Packing and finding paths in sparse random graphs

TL;DR

This paper proves that the logarithmic factor in the query complexity for finding long paths in sparse Erdős-Rényi graphs can be removed, confirming a conjecture and advancing understanding of path packings in random trees.

Contribution

It confirms the conjecture that the log factor can be eliminated in path-finding query complexity in sparse random graphs, and introduces new results on path packings in random trees.

Findings

The log factor in query complexity is removable.

Maximum vertices covered by disjoint paths in random trees is Θ(t/ell).

New bounds on path packings in random labelled trees.

Abstract

Let $G\sim G(n,p)$ be a (hidden) Erd\H{o}s-R\'enyi random graph with $p=(1+ \varepsilon)/n$ for some fixed constant $ \varepsilon >0$. Ferber, Krivelevich, Sudakov, and Vieira showed that to reveal a path of length $\ell=\Omega\left(\frac{\log(1/ \varepsilon)}{ \varepsilon}\right)$ in $G$ with high probability, one must query the adjacency of $\Omega\left(\frac{\ell}{p \varepsilon\log(1/ \varepsilon)}\right)$ pairs of vertices in $G$, where each query may depend on the outcome of all previous queries. Their result is tight up to the factor of $\log(1/ \varepsilon)$ in both $\ell$ and the number of queries, and they conjectured that this factor could be removed. We confirm their conjecture. The main ingredient in our proof is a result about path-packings in random labelled trees of independent interest. Using this, we also give a partial answer to a related question of Ferber, Krivelevich, Sudakov, and Vieira. Namely, we show that when $\ell=o\left((t/\log t)^{1/3}\right)$, the maximum number of vertices covered by edge-disjoint paths of length at least $\ell$ in a random labelled tree of size $t$ is $\Theta(t/\ell)$ with high probability.