The average solution of a TSP instance in a graph
Stijn Cambie
TL;DR
The paper introduces the average $k$-TSP distance $\mu_{tsp,k}(G)$ and the $k$-TSP Wiener index $W_{tsp,k}(G)$, linking them to the $k$-Steiner distance and Steiner Wiener index and extending the framework to weighted graphs and digraphs. A central result is the sharp bound $W_{tsp,k}(G) \le 2W_k(G)$ with precise equality criteria, including special cases for $k=2$ and $k=3$ and a tree condition for $k\ge 4$, along with a complementary bound $W_{tsp,k}(G) \le k\binom{n}{k}\mu(G)$ with its own equality cases. The authors also establish the exact identity $W_{tsp,3}(G)=(n-2)W_3(G)$ for all $G$ and derive tight extremal bounds for $\mu_{tsp,k}(G)$, characterizing graphs achieving equality. They extend the discussion to digraphs, relate these results to known Wiener-type indices, and comment on conjectures such as a DeLaViña–Waller analogue, noting its failure for larger $k$. Overall, the work provides a foundational analytic framework for average TSP distances and their connections to Steiner-type indices, with implications for graph design and analysis of TSP instances.
Abstract
We define the average $k$-TSP distance $μ_{tsp,k}$ of a graph $G$ as the average length of a shortest walk visiting $k$ vertices, i.e. the expected length of the solution for a random TSP instance with $k$ uniformly random chosen vertices. We prove relations with the average $k$-Steiner distance and characterize the cases where equality occurs. We also give sharp bounds for $μ_{tsp,k}(G)$ given the order of the graph.
