🤖 AI Summary
This paper studies the tree cover problem for point sets in Euclidean space: given a set (P subset mathbb{R}^d), can we cover all pairwise distances using a small number of trees such that, for any two points, their distance in at least one tree is at most a constant factor (i.e., stretch (O(1))) of their Euclidean distance? In two dimensions, we construct the first constant-stretch tree cover using only two trees—improving upon prior upper bounds. Our construction integrates geometric divide-and-conquer, planar graph embedding, and combinatorial graph design, critically leveraging planarity. In (d) dimensions, for the stronger variant requiring *every* point pair to achieve low stretch in *at least one* tree, we prove a lower bound of (lceil(d+1)/2
ceil) trees—the first nontrivial dimension-dependent lower bound. Together, these results establish tight structural and dimensional limits on tree covers, unifying the understanding of their inherent trade-offs.
📝 Abstract
For a given metric space $(P,φ)$, a tree cover of stretch $t$ is a collection of trees on $P$ such that edges $(x,y)$ of trees receive length $φ(x,y)$, and such that for any pair of points $u,vin P$ there is a tree $T$ in the collection such that the induced graph distance in $T$ between $u$ and $v$ is at most $tφ(u,v).$ In this paper, we show that, for any set of points $P$ on the Euclidean plane, there is a tree cover consisting of two trees and with stretch $O(1).$ Although the problem in higher dimensions remains elusive, we manage to prove that for a slightly stronger variant of a tree cover problem we must have at least $(d+1)/2$ trees in any constant stretch tree cover in $mathbb R^d$.