Edge density and minimum degree thresholds for $H$-free graphs with unbounded chromatic number

TL;DR

This paper explores the relationship between edge density and minimum degree in graphs that avoid a fixed subgraph $H$, especially near the chromatic threshold, providing sharp bounds and extremal constructions.

Contribution

It establishes sharp upper bounds on edge density for $H$-free graphs with diverging chromatic number near the chromatic threshold, extending understanding of the degree-density trade-off.

Findings

Bounds are sharp up to lower-order terms.

Global edge bounds can limit chromatic number growth.

Extremal constructions match the bounds up to $o(n^2)$.

Abstract

The chromatic threshold $\delta_\chi(H)$ of a graph $H$ is the infimum of $d>0$ such that the chromatic number of every $n$-vertex $H$-free graph with minimum degree at least $dn$ is bounded in terms of $H$ and $d$. A breakthrough result of Allen, B\"ottcher, Griffiths, Kohayakawa, and Morris determined $\delta_\chi(H)$ for every graph $H$; in particular, if $\chi(H)=r\ge 3$, then $\delta_\chi(H) \in\{\frac{r-3}{r-2},~\frac{2 r-5}{2 r-3},~\frac{r-2}{r-1}\}$. In this paper we investigate the trade-off between minimum degree and edge density in the critical window around the chromatic threshold. For a fixed graph $H$ with $\chi(H)=r$, allowing a constant deficit below $\delta_\chi(H)$, we prove sharp (up to lower-order terms) upper bounds on the edge density of $n$-vertex $H$-free graphs whose chromatic number diverges. Equivalently, within this degree regime we show that a suitable global bound on the number of edges forces the chromatic number to remain bounded. Our results thus quantify how global edge density can compensate for a deficit in the local minimum-degree condition near $\delta_\chi(H)$; more specifically, we obtain explicit bounds in two of the three possible cases arising in the trichotomy of $\delta_\chi(H)$. Our extremal constructions -- based on Erd\H{o}s graphs and blowups of Borsuk--Hajnal graphs -- show that these bounds are best possible up to $o(n^2)$ terms.