Approximating Metrics by Tree Metrics of Small Distance-Weighted Average Stretch

TL;DR

This paper investigates how well tree metrics can approximate the sum of pairwise distances in arbitrary and Euclidean metrics, showing that average performance can be close to optimal with constructive methods.

Contribution

It introduces a direct approach to approximate pairwise distance sums with tree metrics, improving guarantees over previous work, especially for Euclidean point sets.

Findings

Tree metrics can preserve the sum of pairwise distances within a small constant factor.

For Euclidean metrics, spanning trees can preserve pairwise sums up to a dimension-dependent constant.

Constructive proofs provide explicit methods for such approximations.

Abstract

We study the problem of how well a tree metric is able to preserve the sum of pairwise distances of an arbitrary metric. This problem is closely related to low-stretch metric embeddings and is interesting by its own flavor from the line of research proposed in the literature. As the structure of a tree imposes great constraints on the pairwise distances, any embedding of a metric into a tree metric is known to have maximum pairwise stretch of $\Omega(\log n)$. We show, however, from the perspective of average performance, there exist tree metrics which preserve the sum of pairwise distances of the given metric up to a small constant factor, for which we also show to be no worse than twice what we can possibly expect. The approach we use to tackle this problem is more direct compared to a previous result of [4], and also leads to a provably better guarantee. Second, when the given metric is extracted from a Euclidean point set of finite dimension $d$, we show that there exist spanning trees of the given point set such that the sum of pairwise distances is preserved up to a constant which depends only on $d$. Both of our proofs are constructive. The main ingredient in our result is a special point-set decomposition which relates two seemingly-unrelated quantities.