Improving dynamic programming for travelling salesman with precedence constraints: parallel Morin–Marsten bounding. (2nd November 2021)

• Record Type: Journal Article
• Authors: Salii, Yaroslav. V.; Sheka, Andrey S.
• Is Part Of: Optimization methods and software. Volume 36:Number 6(2021)
• Page Start: 1128
• Page End: 1154
• Publication Date: 2021-11-02
• Subjects: Precedence constraints -- travelling salesman -- sequential ordering problem -- dynamic programming -- brand-and-bound -- parallelism; 90-08 -- 06A06 -- 90C39; Mathematical optimization -- Periodicals; Algorithms -- Periodicals; 519.7
• DOI: 10.1080/10556788.2020.1817447
• Languages: English
• ISSNs: 1055-6788

Abstract

Abstract : The precedence constrained traveling salesman (TSP-PC), also known as sequential ordering problem (SOP), consists of finding an optimal tour that satisfies the namesake constraints. Mixed integer-linear programming works well with the 'lightly constrained' TSP-PCs, close to asymmetric TSP, as well as the with the 'heavily constrained' (Gouveia, Ruthmair, 2015). Dynamic programming (DP) works well with the heavily constrained (Salii, 2019). However, judging by the open TSPLIB SOP instances, the worst for any method are the 'medium'. We implement a parallel Morin–Marsten branch-and-bound scheme for DP (DPBB). We show how the lower bound heuristic parameterizes DPBB's worst-case complexity and DPBB 'inherits' the abstract travel cost aggregation feature of the DP, permitting its direct use with both the conventional and bottleneck TSP-PC. The scheme was tested on TSPLIB instances, with best known upper bounds (TSP-PC), or those found by restricted DP (Bottleneck TSP-PC), and lower bounds from a greedy-type heuristic. Our OPEN MP-based parallel implementation achieved 20-fold speedup for larger instances. We close the long-standing kro124p.4.sop (conventional TSP-PC) and both kro124p.4.sop and ry48p.2.sop (Bottleneck TSP-PC).