\}$ is sufficiently large with respect to $r$. They conjectured that in the asymmetric case, we have $R_{3}^{\mathbf{t}}(s,n)<n^{O(1)}$ for fixed $s$. We refute this conjecture by showing that $R_{3}^{\mathbf{t}}(4,n)>n^{(\log n)^{1/3-o(1)}}$ for some complexity $\mathbf{t}$. In addition, motivated by results of Bukh and Matou\v{s}ek and Basit, Chernikov, Starchenko, Tao and Tran, we study the complexity of the Ramsey problem when the defining polynomials are linear, that is, when $D=1$. In particular, we prove that $R_{r}^{d,1,m}(n,n)\leq 2^{O(n^{4r^2m^2})}$, while from below, we establish $R^{1,1,1}_{r}(n,n)\geq 2^{\Omega(n^{\lfloor r/2\rfloor-1})}$.