Computers & Operations Research 38 (2011) 16781695

Contents lists available at ScienceDirect

Computers & Operations Research

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

The vendor location problem

Yuce
C
- nar, Hande Yaman 
Bilkent University, Department of Industrial Engineering, 06800 Ankara, Turkey

a r t i c l e i n f o

a b s t r a c t

Available online 4 March 2011

The vendor location problem is the problem of locating a given number of vendors and determining the
number of vehicles and the service zones necessary for each vendor to achieve at least a given prot.
We consider two versions of the problem with different objectives: maximizing the total prot and
maximizing the demand covered. The demand and prot generated by a demand point are functions of
the distance to the vendor. We propose integer programming models for both versions of the vendor
location problem. We then prove that both are strongly NP-hard and we derive several families of valid
inequalities to strengthen our formulations. We report the outcomes of a computational study where
we investigate the effect of valid inequalities in reducing the duality gaps and the solution times for the
vendor location problem.
& 2011 Elsevier Ltd. All rights reserved.

Keywords:
Location
Vendor location problem
Hierarchical facility location
Valid inequalities
Computational complexity

1. Introduction
With a major beverage company about to launch its own
brand for demijohn water, we recently worked on the following
discrete facility location problem.
Unlike drinks sold in regular bottles, demijohn water has the
distinctive feature of making it hard for customers to switch
brands; every brand has its own containers and customers pay for
the rst container, replacing it when empty with a full one. In this
way, the customer then continues to only pay for the contents of
the bottles; switching brands would mean they would have to
pay for a full bottle again. Suppliers of bottled gas for cooking and
heating purposes also benet from this quasimonopoly once the
customer has made her choice of brands.
Water sold in large containers is the rule rather than the
exception in Turkey: in 2008, 80% of consumption was demijohn
water and the remaining 20% was water bottled in smaller
containers. And the market itself is large: about 8.5 billion liters
per year according to the Association of Packaged Water Producers in Turkey (SUDER [27]) and still expected to grow (by 10%
in 2009).
A recent marketing survey carried out by the beverage company shows that customers value the quality of the water (taste,
hygiene, chemical composition, etc.) and the quality of the service
the most. The quality of the service is strongly related to service
times and the satisfaction is affected by the presence of competitors in the same region who could provide shorter service times.

 Corresponding author. Tel.: 90 312 290 27 68; fax: 90 312 266 40 54.

E-mail addresses: yuce@bilkent.edu.tr (Y. C

- nar),
hyaman@bilkent.edu.tr (H. Yaman).
0305-0548/$ - see front matter & 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.cor.2011.02.011

The number of potential customers in a given region mainly

depends on the distance to the assigned vendor and on the
proximity of competitors. This explains why selling in many
locations could increase the market share. This strategy, however,
has a price: some vendors may not reach a given prot. The
beverage company wanted to ensure that each vendor would earn
enough money and that the company would maximize its
market share.
Inspired by this real-life problem, we dene the vendor location
problem (VLP) as follows. We are given a set of demand points
corresponding to population zones and a set of possible locations
for vendors. Each vendor can only use a given number of vehicles.
We also know the (xed) cost of a vendor ofce (rent, insurance,
salaries of employees at ofce, etc.) at a given location as well as
the cost (including the salary of the driver) and capacity of a
vehicle.
For a given demand point, there is a set of eligible vendors.
Each demand point has a potential demand. The market share
that our company can have depends on the travel times of its
vendors and the proximity of competitors. The prot (sales
revenue minus the transportation cost) therefore depends on
the vendor that serves a demand point.
The VLP is the problem of locating a given number of vendors
and assigning each demand point to at most one vehicle of an
eligible vendor such that capacities of vehicles are not exceeded
and each vendor achieves at least a determined prot. We
consider two objective functions. In ProtVLP, the aim is to
maximize the total prot and in CoverageVLP, the aim is to
maximize the coverage, i.e., the total demand served.
Our problem can be seen as a hierarchical facility location
problem where demand points are in level 0, vehicles are level 1
facilities, and vendors are level 2 facilities. Sahin and Sural [28]

Y. C
- nar, H. Yaman / Computers & Operations Research 38 (2011) 16781695

review hierarchical facility location models and propose a classication scheme. The rst attribute in this scheme is ow pattern.
In a single ow pattern, the ow starts from level 0 and ends at
the highest level by passing through all intermediate levels. In a
multiple ow pattern, ows can travel from any lower level to
any higher level. Our problem has a single ow pattern in the
opposite direction. The second attribute is service varieties. Here
in a nested system, a higher level facility provides all services
provided by a lower level facility; in a non-nested system,
facilities in different levels provide different services. Our system
is a non-nested system. As the third attribute, the authors
consider the spatial conguration. In a coherent system, all
demand that is served by a given lower level facility is served
by the same higher level facility. Since in our system, each vehicle
belongs to a vendor, we have a coherent system. The nal
attribute is the objective. Here the authors consider the three
common objectives: median, covering, and xed charge. ProtVLP
can be considered as a median type problem even though we
maximize prot rather than minimize cost. CoverageVLP is a
maximum covering type problem.
Multi-level facility location problems have been previously
studied by many researchers. Aardal et al. [2] propose some facet
dening and valid inequalities for the polytope associated with
the two level uncapacitated facility location problem. Approximation algorithms are studied by Aardal et al. [1], Ageev [3], Ageev
et al. [4], Bumb [9], Bumb and Kern [10], Gabor and van
Ommeren [14], Guha et al. [16], Meyerson et al. [22], Shmoys
et al. [26], Zhang [32], and Zhang and Ye [33]. Branch and bound
algorithms are given by Kaufman et al. [18], Ro and Tcha [25],
Tcha and Lee [29], and Tragantalerngsak et al. [31]. Barros and
Labbe [7] present various formulations, a Lagrangean relaxation,
and a primal heuristic. Gao and Robinson [12,13] propose dualbased solution procedures. Chardaire et al. [11] present two
formulations, valid inequalities, a Lagrangian relaxation, and a
simulated annealing algorithm. Linear and Lagrangian relaxations
are studied by Bloemhof-Ruwaard et al. [8], Marn [20], Marn and
Pelegrn [21], Pirkul and Jayaraman [24], Tragantalerngsak
et al. [30] for different versions of the problem.
A recent work that is closely related to ours is on the capacity
and distance constrained plant location problem by AlbaredaSambola et al. [5]. In this problem, a set of possible locations is
given. A facility may house a number of identical vehicles. Each
demand point must be assigned to a single vehicle of a facility.
There are capacity restrictions for facilities and restrictions on the
total distance traveled for vehicles. The aim is to determine where
to open facilities, to decide on the number of vehicles for each
facility, and to assign the demand points to vehicles and facilities
with the aim of minimizing the costs of opening facilities, using
vehicles, and assigning demand points to facilities and vehicles.
The authors provide different models and a tabu search algorithm
for this problem. This study is similar to ours in that it is
concerned with assigning demand points to facility vehicles. It
is different from ours in that it has capacity constraints for
facilities and restrictions on the total distance traveled for
vehicles; we have capacity constraints for vehicles and minimum
prot constraints for facilities.
In this paper, we introduce two new two-level facility location
problems, namely ProtVLP and CoverageVLP, which are motivated
by a real life problem. Different from the classical facility location
problems, here we have minimum prot constraints for open
facilities and capacity constraints for their vehicles. We investigate the computational complexity of these problems and prove
that they are strongly NP-hard. We propose integer programming
formulations, valid inequalities, and extra constraints to be able
to use the cutting planes of off-the-shelf integer programming
solvers. We report the outcomes of a computational study where

1679

we use four types of instances that differ in their demand and

prot functions. We investigate the effect of valid inequalities on
linear programming relaxation bounds and solution times for
these different types of instances. Finally, we analyze the optimal
solutions of ProtVLP and CoverageVLP and report how the
differences in demand and prot functions effect the service
regions for an example problem. Hence, the contributions of the
paper are two new facility location problems motivated by a real
life problem, resolution of the status of their computational
complexity, and strong mixed integer programming formulations
for these problems.
The paper is organized as follows. In Section 2, we present
integer programming formulations for ProtVLP and CoverageVLP
and prove that both problems are strongly NP-hard. We propose
some valid inequalities in Section 3. Computational results are
given in Section 4. We analyze the solutions of ProtVLP and
CoverageVLP for two different types of instances in Section 5.
In Section 6, we conclude the paper.

2. Formulations and complexity

In this section, we rst introduce the notation and then
present formulations for ProtVLP and CoverageVLP. Then we
prove that both ProtVLP and CoverageVLP are strongly NP-hard.
Let I be the set of demand points and J be the set of possible
locations for vendors. For a demand point i A I, Ji is the set of
vendors that can serve i. In our problem, we dene Ji to be the set
of vendors whose travel time to i does not exceed a given bound.
We also dene Ij fi A I : j A Ji g for j A J.
We denote with fj the xed cost of the vendor ofce and with
vj the xed cost of a vehicle for a vendor located at j A J. We
assume that these cost values are non-negative. We dene rmin to
be the minimum prot a vendor should achieve.
We denote with p the number of vendors to be located. The
vendor at location j A J may have up to kmax
vehicles. Let
j
Kj f1, . . . ,kmax
g for j A J. The capacity of a vehicle is equal to g.
j
Demand point iA I has demand qij and generates prot rij if it
is served by the vendor at location j A Ji . We assume that qijs are
positive and that rij s are non-negative.
We dene the following decision variables. For i A I, j A Ji , and
kA Kj , xijk is 1 if demand point i is assigned to vehicle k of vendor j
and 0 otherwise, for j A J, and k A Kj , zjk is 1 if vendor j uses vehicle
k and 0 otherwise, and nally, for j A J, yj is 1 if a vendor is located
at location j and 0 otherwise.
Using these variables, the ProtVLP can be modeled as follows:
XX X
XX
X
max
rij xijk 
vj zjk  fj yj
1
i A I j A Ji k A K j

XX

s:t:

j A J k A Kj

xijk r1

jAJ

8i A I

j A Ji k A K j

X
yj p

jAJ

xijk ryj

8iA I, j A Ji

k A Kj

X
qij xijk r gzjk

8j A J, k A Kj

i A Ij

i A Ij

k A Kj

rij

xijk Z

vj zjk rmin fj yj

8j A J

k A Kj

xijk A f0,1g

8i A I, j A Ji , kA Kj

zjk A f0,1g

8j A J, k A Kj

Y. C
- nar, H. Yaman / Computers & Operations Research 38 (2011) 16781695

1680

yj A f0,1g

8j A J

Constraints (2) ensure that a demand point is assigned to at

most one vehicle of one eligible vendor. Constraint (3) states that
the number of vendors to be located is p. If a vendor is not located
at a given location, then a demand point cannot be served by any
of its vehicles due to constraints (4). Constraints (5) are capacity
constraints for vehicles. At the same time, they ensure that
demand points are not assigned to vehicles that are not in use.
Constraints (6) ensure that each vendor makes a prot of at least
rmin units. Constraints (7)(9) state that the variables are binary.
Objective function (1) is the total prot of all vendors.
Note here that constraints zjk r yj for j A J and k A Kj are not
included in the model. Let j A J and k A Kj . If there exists iA Ij with
xijk 1, then constraints (4) force yj to one and constraints (5)
force zjk to one. On the other hand, if xijk 0 for all i A Ij , then there
exists an optimal solution with zjk 0 since vjs are non-negative.
Hence constraints zjk r yj for j A J and k A Kj are not necessary for
the validity of the model. We do not include them in the model
not to increase the number of constraints. Later, we use them as
valid inequalities and test their performance.
The CoverageVLP can be modeled as follows:
XX X
max
qij xijk
i A I j A Ji k A Kj

s:t:

229

10

convert some structures in our problem into knapsack structures

so that we can use the lifted cover inequalities of off-the-shelf
integer programming solvers.
3.1. Lower bounds on the number of vehicles
Albareda-Sambola et al. [5] propose the optimality cuts
for j A J. These inequalities imply that if a vendor is
open then it should use at least one vehicle. In our problem, since
we have minimum prot constraints, in some cases we can obtain
tighter bounds on the number of vehicles to be used by a vendor.
Note that the resulting inequalities are valid inequalities for our
problem rather than optimality cuts.
For j A J and a positive integer m, consider the following
problem:
P

k A Kj zjk Z yj

dj m max

m
XX

rij aik 

i A Ij k 1
m
X

s:t:

m
X

vj bk fj

11

k1

aik r 1 8i A Ij

12

k1

X
qij aik r gbk

8k 1, . . . ,m

13

i A Ij

Here the objective function (10) is the total demand served.

To conclude this section, we investigate the computational
complexity of problems ProtVLP and CoverageVLP.

aik A f0,1g 8i A Ij , k 1, . . . ,m

14

bk A f0,1g 8k 1, . . . ,m

15

Theorem 1. ProtVLP and CoverageVLP are strongly NP-hard.

Here, the variable bk takes value 1 if vehicle k 1, . . . ,m is used

and takes value 0 otherwise, and the variable aik takes value 1 if
demand point i A Ij is assigned to vehicle k 1, . . . ,m and takes
value 0 otherwise. Constraints (12) ensure that each demand
point is assigned to at most one vehicle and constraints (13)
ensure that the sum of demands of demand points assigned to a
given vehicle does not exceed the capacity of the vehicle if the
vehicle is in use and no demand points are assigned to this vehicle
if it is not in use. The objective function is equal to the sum of
prots of demand points that are assigned to some vehicle minus
the sum of costs of using vehicles and the vendor ofce j.
This problem hence maximizes the total prot for vendor j if
vendor j can use at most m vehicles. Let mj be the smallest integer
with dj mj Z rmin . Then for vendor j to achieve a minimum level
of prot of rmin units, it should have at least mj vehicles. If mj is a
positive integer less than or equal to kmax
, then the inequality
j
P
k A Kj zjk Z mj yj is a valid inequality. If mj does not exist or if

Proof. We prove that the decision versions of ProtVLP and

CoverageVLP are NP-complete in the strong sense by a reduction
from the decision version of the bin packing problem.
Given a nite set of items U, a size si A Z for each i A U, a
positive integer bin capacity B, and a positive integer k, the
decision version of the bin packing problem is dened as follows.
P
Is there a partition of set U into U1 , . . . ,Uk such that i A Uu si r B
for all u 1, . . . , k? This problem is NP-complete in the strong
sense (see problem [SR1] in Garey and Johnson [15]).
First note that when vj fj 0 for all j A J and rij qij for all i A I
and j A Ji , problems ProtVLP and CoverageVLP become the same
problem. Hence in the remaining part of the proof, we only
consider CoverageVLP with vj fj 0 for all j A J and rij qij for
all i A I and j A Ji .
We dene the decision version of CoverageVLP as follows. Given
the parameters of the problem and a positive constant F, does
there exist a feasible solution with coverage at least F? This
problem is in NP.
Given an instance of the bin packing problem, let J be a
k,
singleton, I I1 U, p 1, v1 0, f1 0, rmin 0, kmax
1
P
ri1 qi1 si for i A I, g B, and F i A I qi1 . Now there exists a
solution to the decision version of the bin packing problem if and
only if there exists a solution to the decision version of CoverageVLP. Hence, the decision version of CoverageVLP is NP-complete in the strong sense. &

, then vendor j cannot be protable. Hence we can set

mj 4 kmax
j
yj 0.
The above problem is a capacitated facility location problem
with single sourcing, which is an NP-hard problem (see, e.g.,
Neebe and Rao [23], Barcelo and Casanovas [6], Klincewicz and
Luss [19], and Holmberg et al. [17]). As a result, computing the
dj m values may be quite time consuming, hence we propose a
way of computing lower bounds on mj values.
Proposition 1. Let j A J and sj maxi A Ij rij =qij . The inequality


X
r fj
y
zjk Z min
16
sj gvj j
kAK
j

3. Valid inequalities
is valid for F.
In this section, we propose some valid inequalities for both
versions of the VLP.
Let F be the set of solutions that satisfy constraints (2)(9).
We use some substructures in the formulation to derive our valid
inequalities. We also propose some redundant constraints to

Proof. For j A J, sj qij Z rij for all i A Ij . Multiplying constraints (5)

P
P
with sj and summing over k A Kj yields
i A Ij sj qij
k A Kj xijk r
P
P
sj g k A Kj zjk . Since sj qij Z rij for all iA Ij , this implies i A Ij rij
P
P
k A Kj xijk r sj g
k A Kj zjk . Now combining this with constraint (6),

Y. C
- nar, H. Yaman / Computers & Operations Research 38 (2011) 16781695

P
Let j A J. For i A Ij , dene the variable x ij 1 k A Kj xijk . Notice
that x ij is a 01 variable. Now the minimum prot constraint (6)
can be rewritten as
X
X
X
rij Z rij x ij
vj zjk rmin fj yj
20

we obtain
X X
X
X
sj g
zjk Z
rij
xijk Z
vj zjk rmin fj yj
i A Ij

k A Kj

k A Kj

1681

k A Kj

which gives
X
zjk Z rmin fj yj
sj gvj

i A Ij

k A Kj

This implies that if yj 1, i.e., if a vendor is located at location j,

P
then
k A Kj zjk Zrmin fj =sj gvj . Since the left hand side is
integer in a feasible solution, we can round up the right hand side.
If yj 0, then (16) becomes redundant. Hence we can conclude
that inequality (16) is valid for F. &

i A Ij

which is a 01 knapsack inequality.

Now based on this substructure, we can derive cover inequalities that are valid for F.
Proposition 4. Let j A J, S1 D Ij , and S2 DKj with jS2 jvj rmin fj 4
P
i A Ij \S1 rij . The inequality
X
XX
zjk r
xijk jS2 j1yj
21
i A S1 k A Kj

k A S2

For j A J, sj can be computed in OjIj j time.

k A Kj

is valid for F.
3.2. Cover inequalities for vehicle capacity constraints
For i A I, j A Ji , and kA Kj , inequality
xijk rzjk

17

is a valid inequality for F. These inequalities are often dominated

by cover inequalities that may be generated using the knapsack
structure of the capacity constraints (5) for the vehicles. Cover
inequalities that are valid for each of these knapsack constraints
are also valid for F. Let j A J, k A Kj , and C DIj be such that
P
P
i A C qij 4 g. Then the cover inequality
i A C xijk rjCj1zjk is a
valid inequality for F. These inequalities can be strengthened by
lifting.
Most of the integer programming solvers recognize knapsack
constraints and use lifted cover inequalities as cutting planes. So
here we limit our attention to some lifted cover inequalities that
are not many in number so that they can be added to the
formulation before giving it to the solver.
For a given location j A J, we rst consider all demand points
with demand larger than half of the capacity of a vehicle. Then we
know that at most one of these points may be assigned to a given
vehicle of vendor j. This leads to the following set of inequalities.
Proposition 2. For j A J and kA Kj , the lifted cover inequality
X
xijk r zjk

18

i A Ij :qij 4 g=2

is valid for F.
Next, we generate lifted cover inequalities for each demand
point i A Ij with demand not more than half the capacity.
Proposition 3. Let i A Ij be such that qij r g=2. Dene Cij fl A Ij :
qij qlj 4 gg. Then the lifted cover inequality
X
xljk rzjk
19
xijk
l A Cij

Proof. Let j A J. Consider the knapsack inequality (20). Suppose

P
that yj 1. Let S1 D Ij and S2 D Kj . If i A S1 rij jS2 jvj rmin fj 4
P
P
P
jk r jS1 j
i A Ij rij , then the cover inequality
i A S1 x ij
k A S2 zP
jS2 j1 is valid. We can rewrite this inequality as
iA
S1 1
P
P
P
x

z
r
jS
j
jS
j1,
which
simplies
to
1
2
ijk
jk
kAK
k A S2
j
P
P k A S2
zjk r i A S1 k A Kj xijk jS2 j1.
If yj 0, then xijk 0 for all iA Ij and k A Kj and zjk 0 for all
kA Kj . Hence inequality (21) is valid for F. &
4. Computational results
In this section, we report the outcomes of our computational
study. Here, we investigate for which sizes we can solve the
formulations to optimality in reasonable times and the effect of
valid inequalities on the quality of upper bounds of linear
programming relaxations and the solution times.
4.1. The data set and models
We use the data from the demijohn water company. The data
includes 84 demand points, their estimated demands, the distances, and cost parameters. The set of possible locations for the
vendors is the same as the set of demand points. Moreover, there
is the additional restriction that if a vendor is located at a given
demand point, then the demand of this point should be served by
itself. To handle this, we added the constraint
X
xjjk yj 8j A J
22
k A Kj

We can also use this information to break the symmetry. We

impose that if a vendor is located at a demand point, then the
point should use its vehicle indexed as its rst vehicle by adding
the constraints

is valid for F.

xjj1 zj1

Proof. If xijk 1, then as qij qlj 4 g for each l A Cij , none of these
demand points can be served by the same vehicle. If xijk 0, then
P
as qlj qmj 4 g for l and m in Cij, we know that l A Cij xljk rzjk . &

xjj1 yj

Notice that if Cij is empty, then inequality (19) reduces to (17).

3.3. Cover inequalities for the minimum prot constraints
Finally, we propose to model the minimum prot constraints
in a different way so that we can use the lifted cover cuts of
off-the-shelf solvers. To this end, we complement sums of assignment variables and rewrite the minimum prot constraints as
01 knapsack constraints as follows.

8j A J

23

8j A J

24

Let PM0 and CM0 be the models obtained by adding the above
constraints to ProtVLP and CoverageVLP, respectively. Let PM1
and CM1 be the models PM0 and CM0 strengthened with the valid
inequalities (16), which provide lower bounds on the number of
vehicles for each vendor.
The fact that if a vendor is located at a demand point, then the
point should use its rst vehicle can further be used to obtain
stronger lifted cover inequalities for the rst vehicles:
X
i A Ij \fjg:qij qjj 4 g

xij1 0

8j A J

25

Y. C
- nar, H. Yaman / Computers & Operations Research 38 (2011) 16781695

1682

xij1 rzj1

8j A J

26

i A Ij \fjg:qij 4 gqjj =2

xij1

xlj1 rzj1

8j A J, i A Ij \fjg : qij r

l A Ij \fjg:qij qlj 4 gqjj

gqjj
2

27

We add the above cover inequalities for the rst vehicles and
inequalities (18) and (19) for the remaining vehicles to models
PM1 and CM1 and call the resulting models PM2 and CM2,
respectively.
We remove constraints (6) from models PM2 and CM2 and
add the following variables and constraints to obtain models PM3
and CM3:
X
x ij 1
xijk 8i A I,j A Ji
28
k A Kj

i A Ij

k A Kj

rij x ij

x ij A f0,1g

vj zjk rmin fj yj r

rij 8j A J

29

i A Ij

8i A I, j A Ji :

30

The aim is to enable the solver to see the knapsack structure in

the minimum prot constraints so that it can generate cover
inequalities as discussed in Section 3.3.
We add the simple valid inequalities
zjk ryj

8j A J, k A Kj

31

to models PM3 and CM3 to obtain models PM4 and CM4.

Finally, analyzing the results of our computational study, we
also decided to repeat our experiment with additional models for
ProtVLP and CoverageVLP. For ProtVLP, we tested model PM5,
which is obtained by removing the cover inequalities obtained
using vehicle capacity constraints, i.e., inequalities (18), (19),
(25)(27), from model PM4. For CoverageVLP, model CM5 is
obtained by adding only valid inequalities zjk r yj for all j A J and
k A Kj to model CM0.
In Tables 1 and 2, we give the constraints of the different
models for ProtVLP and CoverageVLP, respectively.
To evaluate the performances of the models dened above, we
used the following test set. We let p A f4,6,8g, kmax
kmax A
j
f6,8,10g for all j A J, and rmin A f50,100,150g.
For each value of p, kmax, and rmin , we have four problems with
different demand and prot structures. In A-type problems, we
take qij qi and rij ri for all j A Ji and iA I. So in A-type instances,

the demand and prot are independent of the distance between

the demand point and its vendor. In B-type problems, we take
qij qi and rij cij qi for all j A Ji and iA I where cij is the unit prot
that vendor j gains if it serves demand point i and is a function of
the distance between i and j. In C-type problems, we take qij to be
a function of the distance between i and j and rij cqij for all j A Ji
and iA I where c is the unit prot and does not depend on
distances. In this case, we let qij qi for vendors j that are within
a short traveling time of i and then let qij decrease with the
distance between i and j for other eligible vendors. Precisely, for
iA I, we let Ji fj A J : dij r 10g, where dij is the distance between
the demand point i and the vendor j. For i A I and j A Ji , we let
qij qi minf1,1:50:1dij g. Hence the demand generated by point i
is equal to qi if the vendor j is within 5 km of point i and is equal
to qi 1:50:1dij if j is farther. Finally, in D-type problems, we take
both the demands and the prots as functions of the distances.
Both problems ProtVLP and CoverageVLP are infeasible for
r 150, p 8, and all four demand and prot structures. These
instances are removed from the results.
All models are solved using GAMS 22.5 and CPLEX 11.0.0 on an
AMD Opteron 252 processor (2.6 GHz) with 2 GB of RAM operating under the system CentOS (Linux version 2.6.9-42.0.3.ELsmp).
We have a time limit of 1 h.
4.2. Results for ProtVLP
In Tables 36, we report the results for ProtVLP and the four
types of instances, A, B, C, and D, respectively. For each instance
and model, we report the percentage gap between the upper
bound obtained by solving the linear programming relaxation of
the corresponding model and the best lower bound for the integer
problem in the column LP gap. Then we report the cpu times in
seconds. If the problem is not solved to optimality in 1 h, then we
report the remaining percentage gap in parentheses. Finally, we
report the number of nodes in the branch and cut tree for each
model and instance. The best results are marked bold.
Each table has a summary, where we can see the averages of
linear programming relaxation gaps, nal optimality gaps, cpu
times, number of nodes, the number of instances solved to
optimality with each model, and the number of times each model
was among the best for the considered criterion.
In these tables we observe that the initial model PM0 has huge
duality gaps and adding the valid inequalities (16), which impose

Table 1
Constraints of the models for ProtVLP.
PM0

PM1

PM2

PM3

PM4

PM5

(2)(9)
(22)(24)

(2)(9)
(22)(24)
(16)

(2)(9)
(22)(24)
(16)
(18), (19), (25)(27)

(2)(5), (7)(9)
(22)(24)
(16)
(18), (19), (25)(27)
(28)(30)

(2)(5), (7)(9)
(22)(24)
(16)
(18), (19), (25)(27)
(28)(30)
(31)

(2)(5), (7)(9)
(22)(24)
(16)
(28)(30)
(31)

Table 2
Constraints of the models for CoverageVLP.
CM0

CM1

CM2

CM3

CM4

CM5

(2)(9)
(22)(24)

(2)(9)
(22)(24)
(16)

(2)(9)
(22)(24)
(16)
(18), (19), (25)(27)

(2)(5), (7)(9)
(22)(24)
(16)
(18), (19), (25)(27)
(28)(30)

(2)(5), (7)(9)
(22)(24)
(16)
(18), (19), (25)(27)
(28)(30)
(31)

(2)(9)
(22)(24)

(31)

Table 3
Results for ProtVLP and A-type instances.
Parameters

LP gap (%)

Cpu time (s)/optimality gap (%)

kmax

PM0

PM1

PM2

PM3

50
50
50
50
50
50
50
50
50

6
8
10
6
8
10
6
8
10

4
4
4
6
6
6
8
8
8

76.77
48.05
42.51
58.36
50.91
49.93
55.24
55.09
55.09

76.77
48.05
42.51
58.36
50.91
49.93
55.24
55.09
55.09

29.21
8.56
3.92
9.26
4.80
4.12
2.01
1.93
1.93

29.21
8.56
3.92
9.26
4.80
4.12
2.01
1.93
1.93

100
100
100
100
100
100
100
100
100

6
8
10
6
8
10
6
8
10

4
4
4
6
6
6
8
8
8

76.77
48.05
42.51
58.36
50.91
49.93
55.24
55.24
55.24

76.77
48.05
42.51
58.36
50.91
49.93
55.24
55.24
55.24

29.21
8.56
3.92
9.26
4.77
4.09
1.84
1.84
1.84

150
150
150
150
150
150

6
8
10
6
8
10

4
4
4
6
6
6

76.77
48.05
42.51
61.23
55.77
54.73

76.77
48.05
42.51
61.23
55.77
54.73

Average
Avg. opt. gap (%)
# of solved ins. (/24)
# of best solutions (/24)

55.14

55.14

PM5

PM0

6.48
5.89
3.92
5.04
4.15
4.12
1.91
1.93
1.93

6.48
25.82
42.31
21.57
50.91
49.93
55.24
55.09
55.09

397.61
411.81
244.58
116.99
1180.66
287.11
128.86
262.26
2062.02

29.21
8.56
3.92
9.26
4.77
4.09
1.84
1.84
1.84

6.48
5.89
3.92
5.04
4.12
4.09
1.82
1.84
1.84

6.48
25.82
42.31
21.57
50.91
49.93
55.24
55.24
55.24

29.21
8.56
3.92
11.02
7.06
6.35

29.21
8.56
3.92
11.02
7.06
6.35

6.48
5.89
3.92
6.85
6.76
6.35

8.22

8.22

4.45

10

10

PM4

24

PM2

PM3

PM4

PM5

PM0

PM1

203.00
127.10
153.81
110.11
568.44
772.12
143.02
280.51
996.55

284.25
562.07
32.81
123.83
60.53
149.12
105.59
286.32
1013.52

454.59
2874.28
12.25
196.02
112.02
84.97
161.81
218.60
871.82

261.57
866.48
9.46
166.54
91.72
215.28
130.32
318.31
838.39

177.23
1002.29
43.52
110.79
653.58
407.10
285.60
471.21
1068.35

29387
20792
27254
3847
49637
14542
2767
5330
57867

13208
7760
17360
2575
20099
25647
2380
3551
20087

99.38
907.42
93.30
85.56
565.66
326.92
270.22
3259.19
(0.05)

224.41
330.63
67.80
168.24
593.62
292.73
159.86
1315.94
(0.05)

214.90
1712.14
20.82
108.66
73.19
176.99
330.99
608.58
(0.05)

226.33
718.57
10.61
113.17
80.21
100.61
310.32
460.38
741.72

174.20
630.69
7.48
248.37
117.88
139.74
408.94
467.88
897.38

161.32
859.88
30.33
121.41
312.07
477.56
285.72
472.78
742.00

6639
58284
7063
1947
18574
9777
4837
66262
54635

6.48
25.82
42.31
23.77
55.77
54.73

427.28
282.47
27.77
242.70
366.69
979.08

205.37
883.12
25.76
256.54
249.83
2620.63

340.14
976.47
19.65
141.90
619.59
(0.04)

142.27
961.01
9.84
97.67
52.91
202.62

166.03
164.50
36.53
125.62
139.67
279.85

132.58
398.14
52.64
120.08
108.36
163.29

38.92

691.24
(0.00)
23
2

631.76
(0.00)
22
3

383.94
(0.00)
24
10

287.62
(0.00)
24
4

360.74
(0.00)
24
1

PM1

597.88
(0.00)
23
4

PM2

PM3

PM4

PM5

14983
26693
3119
4269
929
1890
1437
4569
19222

28200
103257
248
4660
1621
1496
2094
2248
17513

13612
23868
85
3846
1683
2853
1617
3228
8925

8503
43420
780
2894
28500
10654
3476
6962
21266

13122
13090
3413
7306
24299
11121
2495
30370
50175

10132
96939
504
3870
1034
4441
4595
10986
47481

12692
39225
274
2791
1032
1315
3669
3431
4803

9235
26042
64
5043
1598
1655
4189
4171
5475

7816
39542
533
2483
12511
14364
2963
4406
5494

34745
20508
738
11248
10633
35910

10641
46832
690
6592
11690
64156

23289
54511
638
4629
23780
115684

6411
38617
160
1580
652
877

7191
5468
555
1757
666
928

5580
22293
661
3557
198
981

23051

17027

19984

11604

5573

10411

Y. C
- nar, H. Yaman / Computers & Operations Research 38 (2011) 16781695

rmin

Number of nodes

1683

1684

Table 4
Results for ProtVLP and B-type instances.
Parameters

LP gap (%)
k

50
50
50
50
50
50
50
50
50

Cpu time (s)/optimality gap (%)

PM0

PM1

PM2

PM3

6
8
10
6
8
10
6
8
10

4
4
4
6
6
6
8
8
8

77.61
48.76
43.51
59.28
52.04
51.17
57.31
56.75
56.75

77.61
48.76
43.51
59.28
52.04
51.17
57.31
56.75
56.75

29.38
8.72
4.32
9.42
5.12
4.53
2.38
2.03
2.04

29.38
8.72
4.32
9.42
5.12
4.53
2.38
2.03
2.04

100
100
100
100
100
100
100
100
100

6
8
10
6
8
10
6
8
10

4
4
4
6
6
6
8
8
8

77.61
48.76
43.51
59.28
52.10
51.23
58.33
57.53
57.31

77.61
48.76
43.51
59.28
52.10
51.23
58.32
57.53
57.31

29.38
8.72
4.32
9.41
5.11
4.52
2.81
2.29
2.15

150
150
150
150
150
150

6
8
10
6
8
10

4
4
4
6
6
6

77.61
48.76
43.51
61.61
56.82
55.87

77.61
48.76
43.51
61.61
56.82
55.87

Average
Avg. opt. gap (%)
# of solved ins. (/24)
# of best solutions (/24)

56.38

56.38

PM4

PM5

PM0

PM1

6.93
6.04
4.32
5.33
4.55
4.53
2.21
2.02
2.04

7.78
26.82
43.31
22.89
52.02
51.17
57.23
56.74
56.75

118.79
1004.05
80.74
378.75
368.67
1969.78
277.26
245.46
765.83

243.27
2726.00
185.76
173.11
119.70
350.27
308.33
285.58
680.23

403.13
858.60
10.28
108.48
149.56
118.53
101.84
147.91
428.27

29.38
8.72
4.32
9.41
5.11
4.52
2.81
2.29
2.15

6.93
6.04
4.32
5.33
4.53
4.52
2.70
2.27
2.15

7.78
26.82
43.31
22.89
52.08
51.23
58.25
57.53
57.31

625.21
371.98
148.09
201.71
626.96
539.51
861.99
(0.21)
(0.06)

392.02
836.81
44.02
256.55
380.98
338.08
1562.33
(0.03)
(0.01)

168.70
356.90
8.64
218.15
398.19
222.25
1284.50
2493.71
(0.05)

29.38
8.72
4.32
10.30
6.78
6.15

29.38
8.72
4.32
10.30
6.78
6.15

6.93
6.04
4.32
6.71
6.50
6.15

7.78
26.82
43.31
24.69
56.80
55.87

1263.19
1382.57
147.35
544.47
(0.01)
111.98

107.18
2057.22
32.47
132.41
125.11
(0.00)

154.69
869.85
22.35
117.53
1006.17
167.07

8.43

8.43

4.73

40.30

951.44
(0.01)
21
1

922.41
(0.00)
21
1

24

PM2

558.98
(0.00)
23
9

Number of nodes
PM3

PM4

PM5

PM0

PM1

435.34
664.81
19.22
302.37
71.49
144.66
172.87
314.23
563.48

394.87
140.27
20.70
206.38
147.88
203.60
178.26
417.90
1128.88

157.66
483.59
54.16
204.76
165.26
1920.13
213.94
306.81
847.33

6510
73179
4358
17782
14233
82051
7455
4151
16686

14967
103608
10526
8094
4161
13989
5374
4682
13179

187.98
782.17
17.50
239.38
75.75
148.77
661.40
743.06
1384.88

204.20
568.03
39.84
428.11
167.60
189.52
506.51
775.63
1296.78

698.42
340.19
68.94
180.65
246.32
755.32
447.86
1250.77
1069.77

49140
25341
10875
10180
49338
28173
30967
80039
70320

217.05
508.54
11.43
120.22
88.80
138.64

214.49
795.61
18.68
167.07
124.49
53.78

156.04
407.60
77.87
94.94
152.24
137.25

333.91
(0.00)
24
6

349.55
(0.00)
24
2

434.91
(0.00)
24
5

PM3

PM4

PM5

26960
66030
278
4562
5163
2111
1530
2970
5266

26043
33633
396
7770
1678
2356
2400
4010
6304

23355
3897
368
5924
2169
4122
1923
4293
9672

6967
16495
690
10605
5203
51737
2837
4268
9794

29850
37881
1288
10857
28355
22321
35979
112763
87082

10168
18964
307
8823
40180
9924
24283
56433
101959

8666
34894
348
9104
1286
2428
6785
7183
9137

14426
22507
576
18007
3042
3587
4620
7549
9011

6520
14522
1074
6085
10095
25125
6645
14744
11935

76293
86130
13428
19639
113245
2345

6924
120951
550
4502
901
89894

7789
42405
685
3275
53107
3173

11128
31613
214
2975
664
743

12527
32705
339
4778
660
573

6577
19499
1622
3757
1605
764

37161

32028

20681

8823

7943

9965

PM2

Y. C
- nar, H. Yaman / Computers & Operations Research 38 (2011) 16781695

rmin

max

Table 5
Results for ProtVLP and C-type instances.
Parameters

LP gap (%)

Cpu time (s)/optimality gap (%)

kmax

PM0

PM1

PM2

PM3

PM4

PM5

PM0

50
50
50
50
50
50
50
50
50

6
8
10
6
8
10
6
8
10

4
4
4
6
6
6
8
8
8

36.40
13.23
13.27
21.78
12.88
12.88
18.37
16.49
16.53

36.40
13.23
13.27
21.78
12.88
12.88
18.37
16.49
16.53

34.53
11.75
11.80
17.42
9.04
9.04
11.74
9.96
10.01

34.53
11.75
11.80
17.42
9.04
9.04
11.74
9.96
10.01

3.99
11.27
11.80
14.10
9.04
9.04
10.90
9.96
10.01

3.99
12.69
13.27
14.80
12.88
12.88
17.56
16.49
16.53

415.66
(0.20)
(1.31)
(0.06)
1029.34
(0.88)
(1.72)
(0.63)
(1.05)

545.82
(0.77)
(2.43)
(0.02)
(0.86)
(0.81)
(1.47)
(0.74)
(1.26)

100
100
100
100
100
100
100
100
100

6
8
10
6
8
10
6
8
10

4
4
4
6
6
6
8
8
8

36.40
13.23
13.27
21.78
12.88
12.88
20.07
16.49
16.53

36.40
13.23
13.27
21.78
12.88
12.88
20.07
16.49
16.53

34.53
11.75
11.80
17.41
8.95
8.95
13.44
9.95
10.00

34.53
11.75
11.80
17.41
8.95
8.95
13.44
9.95
10.00

3.99
11.27
11.80
14.10
8.95
8.95
12.61
9.95
10.00

3.99
12.69
13.27
14.80
12.88
12.88
19.38
16.49
16.53

901.64
3191.66
(3.44)
(0.58)
(0.74)
(0.66)
(1.68)
(0.53)
(2.09)

150
150
150
150
150
150

6
8
10
6
8
10

4
4
4
6
6
6

36.40
13.23
13.27
21.78
14.34
14.34

36.40
13.23
13.27
21.69
14.34
14.34

34.53
11.75
11.80
17.20
10.18
10.18

34.53
11.75
11.80
17.20
10.18
10.18

3.99
11.27
11.80
14.10
10.18
10.18

3.99
12.69
13.27
14.80
14.34
14.34

Average
Avg. opt. gap (%)
# of solved ins. (/24)
# of best solutions (/24)

18.28

18.28

14.49

14.49

10.14

13.23

13

13

24

PM1

PM2

PM3

PM4

908.56
(1.46)
(2.19)
(1.82)
(1.15)
(0.71)
(1.30)
(1.19)
(1.74)

808.90
1823.53
(3.42)
(0.23)
(0.16)
(0.86)
(4.04)
(1.10)
(1.96)

235.64
(1.39)
(1.38)
2466.03
(0.02)
(0.81)
(5.18)
(1.01)
(1.76)

1212.46
(0.81)
(2.10)
(0.02)
(0.47)
(0.59)
(1.32)
(0.30)
(1.31)

690.56
(0.67)
(2.07)
(0.06)
(0.55)
(0.96)
(0.64)
(2.43)
(1.82)

1312.17
(0.75)
(2.41)
(0.41)
(0.06)
(0.76)
(1.49)
(0.81)
(2.36)

350.66
2749.68
(1.16)
1725.05
(1.34)
(1.86)

1091.24
(0.73)
(2.20)
(0.02)
(1.34)
(1.77)

783.26
(0.95)
(2.20)
2271.70
(1.18)
(1.77)

2982.99
(0.83)
7
5

3268.79
(0.89)
3
2

3194.00
(1.49)
4
1

PM5

PM0

PM1

PM2

PM3

PM4

PM5

137.35
(0.39)
(1.16)
(0.35)
(0.68)
(0.58)
(0.68)
(0.55)
(1.18)

21173
263317
153821
89914
33616
60059
29311
22195
21175

29775
210493
125472
81063
52149
55828
25123
21204
21012

25878
33642
46926
22019
25086
24680
18448
15233
12847

16029
55561
64855
25754
65312
31578
17679
14515
11431

4582
113821
58345
25452
41496
22146
24301
14275
13802

5389
195620
115305
66734
57322
67819
57322
21764
18843

227.35
(0.58)
(2.58)
2271.14
(0.52)
(0.78)
(2.55)
(1.58)
(1.17)

157.72
(0.39)
(1.37)
1773.18
(0.48)
(0.81)
1.97
(0.91)
(2.99)

44662
119696
94767
111839
69973
87953
29552
23977
20528

52312
182370
153907
75878
73212
57383
23039
27750
18319

21072
183491
61780
32600
43094
25728
17985
13911
9930

23443
72687
52502
28226
68349
26137
14123
11026
11658

3208
67202
42564
27477
41428
46347
18713
11332
9690

5496
217651
116855
24158
85315
49519
22312
15121
17731

3397.84
3295.62
(2.19)
(0.39)
(0.99)
(1.77)

213.28
3170.46
(1.25)
(0.52)
(1.55)
(1.41)

126.14
(0.57)
(1.16)
1430.80
(1.14)
(1.49)

17131
245267
149194
46143
80373
48787

61934
114423
132028
67638
82554
58336

24382
85248
52135
31197
52963
30143

106412
91799
82438
26376
44136
34224

3816
118125
50298
32231
42435
53502

4747
189367
153399
17171
77732
67901

3293.33
(1.09)
5
3

3057.75
(1.09)
6
3

3001.11
(0.78)
5
11

78518

75133

37934

41510

36941

69608

11

Y. C
- nar, H. Yaman / Computers & Operations Research 38 (2011) 16781695

rmin

Number of nodes

1685

1686

Table 6
Results for ProtVLP and D-type instances.
Parameters

LP gap (%)

Cpu time (s)/optimality gap (%)

rmin

PM0

PM1

PM2

PM3

PM4

PM5

PM0

50
50
50
50
50
50
50
50
50

6
8
10
6
8
10
6
8
10

4
4
4
6
6
6
8
8
8

36.72
13.44
13.48
21.92
13.27
13.27
20.76
17.01
17.03

36.72
13.44
13.48
21.92
13.27
13.27
20.76
17.01
17.03

34.79
11.93
11.98
17.43
9.32
9.32
13.77
10.25
10.27

34.79
11.93
11.98
17.43
9.32
9.32
13.77
10.25
10.27

4.31
11.47
11.98
13.98
9.32
9.32
12.87
10.25
10.27

4.77
12.93
13.48
14.99
13.27
13.27
19.87
17.01
17.03

561.12
1164.72
(3.12)
(0.01)
(1.58)
(1.04)
(3.12)
(1.10)
(1.33)

488.34
455.10
(2.15)
(0.01)
(0.74)
(0.75)
(2.14)
(0.90)
(1.62)

100
100
100
100
100
100
100
100
100

6
8
10
6
8
10
6
8
10

4
4
4
6
6
6
8
8
8

36.72
13.44
13.48
21.92
13.27
13.27
20.60
17.01
17.01

36.72
13.44
13.48
21.92
13.27
13.27
20.60
17.01
17.01

34.79
11.93
11.98
17.42
9.21
9.21
13.74
10.24
10.24

34.79
11.93
11.98
17.42
9.21
9.21
13.74
10.24
10.24

4.31
11.47
11.98
13.98
9.21
9.21
12.86
10.24
10.24

4.77
12.93
13.48
14.99
13.27
13.27
19.86
17.01
17.01

3386.70
1712.80
(2.29)
(0.58)
(0.71)
(0.93)
(1.25)
(0.60)
(1.10)

150
150
150
150
150
150

6
8
10
6
8
10

4
4
4
6
6
6

36.72
13.44
13.48
22.45
14.78
14.78

36.72
13.44
13.48
22.45
14.78
14.78

34.79
11.93
11.98
17.82
10.49
10.49

34.79
11.93
11.98
17.82
10.49
10.49

4.31
11.47
11.98
14.60
10.49
10.49

4.77
12.93
13.48
15.60
14.78
14.78

Average
Avg. opt. gap (%)
# of solved ins. (/24)
# of best solutions (/24)

18.72

18.72

14.81

14.81

10.44

13.73

13

13

24

PM1

PM2

Number of nodes
PM3

PM4

PM5

PM0

PM1

PM2

PM3

PM4

PM5

633.79
(0.28)
(2.67)
(0.60)
(0.73)
(0.86)
(2.17)
(1.36)
(1.69)

1956.61
(1.16)
(2.30)
(0.70)
(0.54)
(0.79)
(2.63)
(1.71)
(1.61)

232.02
(1.09)
(1.64)
1757.17
(0.71)
(0.62)
(4.98)
(1.25)
(2.39)

160.37
2809.47
(2.26)
2004.23
(0.09)
(0.82)
(2.56)
(1.53)
(3.12)

28765
67315
100368
84878
81713
52052
32632
29339
23767

25766
28114
110347
98031
56679
47068
29856
18631
20408

15534
94733
68070
62948
29972
19260
21022
13130
11139

34652
100649
28127
22232
30236
18872
16450
15488
11722

5139
56289
66041
14977
45316
40327
26431
12540
13368

7266
170833
111921
26490
72111
57108
29340
22387
20784

507.64
(1.62)
(2.14)
(0.01)
(0.09)
(1.19)
(0.47)
(0.78)
(1.22)

1106.28
3393.36
(3.88)
(0.01)
(0.32)
(0.81)
(0.04)
(0.76)
(1.29)

1203.65
(0.21)
(1.46)
(0.58)
3097.56
(0.73)
(2.09)
(2.54)
(2.77)

698.44
3577.48
(2.42)
1718.09
(0.91)
(0.81)
(2.89)
(1.35)
(3.31)

164.79
2669.71
(0.82)
1618.76
(0.64)
(0.73)
(1.35)
(1.71)
(1.22)

121859
143447
150628
90554
72710
48044
28266
25717
22651

20229
187751
127016
77487
67244
44638
29342
19918
17197

30277
115416
61970
56953
35361
31362
28226
13032
11800

20402
91689
70603
24518
40838
31477
15084
10990
10205

13817
133343
47248
14288
80211
21707
18162
14204
11563

7294
154873
164907
17998
71418
52541
19170
20374
18160

419.46
2384.95
(2.10)
(0.72)
(1.40)
(1.86)

290.98
(0.62)
(2.74)
(0.01)
(1.33)
(1.61)

801.11
(1.68)
(2.99)
(0.00)
(1.48)
(1.45)

1717.41
932.03
(1.81)
(2.17)
(1.35)
(1.44)

224.96
(0.53)
(2.42)
3056.64
(0.83)
(1.53)

178.18
1703.53
(2.40)
1750.43
(1.41)
(1.71)

23992
118142
118407
59335
80001
48006

14481
116417
194520
83396
83067
61615

38333
99826
49374
46790
53054
24584

27987
32216
75712
19144
39108
26854

4510
110576
64753
25628
43934
36213

7150
113171
86667
22813
53446
61914

3101.29
(1.04)
6
4

3072.64
(0.92)
4
3

3247.35
(1.05)
4
1

3221.22
(1.19)
5
5

3019.44
(1.24)
7
4

2794.21
(0.93)
9
8

68858

65801

43007

33965

38358

57922

10

Y. C
- nar, H. Yaman / Computers & Operations Research 38 (2011) 16781695

max

Y. C
- nar, H. Yaman / Computers & Operations Research 38 (2011) 16781695

lower bounds on the number of vehicles, has almost no impact on

these gaps. The average gaps are 55.14%, 56.38%, 18.28%, and
18.72% for A, B, C, and D instances, respectively. Here we remark
that even though they are still very large, the instances of types C
and D (instances where demands depend on the distances) have
much smaller gaps compared to the instances of types A and B.
This may be due to the fact that capacity constraints for vehicles
are tighter for instances of type A and B. Indeed, the average gaps
for the model with cover inequalities, PM2, are 8.22%, 8.43%,
14.49%, and 14.81% for A, B, C, and D, respectively. Here we see
that these inequalities have reduced the gaps considerably for Aand B-type instances whereas their effect was much smaller for C
and D-type instances. After all the valid inequalities are added,
with model PM4, the average gaps are 4.45%, 4.73 %, 10.14%, and
10.44% for A, B, C, and D, respectively. Here we see that the valid
inequalities are more effective in improving the quality of linear
programming upper bounds for A- and B-type instances.
Except for D-type instances, model PM4 gives the smallest
number of nodes on the average. If we compare the average
number of nodes for the original model PM0 and the ones for
model PM4, we observe that the reductions are 75.82%, 78.63%,
52.95%, and 44.29% for A, B, C, and D instances, respectively. We
can conclude that our valid inequalities are more effective in
reducing the size of the branch and cut tree for A- and B-type
instances.
For A-type instances, models PM1 and PM2 could solve 23
instances, model PM3 22 instances, and models PM4, PM5, and
PM6 24 instances to optimality in 1 h. The remaining gaps are
quite small for the unsolved instances. The best average cpu time
is given by model PM4 and is 58.39% less than the average cpu
time of the original model PM0. Model PM4 has given the best cpu
for only four instances, whereas model PM3 has given the best
cpu for 10 instances out of 24. This model has the best average
cpu time for B-type instances. It is interesting to note that for
these instances, model PM2 has given the best cpu time for nine
instances and model PM3 has given the best cpu time for six
instances. But one of the instances could not be solved to
optimality with model PM2. Models PM0 and PM1 could not
solve three instances to optimality. The average cpu time of PM3
is 64.90% less than the average cpu time of PM0. For these
instances, we see that both cover inequalities based on the vehicle
capacities and the knapsack inequalities for minimum prot
constraints are quite effective in reducing the cpu times on the
average.
ProtVLP is harder for instances of types C and D, where the
demands are functions of distances. Here our valid inequalities
are not useful in reducing cpu times and nal gaps for unsolved
instances. We see that models PM0 and PM5 are the best in terms
of cpu times and the number of instances solved to optimality, for
C and D instances, respectively. The largest nal gap for C
instances is 1.41%, and for D instances it is 2.14%.

4.3. Results for CoverageVLP

We report the results for CoverageVLP and the four types of
instances, A, B, C, and D in Tables 710, respectively. CoverageVLP
turned out to be easier to solve compared to ProtVLP for our
instances. First of all, the duality gaps were smaller for the
original formulation. The average gaps are 16.62%, 16.71%,
5.02%, and 5.03% for A, B, C, and D instances, respectively. Again,
the instances of types C and D have smaller gaps. Our valid
inequalities reduced the average duality gaps to 1.22%, 1.26%,
1.06%, and 1.09% for A, B, C, and D instances, respectively. Even
though it looks like the reduction in the duality gaps is mostly due
to the use of valid inequalities (31), the differences between the

1687

average gaps of models CM4 and CM5 show that some of the
remaining valid inequalities are also effective in strengthening
the original model for A- and B-type instances.
In terms of number of nodes, CM4, the model with all valid
inequalities, has given the best average results, decreasing the
number of nodes by 78.77%, 75.51%, 72.79%, and 86.52% compared to CM0 for A, B, C, and D instances, respectively.
Only model CM4 could solve all 24 type A instances to
optimality in 1 h of cpu time. Its average cpu is 78.14% less than
the average cpu of the original model CM0. Similar results are
obtained for B-type instances. For both types of instances, CM4
performs much better than all other models in terms of average
cpu times.
All our models solve the 24 C-type instances to optimality
within the time limit. Among these, CM5 has the best average cpu
time. Our model with all valid inequalities has an average cpu
time of 96.23 s, whereas model CM5 has an average cpu time of
68.41 s. Hence for these instances, we can conclude that even
though the valid inequalities are effective in reducing the duality
gaps and the sizes of the branch and cut trees, other than the
simple inequalities zjk ryj for all j A J and kA Kj , they are not very
useful in reducing the cpu times.
Finally, for D-type instances, the model CM4 gives the best
average cpu time, which is 73.04% less than the average cpu time
for the original model. It is interesting to note that for these
instances, model CM5 could not solve two problems to optimality.
4.4. Improvements in linear programming bounds
Here, we report the percentage improvement in linear programming bounds obtained by adding families of valid inequalities. We rst solve the linear programming relaxation of the
model without any valid inequalities. Then we add each family of
valid inequalities separately to the original model. We use the
inequalities (16), which impose lower bounds on the number of
vehicles, cover inequalities (18), (19), (25)(27), and the simple
valid inequalities (31). We compute the percentage improvements in the linear programming bounds. The averages are
reported in Table 11.
Here we observe that the inequalities (16), which impose
lower bounds on the number of vehicles, do not improve the
linear programming bounds. The cover inequalities result in
signicant improvements for ProtVLP, especially for A- and
B-type instances. However, they are not as useful for CoverageVLP.
The valid inequalities (31) improve the linear programming
bounds for all problems, more for A- and B-type instances and
less for C- and D-type instances.
4.5. Comparison of prot and coverage values
In Table 12, we report the best prot and coverage values for
all ranges of parameters considered in our experiment. Here, we
observe that for a given rmin value, best prot and coverage values
are achieved with medium or large kmax and p values. We depict
the prot values for A-type instances in Fig. 1. Similar behavior is
observed for the other types of instances.
For CoverageVLP, the best coverage values are achieved with
p 8 and kmax 8,10 for rmin 50,100 and with p 6 and
kmax 10 for rmin 150. Increasing p and kmax has a signicant
effect on the best coverage values.
5. Analysis of example optimal solutions
In this section, we analyze the optimal solutions for problems
ProtVLP and CoverageVLP for an example instance with

1688

Table 7
Results for CoverageVLP and A-type instances.
Parameters

LP gap (%)
k

50
50
50
50
50
50
50
50
50

Cpu time (s)/optimality gap (%)

CM0

CM1

CM2

CM3

6
8
10
6
8
10
6
8
10

4
4
4
6
6
6
8
8
8

54.24
23.05
5.77
30.68
9.27
0.00
9.21
0.00
0.00

53.24
23.05
5.77
30.68
9.27
0.00
9.21
0.00
0.00

53.83
22.71
5.66
30.34
9.27
0.00
9.21
0.00
0.00

53.83
22.71
5.66
30.34
9.27
0.00
9.21
0.00
0.00

1.91
1.58
0.62
1.31
1.15
0.00
0.44
0.00
0.00

100
100
100
100
100
100
100
100
100

6
8
10
6
8
10
6
8
10

4
4
4
6
6
6
8
8
8

54.24
23.05
5.77
30.68
9.44
0.00
9.78
0.00
0.00

54.24
23.05
5.77
30.68
9.44
0.00
9.78
0.00
0.00

53.70
22.69
5.66
30.34
9.44
0.00
9.78
0.00
0.00

53.70
22.69
5.66
30.34
9.44
0.00
9.78
0.00
0.00

150
150
150
150
150
150

6
8
10
6
8
10

4
4
4
6
6
6

54.24
23.05
5.77
31.32
12.38
6.99

54.24
23.05
5.77
31.32
12.38
6.99

53.62
22.69
5.66
29.10
10.35
5.08

Average
Avg. opt. gap (%)
# of solved ins. (/24)
# of best solutions (/24)

16.62

16.62

16.21

CM0

CM1

1.91
7.26
5.64
5.78
9.27
0.00
9.21
0.00
0.00

261.90
255.51
17.32
85.34
71.74
6.21
1329.72
3.79
5.16

175.16
811.10
10.26
61.67
49.29
3.54
1996.49
16.66
13.68

1.91
1.58
0.62
1.31
1.32
0.00
0.96
0.00
0.00

1.91
7.26
5.64
5.78
9.44
0.00
9.78
0.00
0.00

219.06
571.61
22.05
44.16
58.18
8.88
3412.02
(0.88)
131.76

53.62
22.69
5.66
29.10
10.35
5.08

1.91
1.58
0.62
1.78
3.69
5.08

1.91
7.26
5.64
6.30
12.38
6.99

16.21

1.22

4.97

CM4

24

CM5

CM2

Number of nodes
CM3

CM4

CM5

CM0

CM1

CM2

CM3

CM4

CM5

374.14
801.58
14.79
89.38
39.93
1.74
(0.29)
4.00
5.78

246.76
260.04
24.50
88.92
99.65
2.69
(0.30)
13.34
6.94

146.60
51.14
15.63
79.11
68.32
1.71
106.77
12.14
5.64

113.41
203.96
8.96
53.68
42.57
0.61
148.13
33.74
3.02

19701
17862
863
4634
1907
211
233085
0
80

13054
53197
811
2807
843
20
32144
250
150

25542
71304
745
3692
685
0
40052
0
0

11277
15729
654
2097
1333
0
31589
8
0

8610
2597
715
1910
570
0
625
10
0

6198
8852
629
1710
752
0
3298
380
134

209.80
696.36
19.96
63.87
67.67
4.77
(0.63)
220.31
625.47

183.54
1687.35
7.91
75.78
87.82
2.19
(0.57)
25.54
73.83

339.86
543.48
15.94
112.94
214.85
3.38
(0.57)
142.29
20.53

131.53
78.93
7.65
97.16
32.86
1.66
952.09
199.87
22.89

88.21
888.37
16.25
52.00
50.73
1.28
3014.50
(0.33)
202.88

16737
32944
1326
1473
1426
381
29703
12777
1516

15631
29950
945
2951
2435
67
30739
1615
7030

9254
66906
257
2647
830
0
30454
40
519

13453
35824
690
2710
4594
0
27678
557
103

6647
6046
544
2328
425
0
8222
800
30

2866
49649
1141
1069
1163
0
35885
14182
2267

105.72
947.77
19.76
387.94
765.14
188.60

189.55
88.44
23.46
96.44
(0.34)
916.46

262.52
238.58
20.99
108.43
759.93
189.82

137.16
113.91
13.13
130.49
171.76
158.56

92.46
235.52
13.19
146.91
505.64
278.16

8900
44811
1085
21164
24278
3086

15466
5313
1275
2363
78776
18242

18429
11125
968
3592
18036
642

8928
40893
576
1955
963
480

6324
6276
611
2730
809
491

5142
9365
800
1854
18144
4892

521.64
(0.04)
23
2

565.15
(0.04)
22
1

510.66
(0.04)
22
2

114.03
(0.00)
24
7

408.10
(0.01)
23
8

11248

13170

12738

8420

2388

7099

10

242.02
1361.99
8.33
87.50
198.72
135.33
473.76
(0.04)
22
4

Y. C
- nar, H. Yaman / Computers & Operations Research 38 (2011) 16781695

rmin

max

Table 8
Results for CoverageVLP and B-type instances.
Parameters

LP gap (%)

Cpu time (s)/optimality gap (%)

kmax

CM0

CM1

CM2

CM3

50
50
50
50
50
50
50
50
50

6
8
10
6
8
10
6
8
10

4
4
4
6
6
6
8
8
8

54.24
23.05
5.77
30.68
9.27
0.00
9.21
0.00
0.00

54.24
23.05
5.77
30.68
9.27
0.00
9.21
0.00
0.00

53.75
22.69
5.66
30.34
9.27
0.00
9.21
0.00
0.00

53.75
22.69
5.66
30.34
9.27
0.00
9.21
0.00
0.00

1.91
1.58
0.62
1.31
1.15
0.00
0.44
0.00
0.00

1.91
7.26
5.64
5.78
9.27
0.00
9.21
0.00
0.00

100
100
100
100
100
100
100
100
100

6
8
10
6
8
10
6
8
10

4
4
4
6
6
6
8
8
8

54.24
23.05
5.77
30.68
9.44
0.00
9.78
0.00
0.00

54.24
23.05
5.77
30.68
9.44
0.00
9.78
0.00
0.00

53.65
22.69
5.66
30.28
9.44
0.00
9.78
0.00
0.00

53.65
22.69
5.66
30.28
9.44
0.00
9.78
0.00
0.00

1.91
1.58
0.62
1.31
1.32
0.00
0.96
0.00
0.00

1.91
7.26
5.64
5.78
9.44
0.00
9.78
0.00
0.00

150
150
150
150
150
150

6
8
10
6
8
10

4
4
4
6
6
6

54.24
23.05
5.77
31.32
13.14
8.34

54.24
23.05
5.77
31.32
13.14
8.34

53.62
22.69
5.66
28.23
10.29
5.63

53.62
22.69
5.66
28.23
10.29
5.63

1.91
1.58
0.62
1.76
4.12
5.55

1.91
7.26
5.64
6.30
13.14
8.34

Average
Avg. opt. gap (%)
# of solved ins. (/24)
# of best solutions (/24)

16.71

16.71

16.19

16.19

1.26

5.06

CM4

24

CM5

CM0

CM1

CM2

193.65
217.40
18.51
85.60
46.20
2.78
993.46
71.84
10.22

249.32
250.88
16.73
70.00
58.52
3.43
2832.36
73.22
6.66

443.58
1371.16
14.12
108.38
70.34
1.96
(0.31)
9.42
4.26

229.15
326.72
8.75
91.01
126.16
1.58
997.20
26.15
5.42

193.65
217.40
17.30
67.19
108.21
11.27
(0.44)
(0.88)
725.10

351.52
2141.64
22.74
63.94
79.64
14.70
(0.46)
1016.32
343.28

270.66
300.77
11.46
113.23
145.11
2.25
(0.47)
2016.47
42.00

303.63
1272.10
20.61
106.41
173.98
2.51
(0.36)
686.79
11.03

233.72
598.60
23.08
52.67
312.13
269.56

227.44
160.18
9.60
162.01
879.47
348.89

259.32
232.00
10.04
62.76
195.80
157.98

528.53
(0.02)
23
3

579.71
(0.03)
22
2

371.14
(0.02)
23
2

113.95
1028.98
23.13
100.34
310.53
545.08
512.58
(0.06)
22

CM3

CM4
182.90
100.75
19.83
85.15
35.54
1.49
106.86
17.15
5.18

CM5

CM0

CM1

CM2

CM3

CM4

CM5

93.13
653.08
5.49
92.55
43.66
1.62
85.75
23.76
6.82

16136
15880
1125
5087
1618
20
13939
845
220

19605
24977
898
3068
962
49
34035
786
59

26851
58418
658
4460
666
0
38492
20
0

9866
19454
622
2661
920
0
10581
50
0

5938
6358
968
2254
410
0
974
10
0

5315
28674
498
3648
733
0
874
400
190

58.47
446.48
11.45
65.47
100.91
5.31
2009.85
1789.27
(1.14)

16136
15880
978
3114
3842
418
29528
20365
7686

24076
129486
1065
2682
2313
492
29173
11097
3190

15971
23257
714
6621
1113
0
31932
6476
100

10857
73610
653
2508
1836
0
21391
1574
20

7955
5558
968
1761
521
0
15394
1477
19

2824
24236
902
2360
4664
89
18291
10274
15575

95.73
71.69
8.53
64.42
107.40
653.42

73.11
18.17
18.17
96.12
450.92
208.04

7836
68400
1200
7090
9465
5125

18333
32616
1301
1542
3899
2907

12214
8133
491
7745
14650
773

10286
13838
466
1524
903
438

3105
4084
645
1489
457
1353

4149
5525
776
2055
4974
2306

190.60
(0.00)
24
9

420.15
(0.05)
23
8

10497

14525

10823

7669

2571

5806

16

138.82
77.90
15.67
76.37
52.91
1.48
2115.30
481.46
58.59

Y. C
- nar, H. Yaman / Computers & Operations Research 38 (2011) 16781695

rmin

Number of nodes

1689

1690

Table 9
Results for CoverageVLP and C-type instances.
Parameters

LP gap (%)
k

50
50
50
50
50
50
50
50
50

Cpu time (s)/optimality gap (%)

CM0

CM1

CM2

CM3

6
8
10
6
8
10
6
8
10

4
4
4
6
6
6
8
8
8

25.73
2.12
0.00
9.33
0.00
0.00
3.07
0.04
0.04

25.73
2.12
0.00
9.33
0.00
0.00
3.07
0.04
0.04

25.71
2.12
0.00
9.32
0.00
0.00
3.01
0.02
0.02

25.71
2.12
0.00
9.32
0.00
0.00
3.01
0.02
0.02

1.19
1.73
0.00
4.20
0.00
0.00
1.34
0.02
0.02

100
100
100
100
100
100
100
100
100

6
8
10
6
8
10
6
8
10

4
4
4
6
6
6
8
8
8

25.73
2.12
0.00
9.33
0.00
0.00
3.56
0.88
0.88

25.73
2.12
0.00
9.33
0.00
0.00
3.56
0.88
0.88

25.71
2.12
0.00
9.32
0.00
0.00
3.13
0.40
0.40

25.71
2.12
0.00
9.32
0.00
0.00
3.13
0.40
0.40

150
150
150
150
150
150

6
8
10
6
8
10

4
4
4
6
6
6

25.73
2.12
0.06
9.33
0.13
0.13

25.73
2.12
0.06
9.33
0.13
0.13

25.71
2.12
0.06
8.99
0.06
0.06

5.02

5.02

4.93

Average
Avg. opt. gap (%)
# of solved ins. (/24)
# of best solutions (/24)

12

CM0

CM1

CM2

CM3

CM4

CM5

CM0

CM1

CM2

CM3

CM4

1.19
1.73
0.00
4.40
0.00
0.00
2.54
0.04
0.04

432.22
17.82
0.45
786.90
0.48
0.53
56.66
2.06
5.09

900.92
32.34
0.50
485.58
0.56
0.67
62.70
1.45
4.43

3165.87
34.28
0.81
725.56
0.93
0.90
96.86
0.64
0.78

1843.91
45.52
0.94
1299.77
0.88
0.96
144.78
0.79
1.02

150.42
22.19
0.91
533.40
0.96
0.73
82.68
0.82
1.00

158.94
43.85
0.46
261.94
0.46
0.67
105.36
1.38
6.75

25535
3541
0
43195
0
0
3140
163
456

48016
8041
0
15868
0
0
3299
40
418

165456
5304
0
14914
0
0
2386
0
0

43576
5359
0
19261
0
0
2468
0
0

5481
3325
0
5390
0
0
1164
0
0

7935
7035
0
4445
0
0
5045
20
486

1.19
1.73
0.00
4.20
0.00
0.00
1.76
0.38
0.40

1.19
1.73
0.00
4.40
0.00
0.00
3.09
0.88
0.88

342.98
85.01
1.19
284.11
0.42
4.08
196.36
11.23
13.58

379.13
25.16
0.92
541.96
1.56
0.60
273.46
11.54
18.82

477.49
14.24
0.87
388.66
1.02
1.26
191.69
10.08
25.41

1277.67
29.73
0.87
1703.63
1.02
1.08
180.66
6.99
17.09

91.70
15.64
1.22
475.63
0.81
1.21
192.90
1.41
4.97

140.39
92.02
0.56
237.37
0.53
2.44
170.01
12.65
14.22

20271
11960
0
10200
0
330
8225
482
533

19062
3464
9
13656
57
0
17602
581
709

17330
1240
0
10011
0
0
4506
43
415

30092
4053
0
23266
0
0
2313
28
170

2723
1927
0
4724
0
0
1987
0
12

7229
27160
0
4902
0
39
8702
474
525

25.71
2.12
0.06
8.99
0.06
0.06

1.19
1.73
0.06
4.20
0.06
0.06

1.19
1.73
0.06
4.40
0.13
0.13

367.43
73.07
1.48
572.66
4.86
5.03

383.30
38.82
5.31
417.78
2.34
6.50

1142.58
12.76
3.39
585.98
3.61
1.32

2294.78
61.09
4.54
1927.39
1.16
1.36

172.68
79.50
2.66
473.40
1.06
1.56

91.51
11.40
7.63
267.09
8.42
5.78

17763
7779
528
15987
283
457

18110
4756
667
10033
64
402

61517
1314
972
10880
20
0

64122
6003
558
19465
0
0

5660
9639
230
4232
0
0

4664
1095
1002
6283
474
332

4.93

1.06

1.24

135.70
(0.00)
24
5

149.85
(0.00)
24
1

7118

6869

12346

9197

1937

3660

10

20

12

CM4

24

CM5

Number of nodes

13

286.96
(0.00)
24
4

451.98
(0.00)
24

96.23
(0.00)
24
6

68.41
(0.00)
24
8

CM5

Y. C
- nar, H. Yaman / Computers & Operations Research 38 (2011) 16781695

rmin

max

Table 10
Results for CoverageVLP and D-type instances.
Parameters

LP gap (%)

Cpu time (s)/optimality gap (%)

kmax

CM0

CM1

CM2

CM3

50
50
50
50
50
50
50
50
50

6
8
10
6
8
10
6
8
10

4
4
4
6
6
6
8
8
8

25.73
2.12
0.00
9.33
0.00
0.00
3.20
0.04
0.04

25.73
2.12
0.00
9.33
0.00
0.00
3.20
0.04
0.04

25.71
2.12
0.00
9.32
0.00
0.00
3.12
0.02
0.02

25.71
2.12
0.00
9.32
0.00
0.00
3.12
0.02
0.02

1.19
1.73
0.00
4.20
0.00
0.00
1.46
0.02
0.02

1.19
1.73
0.00
4.40
0.00
0.00
2.67
0.04
0.04

100
100
100
100
100
100
100
100
100

6
8
10
6
8
10
6
8
10

4
4
4
6
6
6
8
8
8

25.73
2.12
0.00
9.33
0.00
0.00
3.68
1.21
1.21

25.73
2.12
0.00
9.33
0.00
0.00
3.68
1.21
1.21

25.71
2.12
0.00
9.32
0.00
0.00
2.99
0.64
0.64

25.71
2.12
0.00
9.32
0.00
0.00
2.99
0.64
0.64

1.19
1.73
0.00
4.20
0.00
0.00
1.82
0.64
0.64

1.19
1.73
0.00
4.40
0.00
0.00
1.82
0.68
1.21

150
150
150
150
150
150

6
8
10
6
8
10

4
4
4
6
6
6

25.73
2.12
0.06
9.33
0.13
0.13

25.73
2.12
0.06
9.33
0.13
0.13

25.71
2.12
0.06
8.98
0.04
0.04

25.71
2.12
0.06
8.98
0.04
0.04

1.19
1.73
0.06
4.20
0.04
0.04

1.19
1.73
0.06
4.40
0.13
0.13

Average
Avg. opt. gap (%)
# of solved ins. (/24)
# of best solutions (/24)

5.03

5.03

4.94

4.94

1.09

1.20

13

13

CM4

24

CM5

14

CM0

CM1

CM2

CM3

CM4

CM5

353.54
47.49
0.41
565.19
1.16
2.04
48.02
2.39
4.06

609.85
19.55
0.42
349.39
0.51
1.88
36.32
2.92
3.19

598.64
35.59
0.97
510.66
0.90
0.98
113.23
1.06
0.86

758.12
11.86
0.92
2235.29
1.09
1.11
113.57
0.86
0.86

259.94
15.30
0.99
490.20
0.91
1.20
104.06
0.70
0.80

92.84
47.08
0.52
669.14
1.30
1.25
119.09
3.69
5.31

334.94
22.64
0.56
269.34
3.58
5.05
420.18
(0.23)
(0.32)

430.85
36.73
0.48
637.17
1.58
5.59
321.45
(0.05)
(0.31)

779.60
17.40
0.87
633.62
1.17
5.35
206.31
(0.28)
(0.29)

2111.85
38.84
0.86
1841.72
1.60
1.24
415.81
41.16
123.07

243.26
107.88
0.88
468.53
1.19
1.32
311.55
86.22
85.88

110.99
45.60
0.54
186.58
2.98
2.94
395.33
(0.31)
(0.32)

345.84
60.46
6.56
363.12
1.72
1.60

447.40
40.80
5.40
985.60
2.53
1.34

1900.28
13.44
2.53
1289.40
0.92
1.46

187.28
39.14
3.87
697.38
1.04
1.37

434.89
(0.01)
22
4

482.94
(0.02)
22
4

129.62
(0.00)
24
3

1801.76
14.85
4.72
431.83
2.34
0.72
480.71
(0.02)
22
2

454.49
(0.00)
24
6

CM0

CM1

CM2

CM3

CM4

CM5

17552
7082
0
15258
70
69
3000
156
363

34889
2734
0
10412
0
51
1902
270
255

17870
4079
0
18273
0
0
3122
0
0

20915
1497
0
36960
0
0
1808
0
0

7610
1449
0
5097
0
0
1366
0
0

4460
5086
0
8462
0
39
6319
342
400

15638
6179
0
9873
365
486
35763
104629
117575

23082
5577
0
14758
46
500
16979
69792
65763

27070
1087
0
10579
0
474
8808
94567
60112

47622
4738
0
26504
0
0
4794
515
1633

7012
11227
0
5138
0
0
3219
952
1493

4881
6980
0
3494
474
61
18690
97678
118411

105.06
70.16
6.35
187.72
7.82
0.81

94078
1580
827
10803
141
0

17714
5394
559
8963
50
23

18728
6144
569
23610
20
0

54512
1782
161
14116
0
0

4178
4801
254
5708
0
0

5010
8268
760
3302
240
0

385.98
(0.03)
22
5

18395

11655

12296

9065

2479

12223

12

15

Y. C
- nar, H. Yaman / Computers & Operations Research 38 (2011) 16781695

rmin

Number of nodes

1691

Y. C
- nar, H. Yaman / Computers & Operations Research 38 (2011) 16781695

1692

rmin 100, kmax

8, and p 6 for A- and D-types. The solutions
j

An optimal solution for ProtVLP for an A-type instance is

given in Fig. 2. Here we can see that as the demands and prots of
demand points are independent of the distances to the vendors,
the service regions are quite dispersed. For instance, demand
point 14, which is assigned to the vendor at location 7, is
surrounded by three other demand points that are all served by
the vendor at location 15. Similarly, demand point 57 is served by
the vendor at location 65 even though there is another vendor at a
neighboring location.
We see an optimal solution for ProtVLP for a D-type instance
in Fig. 3. Here we observe that the service regions of vendors are
rather compact and the vendors are located more centrally in
their regions.
Optimal solutions for CoverageVLP for A- and D-type instances
are given in Figs. 4 and 5, respectively. We see a similar pattern
here, i.e., the service regions are more compact in the solution for
the D-type instance.

are depicted in Figs. 25. In all these gures, the locations of

vendors are denoted by rectangles and their service regions are
marked by different colors. Demand points that are not served by
any of the vendors are not colored. The areas that are not
population zones are not numbered.
Table 11
Improvements in linear programming bounds.
Type

A
B
C
D

ProtVLP

Coverage VLP

(16)

(18), (19), (25)(27)

(31)

(16)

(18), (19), (25)(27)

(31)

0
0
0
0

30.23
30.65
3.22
3.31

9.77
9.64
3.80
3.75

0
0
0
0

0.33
0.43
0.09
0.10

8.24
8.24
3.09
3.09

Table 12
Best prot and coverage values.
Parameters

ProtVLP

rmin

50
50
50
50
50
50
50
50
50

max

Coverage VLP

A-type

B-type

C-type

D-type

A-type

B-type

C-type

D-type

6
8
10
6
8
10
6
8
10

4
4
4
6
6
6
8
8
8

832.8
1006.8
1053.3
1094.2
1152.7
1160.2
1070.6
1071.6
1071.6

806.1
974.2
1016.9
1057.7
1113.2
1119.6
1031.5
1035.3
1035.3

852.8
1033.8
1033.8
1158.7
1250.7
1250.7
1250.7
1271.1
1270.6

828.8
1005.0
1005.0
1128.0
1214.3
1214.3
1197.9
1236.4
1236.2

1413
1790
2095
2042
2450
2677
2508
2739
2739

1413
1790
2095
2042
2450
2677
2508
2739
2739

1423
1761
1799
2069
2263
2263
2435
2509
2509

1423
1761
1799
2069
2263
2263
2432
2509
2509

100
100
100
100
100
100
100
100
100

6
8
10
6
8
10
6
8
10

4
4
4
6
6
6
8
8
8

832.8
1006.8
1053.3
1094.2
1152.7
1160.2
1070.6
1070.6
1070.6

806.1
974.2
1016.9
1057.7
1112.8
1119.2
1024.8
1030.2
1031.6

852.8
1033.8
1033.8
1158.7
1250.7
1250.7
1231.6
1271.1
1270.6

828.8
1005.0
1005.0
1128.0
1214.3
1214.3
1197.9
1236.4
1236.4

1413
1790
2095
2042
2446
2677
2495
2739
2739

1413
1790
2095
2042
2446
2677
2495
2739
2739

1423
1761
1799
2069
2263
2263
2422
2488
2488

1423
1761
1799
2069
2263
2263
2420
2480
2480

150
150
150
150
150
150

6
8
10
6
8
10

4
4
4
6
6
6

832.8
1006.8
1053.3
1074.7
1116.7
1124.2

806.1
974.2
1016.9
1042.4
1079.3
1085.9

852.8
1033.8
1033.8
1158.7
1234.7
1234.7

828.8
1005.0
1005.0
1122.0
1198.3
1198.3

1413
1790
2095
2032
2382
2502

1413
1790
2095
2032
2366
2471

1423
1761
1798
2069
2260
2260

1423
1761
1798
2069
2260
2260

1400.0

50
100
150

profit

1200.0

1000.0

800.0

600.0

4, 6 4, 8 4, 10

6, 6 6, 8 6, 10
p, kmax

Fig. 1. Best prot values for A-type instances.

8, 6 8, 8 8, 10

Y. C
- nar, H. Yaman / Computers & Operations Research 38 (2011) 16781695

1693

Fig. 2. Optimal solution of ProtVLP for an A-type instance.

Fig. 3. Optimal solution of ProtVLP for a D-type instance.

In summary, comparing these solutions, we see that demand

points assigned to the same vendor lie around the vendor node for
both ProtVLP and CoverageVLP type D problems, whereas some

demand points serviced from the same vendor are separated from
the group in ProtVLP and CoverageVLP for type A problems. This
is expected since in A-type problems, prots and demands do not

1694

Y. C
- nar, H. Yaman / Computers & Operations Research 38 (2011) 16781695

Fig. 4. Optimal solution of CoverageVLP for an A-type instance.

Fig. 5. Optimal solution of CoverageVLP for a D-type instance.

Y. C
- nar, H. Yaman / Computers & Operations Research 38 (2011) 16781695

depend on the distances between demand points and their

vendors.
Moreover, the number of demand points served is larger in
type D problems compared to type A problems. This is again
expected as the prots and demands decrease as distances
increase in D-type instances.
The total prots are 1152.70, 1214.32, 1032.20, and 992.36
and the amounts of demand covered are 2180, 2229, 2446, and
2263 for ProtVLP for type A, ProtVLP for type D, CoverageVLP for
type A, and CoverageVLP for type D instances, respectively.

6. Conclusion
In this study, motivated by a real life application, we introduced the vendor location problem. We considered two versions
of the problem with different objective functions. We proved that
both versions of the problem are strongly NP-hard and suggested
valid inequalities to strengthen the integer programming formulations and to reduce the solution times.
Our computational experiments showed that the bounds of
the linear programming relaxations of the problem with prot
maximization objective are quite poor in quality and it is very
difcult to solve these problems to optimality with integer
programming solvers. Our valid inequalities strengthened our
formulations signicantly and reduced the computation times,
however their effect was highly dependent on the instance. We
also observed that the problem with the coverage objective was
relatively easier to solve and valid inequalities were also useful in
reducing the solution times for the instances of this problem.
We solved instances with different demand and prot functions and observed that the problems with prot maximization
objective, where the demands change as a function of the
distances between the demand points and their vendors are more
difcult to solve compared to others. For some of these instances,
we could not reach an optimal solution with any of our models.
Even though the nal gaps are not very large, still, we believe that
alternative methods can be developed for these kinds of
problems.

Acknowledgment
This research is supported by TUBITAK Project no. 107M460.
References
[1] Aardal K, Chudak FA, Shmoys DB. A 3-approximation algorithm for the k-level
uncapacitated facility location problem. Information Processing Letters
1999;72:1617.
[2] Aardal K, Labbe M, Leung J, Queyranne M. On the two-level uncapacitated
facility location problem. INFORMS Journal on Computing 1996;8:289301.
[3] Ageev A. Improved approximation algorithms for multilevel facility location
problems. Operations Research Letters 2002;30:32732.
[4] Ageev A, Ye Y, Zhang J. Improved combinatorial approximation algorithms for
the k-level facility location problem. SIAM Journal on Discrete Mathematics
2004;18:20717.
[5] Albareda-Sambola M, Fernandez E, Laporte G. The capacity and distance
constrained plant location problem. Computers & Operations Research
2009;36:597611.

1695

[6] Barcelo J, Casanovas J. A heuristic Lagrangian algorithm for the capacitated

plant location problem. European Journal of Operational Research 1984;15:
21226.
[7] Barros AI, Labbe M. A general model for the uncapacitated facility and depot
location problem. Location Science 1994;2:17391.
[8] Bloemhof-Ruwaard JM, Salomon M, van Wassenhove LN. On the coordination
of product and by-product ows in two-level distribution networks: model
formulations and solution procedures. European Journal of Operational
Research 1994;79:32539.
[9] Bumb A. An approximation algorithm for the maximization version of the
two level uncapacitated facility location problem. Operations Research
Letters 2001;29:15561.
[10] Bumb A, Kern W. A simple dual ascent algorithm for the multilevel facility
location problem. Lecture Notes in Computer Science 2001;2129:5563.
[11] Chardaire P, Lutton J-L, Sutter A. Upper and lower bounds for the two-level
simple plant location problem. Annals of Operations Research 1999;86:
11740.
[12] Gao JJ, Robinson Jr. EP. A dual-based optimization procedure for the twoechelon uncapacitated facility location problem. Naval Research Logistics
1992;839:191212.
[13] Gao JJ, Robinson Jr. EP. Uncapacitated facility location: general solution
procedure and computational experience. European Journal of Operational
Research 1994;76:41027.
[14] Gabor AF, van Ommeren JKCW. A new approximation algorithm for the
multilevel facility location problem. Discrete Applied Mathematics 2010;
158:45360.
[15] Garey MR, Johnson DS. Computers and intractability, a guide to the theory of
NP-completeness. New York: Freeman; 1979.
[16] Guha S, Meyerson A, Munagala K. Hierarchical placement and network
design problems. In: Proceedings of the 41st annual IEEE symposium on
foundations of computer science; 2000. p. 603612.

[17] Holmberg K, Ronnqvist

M, Yuan D. An exact algorithm for the capacitated
facility location problems with single sourcing. European Journal of Operational Research 1999;113:54459.
[18] Kaufman L, Vanden Eede M, Hansen P. A plant and warehouse location
problem. Operational Research Quarterly 1997;28:54757.
[19] Klincewicz J, Luss H. A Lagrangian relaxation heuristic for capacitated facility
location with single-source constraints. Journal of the Operational Research
Society 1986;37:495500.
[20] Marn A. Lower bounds for the two-stage uncapacitated facility location
problem. European Journal of Operational Research 2007;179:112642.
[21] Marn A, Pelegrn B. Applying Lagrangean relaxation to the resolution of twostage location problems. Annals of Operations Research 1999;86:17998.
[22] Meyerson A, Munagala K, Plotkin S. Cost-distance: two-metric network
design. In: Proceedings of the 41st IEEE symposium on foundations of
computer science, FOCS; 2000. p. 62430.
[23] Neebe A, Rao M. An algorithm for the xed-charge assigning users to sources
problem. Journal of the Operational Research Society 1983;34:110713.
[24] Pirkul H, Jayaraman V. Production, transportation, and distribution planning
in a multi-commodity tri-echelon system. Transportation Science 1996;30:
291302.
[25] Ro H, Tcha D. A branch-and-bound algorithm for the two-level uncapacitated
facility location problem with some side constraints. European Journal of
Operational Research 1984;18:34958.
[26] Shmoys DB, Tardos E, Aardal K. Approximation algorithms for facility location
problems. In: Proceedings of the 29th annual ACM symposium on theory of
computing; 1997. p. 26574.
[27] SUDER /http://www.suder.org.tr/sektor.htmlS; 2009.

[28] S- ahin G, Sural

H. A review of hierarchical location models. Computers &
Operations Research 2007;34:231031.
[29] Tcha D, Lee B. A branch-and-bound algorithm for the multi-level uncapacitated facility location problem. European Journal of Operational Research
1984;18:3543.
[30] Tragantalerngsak S, Holt J, Ronnqvist M. Lagrangian relaxation heuristic for
two-echelon single-source capacitated facility location problem. European
Journal of Operational Research 1997;102:61125.
[31] Tragantalerngsak S, Holt J, Ronnqvist M. An exact method for the twoechelon, single-source, capacitated facility location problem. European Journal of Operational Research 2000;123:47389.
[32] Zhang J. Approximating the two-level facility location problem via a quasigreedy approach. Mathematical Programming 2006;108:15976.
[33] Zhang J, Ye Y. A note on the maximization version of the multi-level facility
location problem. Operations Research Letters 2002;30:3335.