Erdős Problem #1092 — Let fr(n)f_r(n) be maximal such that, if a graph GG has the property that every subgraph HH on mm vertices is the union of a graph with chromatic number rr and a graph with ≤fr(m)\leq f_r(m) edges, then…

About 50 years old · traced to

Let fr(n)f_r(n) be maximal such that, if a graph GG has the property that every subgraph HH on mm vertices is the union of a graph with chromatic number rr and a graph with ≤fr(m)\leq f_r(m) edges, then GG has chromatic number ≤r+1\leq r+1. Is it true that f2(n)≫nf_2(n) \gg n? More generally, is fr(n)≫rnf_r(n)\gg_r n?

References

Progress summary

Refreshed
Claimed solved

Rödl’s published 1982 construction disproves the proposed linear bound in every fixed number of colours, so the problem is resolved negatively.

The problem, posed by Erdős, Hajnal, and Szemerédi, asks whether a linear edge-error allowance forces global (r+1)(r+1)-colourability. Rödl’s construction gives a negative answer: for every fixed r≥2r\ge 2, one has fr(n)=o(n)f_r(n)=o(n), in the relevant strong sense.

Rödl’s 1982 disproof

Rödl’s graph K∗(n,k)K^*(n,k) has chromatic number k+2k+2, while every subgraph becomes bipartite after deleting at most ϵ∣V(H)∣\epsilon |V(H)| edges. This contradicts any bound f2(n)≥cnf_2(n)\ge cn and, by joining with a clique of size r−2r-2, yields fr(n)=o(n)f_r(n)=o(n) for every fixed r≥2r\ge 2. The source is Rödl, “Nearly bipartite graphs with large chromatic number,” Combinatorica 2.4 (1982), Theorem 1.51.5.

Current status (as of May 2026): Rödl’s published construction settles the question negatively, with fr(n)=o(n)f_r(n)=o(n) for every fixed r≥2r\ge 2; a formalized version remains incomplete.

Sources

Solutions 0

No solutions have been posted yet.