Documente Academic
Documente Profesional
Documente Cultură
The integrality of the lot size in the basic EOQ and EPQ models:
Applications to other production-inventory models
Juan Garca-Laguna a, Luis A. San-Jos b,*, Leopoldo E. Crdenas-Barrn c, Joaqun Sicilia d
a
a r t i c l e
i n f o
Keywords:
EOQ/EPQ inventory models
Integer lot size
Backordering
a b s t r a c t
In this paper we present a method to obtain the solution of the classic economic order
quantity (EOQ) and economic production quantity (EPQ) models when the lot size must
be an integer quantity. This approach is operatively very simple and allows obtaining a rule
to discriminate between the situation in which the optimal solution is unique and when
there are two optimal solutions. Also, this method is applicable to the resolution of other
production-inventory models. We expose some of them and illustrate the use of the
method with numerical examples.
2010 Elsevier Inc. All rights reserved.
1. Introduction
Economic lot size models have been studied extensively since Harris [1] presented the famous EOQ formula in 1913. Five
years later, the economic production quantity (EPQ) inventory model was proposed by Taft [2]. Both models have been
extensively used by industries and enterprises in the management of their inventories. Roach [3] presents an original and
interesting historical discussion on the likely derivation of Harriss EOQ model. He thinks that possibly/probably Harris
[1] was inuenced by the Kelvins Law. Maybe Harris used Kelvins Law in developing his famous EOQ model.
However, in recent years, both academicians and researchers have shown an increasing level of interest in nding alternatives ways to solve inventory models. For example, some scholars have attempted to obtain optimal solutions of the inventory models without using the classical optimization technique: differential calculus. Instead, they have obtained formulae
for inventory models using either a tabular, graphical or an algebraic approach (see, for instance, Teng and Wee [4] and Teng
and Crdenas-Barrn [5]).
In all previous approaches the decision variable is considered to be a continuous variable. But, in many real life situations
the decision variables must be an integer number. Therefore, in this paper we present a novel approach to determine the lot
size and the number of orders or production runs when these must be integer values. It is important to point out that this
method is very simple and fast.
Let us present a brief summary of the basic EOQ and EPQ inventory models. In the case of the basic EOQ model of
Harris [1], it is very well-known that the assumptions of this model are: (i) the inventory system is based in a single item
which operates over an innite planning horizon; (ii) the rate of demand is a known constant D > 0; (iii) the inventory
* Corresponding author.
E-mail addresses: laguna@eio.uva.es (J. Garca-Laguna), augusto@mat.uva.es (L.A. San-Jos), lecarden@itesm.mx (L.E. Crdenas-Barrn), jsicilia@ull.es
(J. Sicilia).
0096-3003/$ - see front matter 2010 Elsevier Inc. All rights reserved.
doi:10.1016/j.amc.2010.02.042
1661
is continuously revised; (iv) the ordering cost is xed regardless of the lot size K > 0; (v) the holding cost is a linear function of the average inventory h > 0; (vi) the lot size per cycle is an unknown constant and it is the decision variable q > 0;
(vii) the shortages are not allowed. Taking into account these assumptions, the objective function of the EOQ inventory model and its constrain are:
Minimize
Cq K Dq h 2q
subject to
q>0
Minimize
Cq K Dq ha 2q
subject to
q>0
1:1
p
The optimal solution can be obtained
easily by using differential calculus and is given by q 2KD=h. The optimal value
p
for the total inventory cost is Cq 2KDh. The aforesaid mathematical expression for q is the well-known Wilson [6]s
formula.
In the case of the basic EPQ model of Taft [2], it is very well-known that the assumptions of the EOQ model are hold. Moreover, it is necessary to add the following assumption: the production or replenishment rate P > 0 is constant and greater
than demand rate P > D. Taking into account these assumptions, the objective function for the EPQ inventory model and its
constrain are:
1:2
where a 1 D=P represents the fraction of time the production process spends actually p
idling
[7, p. 50].
2KD=h
a, and the optimal value
Also, the optimal solution can be obtained
with
differential
calculus
and
is
given
by
q
p
for the total inventory cost is Cq 2KDha.
It is important to point out that optimization by differentiation is performed under the assumption that the variables are
continuous. But, when the lot size should be an integer value and if one does a simple rounding, therefore may lead to a suboptimal inventory policy. Also, we cannot apply the differential calculus to nd an optimal value for one variable when it is
restricted to an integer value.
The remainder of the paper is organized as follows: Section 2 presents the basic EOQ and EPQ models with integer lot size.
Section 3 exposes the EOQ and EPQ models with shortages and integer lot size. Section 4 contains the basic EOQ model with
nite planning horizon. Section 5 contains the application of our method to the basic EOQ model with two warehouses. Finally, Section 6 provides some conclusions and future research directions.
2. The basic EOQ/EPQ models with integer lot size
When the lot size must be an integer quantity, then it is obvious that the mathematical formulation of the EOQ model is:
Minimize
Cq K Dq h 2q
subject to
2:1
In this case, as the lot size q is restricted to be an integer number, we cannot apply the differential calculus to nd the
optimal lot size q. Since C(q) is a convex function over the continuous interval 0; 1, the intuitive and classical method
to solve this model consists of comparing the values Cbq c and Cdq e, where bxc maxfy integer : y 6 xg and
dxe minfy integer : y P xg. Obviously, if Cbq c Cdq e we have two optimal solutions; otherwise, there is a unique
solution.
In the environment of Econometrics and Engineering, a usual method consists of using the marginal analysis. Taking into
account that C(q) is a convex function over the set of the positive integer numbers N f1; 2; 3; . . .g, the expression
ql minfq 2 N : Cq 1 P Cqg give us the optimal solution if it is unique, and the lower of them, when there are two
optimal solutions. Similarly, the expression qu maxfq 2 N : Cq 1 P Cqg, provide the optimal solution if it is unique,
and the upper when there are two optimal solutions. The two previous expressions are equivalents to ql
minfq 2 N : qq 1 P 2KD=hg and qu maxfq 2 N : qq 1 P 2KD=hg, respectively. In each case, the expressions before
mentioned are necessary and sufcient conditions for the optimal solution of the lot size when this is restricted to be an
integer value.
and
taking into account that the unique positive solution of the quadratic equaNow, if we consider the expression for ql p
tion q2 q 2KD=h 0 is the value 0:5 0:25 2KD=h, it follows that the optimal solution, or the smallest of them, can
be expressed as:
q
ql 0:5 0:25 2KD=h
2:2
q
qu 0:5 0:25 2KD=h
2:3
1662
p
solution qu ql if and only if x 0:5 0:25 2KD=h (or x 1 0:5 0:25 2KD=h) is not an integer number.
Otherwise, there are two optimal solutions, ql and qu ql 1.
Finally, note that to obtain the integer lot size we only need to apply one of the two previous expressions, that is, ql or qu .
To illustrate the previous results the following two examples are introduced.
Example 1. We consider an inventory system with the following parameter values: K = 6500 $ per lot, D = 89,000 units per
year and h = 13 $ per unit per year. In this case, we have
l
pm
ql 0:5 0:25 2650089; 000=13 d9433:4811e 9434 units
89; 000
9434
13
122641:7547 $ per year
Cql 6500
9434
2
The reader can note that there is a unique optimal solution for this example, because x = 9433.4811 is not integer.
Example 2. Now, we consider an inventory system with the following parameter values: K = 10.1 $ per lot, D = 1000 units
per year and h = 2 $ per unit per year. We have
l
pm
ql 0:5 0:25 210:11000=2 d100e 100 units
Then, there are two optimal solutions, which are ql 100 units and qu b101c ql 1 101 units. Obviously,
1000
100
Cql 10:1
2
201 $ per year
100
2
1000
101
Cqu 10:1
2
201 $ per year
101
2
Remark. It is important to point out that the basic EOQ model with integer lot size previously analyzed does not coincide
exactly with the EOQ model where demand must be an integral number of units (see, for instance, [8, p. 41]). In this case, the
usual formulation for the model considers the additional assumption that a customer arrives to the system exactly at the
same moment that arrives the lot size. Thus, we have a small saving in the holding cost that is associated to the
expression h=2. Obviously, if we apply the previous procedure to this new model, one obtains the same mathematical
expression for the optimal lot size that one was obtained in the EOQ model with integer lot size, although the optimal costs
differ in the quantity h=2.
We can also study the EPQ model of Taft when the order quantities are restricted to be positive integers. In this case, we
may apply the same method as in the EOQ model with integer lot size to obtain the optimal production-inventory policy.
Now, we would have to consider the values ql minfq 2 N : qq 1 P 2KD=hag and qu maxfq 2 N : qq 1 P
l
pm
p
2KD=hag. Thus, it follows that the unique optimal solution is ql 0:5 0:25 2KD=ha if 0:5 0:25 2KD=ha
is not an integer number. Otherwise, we have two optimal solutions, ql and qu ql 1.
On the other hand, in some instances, the buyer has to purchase a lot that must be a multiple of some integer value due to
restrictions imposed by the supplier. For example, the products can be sold in a multiple of 100 because the container sizes
or boxes where the products are packaged contain exactly 100 units. In this case, our previous results can be easily extended.
Suppose that the container size is for a units, where a is an integer value. Considering the EOQ inventory model, then the lot
size should be calculated as:
q
q
2
2
ql a 0:50 0:25 2KD=ha or qu a 0:50 0:25 2KD=ha
Moreover,
analysis similar
to theqprevious
case shows
that there is a unique optimal solution if and only if
q
2
2
0:50 0:25 2KD=ha (or 0:50 0:25 2KD=ha is not an integer number. Otherwise, we have two optimal solutions, ql and qu ql a. In Table 1, the mathematical expressions for the integer lot size of the EOQ/EPQ inventory models
when the optimal lot size is restricted to be a integer value and when this is a multiple of an integer a are given.
1663
Table 1
The integer lot size for the EOQ/EPQ inventory models.
Inventory model
l
pm
ql 0:50 0:25 2KD=ha
Example 3. Consider an inventory system with the following parameter values: K = 6900 $ per lot, D = 65,000 units per year
and h = 21 $ per unit per year and the lot size q must be a multiple of a = 130 units. In this case, we have
r
ql 130 0:50 0:25 2690065; 000= 211302 130d49:7764e 6500 units
65; 000
6500
Cql 6900
21
137; 250 $ per year
6500
2
The reader can check that, in this example, the optimal solution is unique.
Example 4. Suppose an inventory system with the following parameter values: K = 5.6 $ per lot, D = 1000 units per year and
h = 2 $ per unit per year. Moreover, the lot size q must be a multiple of a = 10 units. Now
q
ql 10 0:50 0:25 25:61000=2102 10d7e 70 units
Then, there are two optimal solutions which are ql 70 units and qu ql a 80 units. Obviously,
1000
70
Cql 5:6
2
150 $ per year
70
2
1000
80
Cqu 5:6
2
150 $ per year
80
2
3. The EOQ/EPQ models with backlogged shortages and integer lot size
In the Harris EOQ model and in the Tafts EPQ model previously considered, shortages are not allowed. Both models can
be extended considering planned shortages. In the case of the EOQ inventory model, instead of the assumption (vii), we consider the following assumption: the shortages are allowed, fully backordered and there is a backorder cost linearly depending on time. Thus, if x is the cost per unit backordered and unit of time ($ per unit and per unit of time) and b is the maximal
backordered quantity in each cycle, it follows that the objective function and its constrains for the EOQ inventory model with
shortages are:
2
Minimize
b
Cq; b K Dq h qb
x 2q
2q
subject to
q > 0;
3:1
06b6q
q
q 2KDh x=hx; b hq =h x
q
Cq ; b 2KDhx=h x
On the other hand, the EPQ inventory model with shortages fully backlogged is given by the following minimization
problem:
2
Minimize
b
Cq; b K Dq h qa2qb
a x 2qa
subject to
q > 0;
3:2
0 6 b 6 qa
q
q 2KDh x=hax; b haq =h x
q
Cq ; b 2KDhax=h x
The above variables are, in general, continuous values. However, in some situations the supplier sells the lots in integer
values and the buyer sells to his/her clients the products as continuous values. For example, the buyer purchases in a discrete
amount (kilograms, tons, pounds, liters, etc.) and sells the products in fractions (as continuous). Then, now, we have the EPQ
1664
model with shortages, considering the lot size (q) as integer variable and the shortages level (b) as continuous variable. In a
similar way, it can be derived the result for q in the EOQ with shortages because both models only differ by the constant
parameter a. Thus, the problem that should be solved is
2
Minimize
b
Cq; b K Dq h qa2qb
a x 2qa
subject to
3:3
Note that the above minimization problem is a mixed integer nonlinear optimization problem. In order to obtain the optimal q as integer value, we will follow a sequential optimization procedure in two phases: Stage I and Stage II. Stage I consists
on the optimization of the continuous variable (b), which can be carried out through an algebraic method (see, for instance,
Crdenas-Barrn [911]). The algebraic method requires only the knowledge of simple factorization together complete the
perfect trinomial and squared binomial; which are considered as elementary mathematics. Stage II consists of arranging the
total inventory cost function in such way that contains only the integer lot size q. Then, we apply our method for optimization a function with an integer variable.
Stage I. Optimization of shortages level (b)
Consider the function of the total cost of inventory system in the EPQ with shortages:
2
Cq; b K
D
qa b2
b
x
h
q
2qa
2qa
Cq; b
h xb
qa
D
hb h
K
q
2qa
2
It is easy to see that C(q,b) is a quadratic function of b which represents a parabola in the plane. Also the function C(q,b) is
restricted to 0 6 b 6 qa. Thus, for a xed value q > 0, the above function attains its minimum at the point
bq hqa=h x
Substituting the formula bq into equation C(q,b),we get the following minimization problem:
Minimize
Cq; bq K Dq q2a
subject to
hx
hx
3:4
Then, applying our method in order to nd the optimal integer lot size, we have:
ql d0:50
q
q
0:25 2KDh x=haxe or qu b0:50 0:25 2KDh x=haxc
p
Moreover, there is a unique optimal solution when ql qu , that is, if and only if 0:50 0:25 2KDh x=hxa (or
p
0:50 0:25 2KDh x=hxa) is not an integer number. Otherwise, there are two optimal solutions, ql and qu ql 1.
In Table 2, the mathematical expressions for the optimal integer lot size of EOQ and EPQ inventory models with shortages,
in both cases when the lot size is restricted to be a integer value and when this is a multiple of a, are shown.
Example 5. Consider an inventory system with the following parameter values: D = 2000 units per year, P = 5000 units per
year, K = 10 $ per lot, h = 2 $ per unit per year and x = 3 $ per unit per year. Now, we have
Table 2
The integer lot size for the EOQ/EPQ inventory models with shortages.
Inventory
model
EOQ with shortages, q integer,
b continuous and 0 6 b 6 q
EPQ with shortages, q integer,
b continuous and 0 6 b 6 qa
bl hql =h x
l
pm
ql 0:50 0:25 2KDh x=hxa
bl haql =h x
bl hql =h x
l
pm
ql a 0:50 0:25 2KDh x=hxaa2
bl haql =h x
1665
l
pm
ql 0:50 0:25 21020005=20:63 d235:202791e 236 units
q
ql 5 0:50 0:25 2103425=230:652 5d19e 95 units
Then, there are two optimal solutions which are ql ; bl , with ql 95 units and qu ; bu , with qu
Moreover,
bu 21000:6=5 24 units
and, obviously,
342
950:6 23
Minimize
Cn K Hn h DH
2n
subject to
4:1
being H the planning horizon, D the demand rate, K the ordering cost, h the holding cost per unit and per unit time and n the
number of orders in the planning horizon of length H. Note that q=DH/n is the corresponding lot size and it can be integer or
continuous.
q
Analysis similar to that in the previous section shows that the optimal solution is nl 0:50 0:25 Dh=2KH2 or
q
and, moreover, there is a unique optimal solution, if and only if, 0:50
nu 0:50 0:25 Dh=2KH2
q
q
2
2
0:25 DhH =2K (or 0:50 0:25 DhH =2K) is not an integer number. Otherwise, there are two optimal solutions
for n, which are: nl and nu nl 1.
Example 7. Consider an inventory system with the following parameter values: K = 9700 $ per lot, D = 98000 units per
year, h = 26 $ per unit per year and H = 11 years. In this case, we have
q
nl 0:50 0:25 9800026=29700112 d125:5651e 126 orders
ql 9800011=126 8555:5556 units
126
9800011
Cnl 9700
26
222331:3131 $ per year
11
2126
The reader can note that there is a unique optimal solution for this example.
Example 8. Suppose an inventory system with the following parameter values: K = 10 $ per lot, D = 330 units per year,
h = 2 $ per unit per year time and H = 2 years. Now
q
nl 0:50 0:25 3302=21022 d11e 11 orders
1666
Then, there are two optimal solutions which are nl 11 orders and nu nl 1 12 orders. Moreover,
Minimize
D
Cq; n K Dq h 2q K o nq
ho n1
q
2
subject to
5:1
where D is the demand of the system, K and h are the ordering and holding costs, respectively, for the retailer, K o and ho are
the same costs for the wholesaler, q is the lot size of the retailer and n is an integer number such that the wholesaler lot size
is given by qo nq (see, for instance, [15, p. 136] or [16, p. 852]). Additionally, h must be greater than ho (see, for instance, [7,
p. 122]).
For every xed positive integer value of n, we get a univariate function in q, which is similar to the considered in the basic
EOQ model. Thus, the optimal solution is
qn
q
2K K o =nD=h ho nho
5:2
Cqn ; n
q
2K K o =nDh ho nho
5:3
Cqn ; n
q
2DKh ho K o ho gn
5:4
where
gn Kho n
K o h ho
n
5:5
Obviously, to minimize Cqn ; n is equivalent of minimizing g(n). The structure of the function g(n) is similar to the objective function of the problem (2.1). Thus, if we now consider n as a variable with n 2 N f1; 2; 3; :::g, in the same way that in
Section 2, we see that the optimal solution for n is
where A1 0:50
q
0:25 K o h ho =Kho
5:6
As in the previous sections, there is a unique optimal solution, if and only if, A1 is not an integer number. Otherwise,
there are two optimal solutions for the variable n, which are: nl and nu nl 1. The procedure solution for this problem
proposed by Axster [15] and Hillier and Lieberman [16] does not identify the situation when the problem has two optimal
solutions for the variable n. Finally, after than the optimal value of n is determined, the optimal value of q is calculated using
the formula (5.2).
Example 9. Consider an inventory system with the following parameter values: D = 19,650 units per year, K = 98 $ per lot,
h = 13 $ per unit per year, K o 2003 $ per lot and ho 11 $ per unit per year. In this case, we have
l
pm
nl 0:50 0:25 200313 11=9811 d1:491517261e 2
p
qn 298 2003=219650=13 11 211 1341:801494 units
p
Cqn ; n 298 2003=21965013 11 211 32203:23586 $ per year
The reader can note that there is a unique optimal solution for this example.
Example 10. Suppose an inventory system with the following parameter values: D = 19690 units per year, K = 10 $ per lot,
h = 2 $ per unit per year, K o 1100 $ per lot and ho 1 $ per unit per year. Now
1667
l
pm
nl 0:50 0:25 11002 1=101 d10e 10
Then, there are two optimal solutions for n which are nl 10 and nu nl 1 11. Thus,
p
210 1100=1019690=2 1 101 655:4387843 units
p
210 1100=1119690=2 1 111 600:8188856 units
qn10
qn11
Obviously,
p
210 1100=10196902 1 101 7209:826628 $ per year
p
Cqn ; n 11 210 1100=11196902 1 111 7209:826628 $ per year
Cqn ; n 10
5.2. The EOQ model in a warehouse-retailer system with integer lot size
Next, we consider that, besides the variable n, also the variable q must be an integer number. In this case, the formulation
of the model must be
Minimize
D
Cq; n K Dq h 2q K o nq
ho n1q
2
subject to
5:7
Now, the solution of the model is not easy. First, it is necessary to point out that the optimal solution cannot be determined to round down or to round up the optimal solution of q and n obtained when we consider both continuous variables.
Second, the optimal solution cannot be obtained either using the procedure previously exposed when q was a positive continuous variable and n a positive integer variable; that is, rounding down or rounding up the solution for q given by that
procedure.
To solve this model we will use a heuristic relying on the comparison between some lower bounds of the new considered
problem and the value of the objective function in some points which are near of the optimal solutions obtained using the
following models:
(i) The model given in (5.1).
(ii) The model given in (5.1), but considering that q must be a positive integer variable and n a positive continuous variable. That is, we have the problem
Minimize
D
Cq; n K Dq h 2q K o nq
ho n1q
2
5:8
The solution of the model given in (5.8) can be obtained by a method which is similar to one described in the previous
section. Thus, for every xed positive integer value of q, we get a univariate function in n, which is similar to the considered
in the basic EOQ model. Thus, the optimal solution is
nq
q
2K o D=ho q2
5:9
Cq; nq K
D
q p
h hO 2K o Dho
q
2
5:10
If we now consider q variable with q 2 N f1; 2; 3; . . .g, in the same way that in Section 2, we see that the optimal solution for q is
q
0:25 2KD=h ho
5:11
As in the previous sections, there is a unique optimal solution, if and only if, A2 is not an integer number. Otherwise, there
are two optimal solutions for the lot size q.
On the other hand, if we consider the model given by (5.1), but replacing the restriction n > 0 and integer to n > 0 and
continuous, or if we consider the model given by (5.8), but replacing the restriction q > 0 and integer to q > 0 and continuous, then we obtain the following model:
1668
Minimize
D
Cq; n K Dq h 2q K o nq
ho n1q
2
subject to
5:12
It is easy to solve this model using the one of the two previously developed procedures. Now, the optimal solution is
q
q
q
2KD=h ho and n K o h ho =Kho
5:13
Cq ; n
pq p
2D
Kh ho K o ho
5:14
5.2.1. Heuristic procedure for solving the EOQ model with integer lot size in a warehouse-retailer system
Relying on the solutions obtained previously, next we obtain a heuristic procedure for solving the model exposed in (5.7).
We will proceed in two stages.
5.2.1.1. Stage 1. Solve the model exposed in (5.1). Round down or round up the obtained optimal solution for q and select the
best of them. That is, if qc ; ni is an optimal solution obtained for the model (5.1), we consider qc ; ni and qc ; ni , and we
select the best one. To make this, we use the same procedure exposed
Section 2, but now in the context here considered.
pin
That is, instead of the optimal continuous value of q given by qn 2K K o =nD=h ho nho , we consider
q
ql 0:50 0:25 2K K o =nD=h ho nho :
q
nl 0:50 0:25 K o h ho =Kho ;
qm
l
ql 0:50 0:25 2DK K o =nl =h ho nl 1
5:15
p
Alternatively, using the above notation given by A1 0:50 0:25 K o h ho =Kho and adding now the new notal
pm
tion B1 n 0:50 0:25 2DK K o =n=h ho n 1 , we can rewrite (5.15) as
nl dA1 e and ql B1 nl
5:16
5.2.1.2. Stage 2. Solve the model exposed in (5.8). Round down or round up the obtained optimal solution
for n and
select the
best of them. That is, if qi ; nc is an optimal solution obtained for the model (5.8), we consider qi ; nc and qi ; nc , and we
select the best one. To make this, we use the same procedure exposed in Section
2, but
now in the context here considered.
p
That is, instead of the optimal continuous value of n given by nq 1=q 2K o D=ho , we consider
q
nl 0:50 0:25 2K o D=ho q 2 :
Thus, we obtain the approximate solution given by
q
q
ql 0:50 0:25 2KD=h ho and nl 0:50 0:25 2K o D=ho ql 2
5:17
ql dA2 e and nl B2 ql
5:18
p
p
where A2 0:50 0:25 2KD=h ho and B2 q 0:50 0:25 2K o D=ho q2 .
Note that, at the moment to apply the rst stage, we can identify six cases that can be occurring. These cases are shown in
Table 3. For the case 1 the problem only has one solution. With regards to cases 2 and 3, the problem has two possible solutions. In the cases 4 and 5, the problem has tree possible solutions. Finally, for case 6, the problem has four possible solutions.
Note that in case 1 only exists a unique solution. In contrast, in cases 26 the problem could have multiple solutions. Similarly occurs when we apply the second stage (see Table 4). Consequently, when we apply the two stages, we can identify 36
cases, which are obtained when mixing in all the possible ways the cases before considered.
Consequently, when both, q and b, are restricted to be integer variables, we have obtained a procedure to solve the problem. It is important to point out that the above procedure is an approximated method because the obtained points q ; n ,
with q and n both integers, are not necessary the optimal solution to the integer problem. However, the solution that is
given by the approximated method can be a very good solution and, sometimes, it is the optimal solution.
In order to get a feel of the quality of the solution, this solution can be compared with a lower bound LB. Thus, one can
determine the gap between approximated solution and the optimal solution. The gap is calculated as:
1669
Conditions
Is A1 integer?
Number of
solutions for n
Conditions
Is B1 n integer?
Number of
solutions for q
Number of solutions
for the problem
No
One
n1
No
One
q11
One
q11 ; n1
No
One
n1
Yes
Two
q11 and q12
Two
q11 ; n1 and q12 ; n1
Yes
Two
n1 and n2
No, if n n1
No, if n n2
Two
q11 for n1
q21 for n2
Two
q11 ; n1 and q21 ; n2
Yes
Two
n1 and n2
No, if n n1
Yes, if n n2
Three
q11 for n1
q21 and q22 for n2
Three
q11 ; n1
q21 ; n2 and q22 ; n2
Yes
Two
n1 and n2
Yes, if n n1
No, if n n2
Three
q11 and q12 for n1
q21 for n2
Three
q11 ; n1 and q12 ; n1
q21 ; n2
Yes
Two
n1 and n2
Yes, if n n1
Yes, if n n2
Four
q11 and q12 for n1
q21 and q22 for n2
p
p
where A1 0:50 0:25 K o h ho =Kho and B1 n 0:50 0:25 2DK K o =n=h ho n 1.
Four
q11 ; n1 and q12 ; n1
q21 ; n2 and q22 ; n2
Table 4
Cases can be occurring to apply second stage.
Case
Conditions
Is A2 integer?
Number of
solutions for q
Conditions
Is B1 n integer?
Number of
solutions for n
Number of solutions
for the problem
No
One
q1
No
One
n11
One
q1 ; n11
No
One
q1
Yes
Two
n11 and n12
Two
q1 ; n11 and q1 ; n12
Yes
Two
q1 and q2
No, if q q1
No, if q q2
Two
n11 for q1
n21 for q2
Two
q1 ; n11 and q2 ; n21
Yes
Two
q1 and q2
No, if q q1
Yes, if q q2
Three
n11 for q1
n21 and n22 for q2
Three
q1 ; n11
q2 ; n21 and q2 ; n22
Yes
Two
q1 and q2
Yes, if q q1
No, if q q2
Three
n11 and n12 for q1
n21 for q2
Three
q1 ; n11 and q1 ; n12
q2 ; n21
Yes
Two
q1 and q2
Yes, if q q1
Yes, if q q2
Four
n11 and n12 for q1
n21 and n22 for q2
Four
q1 ; n11 and q1 ; n12
q2 ; n21 and q2 ; n22
where A2 0:50
p
p
0:25 2KD=h ho and B2 q 0:50 0:25 2K o D=ho q2 .
5:19
If the gap between the approximated method and the solution with the lower bound is small, then we conclude that the
distance or difference between the approximated solution and the unknown optimal solution must be small. But, if there is a
large gap between both solutions, then this leaves us uncertain about of the effectiveness of the approximated method. Then,
for this type of situation is better to use another type of measure of performance for the approximated method. For example,
it is well-known to use the percent penalty as measure of the performance of one approximated method, where the percent
penalty (from now on, p.p.) is given by:
p:p:
gap
100
LB
5:20
Again, if the percent penalty is low (for example, less than 0.015%), we conclude that the gap between the approximated
solution and the unknown optimal solution must be low also.
In addition, it is important to remark that in some cases the solution given by the approximated method is the optimal
solution. The optimal solution can be easily identied when the gap or the percent penalty is equal to zero. However, it is
possible that the gap is not zero but the approximated solution is the optimal solution, because the gap is calculated comparing the cost of the approximated solution with a lower bound instead of the optimal cost.
1670
Obviously, it is very important to select a good lower bounds for help us to have a small gap and percent penalty. That is,
we want to select a lower bound as big as possible. For this, it is convenient to introduce the following notation: let LBo be the
optimal objective value of the model (5.12), that is, when both variables are continuous; let LB1 be an optimal value of the
model (5.1), that is, when q is continuous and n is integer; let LB2 be an optimal value of the model (5.8), that is, when q is
integer and n is continuous and, nally, let LB3 be an optimal value of the model (5.7), that is, when both variables are integer. Evidently, we have
5:21
Step 1:
(a) A1 4:52494; n1 d4:5249e 5; B1 n1 199:50062; q11 d199:5006e 200; LB1 1200
(b) Cq11 ; n1 1200
Step 2:
(a) A2 199:50062; q1 d199:50062e 200; B2 q1 4:52494; n11 d4:52494e 5; LB2 1200
(b) Cq1 ; n11 1200
Step 3: qi ; ni q1 ; n11 200; 5
Step 4: gap 0; percent penalty 0%: Therefore;qi ; ni is the optimal solution.
Example 12. Suppose a warehouse-retailer system with the following parameter values: D = 624195 units per year, K = 130
$ per lot, h = 26 $ per unit per year, K o 890 $ per lot and ho 13 $ per unit per year. Now, we obtain
Step 1.
(a) A1 2:16386; n1 d2:16386e 3; B1 n1 3200; q11 3200; q12 3201; LB1 166425:997969
(b) Cq11 ; n1 Cq12 ; n1 166426
Step 2.
(a) A2 3532:756; q1 d3532:756e 3533; B2 q1 2:1640; n11 d2:1640e 3; LB2 166114:994098
(b) Cq1 ; n11 167239:602038
Step 3.
qi ; ni
q11 ; n1 3200; 3
q12 ; n1 3201; 3
1671
Data
EXCEL solution
1a
D = 2000,
K = 10, h = 2,
K o 250; ho 1
qi ; ni q1 ; n11 200; 5
Cqi ; ni 1200
p.p. = 0%
qL ; nL 200; 5
CqL ; nL 1200
p.p. = 0%
qE ; nE 200; 5
CqE ; nE 1200
p.p. = 0%
1b
D = 8900,
K = 65, h = 26,
K o 690; ho 11
qi ; ni q11 ; n1 268; 4
Cqi ; ni 15793:1269;
p.p. = 0.0000714040%
qL ; nL 267; 4
CqL ; nL 15793:1667
p.p. = 0.000323419%
qE ; nE 278; 4
CqE ; nE 15804:4173
p.p. = 0.0715608%
1c
D = 100,
K = 10, h = 31,
K o 1000; ho 1
qL ; nL 8; 55
CqL ; nL 692:2727
p.p. = 0.00854242%
qE ; nE 9; 55
CqE ; nE 695:6313
p.p. = 0.493737%
2a
D = 624195,
qL ; nL 3200; 3
qE ; nE 3516; 3
CqL ; nL 166426
p.p. = 0.00000122042%
CqE ; nE 167162:0751
p.p. = 0.442285%
qL ; nL 8; 50
qE ; nE 8; 49
K = 10, h = 31,
K o 816; ho 1
qi ; ni q1 ; n11 8; 56
Cqi ; ni 692:2143;
p.p. = 0.0000997112%
q11 ; n1 3200; 3
qi ; ni
q12 ; n1 3201; 3
Cqi ; ni 166426
p.p. = 0.00000122042%
q1 ; n11 8; 50
qi ; ni
q1 ; n12 8; 51
Cqi ; ni 649
p.p. = 0.00305132%
CqL ; nL 649
p.p. = 0.00305132%
CqE ; nE 649:1633
p.p. = 0.0282085%
3a
D = 9800,
K = 10, h = 22,
K o 100; ho 1
qi ; ni q21 ; n2 98; 14
Cqi ; ni 3429:2857
p.p. = 0.00000216940%
qL ; nL 97; 14
CqL ; nL 3429:4588
p.p. = 0.00504837%
qE ; nE 98; 14
CqE ; nE 3429:2857
p.p. = 0.00000216940%
3b
D = 340,
K = 16, h = 41,
K o 1500; ho 1
qL ; nL 16; 63
CqL ; nL 1669:9524
p.p. = 0.000113004%
qE ; nE 17; 59
CqE ; nE 1669:9746
p.p. = 0.00144210%
4a
D = 31,
qL ; nL 30; 1
qE ; nE 31; 1
CqL ; nL 61
p.p. = 0.0134400%
CqE ; nE 61
p.p. = 0.0134400%
qL ; nL 1; 61
qE ; nE 2; 31
K = 1, h = 62,
K o 31; ho 1
qi ; ni q1 ; n11 16; 63
Cqi ; ni 1669:9524
p.p. = 0.000113004%
8
< q11 ; n1 30; 1
qi ; ni q12 ; n1 31; 1
:
q21 ; n2 20; 2
Cqi ; ni 61
p.p. = 0.0134400%
8
< q1 ; n11 1; 61
qi ; ni q1 ; n12 1; 62
:
q2 ; n21 2; 31
Cqi ; ni 153
p.p. = 0.00132848%
CqL ; nL 153
p.p. = 0.00132848%
CqE ; nE 153
p.p. = 0.00132848%
5a
D = 827.7,
K = 100, h = 21,
K o 550; ho 1
qi ; ni q11 ; n1 92; 10
Cqi ; ni 2774:4946
p.p. = 0.00136465%
qL ; nL 92; 10
CqL ; nL 2774:4946
p.p. = 0.00136465%
qE ; nE 92; 10
CqE ; nE 2774:4946
p.p. = 0.00136465%
5b
D = 120,
K = 9, h = 31,
K o 976; ho 1
qi ; ni q2 ; n21 9; 54
Cqi ; ni 738:9877
p.p. = 0.000566119%
qL ; nL 8; 60
CqL ; nL 739
p.p. = 0.00223675%
qE ; nE 9; 54
CqE ; nE 738:9877
p.p. = 0.000566119%
K = 130, h = 26,
K o 890; ho 13
2b
D = 100,
K = 10, h = 2,
K o 20; ho 1
4b
D = 61,
Example 13. Consider an inventory system with the following parameter values: D=61 units per year, K = 1 $ per lot,
h = 62 $ per unit per year, K o 31 $ per lot and ho 1 $ per unit per year. Now, we get
Step 1:
(a) A1 42:9885; n1 d42:9885e 43; B1 n1 1:006248; q11 d1:006248e 2; LB1 147:767259
(b) Cq11 ; n1 156:488372
Step 2:
(a) A2 1; q1 1; q2 2; B2 q1 61; n11 61; n12 62; B2 q2 30:2530; n21 d30:2530e 31; LB2 152:997967
(b) Cq1 ; n11 Cq1 ; n12 Cq2 ; n21 153
Step 3:
8
>
< q1 ; n11 1; 61
qi ; ni q1 ; n12 1; 62
>
:
q2 ; n21 2; 31
Step 4: gap 0:00203255; percent penalty 0:00132848%.
1672
Table 5 shows a comparison of the approximated solutions calculated by using our procedure and those obtained using
LINGO and EXCEL optimizers for different values of the input parameters.
An interesting observation that emerges from Table 5 is that our approximated procedure is a good and efcient method
because it nds always a better or equal solution than LINGO and EXCEL optimizers. As well as, all percent penalties are lower than 0.015%.
F.W. Harris, How many parts to make at once, factory?, Mag Manage. 10 (1913) 135136 (p. 152).
E.W. Taft, The most economical production lot, The Iron Age 101 (1918) 14101412.
B. Roach, Origin of the economic order quantity formula; transcription or transformation?, Manage Decis. 43 (2005) 12621268.
J.T. Teng, H.M. Wee, Notes on Optimizing the Economic Lot Size of a Three-stage Supply Chain with Backordering Derived Without Derivatives,
Working Paper, The William Paterson University, 2009.
J.T. Teng, L.E. Crdenas-Barrn, The Economic Lot Size of the Integrated VendorBuyer Inventory System Derived Without Derivatives: A Simple
Derivation, Working Paper, The William Paterson University, 2009.
R.H. Wilson, A scientic routine for stock control, Harvard Business Review 13 (1934) 116128.
P.H. Zipkin, Foundations of Inventory Management, McGraw-Hill, Singapore, 2000.
G. Hadley, T.M. Whitin, Analysis of Inventory Systems, Prentice-Hall, Englewood Cliffs, 1963.
L.E. Crdenas-Barrn, The economic production quantity (EPQ) with shortage derived algebraically, Int. J. Prod. Econ. 70 (2001) 289292.
L.E. Crdenas-Barrn, Optimizing inventory decisions in a multi-stage multi-customer supply chain: a note, Transport. Res. E-Log. 43 (2007) 647654.
L.E. Crdenas-Barrn, Optimal manufacturing batch size with rework in a single-stage production system a simple derivation, Comput. Ind. Eng. 55
(2008) 758765.
L.B. Schwarz, Economic order quantities for products with nite demand horizons, AIIE Trans. 4 (1972) 234237.
H.L. Lee, S. Nahmias, Single-product single-location models, in: S.C. Graves, A.H.C. Rinnoy, P.H. Zipkin (Eds.), Handbooks in Operations Research and
Management Science, Logistics of Production and Inventory, vol. 4, North-Holland, Amsterdam, 1993, pp. 355.
J. Brimberg, W.J. Hurley, A note on the assumption of constant order size in the basic EOQ model, Prod. Oper. Manage. 15 (2006) 171172.
S. Axster, Inventory Control, Kluwer Academic Publishers, Boston, 2000.
F.S. Hillier, G.J. Lieberman, Introduction to Operations Research, McGraw-Hill, New York, 2005.