Improved approximation ratios for two Euclidean maximum spanning tree problems

Abstract: We study the following two maximization problems related to spanning trees in the Euclidean plane. It is not known whether or not these problems are NP-hard. We present approximation algorithms with better approximation ratios for both problems. The improved ratios are obtained mainly by employing the Steiner ratio, which has not been used in this context earlier. (i) Longest noncrossing spanning tree: Given a set of points in the plane, the goal is to find a maximum-length noncrossing spanning tree. Alon, Rajagopalan, and Suri (SoCG 1993) studied this problem for the first time and gave a $0.5$-approximation algorithm. Over the years, the approximation ratio has been successively improved to $0.502$, $0.503$, and to $0.512$ which is the current best ratio, due to Cabello et al.. We revisit this problem and improve the ratio further to $0.519$. The improvement is achieved by a collection of ideas, some from previous works and some new ideas (including the use of the Steiner ratio), along with a more refined analysis. (ii) Longest spanning tree with neighborhoods: Given a collection of regions (neighborhoods) in the plane, the goal is to select a point in each neighborhood so that the longest spanning tree on selected points has maximum length. We present an algorithm with approximation ratio $0.524$ for this problem. The previous best ratio, due to Chen and Dumitrescu, is $0.511$ which is in turn the first improvement beyond the trivial ratio $0.5$. Our algorithm is fairly simple, its analysis is relatively short, and it takes linear time after computing a diametral pair of points. The simplicity comes from the fact that our solution belongs to a set containing three stars and one double-star. The shortness and the improvement come from the use of the Steiner ratio.