Most, And Least, Compact Spanning Trees of a Graph

Abstract: We introduce the concept of Most, and Least, Compact Spanning Trees - denoted respectively by $T<sup>*(G)$ and $T<sup>#(G)$ - of a simple, connected, undirected and unweighted graph $G(V, E, W)$. For a spanning tree $T(G) \in \mathcal{T}(G)$ to be considered $T<sup>*(G)$, where $\mathcal{T}(G)$ represents the set of all the spanning trees of the graph $G$, it must have the least average inter-vertex pair (shortest path) distances from amongst the members of the set $\mathcal{T}(G)$. Similarly, for it to be considered $T<sup>#(G)$, it must have the highest average inter-vertex pair (shortest path) distances. In this work, we present an iteratively greedy rank-and-regress method that produces at least one $T<sup>*(G)$ or $T<sup>#(G)$ by eliminating one extremal edge per iteration. The rank function for performing the elimination is based on the elements of the matrix of relative forest accessibilities of a graph and the related forest distance. We provide empirical evidence in support of our methodology using some standard graph families: complete graphs, the Erd\H{o}s-Renyi random graphs and the Barab\'{a}si-Albert scale-free graphs; and discuss computational complexity of the underlying methods which incur polynomial time costs.