Recently several practical variants of the classical traveling salesman problem &re proposed. These variants include the traveling purchaser problem, the prize collecting traveling salesman problem, the orienteering problem, the median shortest path problem, and the covering 8ale~man problem. The most important common chaxacteristic of these problems is that the salesman travels the subset of customers, i.e., the tottr doos not necessarily visit all nodes as in the traveling salesman problem. We call this class of problems the subtour problem. In this paper, we derive several valid inequalities and show some of them axe facets of the polytope of the subtour problem. Further we preBent several integer or mixed integer programming formulations of the subtour problem and compare these formulations in terms of bounds obtained by linear programming relaxations.Key word~, subtour problem, valid inequality, integer programming formulation, traveling salesman problem
IntroductionThe traveling sa[esman problem (TSP) is the problem of finding a miniraum cost (distance) tour which visits a[1 cities (nodes) in a graph, and has wide apphcations in many practica[ areas such as vehicle routing and schednling, machine scheduling and sequencing, VLSI network design, and clustering data array, etc. Many researchers have investigated the TSP, and have derived efficient solution methods; some are approximate a[gorithm.% the others are opt]miTation a[gorithms. Resent progress of such solution methods includes 1. heuristics with the worst-case guarantees, 2. heuristics with the probabilistic guarantees, and 3. exact schemes based on the polyhedra[ computation.The results of these progresses &re snmmarized in a recently published book [29]. The most impressive exact a[gorithm is a combination procedure of the cutting plane and branch-and-bound methods in which polyhedra[ theory plays a centra[ role to compute tight lower bounds. The TSP with severa[ thousands of nodes can be solved based on this approach [37]. In contrast to it, the variants of the TSP cannot be solved when the problem size is relatively large; this indicates that polyhedra[ ana[ysis is necessary for the exact methods for the variants of the TSP.In this paper, we ana[yze the polyhedra[ aspects of a variant of the TSP termed the subtour problem.In the subtour problem, the salesman travels the subset of cities, i.e., the tour does not necessarily visit a[1 nodes as in the TSP. This variant includes the traveling purchaser problem, the prize collecting travehng sa[esman problem, the orienteering