
ABSTRACT The min‐cost group Steiner tree problem ( GSTP ) is a well‐known problem in graph theory, not only being NP‐complete but also not admitting a constant‐factor approximation even under strong restrictions. Given a graph and groups of vertices in , the problem GSTP asks for a minimum‐cost subtree of that contains at least one vertex from each group. In this paper, we address two variants of the GSTP : First, we consider groups of edges rather than vertices. Perhaps surprisingly, this problem is not equivalent to the vertex group variant. For instance, in the edge version it is even NP‐complete to decide whether there exists a feasible solution. Next, we consider the restriction that all of the (vertex or edge) groups are connected. We show that these variants can be solved in polynomial time, provided both the treewidth of and the vertex frequency (the number of groups a vertex is contained in) are bounded. However, the problems remain NP‐complete when only one of these parameters is bounded. Furthermore, we give various approximation algorithms for the connected edge GSTP on trees.
ABSTRACT This study introduces the two‐echelon multicommodity location routing problem with stochastic and correlated demands. We propose a two‐stage stochastic programming formulation, with second‐echelon facilities location and commodity allocation decisions defining the first stage, while recourse decisions, which are made in the second stage, establish how the observed demands are distributed. The overall objective is to minimize the cost of the first‐stage design decisions plus the total expected routing cost incurred in the second stage. We propose a Progressive Hedging matheuristic that heuristically solves the formulation, incorporating a series of algorithmic enhancements to accelerate the exploration of the solution space. These enhancements include: (1) population structures to obtain alternative and diverse solutions for the scenario subproblems that need to be solved throughout the search process; (2) alternative strategies to define the reference solutions that are used to guide and accelerate the overall search; and (3) a reset procedure that reduces the risk of the method becoming trapped in local optima. We assess the efficiency and effectiveness of all proposed strategies through extensive computational experiments, evaluating their capability to generate high‐quality solutions across various problem characteristics and demand correlations.
ABSTRACT This paper introduces a novel modeling approach for arc routing problems (ARPs) which is easily applied to multiple ARP variants (e.g., capacitated ARPs, multiple depot ARPs, heterogeneous fleets). Our approach tracks the order in which arcs are serviced, rather than explicitly tracking deadheading, and couples the sequence of serviced arcs to the cumulative demand collected along the route. As each route is defined solely by the order of serviced arcs, our approach does not use vehicle indices to represent individual vehicles/routes, but rather the vehicle type, which makes our approach agnostic to vehicle fleet size and thereby suitable for solving ARPs with large fleets and capable of solving ARPs with heterogeneous fleets with unknown distributions of vehicle type without significantly increasing model size compared to ARPs on the same graph with a homogeneous fleet. We also propose a novel method of formulating “return‐to‐facility” (RtF) constraints in location arc routing problems (LARPs) by using a “tagging” system built into the flow variables to ensure each vehicle route returns to the depot it departed from. Our “tagging” method does not add depot indices to flow and binary variables, nor does it add constraints with separate binaries assigning arcs to depots, leading to a more compact model for LARPs. Using a proposed four‐step solution algorithm, we can solve several benchmark LARP instances to optimality, and are often able to obtain solutions of comparable or better objective value than those found by heuristics. We demonstrate the application of our modeling approach not only on the basic CARP and LARP, but also on the LARP with capacitated depots, CARP with time windows, and multi‐depot CARP.
ABSTRACT Sum coloring of a graph is a vertex coloring using positive integers (colors) that minimizes the total sum of the colors assigned to all vertices, where adjacent vertices must have different colors. The chromatic sum is the smallest total among all proper colorings of the graph . The need for considering such a graph coloring model follows from its application to solve problems of scheduling equal length tasks (e.g., university timetables). It stems from the fact that finding a minimum chromatic sum coloring corresponds to finding a schedule that minimizes the mean flow time and total completion time of tasks. Unfortunately, evaluating is NP‐hard. Therefore, we are forced to use polynomial‐time approximation algorithms. In this article, we analyze four of the most popular greedy graph coloring algorithms: , , and , standing for Largest First , Smallest Last , Saturation Largest First ( DSatur ) and Greedy Independent Sets , along with their improvements adapted to the needs of chromatic sum coloring. For any such algorithm, we give its description, the smallest , that is, hard‐to‐color (for which the algorithm always yields a sum greater than ) and , that is, slightly hard‐to‐color (where a total sum of colors greater than is possible under the worst‐case vertex ordering) graphs, as well as an estimation of its computational complexity and performance function. Finally, we show that a certain 8‐vertex graph is for the whole collection of algorithms under consideration.
ABSTRACT The Traveling Salesman problem with pickup, delivery, and ride‐time constraints (TSPPD‐RT) requires a depot‐based vehicle to visit multiple known requests. Each request specifies a pickup vertex, corresponding delivery vertex, and maximum travel time between them ( ride‐time ). The goal is to find a tour that (i) starts and finishes at the depot, (ii) visits all requests, (iii) satisfies ride‐time constraints for each request, and (iv) minimizes the vehicle's total travel cost. Additionally, undelivered requests must remain within the vehicle's capacity limit. The TSPPD‐RT is a fundamental optimization problem in logistics and routing. This paper introduces an exact method for solving an extended version of the TSPPD‐RT with time window constraints (TSPPD‐TWRT), incorporating both ride‐time constraints (RTCs) and hard ride‐time constraints (HRTCs). RTCs consider only travel and service times, while in the hard case, they also include waiting times, the former being a relaxation of the latter. This paper presents two mathematical formulations: an arc‐based formulation and a set‐partitioning formulation based on the ‐tour relaxation proposed by [1]. Using these formulations, we develop MIP‐based and branch‐price‐and‐cut solution approaches. Computational experiments on datasets from the Dial‐A‐Ride Problem and the Pickup and Delivery Problem with Time Windows demonstrate the effectiveness of the proposed methods. Results indicate that the ‐tour formulation is particularly efficient, solving instances with up to 60 requests within reasonable computational times.