Computers & Operations Research 161 (2024) 106426

Contents lists available at ScienceDirect

Computers and Operations Research

journal homepage: www.elsevier.com/locate/cor

A branch-and-cut algorithm for the one-commodity pickup and delivery

location routing problem
Bencomo Domínguez-Martín a , Hipólito Hernández-Pérez a , Jorge Riera-Ledesma b ,
Inmaculada Rodríguez-Martín a ,∗
a
Departamento de Matemáticas, Estadística e Investigación Operativa, Universidad de La Laguna, Spain
b Departamento de Ingeniería Informática y de Sistemas, Universidad de La Laguna, Spain

ARTICLE INFO ABSTRACT

Keywords: This paper introduces a new problem that combines characteristics of the Location and Routing Problem and
Routing the One-commodity Pickup and Delivery Traveling Salesman Problem. We are given a set of customers that
Location provide or demand given amounts of a product, and a set of potential facility locations that can be opened or
Pickup and delivery
not in order to give service to the customers. Each facility has an opening cost and is the depot of a vehicle with
Branch-and-cut
capacity 𝑄. The problem consists in deciding which facilities to open, assigning customers to open facilities,
and designing the routes that connect each facility with its customers. The objective is to minimize the sum
of the cost of the routes and the facilities. This NP-hard problem has not been previously studied. We propose
for it two mathematical formulations, compare them, and present a branch-and-cut algorithm able to solve
instances with up to 100 nodes.

1. Introduction node represents its demand, the routing costs are equal to the rounded
Euclidean distances, the opening cost for all the facilities is 100, and
This paper addresses a problem that combines aspects of two clas- all vehicles have capacity 20. We can see that only two facilities are
sical optimization problems: the Location and Routing Problem and open in the optimal solution (18 and 19). The load of the vehicle that
the One-commodity Pickup and Delivery Traveling Salesman Problem. traverses an arc is the figure next to the arc. Fig. 2 shows the optimal
In particular, the One-commodity Pickup and Delivery Location Routing solution for the same instance when the capacity of the vehicles is set to
Problem (1-PDLRP) studied in this paper is defined as follows. We are 10. In this case, we observe that three facilities have to be open in order
given a set of potential site locations where a facility, consisting of a to serve the customers (16, 17, and 18), and that there is an increment
depot or base for a capacitated vehicle, can be established, and a set of around 7.3% in the solution cost. Note also the effect of the capacity
of customers that provide or demand certain amounts of a commodity. restrictions on the routes.
Establishing a facility at a certain location has a fixed cost. In order The 1-PDLRP is an extension of the One-commodity Pickup and
to serve the customers, each of them must be assigned to an open Delivery Traveling Salesman Problem in which not one but several
facility and it must be in the route of the vehicle based at that facility. potential facility locations are considered. So it can have similar ap-
All customers have to be visited exactly once, and the commodity
plications, i.e., applications in transportation problems where only one
collected at a customer can be supplied to another. The objective of
type of commodity is transported, and the cargo collected at pickup
the problem is to decide the facilities to open and to design the routes
customers can be delivered to delivery customers. The typical example
of the vehicles, so that all customers demands are satisfied and the sum
is encountered in a bike sharing system (see Dell’Amico et al., 2014),
of the routing costs and facility costs is minimized.
and nowadays it could comprise also an electric scooters sharing sys-
To further illustrate the problem, we show in Fig. 1 the optimal
tem. In any case, consider, for example, a situation where a company
solution of the 1-PDLRP for a particular instance named n20 with 20
in this sector operates in a large city. The company has many small
nodes located in the plane, 15 customers (nodes 0 to 14) and 5 potential
facility locations (nodes 15 to 19). The number next to a customer parking places for its bikes/scooters distributed all over the city. The

∗ Correspondence to: Departamento de Matemáticas, Estadística e Investigación Operativa, Facultad de Ciencias, Universidad de La Laguna, P.O. box
456, 38200 La Laguna, Tenerife, Spain.
E-mail addresses: [email protected] (B. Domínguez-Martín), [email protected] (H. Hernández-Pérez), [email protected] (J. Riera-Ledesma),
[email protected] (I. Rodríguez-Martín).

https://doi.org/10.1016/j.cor.2023.106426
Received 29 December 2022; Received in revised form 6 September 2023; Accepted 9 September 2023
Available online 16 September 2023
0305-0548/© 2023 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC license (http://creativecommons.org/licenses/by-
nc/4.0/).
B. Domínguez-Martín et al. Computers and Operations Research 161 (2024) 106426

Fig. 1. Optimal solution of instance n20 with 𝑄 = 20 (cost: 3587).

users have to choose through a web site or app where they want to The One-commodity Pickup and Delivery Traveling Salesman Prob-
collect and drop off a bike or scooter, usually at least a day before using lem is a well known routing problem where each customer provides or
the service, so this information is known in advance by the company. requires given amounts of a single product, and these demands must
It may happen that, at the beginning of the day, some parking places be served by a capacitated vehicle located at a depot. The product col-
need to be provided with extra units in order to cover the demand of lected at pickup customers can be delivered to delivery customers. The
that day, while in some other parking places there are more units than objective is to design a minimum cost Hamiltonian route for the vehicle
requested. The units have then to be transported from some parking such that all the demands are satisfied. This problem was introduced
places to others using a vehicle. Let us suppose that the company has a in Hernández-Pérez and Salazar-González (2004a), where a branch-
main depot, maybe in an industrial area outside the city, and a fleet of and-cut algorithm capable of solving to optimality instances with up
small vehicles, such as vans, to transport the bikes/scooters. In a large to 60 customers was presented. This branch-and-cut algorithm was
city, it may be convenient to send the vans directly to certain locations enhanced in Hernández-Pérez and Salazar-González (2007) including
early in the morning, so that each van can then make a route from new valid inequalities. On the other hand, heuristic algorithms for the
that point to serve the parking places in the area. The cost of sending a 1-PDTSP have been proposed in Hernández-Pérez and Salazar-González
van directly from the central depot to a location can be considered the (2004b), Zhao et al. (2009) and Mladenović et al. (2012).
cost of establishing a facility in that point. The objective is to decide Several recent papers have studied a generalization of the 1-
to which points to send the vans, assign the parking places to the vans, PDTSP consisting in allowing to visit each location more than once.
and design the routes for the vans, so that all the parking places are This problem, known as the Split-Demand 1-PDTSP, has also been
provided with the necessary number of bikes/scooters to satisfy the applied to rebalancing bike sharing systems (see Dell’Amico et al.,
2014, Chemla et al., 2013, Erdoğan et al., 2015, Salazar-González
users that day, at minimum cost.
and Santos-Hernández, 2015, and Cruz et al., 2017), and has several
As the technology advances, other applications for the 1-PDLRP
variants depending, for example, on whether customers can be used to
will for sure appear, for example, in transportation systems with aerial
temporarily store product (see Hernández-Pérez and Salazar-González,
drones (see Kim et al., 2017), where drone bases and routes have to be
2022).
determined, or in the transportation of goods with automatic guided
Another variant of the 1-PDTSP is the single-vehicle Two-Echelon
vehicles or robots inside a factory (see Almeida et al., 2010).
One-Commodity Pickup and Delivery Problem, where not all customers
need to be visited by the vehicle but, instead, they can be allocated to
1.1. Related literature visited customers (see Hernández-Pérez et al., 2021). More variants of
pickup and delivery problems can be found in the extensive reviews
As commented before, the 1-PDLRP is an extension of the One- by Berbeglia et al. (2007), Parragh et al. (2008a), Parragh et al.
commodity Pickup and Delivery Traveling Salesman Problem (1-PDTSP) (2008b), Battarra et al. (2014), and Koç et al. (2020).
that combines the features of this problem and the Location and On the other hand, the Location and Routing Problem is another
Routing Problem (LRP). classical optimization problem where, given a set of potential facility

2
B. Domínguez-Martín et al. Computers and Operations Research 161 (2024) 106426

Fig. 2. Optimal solution of instance n20 with 𝑄 = 10 (cost: 4141).

locations with opening costs, a fleet of identical vehicles, and a set of and described a two-phase heuristic approach based on simulated
customers with known demands, the objective is to open a subset of annealing.
facilities, assign customers to them, and determine vehicle routes, in Neither of the two problems, the LRPSPD and the 1-PDLRP, is a
order to minimize the sum of the costs of the routes and the open
special case of the other. To see this, consider a simple case with a sin-
facilities. The version of this problem with a single vehicle at each
gle depot and a vehicle of capacity 5, and three customers: customer 1
facility was studied by Labbé et al. (2004), who presented an exact
branch-and-cut algorithm to solve the case with unit demands, and demands 5 units of product, customer 2 supplies 5 units, and customer
by Albareda-Sambola et al. (2005), who developed a heuristic method. 3 demands 5 units. If the product collected and supplied is the same,
Many other different variants of the LRP can be found in the literature, and it can be taken from pickup to delivery customers, that is, if the
and we refer the interested reader to the surveys by Nagy and Salhi problem is a 1-PDLRP, then a feasible solution is a route in which the
(2007) and Prodhon and Prins (2014). vehicle leaves the depot with 5 units of product, and visits customers
In the literature, we have found only one problem that combines si- 1, 2, and 3, in that order, before returning to the depot. The vehicle
multaneous pickup and delivery and LRP. This problem is the Location- is empty after visiting customer 1, full loaded after visiting customer
Routing Problem with Simultaneous Pickup and Delivery (LRPSPD) 2, and empty again after visiting customer 3. However, if we consider
introduced by Karaoglan et al. (2011). The LRPSPD is a LRP where
that the product to be collected is different from the product to be
each customer has a pickup and a delivery demand, simultaneously,
delivered (that is, if the problem to be solved is the LRPSPD), then
and both demands have to be satisfied by the vehicle visiting the
there is no feasible solution for this example. So, the 1-PDLRP cannot
customer. It can be considered that each customer demands a given
product, and supplies another product. The classic example given in be transformed into the LRPSPD.
the literature relates to the distribution of a drink, in which full bottles The particular case of the LRPSPD with a single vehicle is known in
have to be delivered and empty bottles have to be collected. Another the literature as the Traveling Salesman Problem with Pickup and Delivery
example arises in a rural area where a vehicle from a local store or TSPPD (see Mosheiov, 1994). Hernández-Pérez and Salazar-González
distributes liquid plant fertilizer to the farms, and collects vegetables (2004a) showed that the TSPPD can be transformed into the 1-PDTSP
from them, simultaneously. In any case, note that in the LRPSPD two (the special case of the 1-PDLRP with one vehicle). The key point in
different types of products have to be transported: one of the products
this transformation is that, in the TSPPD, the total amount of demand
is transported from the depots to the customers, and the other one is
that has to be delivered by the vehicle, and the total amount of demand
collected at the customers and transported to the depots. However, in
the 1-PDLRP there is a single type of commodity to be transported, and that has to be collected, are known. However, in the general LRPSPD,
the product collected at pickup customers can be supplied to delivery this information is not known since there are several vehicles, and
customers. Karaoglan et al. (2011) proposed an exact branch-and-cut the customers to be visited by each of them are ignored in advance.
algorithm to tackle the LRPSPD. In a second paper, Karaoglan et al. Therefore, it is not possible to use the same mechanism to transform
(2012) presented two other mathematical formulations for the problem, the LRPSPD into the 1-PDLRP.

3
B. Domínguez-Martín et al. Computers and Operations Research 161 (2024) 106426

1.2. Contribution We can model our problem using these variables as follows:
∑ ∑
To the best of our knowledge, the 1-PDLRP, as stated in this paper, (𝑃𝑓 𝑙𝑜𝑤 ) min 𝑑𝑗 𝑦𝑗 + 𝑐𝑒 𝑥 𝑒 (1)
𝑗∈𝐽 𝑒∈𝐸
has not been studied before, although related problems appear in
the literature. In this sense, the study of the 1-PDLRP contributes to s.t. 𝑥(𝛿(𝑖)) = 2 ∀𝑖 ∈ 𝐼 (2)
close the research gap in a field, the transportation and distribution 𝑥(𝛿(𝑗)) = 2𝑦𝑗 ∀𝑗 ∈ 𝐽 (3)
of goods, which is undoubtedly of great interest. We present in this ∑
paper two mathematical models for the problem, a classical one that 𝑧𝑖𝑗 = 1 ∀𝑖 ∈ 𝐼 (4)
𝑗∈𝐽
uses continuous flow variables as well as integer routing variables, and
an alternative one based only on the integer variables. We show that 𝑧𝑖𝑗 ≤ 𝑦𝑗 ∀𝑖 ∈ 𝐼, 𝑗 ∈ 𝐽 (5)
the second model is the best option to develop an exact branch-and-cut 𝑥(𝛿(𝑆)) ≥ 2 ∀𝑆 ⊆ 𝐼 (6)
algorithm. The 1-PDLRP can be seen as a simplified version of more
realistic problems, but the development of good models and efficient 𝑥𝑖𝑗 ≤ 𝑧𝑖𝑗 ∀𝑖 ∈ 𝐼, 𝑗 ∈ 𝐽 (7)
exact algorithms for it can serve to provide good approximations to 𝑥𝑖𝑖′ + 𝑧𝑖𝑗 + 𝑧𝑖′ 𝑗 ′ ≤ 2 ∀𝑖, 𝑖′ ∈ 𝐼, 𝑖 ≠ 𝑖′ ,
more complex problems, and can help to asses the effectiveness of
∀𝑗, 𝑗 ′ ∈ 𝐽 , 𝑗 ≠ 𝑗 ′ (8)
heuristic algorithms capable of dealing with larger instances.
𝑥𝑒 ∈ {0, 1} ∀𝑒 ∈ 𝐸 (9)
1.3. Organization 𝑦𝑗 ∈ {0, 1} ∀𝑗 ∈ 𝐽 (10)
𝑧𝑖𝑗 ∈ {0, 1} ∀𝑖 ∈ 𝐼, 𝑗 ∈ 𝐽 (11)
The remainder of this paper is organized as follows. First, in Sec- ∑ ∑
tion 2 we formally describe the problem and present two mathematical 𝑓[𝑖,𝑗] − 𝑓[𝑗,𝑖] = 𝑞𝑖 ∀𝑖 ∈ 𝐼 (12)
formulations, one with continuous flow variables and integer variables, 𝑗∈𝑉 ⧵{𝑖} 𝑗∈𝑉 ⧵{𝑖}

and another one with only integer variables, as well as valid inequal- 𝑄
0 ≤ 𝑓[𝑖,𝑗] ≤ 𝑥 ∀𝑖, 𝑗 ∈ 𝑉 , 𝑖 < 𝑗 ∈ 𝑉 (13)
ities for them. In Section 3 we describe a branch-and cut algorithm 2 (𝑖,𝑗)
𝑄
to solve the problem. We report the results of our computational 0 ≤ 𝑓[𝑖,𝑗] ≤ 𝑥(𝑗,𝑖) ∀𝑖, 𝑗 ∈ 𝑉 , 𝑖 > 𝑗 ∈ 𝑉 (14)
2
experiments in Section 4. Finally, the paper ends with conclusions in
Section 5. The objective function (1) to minimize is the total cost of the
solution. Constraints (2) and (3) are the degree constraints for the
2. Mathematical models customers and facilities, respectively. They ensure that each customer is
visited by exactly one vehicle, and that the same happens to facilities
In order to formally describe the 1-PDLRP, we need to introduce but only if they are open. Constraints (4) enforce each customer 𝑖 to
some notation. Let 𝑉 = 𝐼 ∪ 𝐽 be the set of locations, with 𝐼 being the be assigned to exactly one facility, and constraints (5) avoid that a
set of customers’ locations and 𝐽 the set of potential facility locations. customer is assigned to a closed facility. Constraints (6) are connectivity
Let 𝐸 = {(𝑖, 𝑗) ∶ 𝑖, 𝑗 ∈ 𝑉 , 𝑖 < 𝑗} be the set of edges linking all possible constraints, similar to the subtour elimination constraints for the TSP
pairs of locations in 𝑉 . The 1-PDLRP is defined on the undirected graph (see Padberg and Rinaldi, 1991). They ensure that at least a vehicle
𝐺 = (𝑉 , 𝐸). The cost of traversing an edge 𝑒 ∈ 𝐸 is denoted by 𝑐𝑒 . Each must enter and leave each set of customers 𝑆 ⊆ 𝐼. Constraints (7) state
customer 𝑖 ∈ 𝐼 requires or supplies 𝑞𝑖 units of the commodity: if 𝑞𝑖 > 0, that, if a customer 𝑖 is not assigned to a facility 𝑗, then the edge (𝑖, 𝑗)
the customer is a pickup customer (i.e., a customer where the vehicle cannot be routed. Constraints (8) forbid customers assigned to different
picks up product), and if 𝑞𝑖 < 0, the customer is a delivery customer facilities to be in the same vehicle route. Constraints (12) are the flow
(i.e., a customer where the vehicle delivers product). All customers conservation conditions for the customers. Finally, (9)–(11) and (13)–
must be visited by a vehicle. Each potential facility location 𝑗 ∈ 𝐽 has (14) are the value restrictions for the different kinds of variables. We
an associated opening cost 𝑑𝑗 , and it can serve customers using a vehicle will refer to model (1)–(14) as 𝑃𝑓 𝑙𝑜𝑤 .
with capacity 𝑄. Note that the upper bound for variables 𝑓[𝑖,𝑗] is 𝑄2 𝑥(𝑖,𝑗) , and not
We also use some additional notation. For each subset 𝑆 ⊆ 𝑉 , 𝛿(𝑆) 𝑄𝑥(𝑖,𝑗) , in constraints (13), and 𝑄2 𝑥(𝑗,𝑖) , not 𝑄𝑥(𝑗,𝑖) , in constraints (14). As
is the set of edges with one endpoint in 𝑆 and the other one in 𝑉 ⧵ 𝑆, explained in Hernández-Pérez and Salazar-González, 2004a, variables
and 𝐸(𝑆) is the set of edges with both endpoints in 𝑆. If 𝑆 = {𝑖}, we 𝑓[𝑖,𝑗] do not represent the load of the vehicle when traversing arc [𝑖, 𝑗],
∑
simply use 𝛿(𝑖). For brevity in notation, we write 𝑥(𝐸 ′ ) = 𝑒∈𝐸 ′ 𝑥𝑒 for but are instead a certificate that guarantees that the variables 𝑥𝑒 define
′
all 𝐸 ⊆ 𝐸. a feasible 1-PDLPR solution. Anyway, it is possible to calculate the load
The aim of the one-commodity Pickup and Delivery Location Rout- of the vehicle traversing an arc [𝑖, 𝑗] from the values of the 𝑓𝑎 variables
ing Problem is to decide which facilities to open, to assign each as 𝑓[𝑖,𝑗] + 𝑄2 − 𝑓[𝑗,𝑖] . Note as well that the initial load of a vehicle when
customer to an open facility, and to design vehicle routes to serve the leaving a facility is unfixed, and so it can be any value between 0 and
customers assigned to each facility. We assume that only one vehicle is 𝑄.
available at each facility, each vehicle route starts and ends at the same
It is possible to derive an alternative model for the 1-PDLRP by
facility, and each customer must be visited by exactly one vehicle. The
using Benders’ decomposition to project the continuous flow variables
objective is to find a feasible solution which minimizes the sum of the
𝑓𝑎 in model (1)–(14) into the space of the binary variables, as was done
routing costs and the opening costs.
in Hernández-Pérez et al., 2021. By doing so we obtain the follow-
The 1-PDLRP can be mathematically formulated by defining the
ing constraints, which are similar to the rounded fractional capacity
following variables. For each facility location 𝑗 ∈ 𝐽 , let 𝑦𝑗 be a binary
inequalities for the Capacitated VRP (Gouveia, 1995):
variable that takes value 1 if the facility at location 𝑗 is opened, and 0
⌈ ∑ ⌉
otherwise. For each edge 𝑒 ∈ 𝐸, variable 𝑥𝑒 takes value 1 if a vehicle | 𝑖∈𝑆 𝑞𝑖 |
𝑥(𝛿(𝑆)) ≥ 2 ∀𝑆 ⊆ 𝐼. (15)
traverses 𝑒, and 0 otherwise. For each customer 𝑖 ∈ 𝐼 and each facility 𝑄
𝑗 ∈ 𝐽 a binary variable 𝑧𝑖𝑗 takes value 1 if 𝑖 ∈ 𝐼 is assigned to 𝑗 ∈ 𝐽 ,
and value 0 otherwise. Moreover, let 𝑓𝑎 be a positive real flow variable Hence, constraints (1)–(11) together with (15) provide a valid and
associated with each ordered pair of nodes or arc 𝑎 = [𝑖, 𝑗], with 𝑖, 𝑗 ∈ 𝑉 , pure integer linear programming formulation for the 1-PDLRP. We will
𝑖 ≠ 𝑗. refer to this model as model 𝑃𝐼𝐿𝑃 .

4
B. Domínguez-Martín et al. Computers and Operations Research 161 (2024) 106426

2.1. Valid inequalities and 𝑖′ such that 𝑥∗(𝑖,𝑖′ ) > 0, let 𝑆 = {𝑗 ∈ 𝐽 ∶ 𝑧∗𝑖𝑗 ≥ 𝑧∗𝑖′ 𝑗 }. This set 𝑆
produces the maximum violation of constraint (16) for customers 𝑖 and
We present next two families of valid inequalities that help to 𝑖′ , if there is violation.
strengthen the linear relaxation of the models described in the former
section for the 1-PDLRP. They are both taken from the literature, and 3.2. Separation of constraints (6)
validity proofs are omitted as they are very similar to the ones that
appear in the cited references. This exact procedure is inspired in the separation algorithm for
The first family of inequalities were used in Rodríguez-Martín et al. the subtour elimination constraints of the traveling salesman problem.
(2014) to solve the Hub Location and Routing Problem: Given a fractional solution (𝑥∗ , 𝑦∗ , 𝑧∗ ), for each given customer 𝑖 ∈ 𝐼, we
∑ ∑
𝑥𝑖𝑖′ ≤ 𝑧𝑖𝑗 + 𝑧𝑖′ 𝑗 ′ ∀𝑖, 𝑖′ ∈ 𝐼, 𝑖 ≠ 𝑖′ , ∀𝑆 ⊂ 𝐽 . (16) build a graph 𝐺′ = (𝑉 ′ , 𝐸 ′ ) with 𝑉 ′ = 𝑉 ∪{𝑡}, where 𝑡 is a dummy vertex.
𝑗∈𝐽 ⧵𝑆 𝑗 ′ ∈𝑆 The edge set 𝐸 ′ contains all edges 𝑒 ∈ 𝐸 such that 𝑥∗𝑒 > 0, with capacity
equal to 𝑥∗𝑒 , plus all edges connecting 𝑡 with vertices 𝑗 ∈ 𝐽 , with infinite
These inequalities say that, if customer 𝑖 is assigned to a facility in
capacity. Then, a set 𝑆 ⊂ 𝑉 ′ with 𝑡 ∉ 𝑆 and 𝑖 ∈ 𝑆 defines a violated
𝑆, and customer 𝑖′ is assigned to a facility in 𝐽 ⧵ 𝑆, then the edge (𝑖, 𝑖′ )
inequality (6) if the capacity of the cut 𝛿(𝑆) is smaller than 2. So, the
cannot be in a vehicle’s route. Note as well that (16) dominate (8) since
separation procedure consists of solving a series of min-cut/max-flow
they can be rewritten, using (4), as
∑ ∑ problems.
𝑥𝑖𝑖′ + 𝑧𝑖𝑗 + 𝑧𝑖′ 𝑗 ′ ≤ 2 ∀𝑖, 𝑖′ ∈ 𝐼, 𝑖 ≠ 𝑖′ , ∀𝑆 ⊂ 𝐽 , (17)
𝑗∈𝑆 𝑗 ′ ∈𝐽 ⧵𝑆 3.3. Separation of constraints (15)
and (17) are clearly stronger than (8).
The second family of valid inequalities were derived in Labbé et al. Inequalities (15) are similar to the rounded capacity constraints
(2004) for the Plant Cycle Location Problem: for the capacitated vehicle routing problem, for which no polynomial
time separation algorithm has been found until now. We have designed
⎛ ∑ ∑ ⎞
two heuristic separation algorithms for them, similar to the ones used
𝑥(𝛿(𝑆)) ≥ 2 ⎜ 𝑧𝑖𝑗 + 𝑧𝑖′ 𝑗 ′ ⎟ ∀𝑆 ⊂ 𝑉 , 𝑖 ∈ 𝑆 ∩ 𝐼, 𝑖′ ∈ 𝐼 ⧵ 𝑆.
⎜𝑗∈𝐽 ⧵𝑆 ⋂ ⎟ in Hernández-Pérez et al. (2021), that are applied one after the other.
⎝ 𝑗 ′ ∈𝐽 𝑆 ⎠
The first heuristic is in fact an exact separation procedure for the
(18) fractional capacity constraints
∑
Constraints (18) state that a subset of nodes 𝑆 must be connected to | 𝑖∈𝑆 𝑞𝑖 |
𝑥(𝛿(𝑆)) ≥ 2 ∀𝑆 ⊂ 𝐼. (19)
its complement by at least two edges for each customer in 𝑆 assigned 𝑄
to a facility not in 𝑆, and for each customer outside 𝑆 assigned to a If one of these inequalities is violated, then the corresponding rounded
facility inside 𝑆. constraint (15) is also violated. Note that inequality (19) is equivalent
to both
3. Branch-and-cut algorithms ∑
𝑖∈𝑆 +𝑞𝑖
𝑥(𝛿(𝑆)) ≥ 2 ∀𝑆 ⊂ 𝐼,
As the two formulations we have presented for the 1-PDLRP contain 𝑄
an exponentially large number of constraints, a natural way to solve the and
problem exactly is to use the branch-and- cut technique. This approach ∑
𝑖∈𝑆 −𝑞𝑖
combines a branch-and-bound method for exploring a decision tree, 𝑥(𝛿(𝑆)) ≥ 2 ∀𝑆 ⊂ 𝐼,
𝑄
and a cutting plane method for computing bounds. The performance
which, in turn, are equivalent to
of the whole algorithm depends mainly on the bounds given by the
linear relaxations, and on the efficiency of the separation procedures. ∑ ∑ ∑
𝑖∈𝑆∶𝑞𝑖 <0 −𝑞𝑖 𝑖∈𝐼⧵𝑆∶𝑞𝑖 >0 𝑞𝑖 𝑖∈𝐼∶𝑞𝑖 >0 𝑞𝑖
We outline next the main features of our branch-and-cut algorithms. 𝑥(𝛿(𝑆)) + 2 +2 ≥2 ∀𝑆 ⊂ 𝐼,
𝑄 𝑄 𝑄
We keep the solver default settings to control the branching variable
selection, and to decide which node of the branch-and-bound search (20)
tree to explore next. The first LP solved at the root node of the branch- and
and-cut algorithm depends on the model used. In the case of the flow ∑ ∑ ∑
model 𝑃𝑓 𝑙𝑜𝑤 , it is the linear relaxation of model (1)–(13), excluding 𝑖∈𝑆∶𝑞𝑖 >0 𝑞𝑖 𝑖∈𝐼⧵𝑆∶𝑞𝑖 <0 −𝑞𝑖 𝑖∈𝐼∶𝑞𝑖 <0 −𝑞𝑖
𝑥(𝛿(𝑆)) + 2 + ≥2 ∀𝑆 ⊂ 𝐼,
(8) and the exponential family of connectivity constrains (6). In the 𝑄 𝑄 𝑄
case of model 𝑃𝐼𝐿𝑃 , the first LP solved is the linear relaxation of model (21)
(1)–(11), excluding also inequalities (6) and (8).
In the cutting plane phase, given a fractional solution (𝑥∗ , 𝑦∗ , 𝑧∗ ) of respectively. As the coefficients in the left hand sides of (20) and
model 𝑃𝑓 𝑙𝑜𝑤 or 𝑃𝐼𝐿𝑃 , separation routines are sequentially applied to (21), and the constant values in their right hand sides (let us call 𝐶
find violated cuts. The number of violated cuts of each family added that constant value), are all positive numbers, these constraints can
to the model in each cutting planes phase is limited to 100, and a be separated exactly solving a min-cut problem on an appropriately
separation procedure is applied only if the previous one has failed to defined support graph. If (20) or (21) is violated, so does (19). The
find violated cuts. In the following we give the details of the separation support graph is defined as 𝐺′ = (𝑉 ′ , 𝐸 ′ ) with 𝑉 ′ = 𝑉 ∪ {𝑠, 𝑡}, with 𝑠
procedures for the different types of valid inequalities and exponential and 𝑡 being dummy nodes, and edge set 𝐸 ′ composed of all the edges
number families of constraints, in the order they are applied. These (𝑖, 𝑗) ∈ 𝐸 with capacity 𝑥∗ . In the case of (20), we add to 𝐸 ′ all edges
procedures are analogous to others in the literature (see Labbé et al., (𝑖, 𝑡), if 𝑖 is a pickup customer, with capacity 2𝑞𝑖 ∕𝑄, all edges (𝑖, 𝑠), if
2004, Rodríguez-Martín et al., 2014, and Hernández-Pérez et al., 2021). 𝑖 is a delivery customer, with capacity −2𝑞𝑖 ∕𝑄, and all edges (𝑗, 𝑠), for
𝑗 ∈ 𝐽 , with a capacity infinite. Similarly, in the case of (21), we add to
3.1. Separation of constraints (8) and (16) 𝐸 ′ all edges (𝑖, 𝑡), if 𝑖 is a delivery customer, with capacity −2𝑞𝑖 ∕𝑄, all
edges (𝑖, 𝑠), if 𝑖 is a pickup customer, with capacity 2𝑞𝑖 ∕𝑄, and all edges
Inequalities (8) are separated by complete enumeration in polyno- (𝑗, 𝑠), for 𝑗 ∈ 𝐽 , with a capacity infinite. We calculate then a min-cut
mial time. Inequalities (16) can be also separated exactly using the set 𝑆 ′ ⊂ 𝑉 ′ with 𝑡 ∈ 𝑆 and 𝑠 ∉ 𝑆. If the value of that cut is less than
following 𝑂(|𝐼|2 |𝐽 |) algorithm: for each pair of different customers 𝑖 𝐶, then 𝑆 = 𝑆 ′ ⧵ {𝑡} defines a violated inequality (19).

5
B. Domínguez-Martín et al. Computers and Operations Research 161 (2024) 106426

Table 1
Model comparison for instances with 𝑛 = 30, |𝐽 | = 5, and 𝑑𝑗 = 100.
Q Name 𝐵&𝐶𝑓 𝑙𝑜𝑤 𝐵&𝐶0 𝐵&𝐶1 𝐵&𝐶2
Time %-gap Time %-gap Time %-gap Time %-gap
10 A 236.31 9.94 2.00 3.02 1.14 2.49 1.30 2.47
B 161.05 9.10 6.38 3.34 0.81 3.34 0.78 3.34
C 64.30 8.12 2.45 1.73 1.28 1.71 1.02 1.71
D 26.92 10.04 2.98 3.77 0.73 3.57 0.73 3.57
E 14.44 6.83 3.06 18.79 0.47 18.79 0.77 2.14
F 265.72 9.65 3.27 2.52 1.64 4.45 1.75 4.45
G 855.48 9.28 20.64 11.21 5.22 2.29 2.17 2.29
H 192.47 10.06 7.95 4.77 8.83 4.44 2.75 4.44
I 34.05 9.65 1.31 2.94 0.36 2.00 0.33 2.00
J 271.88 11.71 1.38 3.75 1.13 3.75 1.11 3.75
Aver 212.26 9.44 5.14 5.58 2.16 4.68 1.27 3.02
20 A 13.98 6.46 0.39 2.56 0.34 2.56 0.33 2.56
B 3.34 2.77 0.13 0.00 0.06 0.73 0.06 0.00
C 2.23 4.13 0.70 3.26 0.41 3.26 0.42 3.26
D 11.30 6.86 0.61 3.51 0.44 3.51 0.47 3.51
E 2.52 3.85 0.47 2.06 0.25 2.06 0.33 1.63
F 17.55 4.98 0.28 2.57 0.16 2.57 0.17 2.57
G 258.22 9.14 0.98 3.19 0.73 3.19 1.05 2.90
H 6.14 5.08 3.09 4.61 2.34 3.86 0.53 3.86
I 0.38 0.00 0.03 0.00 0.03 0.00 0.03 0.00
J 49.58 4.69 0.08 0.00 0.08 0.00 0.08 0.00
Aver 36.52 4.80 0.68 2.18 0.48 2.17 0.35 2.03
40 A 0.36 0.00 0.08 0.92 0.06 0.92 0.06 0.92
B 0.28 0.00 0.03 0.00 0.02 0.00 0.02 0.00
C 0.34 0.00 0.13 0.00 0.06 0.00 0.06 0.00
D 0.31 0.00 0.03 0.00 0.02 0.00 0.02 0.00
E 0.63 1.80 0.20 1.46 0.17 1.46 0.17 0.82
F 1.11 2.71 0.36 2.19 0.20 2.19 0.20 2.19
G 6.30 5.06 0.27 1.85 0.30 1.85 0.30 1.85
H 0.80 1.74 0.06 0.00 0.05 0.00 0.05 0.00
I 0.25 0.00 0.06 0.00 0.05 0.00 0.03 0.00
J 1.14 2.64 0.23 2.40 0.14 2.40 0.14 2.40
Aver 1.15 1.40 0.15 0.88 0.11 0.88 0.10 0.82

The second heuristic designed to find violated constraints (15) is such that 𝑥∗𝑒 > 0. The capacity of the edges in 𝐸 ′ is set to the positive
similar to a procedure used in Hernández-Pérez and Salazar-González value considered for their definition. Then we look for the min-cut set
(2007). It is a greedy algorithm that builds a list of subsets 𝑆 ∈ 𝐼 which 𝑆 in 𝐺′ such that 𝑖 ∈ 𝑆 and 𝑖′ ∉ 𝑆. If the capacity of this cut is less
could result in violated inequalities (15). To this end, first note that (15) than 4, we have found a violated constraint (18).
can be written as
⌈ ∑ ⌉
| 𝑖∈𝑆 𝑞𝑖 | 4. Computational experiments
𝑥(𝐸(𝑆)) ≤ |𝑆| − ∀𝑆 ⊂ 𝐼,
𝑄
The list of candidates is built as follows. It is initialized with the The branch-and-cut algorithms described in Section 3 were imple-
|𝐼| subsets of cardinality two 𝑆 = {𝑖, 𝑖′ } with the largest values 𝑥∗𝑖𝑖′ . mented in C++ and run on a personal computer with an Intel Core i7
Then, each of these subsets is enlarged iteratively, obtaining in each CPU at 3.4 GHz and 16 GB of RAM. We used CPLEX 22.1 as mixed
iteration a new subset with one more element, until we get subsets of integer linear programming solver, and a routine from the Concorde
cardinality |𝐼|∕2. A subset of cardinality 𝑘 is generated from a subset TSP software package to solve the max-flow/min-cut problems.
𝑆 of cardinality 𝑘 − 1 by adding to 𝑆 a customer 𝑖′ such that 𝑖′ ∈ 𝐼 ⧵ 𝑆, We tested the algorithms on a benchmark data set from the liter-
∑ ∗ ′ ature (see, Hernández-Pérez and Salazar-González, 2004a, Hernández-
𝑖∈𝑆 𝑥𝑖𝑖′ > 0, and 𝑆 ∪{𝑖 } is not in the list. If there are several∑customers
𝑖′ verifying these conditions, the one which maximizes 𝑖∈𝑆 𝑥∗𝑖𝑖′ is Pérez et al., 2009, or Hernández-Pérez et al., 2021) These are randomly
inserted. When the list is completed, we check each of its elements 𝑆 generated instances with node coordinates in the square [−500, 500] ×
for possible violation of a constraint (15). [−500, 500], and with customers demands in [−10, 10]. The travel cost
𝑐𝑖𝑗 is computed as the rounded Euclidean distance between nodes 𝑖 and
3.4. Separation of constraints (18) 𝑗. The number of nodes 𝑛 varies in {30, 40, 50, 60, 100} and the vehicle
capacity 𝑄 takes values in {10, 20, 40}, which correspond to a tight,
Inequalities (18) can be written as medium, and relaxed capacity, respectively. There are 10 instances for
( ) each combination of 𝑛 and 𝑄, named from A to J. The data is available
∑ ∑ ∑ ∑ at the pickup-and delivery site http://webpages.ull.es/users/hhperez/
𝑥(𝛿(𝑆)) + 2 𝑧𝑖𝑗 + 2 𝑧𝑖′ 𝑗 ′ ≥ 2 𝑧𝑖𝑗 + 𝑧𝑖′ 𝑗 ′
𝑗∈𝐽 ∩𝑆 𝑗 ′ ∈𝐽 ⧵𝑆 𝑗∈𝐽 𝑗∈𝐽 PDsite. To adapt these instances to the 1-PDLRP, we consider that the
|𝐼| first nodes of each instance are customers, and the following |𝐽 |
= 4, ∀𝑆 ⊂ 𝑉 , 𝑖 ∈ 𝑆 ∩ 𝐼, 𝑖′ ∈ 𝐼 ⧵ 𝑆.
nodes are possible facility locations.
Then, given a fractional solution (𝑥∗ , 𝑦∗ , 𝑧∗ ), to separate these in- Our first experiment consisted in comparing the performance of the
equalities we proceed as follows. For each pair of customers 𝑖, 𝑖′ ∈ 𝐼, branch-and-cut algorithms based on the flow model 𝑃𝑓 𝑙𝑜𝑤 and on the
we define a support graph 𝐺′ = (𝑉 ′ , 𝐸 ′ ), with 𝑉 ′ = 𝑉 and with 𝐸 ′ being pure ILP model 𝑃𝐼𝐿𝑃 , as well as assessing the impact of the different
the set of all edges (𝑖, 𝑗) ∈ 𝐸 such that 𝑗 ∈ 𝐽 and 𝑥∗𝑖𝑗 +2𝑧∗𝑖′ 𝑗 > 0, all edges families of valid inequalities proposed for the problem. The outcome
(𝑖′ , 𝑗) ∈ 𝐸 such that 𝑗 ∈ 𝐽 and 𝑥∗𝑖′ 𝑗 + 2𝑧∗𝑖𝑗 > 0, and all other edges 𝑒 ∈ 𝐸 is displayed in Table 1. This table shows results on instances with

6
B. Domínguez-Martín et al. Computers and Operations Research 161 (2024) 106426

Table 2
Branch-and-cut results for medium-sized instances.
𝑛 𝑄 Name OptVal nOpen Time BBnodes %-gap %-fgap nCuts (6) (15) (16) (18) %-uncapa

30 10 A 5703 3 1.30 87 2.47 0.00 561 4.63 40.11 41.89 13.37 25.32
B 5356 3 0.78 106 3.34 0.00 560 5.89 37.68 56.43 0.00 21.28
C 5420 3 1.02 106 1.71 0.00 580 4.83 39.48 52.59 3.10 22.75
D 6120 3 0.73 87 3.57 0.00 476 4.62 38.45 56.93 0.00 27.63
E 5760 3 0.77 4 2.14 0.00 574 5.92 22.30 12.72 59.06 17.57
F 5660 3 1.75 186 4.45 0.00 902 3.44 35.81 58.09 2.66 24.17
G 8125 3 2.17 156 2.29 0.00 806 3.23 32.75 55.46 8.56 44.02
H 6118 4 2.75 468 4.44 0.00 1101 2.27 27.88 66.58 3.27 27.13
I 4976 2 0.33 25 2.00 0.00 232 9.48 55.17 35.34 0.00 16.44
J 5896 4 1.11 105 3.75 0.00 736 5.84 41.03 53.13 0.00 29.46

Aver 3.10 1.27 133.00 3.02 0.00 652.80 5.02 37.07 48.92 9.00 25.58

20 A 4606 1 0.33 25 2.56 0.00 262 7.63 49.62 42.75 0.00 7.53
B 4479 1 0.06 0 0.00 0.00 68 11.76 19.12 1.47 67.65 5.87
C 4377 2 0.42 21 3.26 0.00 144 18.75 32.64 48.61 0.00 4.34
D 4827 1 0.47 35 3.51 0.00 214 12.15 47.20 40.65 0.00 8.25
E 4837 3 0.33 14 1.63 0.00 240 9.17 13.75 25.83 51.25 1.84
F 4483 1 0.17 9 2.57 0.00 139 14.39 33.81 51.80 0.00 4.26
G 5758 2 1.05 87 2.90 0.00 573 4.54 34.03 23.04 38.39 21.01
H 4800 2 0.53 52 3.86 0.00 344 4.94 15.41 43.90 35.76 7.13
I 4158 1 0.03 0 0.00 0.00 3 100.00 0.00 0.00 0.00 0.00
J 4436 1 0.08 0 0.00 0.00 58 22.41 77.59 0.00 0.00 6.24

Aver 1.50 0.35 24.30 2.03 0.00 204.50 20.57 32.32 27.80 19.30 6.65

40 A 4259 1 0.06 2 0.92 0.00 37 40.54 0.00 59.46 0.00 0.00

B 4216 1 0.02 0 0.00 0.00 4 100.00 0.00 0.00 0.00 0.00
C 4187 1 0.06 0 0.00 0.00 101 12.87 0.00 0.00 87.13 0.00
D 4429 1 0.02 0 0.00 0.00 4 100.00 0.00 0.00 0.00 0.00
E 4748 1 0.17 21 0.82 0.00 49 24.49 0.00 75.51 0.00 0.00
F 4308 3 0.20 57 2.19 0.00 157 15.29 2.55 82.17 0.00 0.37
G 4715 2 0.30 58 1.85 0.00 194 14.95 11.86 73.20 0.00 3.54
H 4466 1 0.05 0 0.00 0.00 15 93.33 6.67 0.00 0.00 0.18
I 4158 1 0.03 0 0.00 0.00 3 100.00 0.00 0.00 0.00 0.00
J 4159 1 0.14 9 2.40 0.00 49 38.78 4.08 57.14 0.00 0.00

Aver 1.30 0.10 14.70 0.82 0.00 61.30 54.02 2.52 34.75 8.71 0.41

40 10 A 6565 3 6.06 692 3.54 0.00 1561 3.72 33.82 45.48 16.98 25.58
B 5877 2 2.30 144 5.05 0.00 899 4.23 40.49 44.16 11.12 17.03
C 7305 2 14.34 474 30.77 0.00 2694 3.82 47.25 48.92 0.00 35.62
D 6108 1 0.47 2 1.49 0.00 273 12.09 49.45 38.46 0.00 14.03
E 6554 4 41.83 3194 6.34 0.00 3424 2.75 27.92 39.52 29.82 20.20
F 6875 4 22.39 1424 4.57 0.00 2727 2.82 40.92 49.06 7.19 30.76
G 6216 2 11.06 919 4.78 0.00 2314 4.24 39.33 51.30 5.14 16.38
H 6519 2 8.03 551 3.59 0.00 1838 4.30 43.53 51.09 1.09 26.42
I 6697 3 7.31 494 30.34 0.00 1931 4.71 37.86 50.28 7.15 28.48
J 6051 1 28.95 3173 5.08 0.00 2616 4.28 39.37 44.84 11.51 23.05

Aver 2.40 14.28 1106.70 9.56 0.00 2027.70 4.69 39.99 46.31 9.00 23.75

20 A 5139 2 0.63 10 0.73 0.00 283 13.07 23.32 63.60 0.00 4.92
B 4974 2 0.27 8 1.44 0.00 84 14.29 1.19 84.52 0.00 1.97
C 5523 1 2.41 324 3.13 0.00 921 6.19 48.75 41.80 3.26 14.85
D 5269 1 0.22 0 0.00 0.00 179 8.38 1.68 20.67 69.27 0.34
E 5374 2 1.56 162 2.98 0.00 848 10.85 31.25 47.88 10.02 2.68
F 5023 1 0.77 105 3.14 0.00 379 11.61 34.30 54.09 0.00 5.24
G 5266 2 0.34 5 1.07 0.00 136 14.71 25.00 60.29 0.00 1.29
H 5033 1 4.27 466 2.29 0.00 1489 7.25 22.83 69.91 0.00 4.69
I 5135 2 0.13 4 9.15 0.00 140 10.71 50.00 39.29 0.00 6.72
J 4773 1 0.03 0 0.00 0.00 35 17.14 82.86 0.00 0.00 2.45

Aver 1.50 1.06 108.40 2.39 0.00 449.40 11.42 32.12 48.21 8.26 4.51

40 A 4886 1 0.14 2 0.26 0.00 11 100.00 0.00 0.00 0.00 0.00

B 4876 2 0.17 0 0.00 0.00 78 6.41 0.00 50.00 43.59 0.00
C 4703 1 0.06 0 0.00 0.00 5 100.00 0.00 0.00 0.00 0.00
D 5251 1 0.05 0 0.00 0.00 13 100.00 0.00 0.00 0.00 0.00
E 5236 1 0.33 29 0.68 0.00 144 27.78 2.08 70.14 0.00 0.11
F 4760 1 0.16 0 0.00 0.00 12 100.00 0.00 0.00 0.00 0.00
G 5198 1 0.27 26 1.34 0.00 139 33.09 0.00 66.91 0.00 0.00
H 4797 1 0.14 4 1.09 0.00 68 44.12 5.88 50.00 0.00 0.00
I 4796 2 0.22 7 2.73 0.00 143 25.17 1.40 73.43 0.00 0.13
J 4656 1 0.05 0 0.00 0.00 6 100.00 0.00 0.00 0.00 0.00

Aver 1.20 0.16 6.80 0.61 0.00 61.90 63.66 0.94 31.05 4.36 0.02

30 nodes, 25 customers and 5 possible facility locations. The opening • 𝐵&𝐶𝑓 𝑙𝑜𝑤 : It solves the 𝑃𝑓 𝑙𝑜𝑤 model (1)–(13). Constraints (6) and
cost 𝑑𝑗 is set to 100 for all the facilities. Columns 𝑄 and Name show (8) are dynamically incorporated when violated.
the capacity of the vehicle and the name of the instance. For each • 𝐵&𝐶0: It solves the 𝑃𝐼𝐿𝑃 model (1)–(11), (15). Constraints (6),
instance, and for each algorithm, columns Time and %-gap display the (8), and (15) are dynamically incorporated when violated.
total computation time, in seconds, and the percentage gap between • 𝐵&𝐶1: It is similar to 𝐵&𝐶0, but constraints (8) are replaced by
the optimal solution value and the lower bound at the end of the root constrains (16).
node, respectively. Lines Aver show average values for the ten instances • 𝐵&𝐶2: Like 𝐵&𝐶1, but valid inequalities (18) are also separated.
with a given capacity 𝑄. The four branch-and-cut algorithms compared
are:

7
B. Domínguez-Martín et al. Computers and Operations Research 161 (2024) 106426

Table 3
Branch-and-cut results for large instances.
𝑛 𝑄 Name OptVal nOpen Time BBnodes %-gap %-fgap nCuts (6) (15) (16) (18) %-uncapa

50 10 A 6263 2 1.14 16 0.48 0.00 532 8.83 76.50 14.66 0.00 11.32
B 9034 2 126.13 1478 4.17 0.00 6897 3.15 47.14 46.70 3.02 36.44
C 8066 2 71.52 1577 3.82 0.00 3860 3.03 59.07 31.48 6.42 35.26
D 8946 2 26.63 849 3.43 0.00 2963 4.22 46.03 42.02 7.73 33.22
E 8434 4 110.41 2101 3.40 0.00 5840 1.76 42.31 50.50 5.43 30.19
F 7695 2 54.56 1188 8.46 0.00 4051 2.64 51.67 35.57 10.12 29.89
G 6818 2 10.33 582 2.60 0.00 1983 4.74 34.90 60.36 0.00 17.53
H 8200 3 147.25 3881 6.16 0.00 6082 2.76 34.97 50.35 11.92 31.82
I 7510 2 46.09 2126 29.91 0.00 3504 3.40 44.04 51.20 1.37 26.80
J 7519 2 43.50 1687 3.68 0.00 3098 3.03 52.68 34.67 9.62 26.04

Aver 2.30 63.75 1548.50 6.61 0.00 3881.00 3.76 48.93 41.75 5.56 27.85

20 A 5713 1 0.47 23 2.72 0.00 232 17.24 27.16 55.60 0.00 2.78
B 6745 2 11.02 655 3.18 0.00 1834 6.71 45.58 44.00 3.71 14.87
C 6221 1 22.94 967 4.56 0.00 2676 4.07 44.39 50.45 1.08 16.06
D 6782 3 13.53 851 4.45 0.00 2217 5.28 48.22 46.19 0.32 11.91
E 6292 2 2.27 58 0.62 0.00 655 6.56 43.66 49.77 0.00 6.42
F 5891 3 3.61 246 2.72 0.00 1272 7.55 32.31 60.14 0.00 8.42
G 5861 2 1.61 235 4.07 0.00 721 11.37 17.75 70.87 0.00 4.06
H 6033 1 6.63 218 2.41 0.00 1713 5.31 35.32 53.59 5.78 7.33
I 6014 1 8.00 658 12.47 0.00 1729 7.17 25.45 57.32 10.06 8.60
J 5764 1 1.27 169 2.38 0.00 488 11.27 29.30 59.43 0.00 3.52

Aver 1.70 7.13 408.00 3.96 0.00 1353.70 8.25 34.91 54.74 2.10 8.40

40 A 5554 1 0.22 7 0.68 0.00 30 100.00 0.00 0.00 0.00 0.00

B 5825 2 0.66 6 1.22 0.00 80 30.00 61.25 8.75 0.00 1.42
C 5449 1 0.80 67 1.77 0.00 311 15.43 10.29 74.28 0.00 4.17
D 6096 1 0.78 69 2.18 0.00 425 16.00 8.00 76.00 0.00 2.00
E 5888 3 0.44 0 0.00 0.00 129 20.93 6.20 72.87 0.00 0.00
F 5396 1 0.17 0 0.00 0.00 14 92.86 7.14 0.00 0.00 0.02
G 5623 1 0.27 16 1.17 0.00 39 89.74 5.13 5.13 0.00 0.00
H 5620 1 0.94 78 2.30 0.00 350 16.86 11.43 71.71 0.00 0.52
I 5497 1 3.28 360 4.24 0.00 1381 8.25 0.58 84.58 6.59 0.00
J 5561 1 0.13 0 0.00 0.00 17 100.00 0.00 0.00 0.00 0.00

Aver 1.30 0.77 60.30 1.36 0.00 277.60 49.01 11.00 39.33 0.66 0.81

60 10 A 8043 2 t.l. 41789 9.21 2.74 17266 1.92 48.20 24.35 25.52 24.07
B 8322 2 33.31 647 3.34 0.00 3053 4.06 42.29 49.66 4.00 23.32
C 8861 2 36.20 428 3.84 0.00 3252 3.57 42.19 47.54 6.70 29.36
D 11011 3 t.l. 38351 8.05 1.71 17587 2.16 50.56 31.46 15.83 40.88
E 9565 3 1733.70 15040 4.85 0.00 13663 1.81 45.03 38.07 15.10 33.02
F 8733 3 943.69 7497 5.82 0.00 12690 2.06 44.98 41.89 11.06 32.38
G 8579 3 1304.33 10085 6.27 0.00 11879 1.83 47.14 38.69 12.34 28.06
H 7827 3 44.63 1283 21.52 0.00 3618 3.79 42.87 52.18 1.16 24.06
I 9458 4 t.l. 30330 8.30 1.90 17729 1.55 39.09 30.66 28.69 37.75
J 8261 1 1685.47 9769 22.52 0.00 16222 4.22 46.64 44.91 4.22 22.24

Aver 2.60 2738.13 15 521.90 9.37 0.63 11 695.90 2.70 44.90 39.94 12.46 29.51

20 A 6603 1 91.77 5527 5.20 0.00 4035 5.75 31.13 55.61 7.51 7.51
B 6648 2 30.44 826 3.09 0.00 3589 5.74 44.55 47.53 2.17 4.02
C 6868 2 38.33 964 2.07 0.00 3869 5.20 41.79 52.47 0.54 8.87
D 7784 1 61.77 2385 3.91 0.00 3670 5.59 43.51 47.79 3.11 16.37
E 7336 1 117.33 5483 3.68 0.00 4853 4.10 41.15 50.53 4.22 12.66
F 6589 2 73.30 2243 2.00 0.00 4845 4.95 35.11 56.84 3.10 10.38
G 6651 2 23.13 647 2.20 0.00 3091 5.53 32.97 61.50 0.00 7.20
H 6404 2 5.91 544 2.70 0.00 1380 7.25 23.70 66.30 2.75 7.18
I 7105 2 2829.42 30076 6.96 0.00 14311 2.31 45.28 26.58 25.83 17.13
J 6719 2 13.08 416 1.90 0.00 2247 8.14 34.53 57.32 0.00 4.39

Aver 1.70 328.45 4911.10 3.37 0.00 4589.00 5.46 37.37 52.25 4.92 9.57

40 A 6110 1 0.98 131 2.02 0.00 267 27.34 10.86 61.80 0.00 0.05
B 6381 1 0.67 61 0.63 0.00 59 74.58 25.42 0.00 0.00 0.00
C 6271 1 0.30 2 0.51 0.00 178 17.42 25.84 56.74 0.00 0.19
D 6589 1 0.72 29 2.09 0.00 227 27.31 16.74 55.95 0.00 1.20
E 6415 2 0.33 22 1.15 0.00 154 26.62 24.03 49.35 0.00 0.12
F 6033 2 2.05 245 2.20 0.00 577 15.42 36.05 48.53 0.00 2.12
G 6276 1 1.45 163 1.69 0.00 545 11.19 7.16 81.65 0.00 1.66
H 5945 1 0.36 18 0.74 0.00 94 51.06 5.32 43.62 0.00 0.02
I 6029 2 32.06 2615 4.12 0.00 3014 13.24 20.24 66.52 0.00 2.34
J 6431 1 0.31 19 0.42 0.00 34 82.35 17.65 0.00 0.00 0.11

Aver 1.30 3.92 330.50 1.56 0.00 514.90 34.65 18.93 46.42 0.00 0.78

Table 1 clearly shows that even the basic branch-and-cut algorithm example, for the instances with 𝑄 = 10 the average computing time
based on the model 𝑃𝐼𝐿𝑃 (𝐵&𝐶0) outperforms the one based on the and LP gap of 𝐵&𝐶0 are 5.14 and 5.58, respectively; using constraints
model 𝑃𝑓 𝑙𝑜𝑤 , and this is more evident when the vehicle capacity is the (16) reduces those values to 2.16 and 4.68, respectively, and separating
smallest (𝑄 = 10). For those instances, the average computing times of the valid inequalities (18) further improves the results to 1.27 and 3.02,
𝐵&𝐶𝑓 𝑙𝑜𝑤 and 𝐵&𝐶0 are 212.26 and 5.14 s, respectively. When 𝑄 = 40, respectively. The effect of the valid inequalities is greater on the hardest
those figures are 1.15 and 0.15, respectively. 𝐵&𝐶0 also outperforms instances in the table (those with the tightest capacity). The remaining
𝐵&𝐶𝑓 𝑙𝑜𝑤 in terms of gap at the root node. Table 1 also shows the tables in this section show results only for the best of the algorithms
effect of using the valid inequalities (16) and (18). We can see that implemented, that is, 𝐵&𝐶2.
they contribute to reduce both the computing time and the gap. For

8
B. Domínguez-Martín et al. Computers and Operations Research 161 (2024) 106426

Table 4
Branch-and-cut results for very large instances.
𝑛 𝑄 Name OptVal nOpen Time BBnodes %-gap %-fgap nCuts (6) (15) (16) (18) %-uncapa

100 10 A
B 17614 3 t.l. 6402 32.30 30.95 25986 2.48 70.05 23.80 3.67 56.10
C
D 16488 4 t.l. 8557 20.42 18.33 23044 2.14 53.32 38.25 6.29 52.06
E 11810 3 t.l. 8336 8.51 6.59 27046 1.62 55.86 38.92 3.59 32.95
F 19040 2 t.l. 12229 44.38 43.29 25493 2.40 67.91 26.76 2.92 58.92
G
H 19284 4 t.l. 8790 38.95 37.54 24810 2.57 67.42 27.59 2.42 59.62
I 18159 4 t.l. 9261 31.69 30.11 27267 2.11 57.37 36.09 4.42 56.78
J

Aver 3.33 t.l. 8929.17 29.38 27.80 25 607.67 2.22 61.99 31.90 3.89 52.74

20 A 8628 1 t.l. 12072 4.27 1.36 28159 3.98 52.72 42.16 1.14 9.90
B 12959 3 t.l. 16822 33.31 31.81 25577 3.19 72.37 21.82 2.63 40.33
C 9150 3 t.l. 32800 4.00 0.65 20311 2.87 46.88 34.43 15.81 13.33
D 9591 3 2978.23 11738 3.64 0.00 17034 3.23 55.70 39.12 1.95 17.59
E 9567 3 t.l. 17322 12.60 11.14 22930 2.89 62.02 28.09 7.00 17.23
F 9008 1 t.l. 10201 8.09 6.40 27144 4.86 52.97 39.73 2.44 13.18
G 8600 1 2260.05 18394 4.05 0.00 11024 4.91 70.44 23.58 1.08 10.37
H
I 9735 3 t.l. 13055 5.62 3.36 24326 3.87 54.48 39.30 2.35 19.37
J 9509 2 t.l. 20200 10.93 9.19 19590 3.78 61.55 24.98 9.69 19.34

Aver 2.22 6182.03 16 956.00 9.61 7.10 21 788.33 3.73 58.79 32.58 4.90 17.85

40 A 7947 1 24.69 962 1.21 0.00 1406 17.99 33.57 48.44 0.00 2.18
B 8053 1 865.34 15675 2.78 0.00 7057 7.65 56.38 29.08 6.89 3.97
C 8137 1 94.45 2492 2.54 0.00 4041 9.30 29.92 58.62 2.15 2.54
D 8127 1 19.16 356 1.43 0.00 1870 9.20 47.49 43.32 0.00 2.74
E 7982 1 759.25 2997 1.86 0.00 14081 6.38 19.52 73.08 1.01 0.79
F 7949 1 701.98 20357 2.87 0.00 7162 12.33 26.64 57.46 3.57 1.61
G 7872 1 46.06 2332 2.74 0.00 2399 10.46 26.30 63.23 0.00 2.08
H 8005 1 56.05 665 1.69 0.00 3036 9.29 64.99 25.72 0.00 2.72
I 8240 2 298.80 4275 1.82 0.00 6121 10.31 35.27 50.73 3.69 4.75
J 8066 2 624.39 13183 2.77 0.00 7679 9.32 31.01 54.66 5.01 4.91

Aver 1.20 349.02 6329.40 2.17 0.00 5485.20 10.22 37.11 50.43 2.23 2.83

Tables 2 to 4 show the detailed results of our experiments on to serve the customers decreases as the vehicle capacity increases. For
instances with a number of nodes that ranges from 30 to 100. In all example, in most instances with 𝑛 = 30 and 𝑄 = 10 more than 3
cases, the number of possible facility locations is set to 5, and the facilities are opened, while only one facility is generally opened when
opening cost is set to 𝑑𝑗 = 100 for all the facilities. Besides the number 𝑄 = 40. Observe also that, for the instances in our benchmark, 𝑄 = 40
of nodes (𝑛), the vehicles’ capacity (𝑄), the name of the instances is a rather loose capacity that produces solutions very close to the ones
(Name), the total computation time (Time), and the gap at the end of of the uncapacitated 1-PDLRP. As for the valid inequalities separated,
the root node (%-gap), the data displayed for each instance are: cuts (16) are more numerous than (18), and the later are rarely used
when 𝑄 = 40. This happens because the violation of constraints (18) is
• OptVal: Optimal solution value. only checked when no other types of violated cuts haven been found.
• nOpen: Number of opened facilities.
Although 𝑛 = 60 is already a large instance size to be solved opti-
• BBnodes: Number of nodes explored in the branch-and-cut tree.
mally in complex routing problems such as the 1-PDLRP, we performed
• %-fgap: Percentage gap between the value of the best solution
experiments on a set of instances with 100 nodes to test the limits of
found and the lower bound at the end of the computation.
our branch-and-cut algorithm. The results are displayed in Table 4.
• nCuts: Total number of violated cuts.
We can see that the hardest instances are those with 𝑄 = 10. In 6
• (6), (15), (16), and (18): Percentage of violated cuts of each type
out of the 10 cases, the algorithm reaches the time limit of two hours
over the total number of violated cuts.
without proving optimality, and in the remaining 4 cases it fails to find
• %-uncapa: Percentage difference between the optimal solution
a feasible solution. When the vehicle capacity increases to 𝑄 = 20, 2 out
of the 1-PDLRP and the optimal solution of the uncapacitated
of the 10 cases are solved to optimality, feasible solutions are provided
version of the problem.
for 7 other cases, and only in one case the algorithm does not find any
The algorithm was run with a time limit of two hours. If the time limit solution. When 𝑄 = 40, all cases are solved to optimality in an average
is reached, we report ‘t.l.’ in the column Time, we show the value of the time of 349.02 s.
best feasible solution found in column OptVal, and we use that value In all the experiments whose results are shown in Tables 1 to 4, we
to compute the gap at the end of the root node. Blank lines indicate set |𝐽 | = 5 and 𝑑𝑗 = 100 for all 𝑗 ∈ 𝐽 . To analyze the effect of the
that the algorithm was unable to find even a feasible solution within number of possible facility locations (|𝐽 |) and the opening cost (𝑑𝑗 ),
the time limit. Lines Aver show average results, as in Table 1. we performed some additional experiments. The complete outputs are
We can see in Tables 2 and 3 that the proposed algorithm is able presented in Appendix in order not to overload the article. Table 5
to solve to optimality all instances with up to 50 nodes, and that it shows the results for instances with 𝑛 = 50, |𝐽 | ∈ {5, 10}, and 𝑑𝑗 ∈
fails to find the optimum for only three instances with 60 nodes. As {0, 100, 1000}. Note that 𝑑𝑗 = 0 means that there is not any cost to
expected, the difficulty of the problem increases with the size of the pay for opening a facility, and therefore, the only cost to minimize in
instances, and, for a given size, the hardest instances are those with the problem is the routing cost. On the other hand, setting 𝑑𝑗 = 1000
the smallest capacity. For example, in the case of the instances with makes it very costly to open a facility in the instances considered. A
50 nodes, the average computing time needed to solve the problem is first observation is that, in most cases, as expected, for a fixed |𝐽 |, the
63.75 s when 𝑄 = 10, 7.13 s when 𝑄 = 20, and 0.77 s when 𝑄 = 40. number of open facilities decreases when the opening cost increases.
Regarding the solutions, the number of facilities that need to be opened For example, for the instance C with |𝐽 | = 10, 6 facilities are opened

9
B. Domínguez-Martín et al. Computers and Operations Research 161 (2024) 106426

Table 5
Results for instances with 𝑛 = 50, 𝑄 = 10, |𝐽 | ∈ {5, 10}, and different opening costs.
|𝐽 | Name 𝑑𝑗 = 0 𝑑𝑗 = 100 𝑑𝑗 = 1000
OptVal nOpen Time OptVal nOpen Time OptVal nOpen Time
5 A 6063 2 0.28 6263 2 1.14 7304 1 3.78
B 8834 2 7.75 9034 2 126.13 10043 1 27.19
C 7866 2 49.66 8066 2 71.52 9073 1 88.28
D 8656 4 67.13 8946 2 26.63 10746 2 47.92
E 8034 4 72.95 8434 4 110.41 10684 2 875.80
F 7493 3 24.84 7695 2 54.56 9470 1 1931.36
G 6492 4 9.89 6818 2 10.33 8618 2 5.48
H 7857 4 93.55 8200 3 147.25 10068 2 56.70
I 7310 2 15.47 7510 2 46.09 8459 1 7.50
J 7319 2 22.30 7519 2 43.50 8717 1 89.69
Aver 2.90 36.38 2.30 63.75 1.40 313.37
10 A 6097 4 356.67 6497 4 422.48 10097 4 812.72
B 6732 5 67.17 7156 4 23.17 9341 2 42.89
C 6117 6 49.36 6482 3 63.39 8064 1 36.59
D 5914 7 10.38 6422 4 29.66 8434 2 6.44
E 6846 5 72.72 7280 3 39.77 8679 1 66.06
F 6479 7 42.36 7118 6 717.91 9390 2 1695.81
G 5920 7 5.31 6543 5 46.95 7554 1 3.23
H 6910 3 14.30 7210 3 7.36 8261 1 27.05
I 6642 4 35.08 7035 3 44.67 7946 1 4.84
J 6669 5 648.88 7071 2 160.67 8871 2 59.94
Aver 5.30 130.22 3.70 155.60 1.70 275.56

in the optimal solution when 𝑑𝑗 = 0, and only one when 𝑑𝑗 = 1000. Writing – review & editing. Inmaculada Rodríguez-Martín: Con-
Taking a look at the average results, we can say that, in general, for ceptualization, Methodology, Investigation, Software, Formal analysis,
a given |𝐽 |, when the facility opening costs increase so does the time Writing – original draft, Writing – review & editing.
needed to solve the problem, and that the problem gets harder when the
number of possible facility locations augments from 5 to 10. However,
Data availability
these statements are not always true for the individual instances, due
to their random nature.
I have included in the manuscript a link to the data used in the
experiments.
5. Conclusions

We have studied in this paper a new problem that combines aspects Acknowledgments
of two classical and complex routing problems: the Location and Rout-
ing Problem, and the One-commodity Pickup and Delivery Traveling This work has been partially supported by the Ministerio de Cien-
Salesman Problem. The result is a problem where there is a set of cia e Innovación de España (MCIN), Spain through project PID2019-
customers that supply or demand given amounts of a product, and a set 104928RB-I00.
of potential facility locations with associated opening costs. Customers
are served by capacitated vehicles that can perform routes only from
open facilities. The objective is to choose the facilities to open, to Appendix
assign the customers to the open facilities, and to design the vehicle
routes, in order to minimize the total cost. This problem may encounter In this appendix we show the results given by the branch-and-cut
practical applications in different fields, and, to our knowledge, it has algorithm on our whole set of benchmark instances. For each instance
not been addressed before in the literature as stated here. We present and each facility opening cost 𝑑𝑗 ∈ {0, 100, 1000}, Tables 6 to 10 display
two mathematical models for it, and design a branch-and-cut algorithm the characteristics of the instance (|𝐽 |, 𝑄, and Name), and the following
that is able to cope with benchmark instances with up to 100 nodes, as information:
the extensive computational experiments show.
• OptVal: Optimal solution value.
• nOpen: Number of opened facilities.
CRediT authorship contribution statement
• Time: Total computation time, in seconds.
• %-uncapa: Percentage difference between the optimal solution
Bencomo Domínguez-Martín: Conceptualization, Methodology, of the 1-PDLRP and the optimal solution of the uncapacitated
Investigation, Formal analysis, Writing – original draft, Writing – re- version of the problem.
view & editing. Hipólito Hernández-Pérez: Conceptualization, Method-
ology, Investigation, Software, Formal analysis, Writing – original draft, Blank cells in the tables indicate that the algorithm was unable to
Writing – review & editing. Jorge Riera-Ledesma: Conceptualization, find a feasible solution within the time limit of two hours.
Methodology, Investigation, Formal analysis, Writing – original draft,

10
B. Domínguez-Martín et al. Computers and Operations Research 161 (2024) 106426

Table 6
Results for instances with 𝑛 = 30 and different values of 𝑄, |𝐽 |, and 𝑑𝑗 .
|𝐽 | 𝑄 Name 𝑑𝑗 = 0 𝑑𝑗 = 100 𝑑𝑗 = 1000

OptVal nOpen Time %-uncapa OptVal nOpen Time %-uncapa OptVal nOpen Time %-uncapa

5 10 A 5324 4 0.50 21.88 5703 3 1.30 25.32 6736 1 1.56 23.41

B 5056 3 0.86 18.59 5356 3 0.78 21.28 7232 2 1.13 29.26
C 5120 3 0.89 20.72 5420 3 1.02 22.75 6390 1 0.42 20.39
D 5735 5 1.44 24.52 6120 3 0.73 27.63 8181 2 2.02 34.86
E 5386 4 0.38 16.84 5760 3 0.77 17.57 6998 1 0.91 19.29
F 5360 3 5.05 25.22 5660 3 1.75 24.17 7757 2 4.78 32.24
G 7825 3 9.28 44.17 8125 3 2.17 44.02 9291 1 1.25 41.36
H 5718 4 3.56 24.94 6118 4 2.75 27.13 9718 4 3.11 44.87
I 4752 4 0.66 14.60 4976 2 0.33 16.44 5946 1 0.64 14.93
J 5496 4 0.55 26.15 5896 4 1.11 29.46 7135 1 4.02 29.10

Aver 3.70 2.32 23.76 3.10 1.27 25.58 1.60 1.98 28.97

20 A 4453 2 0.52 6.60 4606 1 0.33 7.53 5506 1 0.23 6.30

B 4379 1 0.20 6.01 4479 1 0.06 5.87 5379 1 0.05 4.89
C 4177 2 0.27 2.82 4377 2 0.42 4.34 5301 1 0.27 4.04
D 4718 3 0.64 8.25 4827 1 0.47 8.25 5727 1 0.41 6.95
E 4537 3 0.05 1.28 4837 3 0.33 1.84 5802 1 0.23 2.65
F 4247 3 0.30 5.63 4483 1 0.17 4.26 5383 1 0.09 2.36
G 5468 3 0.64 20.10 5758 2 1.05 21.01 6849 1 0.69 20.46
H 4526 3 0.64 5.17 4800 2 0.53 7.13 6600 2 1.13 18.82
I 4058 1 0.02 0.00 4158 1 0.03 0.00 5058 1 0.02 0.00
J 4336 1 0.38 6.39 4436 1 0.08 6.24 5336 1 0.08 5.19

Aver 2.20 0.36 6.22 1.50 0.35 6.65 1.10 0.32 7.17

40 A 4159 1 0.08 0.00 4259 1 0.06 0.00 5159 1 0.05 0.00

B 4116 1 0.03 0.00 4216 1 0.02 0.00 5116 1 0.02 0.00
C 4059 2 0.08 0.00 4187 1 0.06 0.00 5087 1 0.09 0.00
D 4329 1 0.03 0.00 4429 1 0.02 0.00 5329 1 0.03 0.00
E 4479 3 0.03 0.00 4748 1 0.17 0.00 5648 1 0.08 0.00
F 4008 3 0.05 0.00 4308 3 0.20 0.37 5277 1 0.08 0.40
G 4515 2 0.14 3.23 4715 2 0.30 3.54 5697 1 0.11 4.37
H 4292 2 0.03 0.00 4466 1 0.05 0.18 5366 1 0.06 0.15
I 4058 1 0.03 0.00 4158 1 0.03 0.00 5058 1 0.03 0.00
J 4059 1 0.17 0.00 4159 1 0.14 0.00 5059 1 0.09 0.00

Aver 1.70 0.07 0.32 1.30 0.10 0.41 1.00 0.06 0.49

10 10 A 4434 3 1.52 15.45 4734 3 0.41 17.24 7434 3 0.20 35.19

B 4146 2 0.56 6.08 4346 2 0.22 6.99 6146 2 0.28 19.59
C 4693 4 0.78 18.41 5041 3 0.66 20.71 6728 1 1.47 27.21
D 4519 6 0.08 17.17 5040 4 0.66 23.75 7228 2 6.22 34.38
E 5190 5 0.77 22.85 5607 3 1.86 25.02 7745 2 2.89 31.57
F 4842 7 3.23 20.53 5370 5 5.78 22.76 8213 3 22.95 37.66
G 6377 5 1.14 37.60 6877 5 1.00 39.23 9173 2 1.34 42.78
H 4485 3 0.69 10.48 4785 3 0.23 14.00 7485 3 0.23 33.00
I 4139 5 1.22 6.62 4524 3 1.17 10.15 5612 1 0.58 10.44
J 4036 5 0.83 24.03 4438 4 2.08 24.16 5796 1 0.27 19.27

Aver 4.50 1.08 17.92 3.50 1.41 20.40 2.00 3.64 29.11

20 A 3928 3 1.77 4.56 4214 2 0.70 7.02 6014 2 1.06 19.89

B 3894 2 0.06 0.00 4042 1 0.20 0.00 4942 1 0.19 0.00
C 3892 2 0.36 1.62 4092 2 0.30 2.32 5360 1 1.16 8.64
D 4022 4 0.19 6.94 4252 1 0.53 9.62 5152 1 0.28 7.94
E 4244 2 0.30 5.66 4444 2 0.20 5.40 5557 1 0.09 4.62
F 4124 3 3.30 6.69 4424 3 2.38 6.24 6288 2 1.52 18.58
G 4599 3 0.23 13.48 4899 3 0.39 14.70 6203 1 0.84 15.38
H 4186 2 1.75 4.09 4386 2 1.45 6.18 6186 2 1.59 18.93
I 3865 2 0.05 0.00 4065 2 0.48 0.00 5026 1 0.39 0.00
J 3231 3 0.53 5.11 3531 3 0.47 4.67 4868 1 0.86 3.88

Aver 2.60 0.85 4.81 2.10 0.71 5.62 1.30 0.80 9.79

40 A 3749 2 0.11 0.00 3919 1 0.27 0.03 4819 1 0.14 0.02

B 3894 2 0.05 0.00 4042 1 0.22 0.00 4942 1 0.17 0.00
C 3829 2 0.06 0.00 3997 1 0.67 0.00 4897 1 0.64 0.00
D 3743 1 0.05 0.00 3843 1 0.09 0.00 4743 1 0.05 0.00
E 4004 2 0.05 0.00 4204 2 0.06 0.00 5300 1 0.33 0.00
F 3848 3 0.03 0.00 4148 3 0.31 0.00 5120 1 0.39 0.00
G 4029 2 0.39 1.24 4229 2 0.33 1.18 5264 1 0.16 0.28
H 4015 1 0.09 0.00 4115 1 0.08 0.00 5015 1 0.06 0.00
I 3865 2 0.05 0.00 4065 2 0.45 0.00 5026 1 0.38 0.00
J 3066 3 0.06 0.00 3366 3 0.30 0.00 4687 1 0.28 0.17

Aver 2.00 0.09 0.12 1.70 0.28 0.12 1.00 0.26 0.05

11
B. Domínguez-Martín et al. Computers and Operations Research 161 (2024) 106426

Table 7
Results for instances with 𝑛 = 40 and different values of 𝑄, |𝐽 |, and 𝑑𝑗 .
|𝐽 | 𝑄 Name 𝑑𝑗 = 0 𝑑𝑗 = 100 𝑑𝑗 = 1000

Sol nOpenF Time %-uncapa Sol nOpenF Time %-uncapa Sol nOpenF Time %-uncapa

5 10 A 6265 3 7.88 23.61 6565 3 6.06 25.58 7988 1 15.20 27.57

B 5619 3 8.23 16.78 5877 2 2.30 17.03 7130 1 11.30 17.97
C 7045 5 6.03 34.66 7305 2 14.34 35.62 8240 1 6.38 32.00
D 6008 1 6.30 15.78 6108 1 0.47 14.03 7008 1 4.55 12.23
E 6098 5 14.77 16.40 6554 4 41.83 20.20 8391 2 15.02 26.95
F 6389 5 9.44 27.06 6875 4 22.39 30.76 7934 1 15.47 28.66
G 6016 2 2.30 16.66 6216 2 11.06 16.38 7121 1 0.98 14.37
H 6232 4 4.75 24.63 6519 2 8.03 26.42 8319 2 4.72 31.52
I 6397 3 3.06 28.83 6697 3 7.31 28.48 8545 2 11.30 31.32
J 5758 3 22.81 20.88 6051 1 28.95 23.05 6951 1 1.83 20.07

Aver 3.40 8.56 22.53 2.40 14.28 23.75 1.30 8.67 24.26

20 A 4939 2 0.61 3.10 5139 2 0.63 4.92 6275 1 0.88 7.79

B 4774 2 0.05 2.05 4974 2 0.27 1.97 5967 1 0.34 1.98
C 5346 2 4.30 13.90 5523 1 2.41 14.85 6423 1 3.08 12.77
D 5164 3 0.28 2.01 5269 1 0.22 0.34 6169 1 0.08 0.29
E 5165 3 1.22 1.30 5374 2 1.56 2.68 6326 1 0.63 3.10
F 4912 2 0.69 5.13 5023 1 0.77 5.24 5923 1 0.44 4.44
G 5066 2 0.42 1.03 5266 2 0.34 1.29 6246 1 0.27 2.37
H 4866 2 1.17 3.47 5033 1 4.27 4.69 5933 1 0.42 3.98
I 4935 2 0.34 7.74 5135 2 0.13 6.72 6153 1 0.16 4.62
J 4673 1 0.06 2.50 4773 1 0.03 2.45 5673 1 0.06 2.06

Aver 2.10 0.91 4.22 1.50 1.06 4.51 1.00 0.63 4.34

40 A 4786 1 0.14 0.00 4886 1 0.14 0.00 5786 1 0.14 0.00

B 4676 2 0.05 0.00 4876 2 0.17 0.00 5849 1 0.09 0.00
C 4603 1 0.05 0.00 4703 1 0.06 0.00 5603 1 0.03 0.00
D 5060 2 0.06 0.00 5251 1 0.05 0.00 6151 1 0.06 0.00
E 5098 4 0.56 0.00 5236 1 0.33 0.11 6136 1 0.22 0.10
F 4660 1 0.11 0.00 4760 1 0.16 0.00 5660 1 0.14 0.00
G 5014 3 0.17 0.00 5198 1 0.27 0.00 6098 1 0.19 0.00
H 4697 1 0.11 0.00 4797 1 0.14 0.00 5697 1 0.13 0.00
I 4553 3 0.08 0.00 4796 2 0.22 0.13 5869 1 0.17 0.00
J 4556 1 0.05 0.00 4656 1 0.05 0.00 5556 1 0.05 0.00

Aver 1.90 0.14 0.00 1.20 0.16 0.02 1.00 0.12 0.01

10 10 A 5827 5 10.25 20.65 6192 3 3.38 23.71 7454 1 2.70 24.55

B 4749 4 1.95 10.42 5072 3 1.13 12.18 6645 1 6.91 16.79
C 5812 5 4.61 24.66 6270 4 4.38 28.56 7651 1 4.23 29.70
D 5660 6 11.50 17.39 6026 3 15.84 17.57 8726 3 1.95 32.62
E 5509 8 3.94 14.79 6000 3 9.02 18.43 7533 1 4.77 22.63
F 5549 4 1.72 23.19 5949 4 2.78 24.58 7966 2 6.69 32.19
G 5222 6 1.45 10.26 5615 2 1.70 11.42 6699 1 2.22 12.27
H 5462 7 3.61 18.53 5910 3 6.81 20.76 8610 3 4.47 35.16
I 5389 5 1.61 22.56 5836 4 4.06 23.85 6869 1 1.28 21.69
J 5140 4 5.88 15.31 5370 1 8.77 15.79 6270 1 2.41 13.52

Aver 5.40 4.65 17.78 3.00 5.79 19.69 1.50 3.76 24.11

20 A 4695 3 0.52 1.51 4972 1 1.05 4.99 5872 1 0.78 4.22

B 4254 2 0.03 0.00 4454 2 0.09 0.00 5638 1 0.31 1.93
C 4566 2 0.94 4.10 4766 2 2.56 6.02 6157 1 1.30 12.64
D 4821 3 0.94 3.01 5073 2 0.61 2.09 6873 2 0.48 14.45
E 4819 3 0.89 2.59 5090 2 0.50 3.85 6147 1 0.53 5.19
F 4294 3 0.08 0.75 4535 1 2.22 1.06 5435 1 0.55 0.61
G 4766 3 1.08 1.68 5000 2 0.67 0.52 5978 1 0.17 1.69
H 4591 3 0.05 3.07 4820 2 0.27 2.84 6620 2 0.36 15.66
I 4504 4 0.22 7.35 4743 1 0.44 6.30 5643 1 0.22 4.68
J 4403 4 0.39 1.14 4635 1 0.72 2.44 5535 1 0.53 2.04

Aver 3.00 0.51 2.52 1.60 0.91 3.01 1.20 0.52 6.31

40 A 4624 1 0.22 0.00 4724 1 0.27 0.00 5624 1 0.14 0.00

B 4254 2 0.06 0.00 4454 2 0.08 0.00 5529 1 0.25 0.00
C 4379 1 0.47 0.00 4479 1 0.27 0.00 5379 1 0.25 0.00
D 4676 3 0.06 0.00 4967 2 0.70 0.00 5884 1 0.36 0.07
E 4694 2 0.09 0.00 4894 2 0.20 0.00 5828 1 0.20 0.00
F 4262 3 0.08 0.00 4487 2 0.80 0.00 5402 1 0.56 0.00
G 4686 3 0.33 0.00 4974 2 0.45 0.00 5877 1 0.17 0.00
H 4450 3 0.06 0.00 4683 1 0.45 0.00 5583 1 0.42 0.00
I 4173 4 0.13 0.00 4444 2 0.42 0.00 5379 1 0.17 0.00
J 4353 4 0.72 0.00 4522 1 0.52 0.00 5422 1 0.36 0.00

Aver 2.60 0.22 0.00 1.60 0.42 0.00 1.00 0.29 0.01

12
B. Domínguez-Martín et al. Computers and Operations Research 161 (2024) 106426

Table 8
Results for instances with 𝑛 = 50 and different values of 𝑄, |𝐽 |, and 𝑑𝑗 .
|𝐽 | 𝑄 Name 𝑑𝑗 = 0 𝑑𝑗 = 100 𝑑𝑗 = 1000

OptVal nOpen Time %-uncapa OptVal nOpen Time %-uncapa OptVal nOpen Time %-uncapa

5 10 A 6063 2 0.28 10.42 6263 2 1.14 11.32 7304 1 3.78 11.64

B 8834 2 7.75 36.69 9034 2 126.13 36.44 10043 1 27.19 33.86
C 7866 2 49.66 34.88 8066 2 71.52 35.26 9073 1 88.28 32.53
D 8656 4 67.13 33.28 8946 2 26.63 33.22 10746 2 47.92 36.03
E 8034 4 72.95 30.45 8434 4 110.41 30.19 10684 2 875.80 35.90
F 7493 3 24.84 29.33 7695 2 54.56 29.89 9470 1 2131.36 33.53
G 6492 4 9.89 15.50 6818 2 10.33 17.53 8618 2 5.48 24.31
H 7857 4 93.55 30.41 8200 3 147.25 31.82 10068 2 56.70 35.53
I 7310 2 15.47 27.10 7510 2 46.09 26.80 8459 1 7.50 24.38
J 7319 2 22.30 25.39 7519 2 43.50 26.04 8717 1 89.69 25.88

Aver 2.90 36.38 27.34 2.30 63.75 27.85 1.40 333.37 29.36

20 A 5536 2 1.41 1.90 5713 1 0.47 2.78 6613 1 0.33 2.40

B 6504 3 2.08 14.01 6745 2 11.02 14.87 7669 1 11.02 13.39
C 6049 2 7.47 15.32 6221 1 22.94 16.06 7121 1 7.25 14.03
D 6482 3 2.86 10.91 6782 3 13.53 11.91 7818 1 26.16 12.07
E 5946 4 1.17 6.02 6292 2 2.27 6.42 7518 1 8.53 8.91
F 5591 3 1.36 5.29 5891 3 3.61 8.42 6946 1 6.94 9.37
G 5661 2 2.91 3.09 5861 2 1.61 4.06 6791 1 1.45 3.95
H 5930 3 6.11 7.79 6033 1 6.63 7.33 6933 1 1.16 6.38
I 5861 3 2.36 9.08 6014 1 8.00 8.60 6914 1 1.41 7.48
J 5596 2 4.14 2.41 5764 1 1.27 3.52 6664 1 0.58 3.05

Aver 2.70 3.19 7.58 1.70 7.13 8.40 1.00 6.48 8.10

40 A 5431 2 0.27 0.00 5554 1 0.22 0.00 6454 1 0.16 0.00

B 5625 2 0.27 0.57 5825 2 0.66 1.42 6777 1 0.84 1.99
C 5349 1 0.59 4.24 5449 1 0.80 4.17 6349 1 0.61 3.58
D 5921 3 0.58 2.47 6096 1 0.78 2.00 6996 1 0.47 1.74
E 5588 3 0.75 0.00 5888 3 0.44 0.00 6957 1 0.56 1.57
F 5296 1 0.19 0.02 5396 1 0.17 0.02 6296 1 0.13 0.02
G 5486 2 0.31 0.00 5623 1 0.27 0.00 6523 1 0.20 0.00
H 5468 3 0.63 0.00 5620 1 0.94 0.52 6520 1 0.67 0.44
I 5329 3 0.30 0.00 5497 1 3.28 0.00 6397 1 0.20 0.00
J 5461 1 0.48 0.00 5561 1 0.13 0.00 6461 1 0.13 0.00

Aver 2.10 0.44 0.73 1.30 0.77 0.81 1.00 0.40 0.93

10 10 A 6097 4 356.67 15.94 6497 4 422.48 19.58 10097 4 812.72 39.34

B 6732 5 67.17 23.63 7156 4 23.17 24.39 9341 2 42.89 32.39
C 6117 6 49.36 22.51 6482 3 63.39 23.08 8064 1 36.59 26.64
D 5914 7 10.38 9.22 6422 4 29.66 12.27 8434 2 6.44 22.31
E 6846 5 72.72 24.77 7280 3 39.77 25.43 8679 1 66.06 26.65
F 6479 7 42.36 22.29 7118 6 717.91 27.52 9332 2 1695.81 35.07
G 5920 7 5.31 12.25 6543 5 46.95 16.46 7554 1 3.23 15.73
H 6910 3 14.30 30.13 7210 3 7.36 31.51 8261 1 27.05 29.33
I 6642 4 35.08 27.97 7035 3 44.67 29.15 7946 1 4.84 24.89
J 6669 5 648.88 25.39 7071 2 160.67 28.21 8871 2 59.94 32.63

Aver 5.30 130.22 21.41 3.70 155.60 23.76 1.70 275.56 28.50

20 A 5311 4 10.53 3.50 5535 2 31.39 5.60 7335 2 13.06 16.50

B 5300 3 1.28 3.00 5600 3 1.34 3.38 6884 1 6.44 8.27
C 5045 3 2.77 6.05 5326 2 3.64 6.38 6618 1 9.58 10.61
D 5399 4 1.36 0.56 5743 2 3.05 1.90 6753 1 1.20 2.98
E 5473 5 1.83 5.90 5789 3 0.81 6.22 7009 1 8.22 9.17
F 5339 2 15.14 5.69 5535 1 2.16 6.79 6435 1 3.31 5.84
G 5275 5 6.75 1.52 5539 1 4.69 1.32 6439 1 0.81 1.13
H 5328 2 6.75 9.38 5466 1 1.09 9.66 6366 1 0.41 8.29
I 5211 4 2.00 8.19 5425 1 1.81 8.13 6325 1 1.25 5.64
J 5240 2 10.81 5.04 5380 1 35.25 5.65 6280 1 2.64 4.84

Aver 3.40 5.92 4.88 1.70 8.52 5.50 1.10 4.69 7.33

40 A 5146 2 0.19 0.41 5271 1 0.34 0.87 6171 1 0.44 0.75

B 5141 3 0.34 0.00 5411 2 0.55 0.00 6315 1 0.52 0.00
C 4783 3 0.58 0.90 4986 2 1.38 0.00 5916 1 0.45 0.00
D 5369 3 3.28 0.00 5634 2 3.80 0.00 6552 1 0.39 0.00
E 5150 3 0.61 0.00 5429 2 1.42 0.00 6407 1 0.50 0.64
F 5035 3 1.02 0.00 5159 1 4.34 0.00 6059 1 0.52 0.00
G 5195 6 0.63 0.00 5466 1 4.33 0.00 6366 1 0.58 0.00
H 4831 2 1.64 0.06 4938 1 0.97 0.00 5838 1 0.58 0.00
I 4784 2 0.34 0.00 4984 2 1.08 0.00 5968 1 0.31 0.00
J 4976 1 0.22 0.00 5076 1 0.17 0.00 5976 1 0.16 0.00

Aver 2.80 0.88 0.14 1.50 1.84 0.09 1.00 0.44 0.14

13
B. Domínguez-Martín et al. Computers and Operations Research 161 (2024) 106426

Table 9
Results for instances with 𝑛 = 60 and different values of 𝑄, |𝐽 |, and 𝑑𝑗 .
|𝐽 | 𝑄 Name 𝑑𝑗 = 0 𝑑𝑗 = 100 𝑑𝑗 = 1000

OptVal nOpen Time %-uncapa OptVal nOpen Time %-uncapa OptVal nOpen Time %-uncapa

5 10 A 7672 4 t.l. 23.10 8043 2 t.l. 24.07 9762 2 4260.53 28.21

B 8063 5 90.77 22.10 8322 2 33.31 23.32 9322 1 14.97 21.89
C 8661 2 526.14 28.89 8861 2 36.20 29.36 10661 2 68.86 32.85
D 10667 3 t.l. 39.91 11011 3 t.l. 40.88 12869 2 t.l. 42.42
E 9265 3 3213.39 32.92 9565 3 1733.70 33.02 11386 2 2987.11 35.82
F 8403 5 3780.36 30.92 8733 3 943.69 32.38 10129 1 1024.56 32.82
G 8279 3 3503.64 27.46 8579 3 1304.33 28.06 9706 1 1787.88 27.14
H 7498 5 153.73 22.25 7827 3 44.63 24.06 9707 2 6.66 29.49
I 8970 5 1375.88 36.71 9458 4 t.l. 37.75 12421 3 t.l. 45.33
J 8142 3 876.17 22.33 8261 1 1685.47 22.24 9161 1 29.31 20.05

Aver 3.80 2792.01 28.66 2.60 2738.13 29.51 1.70 2457.99 31.60

20 A 6336 3 25.23 6.88 6603 1 91.77 7.51 7503 1 19.45 6.60

B 6393 3 2.73 1.75 6648 2 30.44 4.02 7554 1 2.78 3.61
C 6668 2 1.73 7.63 6868 2 38.33 8.87 7850 1 9.47 8.80
D 7571 4 20.08 15.33 7784 1 61.77 16.37 8684 1 26.63 14.67
E 7182 3 145.55 13.46 7336 1 117.33 12.66 8236 1 42.05 11.28
F 6355 4 25.94 8.65 6589 2 73.30 10.38 7753 1 11.63 12.23
G 6448 3 3.80 6.85 6651 2 23.13 7.20 7684 1 5.94 7.96
H 6134 3 9.86 4.96 6404 2 5.91 7.18 7321 1 1.59 6.52
I 6796 4 2868.84 16.47 7105 2 2829.42 17.13 8905 2 513.88 23.74
J 6519 2 1.00 2.99 6719 2 13.08 4.39 7774 1 3.36 5.79

Aver 3.10 310.48 8.50 1.70 328.45 9.57 1.10 63.68 10.12

40 A 5924 4 0.97 0.41 6110 1 0.98 0.05 7010 1 0.73 0.03

B 6281 1 0.78 0.00 6381 1 0.67 0.00 7281 1 0.64 0.00
C 6171 1 1.80 0.19 6271 1 0.30 0.19 7171 1 0.33 0.17
D 6489 1 5.09 1.22 6589 1 0.72 1.20 7489 1 0.47 1.05
E 6215 2 0.08 0.00 6415 2 0.33 0.12 7337 1 0.56 0.41
F 5833 2 0.56 0.48 6033 2 2.05 2.12 6962 1 1.06 2.26
G 6066 4 0.81 0.99 6276 1 1.45 1.66 7176 1 0.61 1.45
H 5830 2 0.48 0.00 5945 1 0.36 0.02 6845 1 0.39 0.01
I 5823 3 40.13 2.51 6029 2 32.06 2.34 6994 1 4.92 2.90
J 6331 1 0.30 0.11 6431 1 0.31 0.11 7331 1 0.30 0.10

Aver 2.10 5.10 0.59 1.30 3.92 0.78 1.00 1.00 0.84

10 10 A 7122 6 2437.88 19.60 7512 2 4139.16 21.15 9312 2 113.48 26.73

B 7515 8 821.81 20.51 8166 5 3142.80 25.01 10272 2 563.03 31.62
C 7815 5 t.l. 23.22 8181 4 2142.89 24.83 9830 1 82.88 28.28
D 9733 6 t.l. 37.33 10264 4 t.l. 39.59 12031 2 5827.72 40.99
E 8065 6 4372.83 26.00 8594 5 235.11 27.19 9781 1 449.89 26.83
F 7459 8 1332.86 28.81 8002 5 163.25 31.14 11574 3 t.l. 44.24
G 7751 7 1344.14 26.02 8288 4 496.64 27.86 9949 1 287.81 30.75
H 7207 8 518.86 22.17 7823 5 t.l. 26.95 10094 2 1389.72 34.47
I 7334 7 t.l. 27.69 7966 6 7181.78 30.02 10459 2 3738.94 37.15
J 7374 6 56.81 19.70 7890 3 43.30 23.69 9037 1 36.53 23.41

Aver 6.70 3248.52 25.11 4.30 3194.49 27.74 1.70 1969.00 32.45

20 A 6047 4 29.61 5.31 6383 2 180.53 7.21 7365 1 31.56 7.36

B 6313 5 695.67 5.37 6607 2 55.42 7.31 7594 1 17.72 7.51
C 6441 4 2806.56 6.85 6683 2 15.67 7.98 7880 1 35.17 10.53
D 7272 3 610.27 16.12 7419 1 188.61 16.43 8319 1 24.64 14.65
E 6421 5 15.36 7.05 6780 1 35.28 7.71 7680 1 1.47 6.81
F 5899 5 192.80 9.98 6254 3 16.47 11.90 8246 2 43.50 21.73
G 6168 5 8.11 7.04 6485 2 23.78 7.80 7587 1 11.88 9.19
H 5912 5 86.09 5.13 6201 2 17.17 7.84 7282 1 9.63 9.16
I 5772 6 112.63 8.13 6354 5 527.17 12.26 7466 1 19.84 11.96
J 6189 2 45.39 4.33 6389 2 14.39 5.76 7381 1 3.56 6.23

Aver 4.40 460.25 7.53 2.20 107.45 9.22 1.10 19.90 10.51

40 A 5764 5 2.91 0.66 5923 1 2.23 0.00 6823 1 1.55 0.00

B 6002 3 4.06 0.47 6124 1 13.22 0.00 7024 1 2.11 0.00
C 6054 2 2.56 0.89 6178 1 0.81 0.45 7078 1 0.61 0.40
D 6169 1 1.14 1.12 6269 1 0.67 1.10 7169 1 0.41 0.96
E 6001 4 1.86 0.55 6270 1 4.50 0.21 7170 1 0.58 0.18
F 5427 3 34.53 2.16 5704 2 3.64 3.40 6620 1 2.09 2.51
G 5809 3 1.08 1.29 5990 1 4.77 0.18 6890 1 0.63 0.00
H 5609 2 1.31 0.00 5718 1 2.33 0.05 6618 1 0.38 0.05
I 5382 4 21.02 1.47 5661 2 5.22 1.52 6665 1 1.56 1.38
J 5921 1 0.42 0.00 6021 1 0.38 0.00 6921 1 0.31 0.00

Aver 2.80 7.09 0.86 1.20 3.78 0.69 1.00 1.02 0.55

14
B. Domínguez-Martín et al. Computers and Operations Research 161 (2024) 106426

Table 10
Results for instances with 𝑛 = 100 and different values of 𝑄, |𝐽 |, and 𝑑𝑗 .
|𝐽 | 𝑄 Name 𝑑𝑗 = 0 𝑑𝑗 = 100 𝑑𝑗 = 1000

OptVal nOpen Time %-uncapa OptVal nOpen Time %-uncapa OptVal nOpen Time %-uncapa

5 10 A 11900 3 t.l. 35.95 13471 1 t.l. 35.61

B 17614 3 t.l. 56.10
C 13075 5 t.l. 40.17 22724 1 t.l. 61.14
D 16488 4 t.l. 52.06
E 11537 5 t.l. 33.29 11810 3 t.l. 32.95 14599 3 t.l. 39.59
F 19040 2 t.l. 58.92 13976 2 t.l. 37.60
G 12967 1 t.l. 33.62
H 19284 4 t.l. 59.62 13932 1 t.l. 37.65
I 18159 4 t.l. 56.78 14330 1 t.l. 38.95
J 15644 4 t.l. 51.87 13937 2 t.l. 38.51

Aver 4.25 t.l. 40.32 3.33 t.l. 52.74 1.50 t.l. 40.33

20 A 8389 3 357.38 9.14 8628 1 t.l. 9.90 9440 1 64.23 8.11

B 11692 4 t.l. 34.72 12959 3 t.l. 40.33 10458 1 t.l. 17.45
C 8747 4 1923.61 10.56 9150 3 t.l. 13.33 10243 1 3158.05 13.79
D 9291 3 3145.89 16.26 9591 3 2978.23 17.59 11454 2 3756.92 23.14
E 8443 3 6300.08 8.85 9567 3 t.l. 17.23 10468 2 4584.31 15.75
F 8496 2 t.l. 9.85 9008 1 t.l. 13.18 9475 1 1381.14 7.96
G 8636 2 t.l. 11.90 8600 1 2260.05 10.37 9500 1 558.06 9.39
H 9598 3 t.l. 19.91 12183 1 t.l. 28.70
I 9219 5 3913.44 15.95 9735 3 t.l. 19.37 10783 1 t.l. 18.86
J 8652 3 t.l. 12.98 9509 2 t.l. 19.34 10347 1 t.l. 17.17

Aver 3.20 5164.04 15.01 2.22 6182.03 17.85 1.20 4230.27 16.03

40 A 7847 1 37.09 2.87 7947 1 24.69 2.18 8847 1 14.33 1.96

B 7903 3 1296.03 3.42 8053 1 865.34 3.97 8953 1 523.47 3.57
C 7932 3 45.86 1.37 8137 1 94.45 2.54 9037 1 44.95 2.29
D 7995 4 59.36 2.69 8127 1 19.16 2.74 9027 1 15.17 2.47
E 7846 4 146.44 1.91 7982 1 759.25 0.79 8882 1 16.06 0.71
F 7799 2 101.69 1.80 7949 1 701.98 1.61 8849 1 56.11 1.45
G 7688 2 7.81 1.04 7872 1 46.06 2.08 8772 1 13.78 1.87
H 7855 2 8.41 2.14 8005 1 56.05 2.72 8905 1 70.19 2.45
I 8038 3 41.44 3.60 8240 2 298.80 4.75 9189 1 96.11 4.79
J 7863 3 530.28 4.25 8066 2 624.39 4.91 9043 1 553.80 5.23

Aver 2.70 227.44 2.51 1.20 349.02 2.83 1.00 140.40 2.68

10 10 A
B 15844 3 t.l. 46.50
C 17315 4 t.l. 56.04 30281 4 t.l. 71.89
D
E 14211 7 t.l. 47.91 11411 5 t.l. 31.85 14353 1 t.l. 39.55
F 11918 4 t.l. 36.19 14124 1 t.l. 39.72
G 20359 8 t.l. 63.76 13382 1 t.l. 37.12
H 13544 1 t.l. 37.39
I 18162 8 t.l. 58.58 21816 5 t.l. 64.92 15709 3 t.l. 45.56
J 14023 5 t.l. 46.32

Aver 7.67 t.l. 56.75 4.60 t.l. 47.06 2.00 t.l. 45.39

20 A 8275 2 t.l. 9.51 9092 1 t.l. 16.01 9359 1 6059.66 8.79

B 9143 6 t.l. 18.72 9006 4 t.l. 15.88 10583 2 t.l. 19.91
C 8359 4 t.l. 10.14 8743 4 712.84 12.95 9941 1 3204.11 14.38
D 9357 6 t.l. 19.02
E 7965 6 5107.83 7.06 8346 2 637.89 6.82 9525 1 1304.72 8.90
F 8155 5 t.l. 9.71 9329 1 t.l. 8.74
G 8140 3 2373.06 9.35 8443 3 t.l. 10.99 9619 1 t.l. 12.52
H 10418 4 t.l. 28.32 9974 1 t.l. 14.98
I 8958 5 t.l. 16.03 9404 4 t.l. 18.63 11227 2 t.l. 23.83
J 11196 5 t.l. 34.47 9141 3 t.l. 17.65 10762 2 t.l. 21.69

Aver 4.60 6508.09 16.23 3.00 5335.82 14.13 1.33 5974.28 14.86

40 A 7664 3 39.17 2.30 7818 1 45.50 2.33 8718 1 21.97 2.09

B 7537 6 954.70 1.41 7782 1 437.45 2.65 8682 1 94.13 2.37
C 7669 3 53.31 2.06 7808 1 25.72 2.52 8708 1 20.78 2.26
D 7688 6 22.52 1.44 8016 2 1219.55 2.57 8923 1 22.13 2.39
E 7470 7 47.88 0.90 7807 2 510.14 0.38 8796 1 96.13 1.35
F 7394 3 2.94 0.42 7636 2 163.89 0.41 8559 1 26.56 0.53
G 7461 4 34.05 1.10 7665 1 103.22 1.96 8565 1 19.34 1.75
H 7578 3 53.86 1.45 7798 1 88.86 2.80 8698 1 59.03 2.51
I 7772 4 53.67 3.22 8035 1 205.39 4.77 8935 1 10.64 4.29
J 7630 2 521.91 3.84 7763 1 48.44 3.03 8663 1 9.22 2.71

Aver 4.10 178.40 1.81 1.30 284.82 2.34 1.00 37.99 2.22

References Chemla, D., Meunier, F., Wolfler-Calvo, R., 2013. Bike sharing systems: Solving the
static rebalancing problem. Discrete Optim. 10 (2), 120–146.

Albareda-Sambola, M., Díaz, J., Fernández, E., 2005. A compact model and tight bounds Cruz, F., Subramanian, A., Bruck, B.P., Iori, M., 2017. A heuristic algorithm for a single
for a combined location-routing problem. Comput. Oper. Res. 32, 407–428. vehicle static bike sharing rebalancing problem. Comput. Oper. Res. 79, 19–33.
Almeida, F.L., Terra, B.M., Dias, P.A., Gonçalves, G.M., 2010. Transport with automatic Dell’Amico, M., Hadjicostantinou, E., Iori, M., Novellani, S., 2014. The bike sharing
guided vehicles in the factory of the future. In: 2010 IEEE 15th Conference on rebalancing problem: Mathematical formulations and benchmark instances. Omega
Emerging Technologies & Factory Automation. ETFA 2010, IEEE, pp. 1–4. 45, 7–19.
Battarra, M., Cordeau, J.F., Iori, M., 2014. Chapter 6: pickup-and-delivery problems for Erdoğan, G., Battarra, M., Wolfler-Calvo, R., 2015. An exact algorithm for the static
goods transportation. In: Vehicle Routing: Problems, Methods, and Applications, rebalancing problem arising in bicycle sharing systems. European J. Oper. Res. 245,
second ed. SIAM, pp. 161–191. 667–679.
Berbeglia, G., Cordeau, J.F., Gribkovskaia, I., Laporte, G., 2007. Static pickup and Gouveia, L., 1995. A result on projection for the vehicle routing problem. European J.
delivery problems: a classification scheme and survey. TOP 15, 1–31. Oper. Res. 85 (3), 610–624.

15
B. Domínguez-Martín et al. Computers and Operations Research 161 (2024) 106426

Hernández-Pérez, H., Landete, M., Rodríguez-Martín, I., 2021. The single-vehicle two- Labbé, M., Rodríguez-Martín, I., Salazar-González, J.-J., 2004. A branch-and-price
echelon one-commodity pickup and delivery problem. Comput. Oper. Res. 127, algorithm for the plant-cycle location problem. J. Oper. Res. Soc. 55, 513–520.
105152. Mladenović, N., Urošević, D., Hanafi, S., Ilić, A., 2012. A general variable neighborhood
Hernández-Pérez, H., Rodríguez-Martín, I., Salazar-González, J.-J., 2009. A hy- search for the one-commodity pickup-and-delivery travelling salesman problem.
brid GRASP/VND heuristic for the one-commodity pickup-and-delivery traveling European J. Oper. Res. 220, 270–285.
salesman problem. Comput. Oper. Res. 36, 1639–1645. Mosheiov, G., 1994. The travelling salesman problem with pick-up and delivery.
Hernández-Pérez, H., Salazar-González, J.-J., 2004a. A branch-and-cut algorithm for a European J. Oper. Res. 79 (2), 299–310.
traveling salesman problem with pickup and delivery. Discrete Appl. Math. 145, Nagy, G., Salhi, S., 2007. Location-routing: Issues, models and methods. European J.
126–139. Oper. Res. 177 (2), 649–672.
Hernández-Pérez, H., Salazar-González, J.-J., 2004b. Heuristics for the one-commodity Padberg, M.W., Rinaldi, G., 1991. A branch-and-cut algorithm for the resolution of
pickup-and-delivery traveling salesman problem. Transp. Sci. 38, 245–255. large-scale symmetric traveling salesman problems. SIAM Rev. 33, 60–100.
Hernández-Pérez, H., Salazar-González, J.-J., 2007. The one-commodity pickup-and- Parragh, S., Doerner, K., Hartl, R., 2008a. A survey on pickup and delivery problems:
delivery traveling salesman problem: Inequalities and algorithms. Networks 50 (4), Part I: Transportation between customers and depot. J. Betriebswirtschaft 58,
258–272. 21–51.
Hernández-Pérez, H., Salazar-González, J.-J., 2022. A branch-and-cut algorithm for Parragh, S., Doerner, K., Hartl, R., 2008b. A survey on pickup and delivery problems:
the split-demand one-commodity pickup-and-delivery travelling salesman problem. Part II: Transportation between pickup and delivery locations. J. Betriebswirtschaft
European J. Oper. Res. 297 (2), 467–483. 58, 81–117.
Karaoglan, I., Altiparmak, F., Kara, I., Dengiz, B., 2011. A branch and cut algorithm Prodhon, C., Prins, C., 2014. A survey of recent research on location-routing problems.
for the location-routing problem with simultaneous pickup and delivery. European European J. Oper. Res. 238 (1), 1–17.
J. Oper. Res. 211 (2), 318–332. Rodríguez-Martín, I., Salazar-González, J.-J., Yaman, H., 2014. A branch-and-cut
Karaoglan, I., Altiparmak, F., Kara, I., Dengiz, B., 2012. The location-routing problem algorithm for the hub location and routing problem. Comput. Oper. Res. 50,
with simultaneous pickup and delivery: Formulations and a heuristic approach. 161–174.
Omega 40 (4), 465–477. Salazar-González, J.J., Santos-Hernández, B., 2015. The split-demand one-commodity
Kim, S., Lim, G., Cho, J., Côté, M., 2017. Drone-aided healthcare services for patients pickup-and-delivery travelling salesman problem. Transp. Res. B 75, 58–73.
with chronic diseases in rural areas. J. Intell. Robot. Syst. 88, 163–180. Zhao, F., Li, S., Sun, J., Mei, D., 2009. Genetic algorithm for the one-commodity
Koç, Ç., Laporte, G., Tükenmez, İ., 2020. A review of vehicle routing with simultaneous pickup-and-delivery traveling salesman problem. Comput. Ind. Eng. 56 (4),
pickup and delivery. Comput. Oper. Res. 122, 104987. 1642–1648.

16