🤖 AI Summary
This paper studies the minimum number of edges $s(n)$ in an $n$-vertex graph that contains every $n$-vertex tree (i.e., every graph of treewidth 1) as a subgraph, and its generalization $s_k(n)$—the minimum number of edges required to contain all $n$-vertex graphs of treewidth at most $k$. By correcting a critical error in an early proof by Chung and Graham, and integrating extremal graph theory, probabilistic methods, and explicit constructive techniques, the authors establish the first tight upper bound $s(n) le O(n log n log log n)$, extended to $s_k(n) le O(k n log n log log n)$. They further prove a matching lower bound $s_k(n) = Omega(k n log n)$. Together, these bounds determine the asymptotically optimal growth rate for universal graphs accommodating trees and, more generally, low-treewidth graphs—significantly improving prior upper estimates and resolving the problem up to iterated logarithmic factors.
📝 Abstract
Let $s(n)$ be the minimum number of edges in a graph that contains every $n$-vertex tree as a subgraph. Chung and Graham [J. London Math. Soc. 1983] claim to prove that $s(n)leqslant O(nlog n)$. We point out a mistake in their proof. The previously best known upper bound is $s(n)leqslant O(n(log n)(loglog n)^{2})$ by Chung, Graham and Pippenger [Proc. Hungarian Coll. on Combinatorics 1976], the proof of which is missing many crucial details. We give a fully self-contained proof of the new and improved upper bound $s(n)leqslant O(n(log n)(loglog n))$. The best known lower bound is $s(n)geqslant Ω(nlog n)$.
We generalise these results for graphs of treewidth $k$. For an integer $kgeqslant 1$, let $s_k(n)$ be the minimum number of edges in a graph that contains every $n$-vertex graph with treewidth $k$ as a subgraph. So $s(n)=s_1(n)$. We show that $Ω(k nlog n) leqslant s_k(n) leqslant O(kn(log n)(loglog n))$.