Lower Bounds on the Integraliy Ratio of the Subtour LP for the Traveling Salesman Problem

TL;DR

This paper explores high integrality ratio instances of the TSP, develops methods to generate such instances with ratios approaching 4/3, and analyzes their structure and computational hardness.

Contribution

It introduces a procedure to generate Euclidean TSP instances with high integrality ratios, including maximization for Rectilinear TSP, and studies their structural properties and computational difficulty.

Findings

Instances with ratios approaching 4/3 are constructed for various TSP variants.

The paper identifies instances that maximize the integrality ratio under certain assumptions.

A family of hard-to-solve TSP instances is described, requiring extensive computational resources.

Abstract

In this paper we investigate instances with high integrality ratio of the subtour LP. We develop a procedure to generate families of Euclidean TSP instances whose integrality ratios converge to $\frac{4}{3}$ and may have a different structure than the instances currently known from the literature. Moreover, we compute the instances maximizing the integrality ratio for Rectilinear TSP with up to 10 vertices. Based on these instances we give families of instances whose integrality ratio converge to $\frac{4}{3}$ for Rectilinear, Multidimensional Rectilinear and Euclidean TSP that have similar structures. We show that our instances for Multidimensional Rectilinear TSP and the known instances for Metric TSP maximize the integrality ratio under certain assumptions. We also investigate the concept of local optimality with respect to integrality ratio and develop several algorithms to find instances with high integrality ratio. Furthermore, we describe a family of instances that are hard to solve in practice. The currently fastest TSP solver Concorde needs more than two days to solve an instance from the family with 52 vertices.