Advances in Engineering Software 40 (2009) 766776

Contents lists available at ScienceDirect

Advances in Engineering Software

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

Scatter search algorithm for supplier selection and order lot sizing under
multiple price discount environment
R. Mohammad Ebrahim a, J. Razmi a,*, H. Haleh b
a
b

Department of Industrial Engineering, Faculty of Engineering, University of Tehran, Tehran, Iran

Faculty of Industrial Engineering, Isfahan University of Technology, Isfahan, Iran

a r t i c l e

i n f o

Article history:
Received 19 August 2008
Received in revised form 20 November 2008
Accepted 16 February 2009
Available online 26 March 2009
Keywords:
Supplier selection
Lot sizing
AHP
Price discount
MCDM
Scatter search

a b s t r a c t
Supplier selection is one the most important elements of supply chain management. This function
involves evaluation of many factors such as, cost of parts/materials, size of order, quality, and delivery
performance. Therefore, this problem is categorized as multi-criteria decision making problems. Different
approaches have been applied in order to assess and select the suppliers when suppliers offer discount on
the unit price. In practical conditions, buyers may face a situation where different types of discount may
be offered by candidate suppliers. None of the previous studies have considered different discount
schemes simultaneously. In this article a mathematical model is introduced which consider different
types of discount (all-unit cost, incremental discount, and total business volume discount) through
multi-objective formulation for single item purchasing problem. In addition, constraints such as suppliers capacity and demand are taken into consideration in the model. Due to the complexity of the problem
a proposed scatter search algorithm (SSA) is presented to solve this problem. Finally several sample problems have been solved by the proposed SSA and the exact (branch and bound) method. The results illustrate slight relative errors to compare with reasonable saving in computational times.
 2009 Elsevier Ltd. All rights reserved.

1. Introduction
Supplier selection is one the most important components of
production and logistics management in the competitive environment of the global market. As organizations become more dependent on their suppliers the direct and indirect consequences of
poor decisions become more severe [4]. Such decisions involve
the selection of individual suppliers to employ and the determination of order quantities to be placed with the selected suppliers.
Practitioners need to follow strategies to achieve higher quality, reduced costs and shorter lead times to be able to compete in the global market. Within new strategies for purchasing and
manufacturing, suppliers play a key role in achieving corporate
competition [1,9,27].
Many attributes affect a suppliers performance. Dickson [8]
identied 23 criteria that have been considered by purchasing
managers in various supplier selection problems. Also Weber
et al. [26] found that 47 out of 76 articles which they reviewed, addressed more than one criterion for supplier selection decision
making. Hence supplier selection problem is a multiple criteria
decision making (MCDM) problem and it is necessary to make a

* Corresponding author. Tel.: +98 21 88021067; fax: +98 21 88013102.

E-mail addresses: rmebrahim@ut.ac.ir (R.M. Ebrahim), jrazmi@ut.ac.ir (J. Razmi),
haleh24@hotmail.com (H. Haleh).
0965-9978/$ - see front matter  2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.advengsoft.2009.02.003

trade off between conicting quantitative and qualitative criteria

to select the best suppliers.
Suppliers sometimes offer discounts. The motivation for using
discount schemes stems from the fact that it tends to encourage
buyers to procure larger quantities and to obtain operating advantages (such as economies of scale or reducing the cost of transportation) for the buyer. From a coordination perspective, it has been
shown that both the buyer and the supplier can realize higher
overall prots if discounting schemes are used to set transfer prices
[25]. Various kinds of discount schemes which are usually offered
by suppliers are discussed in relevant studies. So, the multi-objective supplier selection problem in presence of discount schemes
becomes more complicated. Usually, two types of discounts
all-units discount and incremental discountare used for cost
reduction [3]. All-units model divides the range of possible order
quantities into intervals with progressively lower unit costs. The
unit cost corresponding to the size of a particular order is applied
to every unit in the order. Dolan [10] and others argue, however,
that the supplier is often better served by the incremental discount
model, which applies a lower unit price only to those units purchased in excess of each successive breakpoint [21]. Finally total
business volume discount model, described by Xia and Wu [28],
is usually used when more than one item is to be purchased.
In this paper we propose a mathematical model for a single
item purchasing problem considering various discount schemes
simultaneously. Considering both qualitative and quantitative

R.M. Ebrahim et al. / Advances in Engineering Software 40 (2009) 766776

criteria in this model and in addition, permitting various types of

discount schemes by suppliers, make this model more practical
in comparison with other previous studies.
The remainder of this paper is organized as follows. In Section 2,
some previous studies and researches on supplier selection problem in presence of discount offers are reviewed. In Section 3, the
developed model of supplier selection problem when suppliers
may offer various discount models is presented. In this model qualitative criteria are considered as well as quantitative criteria. In
Section 4, the complexity of the problem is discussed and then a
scatter search algorithm is proposed and in Section 5 the described
problem and the impacts of varying situations on both formulation
and optimum results are analyzed through some examples. In Section 6 the performance of the proposed SSA is evaluated by solving
some sample problems. Finally in Section 7, conclusions achieved
from this research are discussed.

2. Literature review
Relevant studies in the literature of the subject have been divided into two major categories. First category of studies includes
the papers which consider just one criterion, usually the cost of the
procurement for supplier selection problem while suppliers offer
discounts on the quantity of materials being purchased.
Goossens et al. [13] discussed the procurement problem of buying multiple items from a set of suppliers considering just the minimization of purchasing cost function so other affecting criteria are
not included in their linear mathematical model. In their study
authors assume that all suppliers offer just all-unit discount scheme.
They argue that not only this problem is NP-hard but also there exists no polynomial-time approximation algorithm with a constant
ratio. Burke et al. [5] analyze the impact of supplier pricing schemes
and supplier capacity limitations on the optimal sourcing policy for
a single buyer. They consider the situation where the total quantity
to be procured for a single period is known by the rm and communicated to all suppliers. Each supplier quotes a price and capacity
limit as a maximum quantity that can supply to the buyer. According to this information, the buyer makes a decision for quantity allocation among the suppliers and consequently a subset of suppliers is
selected for order allocation. A variety of supplier pricing schemes
from the constituent group of suppliers is analyzed. Kothari et al.
[14] describe procurement auction with marginal decreasing piece
wise-constant supply curves. All-unit discount is allowed by this
auction. They present fully polynomial-time approximation
schemes for the winner determination problem and the computation of the corresponding payments of this auction. Benton [2] uses
Lagrangian relaxation to evaluate a purchasing managers resource
constrained order quantity decisions given alternative pricing
schedules from multiple suppliers. The author assumes that the
decision maker has a limited budget and storage space for ten items
offered by three suppliers, each quoting three all-units discount
intervals for each item. The objective is to minimize total acquisition
and inventory costs. The manager must choose a single supplier for
all items. Crama et al. [6] describe the purchasing decisions faced by
a multi-plant company. The suppliers of this company offer discount
schedules based on the total quantity of materials purchased. The
schedules simultaneously account both for corporate purchases
and for purchases at the individual plant level. The complexity of
the purchasing decisions is further increased due to the existence
of alternative production recipes for each nal product. They formulate the corresponding cost-minimization problem as a non-linear
mixed 01 programming problem. The above articles study the purchasing problem under only one discount offer by the suppliers. In
addition, their models consider just one objective (usually purchasing cost) which cannot be applied for the real cases.

767

Considering the importance of criteria such as quality of products and services provided by suppliers and also direct and indirect
impact of suppliers performance on organizations performance,
persuade organizations to consider other affective criteria in suppliers evaluation as well as cost. So the second category of papers
discussing supplier selection under discount environment declares
that supplier selection problem is a multi-objective decision making problem.
Rosenthal et al. [20] developed a mixed integer programming
model for supplier selection with bundling, in which a buyer needs
to buy various items from several vendors with limited capacity
and also with different quality and delivery performances which
offer bundled products at discounted prices. They applied single
objective programming in their model. Ghodsypour and OBrien
[11] proposed an integrated AHP and linear programming model
to help managers consider both qualitative and quantitative factors
in their purchasing activity in a systematic approach. In their model, price of the product has been offered in a discount scheme and
buyers limitations on budget, quality, and service have been taken
into account.
Amid et al. [1] consider a weighted additive fuzzy multi-objective model for the supplier selection problem under all-unit price
discounts. They formulate the problem in such a way as to simultaneously consider the imprecision of information and determine
the order quantities to each supplier based on price breaks. The
problem includes the three objective functions: minimizing the
cost, minimizing the rejected items and late deliveries, while satisfying capacity and demand requirement constraints. They argue, in
practice, for supplier selection problems, most of the input information is not known precisely, so they use the fuzzy optimization
theory to deal with this vagueness. A fuzzy weighted additive and
mixed integer linear programming is developed.
Dahel [7] presents a multi-objective mixed integer programming approach to simultaneously determine the number of suppliers to employ and the order quantities to allocate to these
suppliers in a multi-product, multi-supplier competitive sourcing
environment. The objectives to be optimized are: cost, delivery
and quality subject to the capacity constraints of suppliers. In their
selection problem the discount on the total business volume is
considered. Similar to the study of Dahel [7], Xia and Wu [28] discuss the multi-objective supplier selection problem under total
business volume discount environment. They propose an integrated approach of analytical hierarchy process improved by rough
sets theory and multi-objective mixed integer programming to
simultaneously determine the number of suppliers to employ
and the order quantity allocated to these suppliers while purchasing multiple products from a set of suppliers with multiple criteria.
In this study, they assume that suppliers offer price discount on total business volume. The above papers consider multiple objective
problems; however, their model cannot deal with multiple discount schemes simultaneously.
It is obvious that in practical situation buyers face with multiple
sourcing in which suppliers may offer different price discount
schemes. As it has been described above, there is need to develop
a model to consider various discount schemes simultaneously. In
this paper we intend to develop a mathematical model to consider
three discount schemes (all-unit discount, incremental discount
and total business volume discount) and some affecting factors
in a multi-objective, single-item supplier selection problem.

3. Description of the proposed model

Various criteria are proposed for evaluation of suppliers. Dickson [8] identied 23 criteria that have been considered by purchasing managers in various supplier selection problems. Tracy [24]

768

R.M. Ebrahim et al. / Advances in Engineering Software 40 (2009) 766776

discussed the importance of criteria such as quality, on-time delivery and improvement of the performance of products in supplier
selection problem. Min and Galle [17] proposed seven criteria such
that each criterion has some sub-criteria.
By attention to the importance of intangible factors in evaluation of suppliers, a trade off between tangible and intangible factors should be made. Hence, we utilize some MADM methods
such as analytic hierarchy process (AHP) [22] to obtain a total
weight (accounted by AHP) for each supplier which implies that
the total performance of that supplier in intangible factors. So,
one of our objectives in this study is to maximize the total
weighted quantity of purchasing. We will discuss this objective later in this section in detail. We also consider three other objectives
as follows: minimizing the total cost of purchasing under discount
environment, minimizing the total defective items and minimizing
the total number of late delivered items. In the following sub-sections, objective functions and constraints are discussed in detail.
3.1. Assumptions and notations
The following assumptions are considered in denition and the
modeling of the problem.
 The buyer can purchase required quantity from multiple
suppliers.
 The buyer is going to purchase only one item from suppliers.
 Each supplier can offer one of three discount schemes arbitrary.
 The total demand is known and the total purchased quantity is
exactly equal to the total demand.
 The total demand and purchased quantities from suppliers are
integer.
 The buyer is purchasing just for one period.
Notation
index of suppliers
index of discount intervals
purchased quantity from supplier i in discount interval k
(integer variable)
binary variable; if the purchased quantity from supplier i
yik
falls on the interval corresponding to this variable then
yik = 1, otherwise yik = 0
n
total number of suppliers
number of suppliers that offer all-unit discount
n1
number of suppliers that offer incremental discount
n2
lower bound of the discount interval k offered by supplier i
lik
upper bound of the discount interval k offered by supplier i
uik
discounted unit price of the discount interval k offered by
pik
supplier i
defective rate of purchased goods from supplier i
di
late delivery rate of goods purchased from supplier i
hi
total weight of supplier i calculated by AHP
wi
index of the last interval offered by supplier i
Ki
D
total demand
i
k
xik

3.2. Cost function

from supplier i. This discount scheme divides the range of possible

order quantities into intervals with progressively lower unit costs.
The unit cost corresponding to the size of a particular order is applied to all-units in the order. Let k = 1, . . . , Ki be the index for discount intervals offered by supplier i. Corresponding to each
discount interval k for a supplier i, dene [lik, uik] as the minimum
and maximum quantities of that interval and pik as the per unit
price. The cost function under all-unit discount policy can be modeled as follows:
Ki
n1 X
X

3.2.1. All-unit discount

Under this discount model, supplier i discloses the all-units discount scheme [18] which depends on the quantity xi purchased

where at most one of the variables xik for k = 1, . . . , Ki can be positive

and the rest must be equal to zero and it is assumed that
pi1 > pi2 >    > piki .
3.2.2. Incremental discount
Under this discount model, supplier i discloses the traditional
incremental units discounting scheme [18] which is dependent
on the quantity xi purchased from supplier i. This discount scheme
applies a lower unit price only to those units purchased in excess of
each successive breakpoint. As with all-unit discount scheme
k = 1, . . . , Ki is the index for discount intervals offered by supplier
i. Corresponding to each discount interval k for a supplier i, we dened [lik, uik] as the minimum and maximum quantities and pik as
the unit price. The cost function under incremental discount policy
has been modeled as follows:
Ki
n
k1
X
X
X
pik xik  yik ui;k1 yik
pij uij  ui;j1 ;
in1 1 k1

j1

where at most one of the variables xik for k = 1, . . . , Ki can be positive

and the rest must be equal to zero and it is assumed that
pi1 > pi2 >    > piki .
3.2.3. Total business volume discount
Similarly to all-unit and incremental discount schemes, under
this scheme, supplier i offers some price intervals. In two previous
discount schemes, price breaks are a function of the order quantity
but in total business volume discount, price discounts depend on
the total value of sales volume. This discount scheme has got a different formulation when multiple items are being purchased but in
the case of single item purchasing this scheme can be formulated
as same as all-unit scheme. We prove this in the following
theorem.
Theorem 1. In the case of single item purchasing the total business
volume discount scheme can be formulated as same as all-unit
scheme.
Proof. Let Si be the set of all suppliers providing item i offering the
total business volume discount and Rj be the set of discount intervals of supplier j. Let Vjr be the business volume purchased from
supplier j in discount interval r and djr be the percent of discount
in interval r offered by supplier j. So the cost function is:

XX
The total cost of purchasing depends only on the unit price of
item. The price for commodity items usually offered in different
discount schemes which can be seen in one of the following structures: All-unit discount, Incremental discount and total business
volume discount. In this section we propose a model in which all
of these discount models are considered simultaneously.

pik xik ;

i1 k1

djr V jr ;

j2Si r2Rj

where

X
r2Rj

V jr

pij xij ;

j 2 Si ;

i2C j

Cj is the set of items offered by supplier j. Since there is just one

item to be purchased, Cj has just one member so we have
P
r2Rj V jr pj xj and Si = S is the set of all suppliers. Let ajr and bjr be
respectively the lower and upper bound of the interval in which

769

R.M. Ebrahim et al. / Advances in Engineering Software 40 (2009) 766776

Vjr lies. With attention to Vj = pjxj which implies that the business
volume purchased from supplier j is the total quantity purchased
multiplied by the unit price without considering discount, and this
fact that Vj lies in an interval (for example r), so we have

V jr xj pj
ajr 6 V jr 6 bjr

) ajr =pj 6 xj 6 bjr =pj

or ljr 6 xj 6 ujr :

So, xj lies in an interval which depends on r so we can have the following relation:

l 6 xjr 6 ujr ) V jr pj xjr )

XX

j2S

XX

djr pj xjr

XX

r2Rj

j2S

djr V jr

r2Rj

XX
j2S

pjr xjr :ljr 6 xjr 6 ujr ) V jr pj xjr )

r2Rj

j2S

XX

j2S

djr pj xjr

r2Rj

XX
j2S

djr V jr

r2Rj

pjr xjr :

r2Rj

where pjr is the discounted unit price of interval r offered by supplier j. This model is as same as all-unit discount model. h
So, in general, we just include all-unit and incremental discount
schemes in formulating the total cost function.
3.2.4. Total cost function
Let n1 suppliers offer all-unit discount schedules and n2 = n n1
suppliers offer incremental discount schedules. So by adding (1),
(2) we have
Ki
n1 X
X

pik xik

i1 k1

Ki
n
X
X

pik xik  yik ui;k1 yik

k1
X

in1 1 k1

Ki
n X
X

!
pij uij  ui;j1

j1
Ki
n
X
X

xik pik

i1 k1

yik

k1
X

in1 1 k1

pij uij  ui;j1  pik ui;k1 :

j1

Hence, the cost objective function can be stated as follows:

min

Z1

Ki
n X
X
i1 k1

Ki
n
X
X

xik pik

in1 1 k1

k1
X

yik
!

pij uij  ui;j1  pik ui;k1 :

j1

3.5. Total weighted quantity of purchasing function

Applying MADM techniques such as AHP, organizations can involve intangible criteria in supplier selection decision making
problem. Depending on the purchasing strategies; various criteria
can be included in decision making process. Table 1 illustrates
some criteria which can be used in common. Obviously, buyers
can include or exclude some of these criteria according to their
organization policies.
In this paper we use the AHP method to obtain weights of suppliers. The AHP [22] is the well-known and useful method to obtain
weights of each alternative in multiple criteria decision making
problems. The AHP consists of three steps [28] of making hierarchical structure, constructing the matrix of pair wise comparison ratios and calculating the total weights for each alternative. The
hierarchy structure of criteria is seen in Table 1. The matrixes of
pair wise comparison ratios can be constructed using scores 1
to 9 in which score 1 is used in order to show equal importance
of two criteria and score 9 is applied to show the absolute superiority of one criterion to another. In this study we use the standard
method of calculating weights introduced by Saaty [22].
So, the fourth objective is as follows:

max

Z4

min

Z2

In real situation, for supplier selection problems, the weights of

each criteria and objectives are different and it is depend on their
purchasing strategies [25]. Suppose that W1, W2, W3 and W4 are the
weights of objective functions. We use the weighted sum of these
four objective functions as the main objective to be minimized.
Since, coefcients are of different types we have to normalize the
coefcients of decision variables. So their summation will be
meaningless. We use linear normalization method introduced in
[23] to do this.
Linear normalization: r ik maxxikfx g for benet attributes and
i;k ik
mini;k fxik g
for cost attributes, where i is the index of suppliers
rik
xik
and k index for intervals.
Since we intend to minimize the total weighted sum of these
objectives, we regard the total weighted quantity of purchasing
as cost attribute and the three other objectives as benet attributes. So the nal objective function can be stated as:
Ki
n X
n
X
X
W 1 pik W 2 di W 3 hi W 4 wi xik W 1
i1 k1

Ki
X

yik

k1

k1
X

di xik :

In order to reduce the total lead time of manufacturing, the number of production breakdowns, and the holding cost, buyers desire
to minimize the total late delivered items purchased from suppliers. So, the third objective function has been dened as Eq. (5):
Ki
n X
X
i1 k1

in1 1

pij uij  ui;j1  pik ui;k1 :

j1

Criteria

Sub-criteria

Services

Length of guarantee period

Available services during guarantee period
Needed training for use of production

Background of relationships

Length of the relation period

Importance of relations
Level of mutual satisfaction during relations

Organizational level

Technological level
Level of information technology
Capital of the supplier
Flexibility in manufacturing

Communication

Capability of getting in touch by the buyer

Available information about supplier

3.4. Late delivered items function

Z3

Table 1
Proposed criteria for supplier selection decision making.

i1 k1

min

3.6. Final objective function

min

Ki
n X
X

wi xik :

i1 k1

3.3. Defective items function

The buyer usually intends to minimize the number of defective
items for improving product quality and to reduce costs related to
quality such as reworks and cost of guaranty. This minimization
can also leads to more satisfaction of buyers. This objective function can be stated as Eq. (4):

Ki
n X
X

hi xik :

770

R.M. Ebrahim et al. / Advances in Engineering Software 40 (2009) 766776

lik yik 6 xik 6 uik yik

After normalizing the coefcients we have:

n
X

min

Ki
X

n
X

aik xik

i1 k1

Ki
X

bik yik ;

di
hi
minr fwr g
;
W3
W4
wi
maxr fdr g
maxr fhr g
P
s1

bik W 1

j1 pij

maxr;s ; prs ;

P

yik 2 f0; 1g 8i 1; . . . ; n 8k 1; . . . ; K i

s1
j1 prj urj



 pik ui;k1
o:

 ur;j1  prs ur;s1

uij  ui;j1

3.7. Constraints
There are some constraints associated with the supplier selection problem. In the following these constraints are explained
and modeled.
3.7.1. Demand constraint
The rst constraint with which we are faced is the demand constraint that implies the total purchased quantity must be equal to
the total demand of the buyer. This is modeled as follows:
Ki
n X
X

yik 6 1 8i 1; . . . ; n

xik 2 z [ f0g 8i 1; . . . ; n 8k 1; . . . ; K i

pik
P

n
o
s1
maxr;s prs ;
p
u
rj  ur;j1  prs ur;s1
j1 rj

W2

8i 1; . . . ; n 8k 1; . . . ; K i

k1

in1 1 k1

where

aik W 1

Ki
X

xik D:

10

This model is a linear, integer programming model which has

P
P
1 n 2 ni1 K i constraints and 2 ni1 K i variables. There are two
main reasons why we used the weighted sum of objective functions.
Firstly, buyers usually look for a unique decision to follow. Therefore, solving a multi-objective problem leads to a set of solutions
which can confuse the buyer. On the other hand, the importance
of objectives may not be the same. Hence, in order to involve the
importance of objective functions and obtaining a unique optimum
decision the weighted sum of objective functions has been
considered.
Assume all suppliers offer all-unit discounts, in this condition
n2 = 0 and the problem is relaxed. Goossens et al. [13] shows that
the supplier selection problem under all-unit discount policy is
strongly NP-complete and no polynomial-time approximation
algorithm with constant worst-case ratio exists for this problem
(unless P = NP). Burk et al. [5] discuss that purchasing problems
under all-unit or incremental discount policy are categorized
among NP-hard problems. Therefore, we need to apply a heuristic
algorithm for our current problem which certainly can be classied
in the NP-hard problems. In the next section we propose a Metaheuristic algorithm to nd the solutions with a desirable quality
in an appropriate computational time.

i1 k1

3.7.2. Discount intervals constraints

Variables corresponding to intervals must be equal to or between their lower and upper limits. Since at most one of the intervals offered by each supplier must be selected, the binary variable
yik is utilized in these constraints. Thus, if an interval is not selected, the binary variable corresponding to that interval will be
zero and the corresponding quantity variable xik will be zero as
well. These constraints can be stated as Eqs. (11) and (12):

lik yik 6 xik 6 uik yik

Ki
X

8i 1; . . . ; n 8k 1; . . . ; K i ;

yik  1 8i 1; . . . ; n:

11
12

k1

3.7.3. Capacity constraints

This constraint implies that the total purchased quantity from
each supplier must be equal or less than the supply capacity of
the nominated supplier. So we have the following relation:
Ki
X

xik 6 C i

8i 1; . . . ; n:

However, for k = Ki we have xik 6 uiK i C i , so this constraint is

already considered in Eq. (11) and adding the above constraint to
the model is not bounding and is redundant.
3.8. The nal mathematical programming model
The nal proposed mathematical model for supplier selection
under discount policy can be formulated as follows:
Ki
n X
X

aik xik

i1 k1

s:t:

Ki
n X
X
i1 k1

Ki
n
X
X
in1 1 k1

xik D

In this paper scatter search algorithm (SSA) is considered as a

population-based method, such as genetic algorithms. The scatter
search methodology has been introduced by Glover [12] as a heuristic for integer programming, based on strategies presented at a
management conference held in Austin, Texas in 1967. The scatter
search Meta heuristic, because of its capability in solving both
combinatorial and continuous optimization problems, has successfully been applied to a widespread variety of hard optimization
problems such as assignment problems, binary problems and
non-linear optimization [19].
In the original proposal, Glover described scatter search as a
method that uses a succession of coordinated initializations to generate solutions. He introduced the reference set (Ref-Set) of solutions and several guidelines, including that the search takes place
in a systematic way as oppose to the random designs of other
methods [15]. In the following section, structure of the scatter
search is described.
4.1. Scatter search structure

k1

min

4. Applying scatter search algorithm to solve the problem

bik yik

The scatter search methodology is very exible, since each of its

elements can be implemented in a variety of ways and degrees of
sophistication. In this section we illustrate the structure of scatter
search based on the well-known ve methods [16]. The following
template for implementing scatter search consists of ve methods.
1. A diversication generation method to generate a set of diverse
trial solutions, using an arbitrary trial solution (or seed solution) as an input.
2. An improvement method to transform a solution into one or
more enhanced solutions. (Neither the input nor the output
solutions are required to be feasible, though the output solutions will more usually be expected to be so.)

R.M. Ebrahim et al. / Advances in Engineering Software 40 (2009) 766776

3. A reference set (Ref-Set) update method to build and maintain a

reference set consisting of the b best solutions found (where
the value of b is typically small, e.g., no more than 20), organized to provide efcient accessing by other parts of the
method. Solutions gain membership to the reference set according to their quality or their diversity.
4. A subset generation method to operate on the reference set, to
produce a subset of its solutions as a basis for creating combined solutions.
5. A solution combination method to transform a given subset of
solutions produced by the subset generation method into one
or more combined solution vectors.
Fig. 1 shows the interaction among these ve methods.
4.2. Proposed scatter search algorithm
Before describing the proposed SSA we present an algorithm by
which we can equalize the discount intervals offered by suppliers.
By doing this, the proposed scatter search algorithm can work easier. Let [lik, uik] represents kth interval offered by supplier i and the
total number of suppliers is n and K maxi fK i g where Ki is the
maximum supply capacity of supplier i.
Algorithm 1. Equalizing discount intervals
Step
Step
Step
Step
Step
Step

1.
2.
3.
4.
5.
6.

For all i let l1 : li1 = 1

j=1
uj : mini fuij g
lj+1 : uj + 1
If (j <= K) j : j + 1, go to 3
Stop.

Fig. 2. Equalizing discount intervals.

4.2.1. Diversication generation method

Similar to other population based Meta-heuristic methods, the
rst step in the SSA, is to build a set of trial solutions. Since the vector solutions in this problem have the format described in (13), the
vector solutions built in this step must follow this format and number of these solutions should be at least about 50.
For ith solution we begin from jth supplier where (j = i mod n &
if (j = 0) then j = n, and n is the total number of suppliers. Let ith
solution is being built; the sum of quantities assigned to suppliers
must be exactly equal to D. The value of j is calculated and then we
begin from jth supplier. So a random integer value xj from interval [1, Kj] to this supplier. Now D is updated as D = D  xj. Now we
put j = j + 1 & if(j = n + 1) then j = 1. We do the above process until
all suppliers are assigned a value or D = 0. If all suppliers are assigned a quantity but
D is still non-zero, the above process must be repeated until D is
zero. This set is called Set P.
4.2.2. Improvement method
This step is considered to improve the quality of solutions and
to obtain more enhanced solutions. This step consists of two
phases.

In the process of equalizing the intervals it happens for suppliers with lower capacities that all of their offered intervals are
equalized before Algorithm 1 is stopped. In this case, we omit
the index of the corresponding supplier from the set of suppliers
for rest of process. Fig. 2 illustrates this algorithm.
Vector solutions in this problem have the following general
format:

1
S1
Sn
z}|{
z}|{
S @x11 ; . . . ; x1K 1 ; . . . ; xn1 ; . . . ; xnK n A

771

4.2.2.1. Phase 1. Let i = 1. For j = i + 1 to j = n we do the following

process. Assume that the assigned quantity to supplier i falls in
interval ki and similarly the assigned quantity to supplier j falls
in interval kj. So, if aiki xiki biki ajkj xjkj bjkj  aikj xikj
bikj ajki xjki bjki > 0 then the assigned quantities to supplier i
and j, i.e. xiki ; xjkj are exchanged. If xiki and xjkj are zero, then
aiki ; biki ; ajkj ; bjkj will be zero. Now if i n - 1 then i is updated as
i = i + 1 and then again we start from the beginning of the process
with the updated i.

13

Here we set these ve methods for our supplier selection

problem.

4.2.2.2. Phase 2. In this phase, for each vector solution, those assigned quantities (positive quantity variables) which are lower
than a xed determined quantity are considered. This xed quantity which is the lowest acceptable quantity that must be pur-

Fig. 1. Scatter search owchart.

772

R.M. Ebrahim et al. / Advances in Engineering Software 40 (2009) 766776

chased from selected suppliers, can be determined based upon

organization policy. We add up these assigned quantities (which
are under this acceptable threshold) with each other and then assign the resulted value to the best supplier who has extra capacity
more than the above calculated summation quantity. In order to
evaluate these suppliers and select the best one we assign the calculated summation quantity to these suppliers and select the supplier that leads to larger decrease in objective function.
4.2.3. Reference set (Ref-Set) updating method
In scatter search algorithm, Ref-Set consists of two subsets. In
the rst subset, solutions which have higher quality or better
objective value enter into this subset. Members of this subset
are sorted with respect to their objective value in such way that
the member with the best objective value is considered as the rst

member of this subset and so on. In order to update this subset

the objective value of all solutions which are available must be
computed and then b1 solutions with the best objective value
are selected and are sorted as discussed above. The second subset
of Ref-Set is the set of diverse solutions. Combination of these
solutions with other solutions in Ref-Set may result in solutions
with good quality. For updating this subset the remaining of solutions which have not entered the rst subset of Ref-Set are considered. We dene a distance index as follows in order to measure
distance of these solutions with solutions in the rst subset of RefSet.

v
u n Kr
uX X
distij t
xirk  xjrk 2 ;
r1 k1

Begin
Initializing (Set P):
for (q:=1) to (q:=200) do{
p:=q (mod) n;
if (p=0) then p:=n;
for (p:=1) to (p:=n) do
{ xp:=random (1, min{ capacity of sup. p & D});
D:=D- xp;
p:= p+1;} While( D>0)
Determine the corresponding interval for x p;}
// end of Initializing (Set P);
Impv. Method:
Phase 1:
for (i:=1) to (i:=n-1) do
{ for (j:=i+1) to (j:=n) do {
if (capacity of sup. i >=xj & capacity of sup. j>=xi) then (swap the values of x i and
xj);
if ( func. value( new)< func. value( old)) then (keep this swapping fixed)
otherwise (discard it); }
Phase 2:
Determine a minimum order limit (MOL);
sum:=0;
for (i:=1) to (i:=n) do {
if (xi <= MOL) then ( sum:=sum+ xi);}
for (i:=1) to (i:=n) do { allocate sum to sup. i if possible;
if ( func. value( new)< func. value( old)) then (keep this allocation fixed)
otherwise (discard it);
// end of Impv. Method;
Ref-Set Updating:
for (i:=1) to (i:=maximum index of solutions available) do
{ sort according to Func. Value)}
for (j:=1) to (j:=10) do
{ copy the jth sorted solution in Ref-Set) }
for (i:=11) to (i:=maximum index of solutions available) do
{ compute dist =
ij

Kr

( x
r =1 k =1

i
rk

x rkj )

dist i* = m in {dist i j }
j

//j is the index of solution in Ref-Set

and sort according to dist factor}
for (j:=11) to (j:=20) do { copy the jth sorted solution in Ref-Set) }
// end of Ref-Set Updating;
Subset Generation Method;
Combination Method:
for (the candidate couple solutions S1 & S2) do ( output S:= (S1 + S2)/2);
// end of Combination Method;
Impv. Method;
Ref-Set Updating;
if ( Ref-Set is changed) then ( go to Combination Method)
Stop;
Fig. 3. The proposed SSA pseudo-code.

773

R.M. Ebrahim et al. / Advances in Engineering Software 40 (2009) 766776

where distij is the distance of solution i from solution j. Now the distance index is measured as follows:


dist i minfdist ij g:
j

The number of b2 solutions with highest distance index value is selected. These are sorted with respect to their distance and placed in
the second subset.
4.2.4. Subset generation method (constructing new solution sets)
In this step, new subsets are constructed from Ref-Set. Let the
Ref-Set have b1 + b2 members, where b1 is the number of members
in the rst section of Ref-Set and b2 is the number of members in
the second section of Ref-Set containing diverse solutions. Therefore, there are 2b1 b2  1 non-empty subsets that can be constructed from Ref-Set. Depend on the combination method; some
of these subsets may be useful. In this paper, in order to optimize
the computational time, we use only the subsets which have just
two members.
4.2.5. Solution combination method
The inputs of this step are outputs of the previous step. Members of the subsets made in the previous step are combined together and consequently new solutions are obtained. In this
paper, the convex combination of each couple of solutions which
are the members of subsets constructed in the previous step is
computed. Since the value of k (parameter of convex combination)
is considered to be 0.5, the outputs are feasible.
In our proposed SSA these ve steps are utilized as shown in
Fig. 3. In the following section some numerical examples are provided to illustrate how the problem formulation and solutions
can be altered by different situations.

general multiple discount problem which this example has been

solved and the results have been analyzed. In order to study how
the proposed model is affected by the discount schemes, in second
example the cost factor has been considered as the only affecting
factor.
Example 1. Suppose a buying company is going to purchase total
demand of 500 units of a raw material and there are four suppliers
as alternatives such that n1 = 2 of them offer all-unit discount
schedules and n2 = 2 of them offer incremental discount schedules.
Table 2 shows discount intervals offered by these suppliers. In
Table 3 rates of defective late delivered items are presented, more
over, the nal weight of each supplier is listed in this table. These
weights are computed according to qualitative criteria such as
criteria in Table 1. The AHP method is applied to calculate these
weights using scores 1 to 9 for computing matrixes of pair wise
comparison ratios. For instance, matrix of pair wise comparison
ratios of criteria services, background of relationships, organizational level, delivery and relationships is constructed
according to Table 4. Similar matrixes can be constructed for subcriteria. Now we have all required data to formulate the problem.
Using (8) and (9), aik, bik can be computed "i = 1, . . . , n "k = 1, . . . , Ki.
Applying Algorithm 1, equalized lower and upper bounds of
discount intervals are computed. Table 5 shows the result of these
computations. Therefore, the mathematical model of this example
will be as follows:

min

Z 2:456189  x1;1 2:456189  x1;2 2:456189

 x1;3    0:514  y4;12 0:514  y4;13 ;

s:t:

X 1;1 x1;2 x1;3    x4;12 x4;13 500;

x1;1 6 50  y1;1 ;
x1;2 6 75  y1;2 ;
..
.

5. Examples with evaluation

x4;12 6 350  y4;12 ;

In the following examples we demonstrate how the proposed

model can be applied for a supplier selection and purchasing planning under multiple discount offers. The rst example expresses a

x4;13 6 375  y4;13 ;

x1;1 P 1  y1;1 ;
x1;2 P 51  y1;2 ;
x1;3 P 76  y1;3 ;
..
.
x4;12 P 301  y4;12 ;

Table 2
Discount intervals of example.
# of supplier

Intervals

Unit prices

1100
101200
201300
301400

623
534
465
383

150
51100
101150
515200
201250
251300
301350
351400
401450
451500

670
610
560
510
470
440
400
360
320
290

180
81140
141200
201280
281350
351400
401500

660
612
550
520
450
370
280

175
76150
151225
226300
301375

650
620
530
460
390

x4;13 P 351  y4;13 ;

y1;1 y1;2    y1;13 y1;14 6 1;
y2;1 y2;2    y2;15 y2;16 6 1;
y3;1 y3;2    y3;15 y3;16 6 1;
y4;1 y4;2    y4;12 y4;13 6 1;
This model is solved using branch and bound (B&B) method using
LINGO package version 8.0 and the proposed SSA. The corresponding solutions are as follows: x3,5 = 125 and x4,13 = 375 and the
remaining variables are equal to zero. This means that the best purchasing strategy is to buy 125 units of total demand from supplier 3
in discount interval 5 and the remaining quantity of 375 units from
Table 3
Data related to each supplier.
# of
supplier

Rate of defective
items (di)

Rate of late delivered

items (hi)

Final weight of
suppliers (wi)

1
2
3
4

0.009
0.02
0.008
0.006

0.01
0.009
0.01
0.008

0.2016
0.3222
0.2734
0.2028

774

R.M. Ebrahim et al. / Advances in Engineering Software 40 (2009) 766776

Table 4
Matrix of pair wise comparison ratios of criteria of Table 1.
Criterion i

Services
Background of
relationships
Organizational
level
Communication

Criterion j
Services

Background of
relationships

Organizational
level

Communication

1
1/4

4
1

2
1/2

7
1/3

1/2

1/2

1/7

supplier 4 in discount interval 13. The corresponding objective

function value under this optimal policy will be 1055.186. The selected suppliers are offering incremental discount schedules and
none of the suppliers offering all-unit discount are selected. It does
not imply that the incremental discount is generally preferred
rather than the all-unit discount. On the other hand the basic unit
price of supplier 1 is lower than the 4th suppliers. In addition the
percentages of discounts for same intervals offered by supplier 1
are generally greater than the 4th suppliers. Therefore, considering
only the cost of purchasing, supplier 1 seems to be more preferable.
The prominent answer to the question why supplier 4 is preferred is
the impact of other criteria. To prove this we discuss the situation
where the buyer is just considering the cost as the main objective
of decision making problem.This problem is also solved by our proposed algorithm within just two iterations and the same results are
obtained. The computational time of the B&B is 1 s while the SSAs is
0.4 s.
Example 2. Given the data in Example 1, we consider a situation in
which the only important factor for the buyer is cost. Therefore
the weight of this criterion will turn to 100% and other three objectives will be omitted from the model. In this way we can examine
the effect of cost and discount models on the optimal decision. The
coefcients of objective function will be unit costs presented in
Table 5 and they can be used directly in objective function without
any normalization. The optimal purchasing decision found by B&B
shows that x1,8 = 220 and x3,10 = 280. As it was expected, supplier 1
is selected in this case instead of supplier 4. Even though supplier 3
is offering incremental discount, is still one of the selected suppliers. That is because of percentage of discounts considered for intervals that is obviously lower than the 2nd suppliers. Running the
SSA using the process illustrated in Fig. 3 leads to the same outputs
within 0.8 second while LINGOs computational time is 1.2 s.Con-

sequently, involving multiple criteria make the prediction of optimal policy and opted suppliers difcult. In addition, in case of
regarding just the cost factor, both discount model and discount
percentages are inuencing the problem and its optimal solution.
Example 3. In this example we intend validate our developed linear integer mathematical programming by comparing it with the
non-linear mathematical programming model described by Burk
et al. [5]. In their model they have assumed that the only criterion
is the cost of purchasing and in addition suppliers are allowed to
offer just one type of discount such as incremental discount.
Therefore, given the data of Example 1, suppose that similar to
Example 2 just the cost function is considered. More over, suppliers
are assumed to offer just incremental discount. This problem is
modeled using both model of Burk et al. and the developed model
of this paper. Running LINGO with B&B solver method, it is seen
that optimal solutions obtained by solving both models are the
same which is purchasing the whole 500 units of demand from
supplier 2 in discount interval 16. Therefore, x2,16 = 500, y2,16 = 1
and other variables are zero. The computational time of B&B for
the model of Burk et al has been 5 min and 33 s while this time
consumed by B&B for solving our model has been 2 s. This problem
is solved by the SSA as well and the same outputs have been
obtained within just 0.7 s.

6. Performance evaluation of the proposed SSA

In order to evaluate the performance of the proposed SSA in
nding the optimal purchasing decision, 24 problems have been
solved by both the SSA and LINGO on a Pentium 2.8 GHz PC with
1 GB RAM. We used LINGO package version 8 to obtain optimal
solutions. In Table 6 there are eight groups of problems such that
each group has three subgroups and the total number of suppliers
in each subgroup is the same. However, the coordinates (n1, n2) for
these subgroups are not equal where n1 is the number of suppliers
offering all-unit discount and n2 is the number of suppliers offering
incremental discount. Furthermore, in each group the offered discount intervals for each supplier cannot be changed from one subgroup to another.
In the third and fourth columns of Table 6, the values of objective function obtained by both SSA method and LINGO are respectively presented. Although we declared that the supplier selection
problem described in this paper is NP-hard and it is not suitable
choice to use exact algorithms to nd optimal solutions, but we
compare the computational time of our proposed SSA with LINGOs

Table 5
Objective coefcients and upper and lower bounds of discount intervals.
Number of interval

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16

Supplier 1

Supplier 2

Supplier 3

Supplier 4

aik

bik

aik

bik

aik

bik

aik

bik

2.456189
2.456189
2.456189
2.456189
2.455229
2.455229
2.455229
2.454553
2.454553
2.454553
2.454553
2.45375
2.45375
2.45375

0
0
0
0
0
0
0
0
0
0
0
0
0
0

2.532259
2.531671
2.531671
2.531671
2.531182
2.531182
2.530692
2.5303
2.5303
2.530007
2.530007
2.529615
2.529223
2.529223
2.528832
2.528538

0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0

2.143844
2.143844
2.143844
2.143374
2.143374
2.142767
2.142767
2.142473
2.142473
2.142473
2.141788
2.141788
2.141004
2.141004
2.140123
2.140123

0
0
0
0.037
0.037
0.122
0.122
0.181
0.181
0.181
0.373
0.373
0.647
0.647
1
1

2.100448
2.100448
2.100154
2.100154
2.100154
2.100154
2.099273
2.099273
2.098588
2.098588
2.098588
2.097902
2.097902

0
0
0.022
0.022
0.022
0.022
0.154
0.154
0.308
0.308
0.308
0.514
0.514

Lower bound (li)

Upper bound (ui)

1
51
76
81
101
141
151
201
226
251
281
301
351
376
401
451

50
75
80
100
140
150
200
225
250
280
300
350
375
400
450
500

775

R.M. Ebrahim et al. / Advances in Engineering Software 40 (2009) 766776

Table 6
Performance evaluation of the proposed SSA.
Total number of suppliers
(n)

n1, n2

Objective function value via

SSA

Objective function value via

LINGO

Relative
Error

SSA computational
time

n1 = 4, n2 = 0
n1 = 0, n2 = 4
n1 = 2, n2 = 2

1277.646
1055.241
1055.186

1277.646
1055.241
1055.186

0
0
0

0.4
0.7
0.4

1.3
3.7
1.0

n1 = 8, n2 = 0
n1 = 0, n2 = 8
n1 = 4, n2 = 4

692.30
345.728
345.728

692.30
345.728
345.728

0
0
0

0.6
0.7
0.5

1.2
3.1
0.9

12

n1 = 12,
n2 = 0
n1 = 0,
n2 = 12
n1 = 6, n2 = 6

1287.7

0.4

0.5

2.3

0.3

2.5

n1 = 15,
n2 = 0
n1 = 0,
n2 = 15
n1 = 8, n2 = 7

1467.0

1467.0

0.6

2.4

1196.87

1196.87

0.5

3.5

1195.24

1195.24

0.6

2.9

1440.0

1440.0

0.4

3.3

1266.93

1266.93

0.5

3.8

1266.93

1266.93

0.7

5.2

n1 = 25,
n2 = 0
n1 = 0,
n2 = 25
n1 = 15,
n2 = 10

1471.026

1471.026

0.5

4.1

1182.496

1182.496

1.2

3.7

1181.671

1181.671

0.6

5.4

n1 = 30,
n2 = 0
n1 = 0,
n2 = 30
n1 = 15,
n2 = 15

1512.68

1500.0

0.0084

0.8

6.2

1443.05

1485.93

0.0297

1.5

7.6

1443.05

1485.93

0.0297

1.8

7.1

n1 = 40,
n2 = 0
n1 = 0,
n2 = 40
n1 = 20,
n2 = 20

1893.7

1887.53

0.0033

1.3

9.3

1497.128

1497.128

0.0

1.5

12.2

1497.46

1524.052

0.0177

1.1

10.8

15

20

25

30

40

n1 = 20,
n2 = 0
n1 = 0,
n2 = 20
n1 = 10,
n2 = 10

974.694
974.15

1287.7
974.694
974.15

to show how fast this algorithm can nd a solution within a short

computational time. In the column entitled Relative Error, the relative errors of outputs of the SSA from exact optimal solutions are
presented. The relative error is measured using the following
relation:

Relative Error Objective function value via SSA

 Objective function value via LINGO=
Objective function value via LINGO:

14

The mean value of these relative errors is 0.0037 and the variance is 7.95809E05. In addition mean computational time of
the SSA is 0.754 and the corresponding variance is 0.1722 that
shows the SSA is adequately fast in nding solutions. Therefore,
we can claim that the SSA is largely reliable and can be regarded
as a suitable alternative for exact algorithms such as B&B in solving
the described supplier selection problem.
7. Summary and conclusions
In supplier selection decision making there are tangible and
intangible conicting factors that affect this decision making problem. Furthermore, in many real situations, suppliers may offer different discount schemes according to their selling policies. Hence,
buyers face a complex decision making problem. This paper intro-

LINGO computational
time

duced a linear integer programming model in which qualitative

and quantitative factors are considered. Furthermore, an extended
cost objective function is modeled as a part of the main decision
making model in which suppliers are allowed to offer any of the
three discount schemes which are all-unit discount, incremental
discount and total business volume discount. We discussed that
for a single-item supplier selection problem the total business volume discount and all-unit discount are the same. The weighted
sum of objective functions is used in development of the mathematical model to consider the importance of objectives and also
obtaining a unique optimal decision for the problem.
Since this problem is NP-hard, we proposed a scatter search
algorithm (SSA) by which this problem can be solved. The performance of this SSA is evaluated through comparing its outputs with
exact optimal solutions while solving 24 sample problems. For
computing exact optimal solutions LINGO package version 8 is
used. Results obtained from this evaluation showed that the proposed SSA can nd solutions with high quality but in a short computational time.
Acknowledgement
We would like to express our appreciation for the University of
Tehran (Grant number 8108023/1/07) for the nancial support of
this study.

776

R.M. Ebrahim et al. / Advances in Engineering Software 40 (2009) 766776

References
[1] Amid A, Ghodsypour SH, OBrien Ch. A weighted additive fuzzy multi-objective
model for the supplier selection problem under price breaks in a supply Chain.
Int J Prod Econom 2007. doi:10.1016/j.ijpe.2007.02.04.
[2] Benton WC. Quantity discount decisions under conditions of multiple items,
multiple suppliers and resource limitations. Int J Prod Res 1991;29(10):195361.
[3] Benton WC, Park SA. Classication of literature on determining the lot size
under quantity discount. Eur J Oper Res 1996;92(2):21938.
[4] Boer L, Labro E, Morlacchi P. A review of methods supporting supplier
selection. Eur J Purchasing Supply Manage 2001;7:7589.
[5] Burke GJ, Carrillo J, Vakharia AJ. Heuristics for sourcing from multiple suppliers
with alternative quantity discounts. Eur J Oper Res 2008;186(1):31729.
[6] Crama Y, Pascual R, Torres A. Optimal procurement decisions in the presence of
total quantity discounts and alternative product recipes. Eur J Oper Res
2004;159:36478.
[7] Dahel N-E. Vendor selection and order quantity allocation in volume discount
environments. Supply Chain Manage 2003;8(4):33542.
[8] Dickson GW. An analysis of vendor selection systems and decisions. J
Purchasing 1966;2(1):517.
[9] Dobler DW et al. Purchasing and materials management: text and cases. New
York: McGraw-Hill; 1990.
[10] Dolan RJ. Quantity discounts managerial issues and research opportunities.
Market Sci 1987;6(l):122.
[11] Ghodsypour SH, OBrien C. An integrated method using the analytical
hierarchy process with goal programming for multiple sourcing with
discounted prices. In: Proceedings of 14th international conference on
production research (ICPR), Osaka, Japan; 1997.
[12] Glover F. Heuristics for integer programming using surrogate constraints.
Decision Sci 1977;8:15666.
[13] Goossens DR, Maas AJ T, Spieksma FCR, Klundert JJ. Exact algorithms for
procurement problems under a total quantity discount structure. Eur J Oper
Res 2007;178:60326.

[14] Kothari A, Parkes D, Suri S. Approximately-strategy proof and tractable multiunit auctions. In: Proceedings 4th ACM conference on electronic commerce
(EC-2003), San Diego, June 2003. ACM, p. 16675.
[15] Mart R. Scatter searchwellsprings and challenges. Eur J Oper Res
2006;169:3518.
[16] Mart R, Languna M, Glover F. Principles of scatter search. Eur J Oper Res
2006;169:35972.
[17] Min H, Galle WP. International supplier selection. Int J Phys Distrib Log
Manage 1994;24(5):2433.
[18] Nahmias S. Production and operations analysis. 4th ed. New York: McGraw/
Irwin; 2001.
[19] Rahimi-Vahed AR, Rabbani M, Tavakkolli Moghaddam R, Torabi SA, Jolai F. A
multi-objective scatter search for a mixed-model assembly line sequencing
problem. Adv Eng Inform 2007;21:8599.
[20] Rosenthal EC, Zydiac JL, Chaudhry SS. Vendor selection with bundling. Decision
Sci 1995;26(1):3548.
[21] Rubin PA, Benton WC. Evaluating jointly constrained order quantity
complexities for incremental discounts. Eur J Oper Res 2003;149:
55770.
[22] Saaty TL. The analytic hierarchy process. NY: McGraw-Hill; 1980.
[23] Shih H-Sh, Shyur H-J, Lee ES. An extension of TOPSIS for group decision
making. Math Comput Model 2007;45:80113.
[24] Tracey M. Empirical analysis of supplier selection and involvement. Supply
Chain Manage: Int J 2001;6(4):17488.
[25] Wang G, Huang SH, Dismukes JP. Product-driven supply chain selection using
integrated multi-criteria decision making methodology. Int J Prod Econ
2004;91:115.
[26] Weber CA, Current JR, Benton WC. Vendor selection criteria and methods. Eur J
Oper Res 1991;50:218.
[27] Willis HT, Huston RC, Pohlkamp F. Evaluation of just in time supplier
performance. Prod Inventory Manage J 1993;34(2):15.
[28] Xia W, Wu Zh. Supplier selection with multiple criteria in volume discount
environments. Omega 2007;35:4945.