European Journal of Operational Research 177 (2007) 982–994

www.elsevier.com/locate/ejor

Production, Manufacturing and Logistics

An optimization approach for supply chain

management models with quantity discount policy
Jung-Fa Tsai *

Department of Business Management, National Taipei University of Technology, No. 1, Sec. 3, Chung Hsiao E. Road,
Taipei 10608, Taiwan

Received 25 April 2004; accepted 8 January 2006

Available online 17 April 2006

Abstract

Owing to the difficulty of treating nonlinear functions, many supply chain management (SCM) models assume that the
average prices of materials, production, transportation, and inventory are constant. This assumption, however, is not prac-
tical. Vendors usually offer quantity discounts to encourage the buyers to order more, and the producer intends to discount
the unit production cost if the amount of production is large. This study solves a nonlinear SCM model capable of treating
various quantity discount functions simultaneously, including linear, single breakpoint, step, and multiple breakpoint
functions. By utilizing the presented linearization techniques, such a nonlinear model is approximated to a linear mixed
0–1 program solvable to obtain a global optimum.
 2006 Elsevier B.V. All rights reserved.

Keywords: Logistics; Optimization; Supply chain management; Quantity discount model

1. Introduction

Supply chain management (SCM) is a control of material flow among vendors, facilities, warehouses and
customers such that the total cost in the supply chain can be minimized [27]. Considerable efforts have been
expended in developing SCM decision models in the last decade. Specifically, the topic on the determination of
optimal supplier-oriented quantity discount policies for a given type of quantity and with a constant demand
has attracted much attention. For instance, Monahan [19] analyzed an all-unit quantity discount policy that
maximized the vendor’s gain without adding cost to the buyer. Lal and Staelin [11] presented a fixed order
quantity decision model which assumed special forms of discount pricing structure with multiple buyers
and constant demands. Lee and Rosenblatt [12] generalized Monahan’s model to increase supplier’s profit
by incorporating constraints imposed on the discount rate and relaxing the assumption of a lot-for-lot supplier
policy. Weng and Wong [29] developed a general all-unit quantity discount model for a single buyer or multi-
ple buyers to determine the optimal pricing and replenishment policy. Weng [28] later presented models for

*
Tel.: +886 2 27712171x3420; fax: +886 2 27763964.
E-mail address: jftsai@ntut.edu.tw

0377-2217/$ - see front matter  2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.ejor.2006.01.034
 J.-F. Tsai / European Journal of Operational Research 177 (2007) 982–994 983

determining optimal all-unit and incremental quantity discount policies and investigated the effect of quantity
discounts on increasing demand and ensuring Pareto-efficient transactions under general price-sensitive
demand functions. Hoffmann [8] also analyzed the impact of all-unit quantity discount on channel coordina-
tion in a system consisting of one supplier and a group of heterogeneous buyers. Chang and Chang [3] pre-
sented a mixed integer optimization approach for the inventory problem with variable lead time, crashing
cost, and price-quantity discount. Yang [31] recently proposed an optimal pricing and ordering policy for a
deteriorating item with price sensitive demand. Related papers on this topic are far too many to be listed here.
However, some reviews are available in the literature such as Min and Zhou [18] and Narasimhan and Maha-
patra [21].
Currently, many optimization models of SCM [20,22,32] assume that the average prices of related expen-
ditures are constant, which implies that the average prices are fixed at a specific value independent of the scale
of quantity. This is far beyond the real situation. In fact, vendors usually offer quantity discounts to encourage
the buyers to order more, and the producer would like to discount the unit production cost if the amount of
production is large. For dealing with the difficulty, this study proposes an SCM model capable of treating var-
ious quantity discount functions. By utilizing the linearization techniques, the nonlinear SCM model can be
approximated to a linear mixed 0–1 program solvable to obtain a global optimum.
The rest of this paper is organized as follows: The next section formulates an SCM model in solving the
four types of quantity discount functions. Then the linear strategies for solving the SCM model to obtain a
global optimum are proposed. A numerical example is illustrated in Section 4 and concluding remarks are
made in the last section.

2. A supply chain management model

There are three traditional stages in the supply chain, including procurement, production and distribution.
Each of these stages may be composed of several facilities located in different countries around the world. The
general structure of a supply chain network has been outlined in Fig. 1. For simplifying presentation, only the
facilities covered in the dashed rectangle, including manufacturers, warehouses, and distribution centers, are
considered in this paper. Basically, there are four kinds of costs involved in this model:

(i) product procurement cost from manufacturers;

(ii) transportation cost from manufacturers to warehouses;
(iii) inventory cost in warehouses;
(iv) transportation cost from warehouses to distribution centers.

(Vendor) (Manufacturer) (Warehouse) (Distribution (Customer)

Center)

Vendor1 Customer1

Warehouse1
Manufacturer1 Distribution Center1
Vendor2 Customer2

Distribution Centern
Manufacturern
Vendorn Warehousen
Customern

Fig. 1. Schema of an SCM model.

984 J.-F. Tsai / European Journal of Operational Research 177 (2007) 982–994

The objective is to minimize the total cost associated with procurement, inventory, and distribution under a
variety of constraints at different timings, e.g., the capacity, lead time, and demands. A simplified SCM opti-
mization model can be presented as follows:
SCM model
(
X
T X X
min Q1tmg  Pcosttmg þ Q2tmwg  T 1costtmwg ð1Þ
t¼1 m;g m;w;g
)
X X
þ Q3twg  Icosttwg þ Q4twdg  T 2costtwdg
w;g w;d;g
X
s.t. Q2tmwg ¼ Q1tmg for all m; g; t; ð2Þ
w
X X
ðtLTmw Þ
Q3twg þ Q2mwg  Q4twdg ¼ Q3tþ1
wg for all w; g; t; ð3Þ
m d
X ðtLTwd Þ
Q4wdg ¼ Dtdg for all d; g; t; ð4Þ
w

Q1tmg 6 Q1 uptmg for all m; g; t; ð5Þ

Q3twg 6 Q3 uptwg for all w; g; t; ð6Þ

where

Decision variables
Q1tmg ordered quantity of the gth product from the mth manufacturer in the tth period;
Q2tmwg quantity of the gth product transported from the mth manufacturer to the wth warehouse in the tth
period;
Q3twg quantity of the gth product in the wth warehouse in the tth period;
Q4twdg quantity of the gth product transported from the wth warehouse to the dth distribution center in the
tth period.

Parameters
Pcosttmg unit procurement cost of the gth product from the mth manufacturer in the tth period;
T 1costtmwg unit transportation cost of the gth product from the mth manufacturer to the wth warehouse in the
tth period;
T 2costtwdg unit transportation cost of the gth product from the wth warehouse to the dth distribution center in
the tth period;
Icosttwg unit stock carrying cost of the gth product in the wth warehouse in the tth period;
Dtdg demands of the gth product from the dth distribution center in the tth period;
LTmw lead time from the mth manufacturer to the wth warehouse and LTmw 6 T ;
LTwd lead time from the wth warehouse to the dth distribution center and LTwd 6 T;
Q1 uptmg maximum supplied level of the gth product from the mth manufacturer in the tth period;
Q3 uptwg maximum inventory level of the gth product in the wth warehouse in the tth period.

The objective function (1) is to minimize the total cost composed of procurement cost from manufacturers,
transportation cost from manufacturers to warehouses, inventory cost in warehouses, and transportation cost
from warehouses to distribution centers. The total amount of the gth product transported from the mth man-
ufacturer to all warehouses must be equal to the total amount ordered from the mth manufacturer in the tth
period for each manufacturer as shown in (2). The left-hand side of (3) is the stock quantity of the gth product
at the beginning of the tth period, plus the total amount flowing in from all manufacturers, minus the total
amount flowing out to all distribution centers. This value should be equal to the stock quantity of the gth
product in the (t + 1)th period, as written on the right-hand side. The total amount of the gth product shipped
 J.-F. Tsai / European Journal of Operational Research 177 (2007) 982–994 985

Price
P1
r1
P2 r2

Pn-1 rn-1
Pn

Q1 Q2 Qn-1 Qn
Quantity

Fig. 2. Multiple breakpoint function.

from warehouses to distribution centers must be equal to the total final demands in the tth period as seen in
(4). The constraints of maximum supplied quantity of manufacturers and inventory level of warehouses are
given in (5) and (6).
Note that in this SCM model there are four price variables: Pcost, T1cost, Icost and T2cost. These four
variables can be expressed as a function of quantity Q. A general form usable in practice is a multiple break-
point function. A single breakpoint function is a special case of a multiple breakpoint function and will be
introduced in Appendix A as well as another two special cases, a linear and a step function. A multiple break-
point quantity discount function depicted in Fig. 2 can be usually represented as follows:
8
> P 1 þ r1 ðQ  Q1 Þ if Q1 6 Q < Q2 ;
>
>
>
< P 2 þ r2 ðQ  Q2 Þ if Q2 6 Q < Q3 ;
P ðQÞ ¼ ..
>
> .
>
>
:
P n1 þ rn1 ðQ  Qn1 Þ if Qn1 6 Q 6 Qn ;
where ri is the slope when the quantity ordered is between Qi and Qi+1, and n means there are ‘‘n  1’’ line
segments in P(Q).
Since the price and quantity of the quantity discount functions are variable, the total cost function of the
SCM model is then formulated as a nonlinear form which is difficult to be solved for finding a global solution.
In order to solve the model for obtaining a global solution, this study applies the linearization techniques pre-
sented in the next section to solve the SCM model.

3. Linearization of nonlinear cost functions

It is well known in optimization theory that for a minimization (maximization) problem with a convex fea-
sible region and a convex (concave) objective function, a local optimum is also a global optimum. To enhance
the computational efficiency, we have to recognize the property of a function for determining the necessity of
linearization. First consider the property of convexity of a single-variable function described below:
Remark 1. A function, f(x) = cxa, "x > 0, a P 1, is convex for c > 0, and f(x) is a concave function for c < 0.
Since this study discusses a cost minimization SCM model, we only have to estimate concave functions by
the piecewise linearization techniques and then convert the SCM model to a convex mixed 0–1 programming
problem for finding a global optimum. Consider the following propositions.
Proposition 1. A concave function f(x) shown in Fig. 3, where 0 < a1 6 x 6 am, can be piecewisely approximated
as [1–3,13,15,16]:

: X
m1
sj  sj1
f ðxÞ¼Lðf ðxÞÞ ¼ f ða1 Þ þ s1 ðx  a1 Þ þ ðjx  aj j þ x  aj Þ
j¼2
2
986 J.-F. Tsai / European Journal of Operational Research 177 (2007) 982–994

f (x)

s2

s1

a1 a2 a3 am x
x*

Fig. 3. A linearization function of f(x).

where L(f(x)) is a linearization function of f(x), and aj, j = 1, 2, . . . , m, are the break points of L(f(x)), aj < aj+1;
f ða Þf ðaj Þ
and sj are the slopes of line segments between aj and aj+1, sj ¼ ajþ1 jþ1 aj
and sj1 P sj for j ¼ 1; 2; . . . ; m  1.
The accuracy of the piecewise linear approximation heavily depends on the selection of proper break
points. The larger the number of break points, the higher precision of the solution obtained and the longer
computation time needed. Li and Yu [15] and Chang [2] have suggested how to select break points and pre-
sented the detailed process of linearizing a concave/convex function. Moreover, Li and Yu [15] and Chang [2]
proposed the way to evaluate the error between the optimal value of the initial function and the optimal value
of the approximated function.
Remark 2. If f(x) is a concave function, it is clear that the approximation bounds f(x) from below and
f(x) P L(f(x)).

Remark 3 (Lower bound). Consider the following two programs:

P1: {Minimize f(x) j x, y 2 F}.

P2: {Minimize L(f(x)) j x, y 2 F}.

Program P2 provides a lower bound on Program P1 due to Remark 2.

Now we discuss the way to linearize L(f(x)). To linearize L(f(x)), we only need to deal with the absolute
term with negative coefficient (i.e., sj1 P sj).
Proposition 2. A program {min z = (jx  ajj + x  aj), s.t. x, aj P 0} is equivalent to the following program:

min z0 ¼ 2ðwj  uj aj  x þ aj Þ
s.t. wj P x þ Mðuj  1Þ;

where x P 0, wj P 0, uj 2 {0, 1}, and M is a large constant.

Proof

(i) If x  aj P 0 then the optimal values are wj = 0, uj = 0, and z 0 = 2x + 2aj = z.

(ii) If x  aj < 0 then the optimal values are wj = x, uj = 1, and z 0 = 0 = z. h

According to the above discussion of how to linearize a concave term in the objective function, the follow-
ing illustrates the linear approximation process of an SCM model with a multiple breakpoint quantity dis-
count model.
 J.-F. Tsai / European Journal of Operational Research 177 (2007) 982–994 987

Proposition 3. For a piecewise linear function P(Q) depicted in Fig. 2, P(Q) can be expressed as
X
n1
P ðQÞ ¼ ui ½pi þ ri ðQ  Qi Þ; ð7Þ
i¼1

where
ðP iþ1  P i Þ
ri ¼ for i ¼ 1; 2; . . . ; n  1; ð8Þ
ðQiþ1  Qi Þ
X
n1 X
n1
ui Qi 6 Q 6 ui Qiþ1 ; ð9Þ
i¼1 i¼1
X
n1
ui ¼ 1; ui 2 f0; 1g. ð10Þ
i¼1

Proof. Expressions (9) Pand (10) imply that if and only if ui = 1, then Qi 6 Q 6 Qi+1 for i = 1, 2, . . . , n  1. So,
n1
it is clear that P ðQÞ ¼ i¼1 ui ½pi þ ri ðQ  Qi Þ. h

Remark 4. A product term z = uf(x) is equivalent to the following linear inequalities:

(i) M(u  1) + f(x) 6 z 6 M(1  u) + f(x).

(ii) Mu 6 z 6 Mu.

u 2 {0, 1}, z is an unrestricted in sign variable, and M is a large constant.

An SCM cost minimization model with multiple breakpoint quantity discount function then can be
expressed as
Model 1:
min Q  P ðQÞ
s.t. (7)–(10),
other constraints.
Model 1 is a nonlinear mixed 0–1 program which is difficult to be solved for obtaining a globally optimal
solution. According to the linearization techniques mentioned above, this program can be approximated to a
linear mixed 0–1 program as follows:
X
n1
min ðP i zi  ri Qi zi þ si Þ
i¼1

s.t. Mðui  1Þ þ Q 6 zi 6 Mð1  ui Þ þ Q for i ¼ 1; 2; . . . ; n  1;

 Mui 6 zi 6 Mui for i ¼ 1; 2; . . . ; n  1;
Mðui  1Þ þ Lðri Q2 Þ 6 si 6 Mð1  ui Þ þ Lðri Q2 Þ for i ¼ 1; 2; . . . ; n  1;
 Mui 6 si 6 Mui for i ¼ 1; 2; . . . ; n  1;
(8)–(10),
other constraints,
where zi, si, for i = 1, 2, . . . , n  1, are unrestricted in sign variables, zi = uiQ, si = uiL(riQ2), M is a large con-
stant, and L(riQ2) is a piecewise linear expression of ri Q2 by Proposition 1.
In fact, the SCM model with a multiple breakpoint quantity discount function is the most general model
which can be reduced to the other models with a single breakpoint function, a linear function or a step func-
tion. For instance, if n = 2, then the SCM model with a multiple breakpoint quantity discount function (i.e.,
Model 1) is equivalent to another model with a single breakpoint function; if n = 3 and r2 = 0, then Model 1 is
988 J.-F. Tsai / European Journal of Operational Research 177 (2007) 982–994

equivalent to the model with a linear quantity discount function; if ri = 0 for i = 1, 2, . . . , n  1, then Model 1 is
reduced to the model with a step function (see Appendix A).

4. Numerical example

Consider the SCM model in (1)–(6). Suppose there is a supply chain system composed of two manufactur-
ers A and B (i.e., m = 2), two warehouses C and D (i.e., w = 2), and two distribution centers E and F (i.e.,
d = 2). The time period is 3 (i.e., t = 3). Only one product is produced (i.e., g = 1). Let the lead-time from each
manufacturer to each warehouse be 1, and from each warehouse to each distribution center be 0. The whole
example network is depicted in Fig. 4 and the associated objective function can be represented as
X
3
min ðQ1tA  PcosttA þ Q1tB  PcosttB þ Q2tAC  T 1costtAC þ Q2tAD  T 1costtAD þ Q2tBC  T 1costtBC
t¼1

þ Q2tBD  T 1costtBD þ Q3tC  IcosttC þ Q3tD  IcosttD þ Q4tCE  T 2costtCE þ Q4tCF  T 2costtCF
þ Q4tDE  T 2costtDE þ Q4tDF  T 2costtDF Þ.
To illustrate the applicability of the quantity discount functions mentioned before to this model, we may arbi-
trarily assign Pcost, T1cost, Icost, and T2cost as various quantity discount functions. Here, we assign PcostA,
T1costAD, IcostD, and T2costDE, respectively, as the functions of step, linear, multiple breakpoint, and single
break (Fig. 4 with the bold lines). All the other price variables, PcostB, T1costAC, T1costBC, T1costBD, IcostC,
T2costCE, T2costCF, and T2costDF, are kept as fixed values listed in Tables 1–5.

Manufacturer Warehouse Distribution Center

A C E

B D F

Fig. 4. The example network. (The quantity discount functions are depicted with bold lines.)

Table 1
Product cost (Pcost) and supplied limitation (Q1_up)
Manufacturer A Manufacturer B
Product cost per unit Step function as shown in Fig. 8, where P_p1 = 50, PcostB = 48
P_p2 = 45, P_p3 = 42, P_q1 = 0,
P_q2 = 1000, P_q3 = 1500, P_q4 = 2000
Maximum supplied quantity Q1A_up = 2000 Q1B_up = 1000

Table 2
Transportation cost from manufacturer to warehouse (T1cost)
A!C A!D B!C B!D
Transportation T1costAC = 2 Linear function as shown in Fig. 6, T1costBC = 2 T1costBD = 2
cost per unit where T1_p1 = 3, T1_r = 0.001

Table 3
Inventory cost (Icost) and limitation (Q3_up)
Warehouse C Warehouse D
Inventory cost per unit IcostC = 2 Multiple breakpoint function as shown in Fig. 2,
where I_p1 = 4, I_p2 = 2.5, I_p3 = 1.5, I_r1 =  0.0015,
I_r2 =  0.001, I_q1 = 0, I_q2 = 1000, I_q3 = 1500
Maximum stock level Q3C_up = 1600 Q3D_up = 1300
 J.-F. Tsai / European Journal of Operational Research 177 (2007) 982–994 989

Table 4
Transportation cost from warehouse to distribution center (T2cost)
C!E C!F D!E D!F
Transportation T2costCE = 2 T2costCF = 2 Single breakpoint function as shown in Fig. 7, T2costDF = 2
cost per unit where T2_p1 = 3, T2_p2 = 1.5, T2_q1 = 0,
T2_q2 = 1000, T2_r =  0.0015

Table 5
Distribution center demand (D)
Distribution center E Distribution center F
t=1 t=2 t=3 t=1 t=2 t=3
Demands 0 1400 1000 0 1000 800

Following the linearization techniques of Section 3, a full SCM model is formulated as the following
program:
Example SCM model
X
3
min ððP p1  zt11 þ P p2  zt12 þ P p3  zt13 Þ þ Q2tAD  T 1 p1
t¼1
X
2
þ LðT 1 r  ðQ2tAD Þ2 Þ þ ðI pi  zt2i  I ri  I qi  zt2i þ I sti Þ
i¼1
þ T 2 p1  zt31  T 2 r  T 2 q1  zt31 þ T 2 p2  zt32 þ T 2 st
þ Q1tB  PcosttB þ Q2tAC  T 1costtAC þ Q2tBC  T 1costtBC þ Q2tBD  T 1costtBD
þ Q3tC  IcosttC þ Q4tCE  T 2costtCE þ Q4tCF  T 2costtCF þ Q4tDF  T 2costtDF Þ
s.t.
X
3 X
3
ut1i  P qi 6 Q1tA 6 ut1i  P qiþ1 for t ¼ 1; 2; 3;
i¼1 i¼1
X
3
ut1i ¼ 1; ut1i 2 f0; 1g for t ¼ 1; 2; 3;
i¼1
Q1tA þ Mðut1i  1Þ 6 zt1i 6 Q1tA þ Mð1  ut1i Þ for i; t ¼ 1; 2; 3;
 Mut1i 6 zt1i 6 Mut1i for i; t ¼ 1; 2; 3;
Mðut2i  1Þ þ Q3tD 6 zt2i 6 Mð1  ut2i Þ þ Q3tD for i ¼ 1; 2; t ¼ 1; 2; 3;
 Mut2i 6 zt2i 6 Mut2i for i ¼ 1; 2; t ¼ 1; 2; 3;
Mðut2i  1Þ þ LðI ri  ðQ3tD Þ2 Þ 6 I sti 6 Mð1  ut2i Þ þ LðI ri  ðQ3tD Þ2 Þ
for i ¼ 1; 2; t ¼ 1; 2; 3;
 Mut2i 6 I sti 6 Mut2i for i ¼ 1; 2; t ¼ 1; 2; 3;
I ri ¼ ðI piþ1  I pi Þ=ðI qiþ1  I qi Þ for i ¼ 1; 2;
X 2 X2
ðut2i  I qi Þ 6 Q3tD 6 ðut2i  I qiþ1 Þ for t ¼ 1; 2; 3;
i¼1 i¼1
X
2
ut2i ¼ 1; ut2i 2 f0; 1g for t ¼ 1; 2; 3;
i¼1
X
2 X
2
ut3i  T 2 qi 6 Q4tDE 6 ut3i  T 2 qiþ1 for t ¼ 1; 2; 3;
i¼1 i¼1
990 J.-F. Tsai / European Journal of Operational Research 177 (2007) 982–994

X
2
ut3i ¼ 1; ut3i 2 f0; 1g for t ¼ 1; 2; 3;
i¼1
Mðut3i  1Þ þ Q4tDE 6 zt3i 6 Mð1  ut3i Þ þ Q4tDE for i ¼ 1; 2; t ¼ 1; 2; 3;
 Mut3i 6 zt3i 6 Mut3i for i ¼ 1; 2; t ¼ 1; 2; 3;
Mðut33  1Þ þ LðT 2 r  ðQ4tDE Þ2 Þ 6 T 2 st 6 Mð1  ut33 Þ þ LðT 2 r  ðQ4tDE Þ2 Þ; for t ¼ 1; 2; 3;
 Mut33 6 T 2 st 6 Mut33 for t ¼ 1; 2; 3;
T 2 r¼ ðT 2 p2  T 2 p1 Þ=ðT 2 q2  T 2 q1 Þ;
(2)–(6).
The above expressions of the constraint set present step, liner, multiple breakpoint, and single breakpoint
quantity discount functions. The objective is to minimize the total cost and to obtain the optimal product
flows in each time period. LðI ri  ðQ3tD Þ2 Þ and LðT 2 r  ðQ4tDE Þ2 Þ are piecewise linear expressions of the con-
2 2
cave functions I ri  ðQ3tD Þ and T 2 r  ðQ4tDE Þ , respectively. By referring to Chang [2], this study selects one
break point for linearizing a concave function at each iteration and sets the tolerable error e = 0.05. If the
obtained optimal solution (i.e., x* in Fig. 3) satisfies the condition f(x)  L(f(x)) 6 e, then stop the solution
process.
Solving this model by LINGO [25], the global optimum is obtained. The total cost obtained is $193,670,
and the optimal SCM flows are depicted in Fig. 5.

(i) For t = 1
• Buying 2000 units from manufacturer A with a price of $42 per unit. Then manufacturer A ships 700
and 1300 units to warehouse C and D with the transportation cost $2 and $1.7 per unit, respectively.
• Purchasing 400 units from manufacturer B with a price of $48 per unit. Manufacturer B ships 400
units to warehouse C with the transportation cost $2 per unit.
(ii) For t = 2
• Warehouse C transports 100 and 1000 units to the distribution centers E and F, respectively, with the
transportation cost of $2 per unit.
• Warehouse D delivers 1300 units to the distribution center E with the transportation cost of $1.5 per unit.

Manufacturer Warehouse Distribution Center

2000/42 700/2 100/2

1400
A C E

400/48 400/2 1300/1.7 1300/1.5 1000/2

1000
B D F

t=1 t=2 t=2

500/2
1800/42 500/2 1000
A C E

1300/1.7 1000/1.5

300/2 800
B D F

t=2 t=3 t=3

Fig. 5. Optimal results of the example, where ‘‘/’’ represents ‘‘quantity/unit cost’’.
 J.-F. Tsai / European Journal of Operational Research 177 (2007) 982–994 991

• Buying 1800 units from manufacturer A for a price of $42 per unit. Then manufacturer A ships 500
and 1300 units to warehouse C and D with the transportation cost of $2 and $1.7 per unit, respectively.

(iii) For t = 3
• Warehouse C transports 500 units to the distribution center F with the transportation cost of $2 per unit.
• Warehouse D delivers 1000 and 300 units to the distribution centers E and F with the transportation
cost of $1.5 and $2 per unit, respectively.

According to the computational results of this example, we observe that the quantity discount policies affect
the activities of the SCM model certainly. In product procurement, since manufacturer A provides the step
quantity discount policy and the lowest price (i.e., $42) is lower than the fixed price of manufacturer B
(i.e., $48), the ordered quantity of manufacturer A therefore reaches the maximum supplied level (i.e.,
2000). Moreover, the transportation cost from manufacturer A to warehouse D has linear quantity discount
and warehouse D gives multiple breakpoint quantity discount for the inventory cost. Consequently, the quan-
tity of the product transported from manufacturer A to warehouse D meets the maximum stock level of ware-
house D (i.e., 1300). The transportation cost from warehouse D to distribution center E also provides single
breakpoint quantity discount policy, and the quantity of the product transported from warehouse D to dis-
tribution center E meets the maximum requirement.

5. Conclusions

This study proposes an SCM model capable of treating various quantity discount functions such as linear,
single breakpoint, step, and multiple breakpoint functions. All these quantity discount functions can be merged
into an SCM model flexibly. An SCM model with various quantity discount functions is formulated as a
nonlinear programming problem which is difficult to be solved to obtain a global optimum. By the proposed
linearization techniques, the SCM model can then be approximated to a linear mixed 0–1 programming problem
solvable to obtain a globally optimal solution effectively. Two possible directions for further research are,
firstly, to utilize the proposed method to solve large scale SCM problems with the distributed computation algo-
rithms [14] and, secondly, to compare the solution quality of the proposed method with that of heuristic
algorithms.

Acknowledgements

The author would like to thank the anonymous referees for contributing their valuable comments regarding
this paper, and thus significantly improving the quality of this paper. This research was supported by the
National Science Council, Taiwan R.O.C. (Grant no. NSC 94-2416-H-027-002).

Appendix A

The other three special cases of quantity discount functions derived from the multiple breakpoint quantity
discount function are described in the following:
(i) Linear function: A linear quantity discount function is the simplest and the most commonly assumed
pricing structure [4,23,26] (Fig. 6), which can be formulated as P(Q) = P1 + r · Q where r is a negative slope
indicating that the price of the item decreases upon increasing order quantities. Both P1 and r are constant,
and Q is a variable that represents the quantity ordered.
Let n = 2 in Model 1, a cost minimization model with a linear quantity discount function can be formulated
as follows:
min Q  P ðQÞ
s.t. P ðQÞ ¼ P 1 þ rQ;
other constraints.
992 J.-F. Tsai / European Journal of Operational Research 177 (2007) 982–994

Price

P1
r

Q Quantity

Fig. 6. Linear function.

The above program is a quadratic programming problem solvable by existing methods such as the Wolfe’s
method [30] and the Kuhn–Tucker theorem [7].
(ii) Single breakpoint function: A single breakpoint quantity discount function is similar to the one with
multiple breakpoints except that there is only one breakpoint. A quantity discount schedule with single price
breakpoint is applied widely to calculate the cost of procurement, production, and transportation [5,9,24,28].
As represented in Fig. 7, the basic price is P1, and the price goes down when the quantity variable Q increases.
Once the amount reaches Q2, the price will reach the minimum incentive price, P2, and will not change any-
more. The discount function with single breakpoint can be usually represented as

P 1 þ rðQ  Q1 Þ if Q1 6 Q < Q2 ;
P ðQÞ ¼
P2 if Q P Q2 .
Fazel et al. [5] and Schniederjans and Cao [24] calculate the optimal quantities in both cases, Q 6 Q2 and
Q > Q2, by differentiating the total cost functions, respectively, and select the quantity that yields the lower
total cost as the optimum quantity. A quantity discount that has a single price breakpoint provides a tactical
model for obtaining managerial insights [10].
Let n = 3 and r2 = 0 in Model 1, a cost minimization model with a single breakpoint quantity discount
function can be written as
min Q  P ðQÞ
s.t. P ðQÞ ¼ u1 ½P 1 þ r1 ðQ  Q1 Þ þ u2 P 2 ;
r1 ¼ ðP 2  P 1 Þ=ðQ2  Q1 Þ;
u1 Q1 þ u2 Q2 6 Q 6 u1 Q2 þ u2 Q3 ;
u1 þ u2 ¼ 1 for u1 ; u2 2 f0; 1g;
other constraints.

Price

P1

P2

Q1 Q2
Quantity

Fig. 7. Single breakpoint function.

 J.-F. Tsai / European Journal of Operational Research 177 (2007) 982–994 993

Price
P1

P2
P3

Q1 Q2 Q3 Q4
Quantity

Fig. 8. Step function.

This model is a nonlinear minimization problem solvable by using the proposed method in Section 3.
(iii) Step function: A step quantity discount function is a complicated quantity discount schedule. The step
function is shown in Fig. 8, where Q1, Q2, and Q3 represent the incentive quantity levels set by manufacturers,
and P1, P2, and P3 stand for the basic and discount prices offered, respectively. The step function defined by
Love [17] is expressed as
8
>
> P 1 for Q1 < Q 6 Q2 ;
>
>
< P 2 for Q2 < Q 6 Q3 ;
P ðQÞ ¼ ..
>
> .
>
>
:
P n for Qn < Q;
where P1, P2, . . . , Pn are discount prices which obey the relationships P1 > P2 >    > Pn and Q1, Q2, . . . , Qn
which stand for boundaries of the incremental quantities at state 1 to n. Guder and Zydiak [6] have demon-
strated a fixed cycle approach, which can also be used to heuristically solve the quantity discount case with the
same cost function. Hoffmann [8] has analyzed the impact of all-units quantity discounts on channel coordi-
nation in a system consisting of one supplier and a group of heterogeneous buyers. In his study, an all-units
quantity discount schedule with a step function is considered, but it is not obvious how to use a price scheme
with price breaks.
Let ri = 0 for i = 1, 2, . . . , n  1, in Model 1, a cost minimization model with a step quantity discount func-
tion can be written as
min Q  P ðQÞ
X
n1
s.t. P ðQÞ ¼ P i ui ;
i¼1
X
n1 X
n1
ui Q i 6 Q 6 ui Qiþ1 ;
i¼1 i¼1
X
n1
ui ¼ 1; ui 2 f0; 1g;
i¼1
other constraints.
ui is a 0–1 variable used to indicate which price level should be adopted based on the quantity variable. ui = 1
means
Pn Pi should be selected. Besides, in order to make sure that only one price level is chosen, the constraint
i¼1 ui ¼ 1 is included. The formulated model is a quadratic programming problem which can be efficiently
solved by the existing methods [7,30].

References

[1] C.T. Chang, A modified goal programming model for piecewise linear functions, European Journal of Operational Research 139
(2002) 62–67.
994 J.-F. Tsai / European Journal of Operational Research 177 (2007) 982–994

[2] C.T. Chang, On the posynomial fractional programming problems, European Journal of Operational Research 143 (2002) 42–52.
[3] C.T. Chang, S.C. Chang, On the inventory model with variable lead time and price-quantity discount, Journal of the Operational
Research Society 52 (2001) 1151–1158.
[4] M.S. Chen, C.C. Chuang, An extended newsboy problem with shortage-level constraints, International Journal of Production
Economics 67 (2000) 269–277.
[5] F. Fazel, K.P. Fischer, E.W. Gilbert, JIT purchasing vs. EOQ with a price discount: An analytical comparison of inventory costs,
International Journal of Production Economics 54 (1998) 101–109.
[6] F. Guder, J.L. Zydiak, Fixed cycle ordering policies for capacitated multiple item inventory systems with quantity discounts,
Computers and Industrial Engineering 38 (2000) 67–77.
[7] F.S. Hillier, G.J. Lieberman, Introduction to Operations Research, McGraw-Hill, New York, 1995.
[8] C. Hoffmann, Supplier’s pricing policy in a Just-in-Time environment, Computers and Operations Research 27 (2000) 1357–1373.
[9] K.H. Kim, H. Hwang, An incremental discount pricing schedule with multiple customers and single price break, European Journal of
Operational Research 35 (1988) 72–79.
[10] R. Kohli, H. Park, A cooperative game theory model of quantity discounts, Management Science 35 (1989) 693–707.
[11] R. Lal, R. Staelin, An approach for developing an optimal discount pricing policy, Management Science 30 (12) (1984) 1524–1539.
[12] H.L. Lee, J. Rosenblatt, A generalized quantity discount pricing model to increase supplier’s profits, Management Science 33 (9)
(1986) 1167–1185.
[13] H.L. Li, C.T. Chang, J.F. Tsai, Approximately global optimization for assortment problems using piecewise linearization techniques,
European Journal of Operational Research 140 (2002) 584–589.
[14] H.L. Li, J.F. Tsai, N.Z. Hu, A distributed global optimization method for packing problems, Journal of the Operational Research
Society 54 (2003) 419–425.
[15] H.L. Li, C.S. Yu, A global optimization method for nonconvex separable programming problems, European Journal of Operational
Research 117 (1999) 275–292.
[16] H.L. Li, C.S. Yu, Solving multiple objective quasi-convex goal programming problems by linear programming, International
Transactions in Operational Research 7 (2000) 265–284.
[17] S. Love, Inventory Control, McGraw-Hill, New York, 1979.
[18] H. Min, G. Zhou, Supply chain modeling: Past, present and future, Computers and Industrial Engineering 43 (2002) 231–249.
[19] J.P. Monahan, A quantity pricing model to increase vendor profits, Management Science 30 (6) (1984) 720–726.
[20] J.M. Mulvey, R.J. Vanderbei, S.A. Zenios, Robust optimization of large-scale systems, Operations Research 32 (1995) 264–281.
[21] R. Narasimhan, S. Mahapatra, Decision models in global supply chain management, Industrial Marketing Management 33 (2004)
21–27.
[22] M.J. Rosenblatt, Y.T. Herer, I. Hefter, An acquisition policy for a single item multi-supplier system, Management Science 44 (1998)
96–100.
[23] M.J. Rosenblatt, H.L. Lee, Improving profitability with quantity discounts under fixed demand, IIE Transactions 17 (1985) 388–395.
[24] M.J. Schniederjans, Q. Cao, A note on JIT purchasing vs. EOQ with a price discount: An expansion of inventory costs, International
Journal of Production Economics 65 (2000) 289–294.
[25] L. Schrage, Optimization Modeling with LINGO, Duxbury Press, Chicago, 1998.
[26] S. Tayur, R. Ganeshan, M. Magazine, Quantitative Models for Supply Chain Management, Kluwer Academic Publishers, Dordrecht,
1999.
[27] D.J. Thomas, P.M. Griffin, Coordinated supply chain management, European Journal of Operational Research 94 (1996) 1–15.
[28] Z.K. Weng, Modeling quantity discounts under general price-sensitive demand functions: Optimal policies and relationships,
European Journal of Operational Research 86 (1995) 300–314.
[29] Z.K. Weng, R.T. Wong, General models for the supplier’s all-unit quantity discount policy, Naval Research Logistics 40 (6) (1993)
971–991.
[30] W.L. Winston, Operations Research—Applications and Algorithms, Duxbury Press, Chicago, 1994.
[31] P.C. Yang, Pricing strategy for deteriorating items using quantity discount when demand is price sensitive, European Journal of
Operational Research 157 (2004) 389–397.
[32] C.S. Yu, H.L. Li, A robust optimization model for stochastic logistic problems, International Journal of Production Economics 64
(2000) 385–397.