Erdős Problem 533 #
References:
- erdosproblems.com/533
- [EHSSS94] P. Erdős, A. Hajnal, M. Simonovits, V. T. Sós and E. Szemerédi, Turán-Ramsey theorems and $K_p$-independence numbers, Combin. Probab. Comput. 3 (1994), 297–325.
- [ErRo62] P. Erdős and C. A. Rogers, The construction of certain graphs, Canad. J. Math. 14 (1962), 702–707.
- [BaLe13] J. Balogh and J. Lenz, On the Ramsey–Turán numbers of graphs and hypergraphs, Israel J. Math. 194 (2013), 45–68.
- [LRSS21] H. Liu, C. Reiher, M. Sharifzadeh and K. Staden, Geometric constructions for Ramsey–Turán theory, arXiv:2103.10423 (2021).
Let $\delta > 0$. If $n$ is sufficiently large and $G$ is a graph on $n$ vertices with no $K_5$ and at least $\delta n^2$ edges, must $G$ contain a set of $\gg_\delta n$ vertices spanning no triangle?
Equivalently, writing $\mathrm{RT}_3(n, K_5, m)$ for the maximum number of edges of a $K_5$-free graph on $n$ vertices in which every triangle-free vertex set has fewer than $m$ vertices (the triangle Ramsey–Turán number), is $$\delta_3(5) = \lim_{\epsilon \to 0} \lim_{n \to \infty} \frac{\mathrm{RT}_3(n, K_5, \epsilon n)}{n^2} = 0?$$
This is a problem of Erdős, Hajnal, Simonovits, Sós, and Szemerédi [EHSSS94], who proved $\delta_3(5) \leq 1/12$ and the analogous $\delta_3(4) = 0$, and observed $\delta_3(7) \geq 1/4$ via a construction of Erdős and Rogers [ErRo62].
The answer is no: Balogh and Lenz [BaLe13] disproved it by showing $\delta_3(5) > 0$, and
the exact value $\delta_3(5) = 1/12$ was determined by the matching lower-bound construction of
Liu, Reiher, Sharifzadeh, and Staden [LRSS21] (see erdos_533.variants.lrss_lower).
The upper bound $\delta_3(5) \leq 1/12$ of Erdős, Hajnal, Simonovits, Sós, and Szemerédi [EHSSS94]: for every $\epsilon > 0$ there is a $\delta > 0$ such that for all sufficiently large $n$, every $K_5$-free graph $G$ on $n$ vertices in which every triangle-free vertex set has at most $\delta n$ vertices has at most $(1/12 + \epsilon)n^2$ edges.
The matching lower bound $\delta_3(5) \geq 1/12$, from the construction of Liu, Reiher,
Sharifzadeh, and Staden [LRSS21] (improving the earlier $\delta_3(5) > 0$ of Balogh and Lenz
[BaLe13]): for every $\epsilon, \delta > 0$ and all sufficiently large $n$ there is a
$K_5$-free graph $G$ on $n$ vertices in which every triangle-free vertex set has at most
$\delta n$ vertices, yet which has at least $(1/12 - \epsilon)n^2$ edges. In particular this
refutes erdos_533.
The contrasting positive result $\delta_3(4) = 0$ of Erdős, Hajnal, Simonovits, Sós, and
Szemerédi [EHSSS94]: the $K_4$ analogue of erdos_533 is true. For every $\delta > 0$
there is a $c > 0$ such that for all sufficiently large $n$, every $K_4$-free graph $G$ on
$n$ vertices with at least $\delta n^2$ edges contains a triangle-free vertex set of size at
least $c n$.
The observation $\delta_3(7) \geq 1/4$ of Erdős, Hajnal, Simonovits, Sós, and Szemerédi [EHSSS94], via a construction of Erdős and Rogers [ErRo62]: for every $\epsilon, \delta > 0$ and all sufficiently large $n$ there is a $K_7$-free graph $G$ on $n$ vertices in which every triangle-free vertex set has at most $\delta n$ vertices, yet which has at least $(1/4 - \epsilon)n^2$ edges.
Sanity check for the triangle-free-set condition: in the empty graph every vertex set spans no triangle, since the empty graph has no $3$-clique.