ARTICLE IN PRESS

Int. J. Production Economics 111 (2008) 180–191

www.elsevier.com/locate/ijpe

Production planning with limited inventory capacity and

allowed stockout
X. Liua,b,, Yl. Tub
a
School of Business Administration, Northeastern University, Shenyang, PR China
b
Department of Mechanical and Manufacturing Engineering, University of Calgary, Canada
Received 1 January 2006; accepted 1 January 2007
Available online 28 February 2007

Abstract

In production planning, there is a situation where the production quantity is limited by inventory capacity rather than
production capacity. This situation often happens in petrochemical manufacturing, food processing, glass manufacturing,
etc. Only a few studies can be found in literature for this situation, among which the stockout strategy is not well studied.
In this paper, we consider the production-planning problem with inventory capacity as a limiting factor. We further
consider the problem with the following features: (1) the stockout is allowed, (2) production and lost sale cost functions are
time varying and non-increasing, and (3) inventory capacity is constant. These features have their roots in practice. In this
paper, first we present a stockout model, and then we prove some properties at an optimal solution to our model. Based on
these properties, we further develop an algorithm with polynomial time complexity using the network flow approach. The
paper also provides a case study to further justify practicability of the stockout strategy with limited inventory and to show
the effectiveness of our model and algorithm.
r 2007 Elsevier B.V. All rights reserved.

Keywords: Production planning; Bounded inventory; Stockout; Algorithm

1. Introduction tion planning is to resolve the conflicting resource

The intensive competition among manufacturing
companies today demands that they have to make
right decisions on allocation of resources and
scheduling their operations. With the expansion of
market size, efficiency of production is becoming a
critical point of success in manufacturing compa-
nies. The normal motivation for effective produc-
Corresponding author. Department of Mechanical and Man-
ufacturing Engineering, University of Calgary, 2500 University
Drive NW, Calgary, Alberta, Canada T2N 1N4.
E-mail address: xialiu@ucalgary.ca (X. Liu).
needs and to meet customer demands within a given
period of time. Fruitful solutions have been
generated over last decades for somewhat conven-
tional problems that exclude stockout and its
related issues (e.g., cost function, etc.). However,
the stockout strategy in production planning,
together with the limited inventory and its related
issues, has not been well studied. In the following
two paragraphs, we discuss the background of the
limited inventory and define the stockout strategy in
production planning for the lowest cost.
Generally, in production planning, two pri-
mary limiting factors are production capacity and

0925-5273/$ - see front matter r 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.ijpe.2007.01.010
 ARTICLE IN PRESS
X. Liu, Y. Tu / Int. J. Production Economics 111 (2008) 180–191
inventory capacity, of which the former is more
popular and well discussed. However, in many
process industries such as paper manufacturing,
petrochemical manufacturing, food processing, and
pharmaceutical manufacturing, inventory capacity
may become a limiting factor. For example, in
the refinery industry, production is usually with
sufficiently high capacity (due to a relative stability
in both equipment and skills) and hence, customer
demands for the petrochemical products are usually
restricted by their sizes of oil tanks or barrels
(in other words by inventory or storage capacity).
There are a couple of other situations that produc-
tion is not able to meet customer demands. These
situations include (1) the setup and production
costs are too high and/or (2) the product holding
and storage costs (assuming that products need
to be held in storage for a while) are too high. When
meeting of customer demands is controlled under
the limit of inventory capacity, we call this situation
limited or bounded inventory capacity.
It is known that in production planning two
primary goals are meeting customer demands and
having the lowest cost. These two goals may not
always be consistent. For example, an increase in
the inventory level can certainly maximize the
meeting of customer demands, but the holding
cost may be too high, leading that the total cost is
not the lowest one. Therefore, from the cost point
of view, customer demands may not be satisfied.
The unsatisfied demand means sale loss to a
manufacturing company, and this situation in this
paper is called stockout. The stockout strategy
simply means to allow the stockout situation to
happen.
In general, the issue in production planning in
which inventory capacity is a limiting factor and
there is a stockout situation has not been well
addressed in literature, though this issue makes
sense in practice. In this paper, we will discuss this
issue. For simplicity, we name this issue production
planning with limited inventory capacity and
allowed stockout (PPLICS). We will further con-
sider that demands and costs with PPLICS are time-
varying (non-increasing) functions.
The remainder of this paper is organized as
follows. In Section 2, we will review related work
along with a further introduction to some back-
ground of PPLICS. In Section 3, a model for
PPLICS will be formulated. In Section 4, some
properties of the model will be presented. Based on
these properties, a polynomial time complexity
181
algorithm is proposed in Section 5. In Section 6, a
case study for a real-life problem in a refinery
company by using the proposed approach is
presented. Finally, in Section 7 a conclusion is
drawn along with a discussion of future research.
2. Related work
The capacity production-planning (CPP) problem
can be described as follows. For a finite time
horizon T, there are T periods from 1 to T, denoted
by t (t ¼ 1,2,y,T). In any production period t,
there is a dynamic demand. This demand must
be satisfied by one or more of the following
strategies: (1) production; (2) inventory; (3) back-
logging; (4) outsourcing; and (5) sale loss. Note that
there are limits, respectively, on production, in-
ventory, backlogging, outsourcing, and sale loss
levels. Furthermore, five kinds of costs must
be taken into account: production cost (including
setup cost), holding cost, backlogging cost, out-
sourcing cost, and sale loss cost. There are two
kinds of capacity limits: (1) production capacity
and (2) inventory capacity. When a production
planning takes production capacity as a limiting
factor, it is called capacitated production planning,
and when a production planning takes inventory
capacity as a limiting factor, it is called production
planning with limited inventory capacity or bounded
inventory (see also previous discussion). The
objective of CPP is to determine periods and
quantities supplied during these periods with the
objective of minimizing the total cost aggregated
from the five kinds of costs over the definite time
horizon T.
In literature, the following four families of models
are proposed, corresponding to the five strategies
mentioned before.
 Models without backlogging: Each demand dt
must be entirely delivered at period t by produc-
tion and/or inventory.
 Models with backlogging: Each demand dt must
be entirely delivered at a time period later than t
at the expense of backlogging cost. In other
words, each demand must be satisfied by
production and/or inventory from previous
periods and/or from subsequent periods.
 Lost sale models: There are two kinds of lost sale
models, stockout models and conservation mod-
els. In stockout models, the demand does not have
to be entirely met in all periods. Unmet demands
 ARTICLE IN PRESS
182 X. Liu, Y. Tu / Int. J. Production Economics 111 (2008) 180–191
mean revenue lost or penalty cost; in other words,
a part or whole demand in a period must be lost
when the inventory at that period is zero
(assuming that backlogging is not permitted). In
this kind of models, the lost sale in a period
implies a non-positive inventory. In conservation
models, the demand does not have to be entirely
met in all periods even when the inventory is
strictly positive in these periods. The reason
behind the conservation model is that the current
potential (in the inventory) is reserved for the
future.
 Outsourcing models: Each demand dt must be
entirely delivered at a period t at expense of the
outsourcing cost. In other words, a demand must
be satisfied by production and/or inventory from
previous periods and/or outsourcing.
Wagner and Whitin (1958) did the first study on
uncapacited (no any capacity limiting issue) lot
sizing problem. They proposed an O(T2) algorithm
for the problem based on the Zero Inventory Order
(ZIO) policy with a constant production and
holding cost. Since then, extensive studies have
been reported. Their problem was further shown to
be solvable in O(T log T) time by Aggarwal and
Park (1993), Federgruen and Tzur (1991), and
Wagelmans et al. (1992).
Florian et al. (1980) and Bitran and Yanasse
(1982) showed that the general CPP problem is NP-
hard. Florian and Klein (1971) proposed an O(T4)
dynamic programming algorithm with constant
production capacity and concave production and
holding costs. This algorithm has been further
improved by van Hoesel and Wagelmans (1996)
such that the O(T3) time complexity was achieved
for the problem with constant production capacity,
concave production cost, and linear holding cost.
Swoveland (1975) examined piecewise concave
production and holding or backlogging costs and
developed an algorithm similar to that of Florian
and Klein. Polynomial algorithms also exist for
many other production capacity single item lot
size problem (SILSP) with special cases; for details
refer to Kirca (1990), Chen et al. (1994), Shaw and
Wagelmans (1998), Baker et al. (1978), and Chung
et al. (1994). These algorithms are based on dynamic
programming, branch and bound, or a combination
of both.
With respect to the production-planning problem
with limited inventory capacity, Love (1973) is first
one who presented an O(T3) dynamic programming
algorithm with constant inventory capacity, con-
cave production and holding costs with allowed
backlogging. Gutierrez et al. (2002) proposed an
O(T3) dynamic programming algorithm with con-
stant inventory capacity and concave production
and holding costs. They have not considered back-
logging, and their work is thus a special case of the
model of Love (1973). However, they proposed
some properties of optimal solutions such that their
algorithm is 30 times faster than Love’s algorithm.
Recently, Gutierrez et al. (2005) further proved the
same complexity of the same problem with back-
logging case.
Regarding the production-planning problem with
stockout, Sandbothe and Thompson (1990) con-
sidered a stockout with limited production capacity,
in which cost functions are constant. Specifically,
they proposed a necessary condition for an optimal
solution and obtained an O(T3) algorithm when
production capacity is constant, and an O(2T)
algorithm when production capacity is time-varying
for the first k+1 period with a computable k.
Furthermore, in 1993 they extended their work by
imposing restrictions on both production capacity
and inventory capacity, and they have obtained an
algorithm with the O(2T) time complexity. Based on
the Sandbothe and Thomposon’s work, Aksen et al.
(2003) discussed the uncapacitated production-
planning problem with allowed lost sales. They
considered setup costs as constant, production and
holding costs as time-varying. Their algorithm has
achieved an O(T2) time complexity. Liu et al. (2004)
considered a lost sale problem with bounded
inventory, and further in their problem, the setup
cost is non-increasing, while production, inventory
holding, and lost sales costs are time-varying. They
proposed a strongly polynomial algorithm. Liu
(2004) presented an algorithm with a pseudo-
polynomial time complexity O(TD2) for the CPP
problem with the general cost function structure
and lost sales.
In summary, previous studies have not paid
attention to the PPLICS problem especially with
the time-varying functions of demands and costs.
On a general note, in the case of limited inventory
capacity, the time-varying cost functions are much
more frequently used than the constant functions,
as pointed out by Sandbothe and Thompson
(1990). Therefore, there is a strong need to study
the PPLICS problem with the time-varying func-
tions for both demands and costs as we con-
cluded above.
 ARTICLE IN PRESS
X. Liu, Y. Tu / Int. J. Production Economics 111 (2008) 180–191 183

3. Problem description pt the unit cost of production in period t

ht the unit inventory holding cost in period t
3.1. Problem statement st the stockout cost per unit product (i.e., a
penalty cost) in period t
The problem of PPLICS can be stated as this: dt the demand in period t
each demand dt must be partly or entirely delivered Ct the inventory capacity at the beginning of
at a period t by production and /or inventory and / period t
or stockout. In this paper, we further consider that Xt the production quantity in period t
(1) demands are time-varying, (2) production and It the inventory level at the end of period t
stockout cost functions are non-increasing (i.e., Lt the amount of unmet demands in period t
ptXpt+1 and stXst+1), and (3) inventory capacity is
constant. The problem is to find periods and
In the above, Xt, Lt, and It (t ¼ 1,y,T) are
production quantities within these periods such that
decision variables.
the total cost is minimum.

3.2. Problem assumptions 3.4. Mathematical model

There are the following assumptions underlying
our model for the above problem:
Assumption 1. Inventory capacity is limited or
bounded in all periods.
Assumption 2. Any unmet demand in a period is
considered lost without backlogging or postpone-
ment.
Assumption 3. The stockout cost (as a sort of
‘‘penalty’’ cost) is larger than the production cost.
Assumption 4. Demands are less than inventory
capacities in all periods, and the demands are gone
at the end of the periods.
Assumption 5. Both the initial inventory level and
We begin by giving definitions of the problem:
Definition 1. Define X maxt as the maximum produc-
tion quantity in period t. Note that the quantities of
production at the beginning of that period can not
exceed the inventory capacity of t-th period, so
X maxt ¼ minfC t ; X maxtþ1 þ d t g with X maxT ¼ d T .
The expression ðX maxT ¼ d T Þ ensures that the
inventory level at the last period (IT) is 0—i.e.,
Assumption 5.
P
Definition 2. Define Di;j ¼ jk¼i d k as the accumu-
lated demand from period i to period j.
Based on the above assumptions and definitions,
the model of our problem can be formulated below:
X
T

the holding level (at the end of the last period T) are Minimize ðK t yt þ pt X t þ ht I t þ st Lt Þ (1)
t¼1
zero.
with
Assumption 6. The short-term demand is known. (
1 if X t 40;
Note that Assumption 6 is true because products yt ¼ t ¼ 1; . . . ; T,
such as oil and food are commodities, and they 0 otherwise;
basically follow the population and demography that subject to
are relatively unchanged. Assumption 3 is true because
otherwise we just let customer demands unmet. I t1 þ X t  ðd t  Lt Þ ¼ I t ; t ¼ 1; . . . ; T, (2)

Lt pd t ; t ¼ 1; . . . ; T, (3)
3.3. Notations and variables
I t1 þ X t pC t ; t ¼ 1; . . . ; T, (4)
In order to facilitate our explanation, the follow-
ing notation will be used throughout this paper. X t pyt X maxt ; t ¼ 1; . . . ; T, (5)

T the number of periods indexed from 1 to T I 0 ¼ I T ¼ 0, (6)

Kt the setup cost of production in period
t(t ¼ 1,y,T) X t X0; I t X0; Lt X0; t ¼ 1; . . . ; T. (7)
 ARTICLE IN PRESS
184 X. Liu, Y. Tu / Int. J. Production Economics 111 (2008) 180–191
From (2) and (4), we get
0pI t pC t  d t þ Lt ; t ¼ 1; . . . ; T.
In the above, expression (1) represents the
minimization of the total cost aggregated from the
setup, production, inventory hold, sale loss activ-
ities. The constraint (2) represents the material
balance for the inventory. The constraint (3)
represents that any lost demand in that period
cannot exceed the demand of that period. The
constraint (4) represents the inventory capacity
limit. The constraint (5) represents production is
bounded indirectly due to the bounded inventory
(see Definition 1 about X maxt ). The constraint (6)
represents Assumption 5. Constraint (7) means that
the production quantities, stockout quantities, and
the inventory levels should be non-negative.
3.5. Model translation
In the following, we translate our model into a
concave network flow model in order to make use of
a rich set of network theories to develop our
algorithms to the model. This translation procedure
is similar to the one proposed by Sandbothe and
Thompson (1990); yet the difference is that we get a
network flow model with the variable edges due to
the time varying cost functions in our problem. We
begin by giving definitions of nodes and edges on
the flow network:
Definition 3. If Xt40, the period t is a production
period.
T
Σ
t =1
Xt
T
Σ dt
t =1
M 1
T
Σ Lt
t =1
Fig. 1. A network flow model for stockout problem.
Definition 4. If It40, the period t is a holding
period.
(8)
Definition 5. If It ¼ 0 and It40, the period t is a
stockout period.
Definition 6. Define X mint as the minimum produc-
tion quantity during a production period. Note that
in our model only when ðst  pt ÞX t XK t profit will
be obtained by setting up a production. There-
foreX mint ¼ K t =ðst  pt Þ , where dX e is the smal-
lest integer larger than or equal to X.
Based on the above definitions, our problem can
be formulated as a concave cost network flow
problem as shown in Fig. 1. M is the master supply
node containing the total demand of T periods. P
and L are two transshipment nodes, where node P
transships the total demands via production, and L
transships the total demands via stockout. Note that
the sum of these two transshipments is equal to the
inflow of the total demands at node M. Further,
Nodes 1, 2,y,T represent the demand in each
period such that the demand at node t is dt. Among
these nodes (NOT nodes 1 and T because
I0 ¼ IT ¼ 0), there are three incoming arcs, which
are (1) production arc coming from P, (2) inventory
arc coming from the previous inventory period, and
(3) stockout arc coming from L. There is a pair of
outgoing arcs, where one is inventory arc carrying
the inventory level at the end of that period to the
next node, and the other is demand arc carrying the
met demand. The balancing between the incoming
and outgoing arcs is represented by constraint (2) in
X1 X2 X3 XT
I1 I2 I3 … IT-1
2 3 T
d1 d2 d3
dT
LT
L1 L2 L3
 ARTICLE IN PRESS
X. Liu, Y. Tu / Int. J. Production Economics 111 (2008) 180–191
our model. By associating the costs with the
network, we have the following:
 directed arcs (M, P) and (M, L) ship flows at a
zero cost and have upper bounds of +N;
 directed arcs (P, t) with lower bounds of X mint
and upper bounds X mint (t ¼ 1,y,T) have the
zero shipping cost for the zero arc is flow and the
cost of Kt+ptXt if arc’s flow is Xt40;
 directed arcs (L, t) with upper bounds of dt for
t ¼ 1,y,T ship flows at the cost of stockout st
(labeled by Lt) per unit flow;
 directed arcs (t, t+1) with upper bounds of
Ctdt+Lt (where Ltpdt) for t ¼ 1,y,T ship
flows at a holding cost ht per unit flow; and
 directed arcs with flows dt have the zero unit
costs.
It is easy to know that all constraints, from
Eqs. (2)–(8), in our problem model are linear, and
the costs on all arcs of the network (in Fig. 1) are
concave. Therefore, the network is a concave cost
network as shown by Zangwill (1968, 1969).
4. Properties of the optimal solutions
In this section, we present some properties of an
optimal solution to the network flow model we
formulated above. These properties will be em-
ployed later in developing an algorithm with its
complexity being a strongly polynomial time.
Theorem 1. In an optimal solution, if Xt40, there
will be X mint pX t pX maxt .
Proof. For a contradiction, assume that we have an
optimal solution such that 0oX t oX mint for some t.
According to the definition of X mint (Definition 6),
we obtain the expression that K t þ pt X t 4st X t .
This expression implies that we will get profit even if
we simply lose Xt*, which is obviously not a case.
On the other hand, the situation that Xt*4X maxt
does not make sense in our problem. &
Theorem 2. In an optimal solution, if Lt40, there is
LtIt ¼ 0 for all t. In this case, if and only if the
inventory level is zero, the stockout occurs.
Proof. For a contradiction, assume that Lt*It*40
for some t. There are two paths through which we
start at node M and reach node t+1. Path 1 is
composed of arc (M, L), arc (L, t), and arc (t, t+1).
Path 2 is composed of arc (M, L) and arc (L, t+1).
185
Note that the cost along Path 1 is st+ht, and the
cost along Path 2 is st+1. Since stXst+1 (one
assumption in our problem is that the stockout
cost function is non-increasing), we come up with
st+ht4st+1 which means that Path 2 has a lower
cost than Path 1. Clearly, the lower cost solution
can be obtained by using alternate path, which
implies a contradiction. Note that Path 1 is
associated with the optimal (i.e., lowest cost)
solution (L*, I*). The fact that Path 2 is associated
with a solution that has a lower cost than the
optimal solution (L*, I*) contradicts with our
assumption. &
Theorem 3. In an optimal solution, if Xt40, there is
It1Xt ¼ 0 for all t.
Proof. For a contradiction, assume that It1Xt*40
for some t. In this case, we can increase production
with It1 in period t and decrease production with
It1 in period t0 (where t0 represents any previous
production period before t). Consequently, the
holding costs on arc (t0 ,t) are reduced, and produc-
tion costs are also reduced (because one assumption
in our problem is that production cost function is
non-increasing, i.e., ptXpt+1). This result implies
that Xt* is not an optimal solution which contra-
dicts with the assumption. Furthermore, as long as
Xt40, the above optimal process, resulting in the
reduction of It1, will continue until It1 ¼ 0. &
It is noted that in the above discussion regarding
Theorem 3, we have implied that a production
activity takes place only if the inventory level is
zero. Such a planning strategy is called Zero
Inventory Order (ZIO). Fig. 2 pictorially illustrates
the production-planning process in our problem. In
Fig. 2a, at the beginning of a cycle there is a zero
inventory level period (the first block), so produc-
tion should be taken in the subsequent period (the
middle block in Fig. 2a). After that, there are some
holding periods in the subsequent periods until
inventory level is zero (as shown in Fig. 2b, the
second block). In Fig. 2c, production takes place
… … …
Zero inventory period Production period Non-zero inventory period
Fig. 2. Zero inventory order.
 ARTICLE IN PRESS
186 X. Liu, Y. Tu / Int. J. Production Economics 111 (2008) 180–191
after the zero inventory level period again (the
middle block in Fig. 2c).
The reasoning process for Theorem 3 implies that
an original lost sale problem can be decomposed
into consecutive sub-problems g(t0 ,t) (where t0 is a
production period, t is the first zero holding period
after t0 with 1pt0 otpT, and g(t0 ,t) is the cost of
production, holding, and stockout from t0 to t) (i.e.,
It0 1 ¼ It ¼ 0) and for each Ii40, i ¼ t0 ,t0 +1,yt1.
The problem is thus to minimize g(t0 ,t) from t0 to t.
In what follows, we discuss how to minimize g(t0 ,t)
by proposing three lemmas.
Lemma 1. In an optimal solution, suppose that there
is a period i, t0 piot, Li40, for the production period
t0 . If a stockout occurs, we have i ¼ t (i.e., a stockout
occurs at the last period)
Proof. For a contradiction, assume there is a period
i , Li 40(where). This implies that the inventory
level I i 40. Thus, we have I i Li 40, which contra-
dicts with Theorem 2. &
Lemma 2. Let be a quantity of production from t0 to
t, where 1pt0 otpT. In an optimal solution to
g(t0 ,t), if Xt0 ,t40, there is Xt0 t ¼ min{Ct0 ,Ct0 ,t}.
Proof. For a contradiction, assume that we have
0oX t0 ;t  Dt0 ;t1 oC t0 for some t. From this as-
sumption, we know that the stockout will occur
before t, which contradicts with Lemma 1. Assume
that we have Dt0 ;t1 oX t0 ;t oC t0 oDt0 ;t for some t.
Given that d t 4I t1 40 by Theorem 3, we have
Xt ¼ 0 and hence I t ¼ I t1  d t þ Lt , and further by
Theorem 2, we have Lt I t ¼ Lt ðI t1  d t þ Lt Þ ¼ 0.
Therefore, either Lt ¼ 0, (which yields It to be
negative), or Lt ¼ d t  I t1 (which represents a
feasible decision). Nevertheless, it is clear that the
cost can be reduced by increasing the production arc
(P,t0 ) until the production quantity becomes max-
imumPCt0 with Dt0 ;t1 oX t0 ;t ¼ C t0 oDt0 ;t . Only when
pt0 þ tk¼t0 hk pst , there is Lt ¼ d t  I t1 . On the
other hand, if Dt0 ;t1P oX t0 ;t oDt0 ;t oC t0 , it can be
shown that if pt0 þ tk¼t0 hk pst the cost can be
reduced by letting X t0 ;t ¼ Dt0 ;t in period t0 and
It ¼ Lt ¼ 0. Last, assume Dt0 ;t oX t0 ;t oC t0 from this
assumption, we will have a solution at t+1 periods
during which the inventory is positive, which
contradicts with Theorem 3. &
Lemma 3. In an optimal solution, if I t0 1 ¼ I t0 ¼ 0
for any t0 , we have either Xt0 ¼ 0 and Lt0 ¼ d t0 , or
X t0 ¼ d t0 and Lt0 ¼ 0.
Proof. Since I t0 1 ¼ I t0 ¼ 0, the cost in period t0 is
K t0 þ pt0 X t0 þ st0 Lt0 , where Lt0 ¼ d t0  X t0 . If
K t0 þ pt0 X t0 pst0 X t0 , the minimum cost will be
obtained at the upper bound of production arc
(P,t0 ) with X t0 ¼ d t0 and Lt0 ¼ 0. Otherwise, if
K t0 þ pt0 X t0 4st0 X t0 , the minimum cost can be
obtained at the upper bound of stockout arc (L,t0 )
with Lt0 ¼ d t0 and Xt0 ¼ 0. &
Note that from Lemma 3, we can get the
minimum cost either K t0 þ pt0 d t0 or st0 d t0 (when t0 is
not the first zero holding period).
5. Algorithm development
We are now able to show that our model can be
solved with a forward recursive dynamic program-
ming algorithm in polynomial time.
By definition, t0 is a production period, and t is
the first next zero inventory period, 1pt0 otpT.
The sub-problem g(t0 , t) can be written as
X
t
gðt0 ; tÞ ¼ K t0 þ pt0 X t0 ;t þ hk ðX t0 ;t  D1;k þ D1;t0 1 Þ
k¼t0
þ st ðD1;t  D1;t0 1  X t0 ;t Þ. ð9Þ
For the sake of simplicity, we use the following
notations:
X
k
Hk ¼ ht , (10)
t¼1
X
k
Mk ¼ ht D1;t . (11)
t¼1
With these notations, we can rewrite (9) in the
following form:
gðt0 ; tÞ ¼ K t0 þ pt0 X t0 ;t þ ðX t0 ;t þ D1;t0 1 ÞðH t  H t0 1  st Þ
 M t1 þ M t0 1  st D1;t . ð12Þ
Let f(t) denote the minimum total production,
holding and stockout cost over period from 0 to t,
we have
f ð0Þ ¼ 0. (13)
There are two situations that are not overlapping
and thus need to be separately evaluated in the
algorithm for the lowest cost. One situation is that
at the period t which is presently concerned, the
inventory is zero (Fig. 3b). In this case, according to
Lemma 3, the optimal solution happens either K t þ
pt d t or st d t . The other situation is that the period t is
not the first zero inventory (Fig. 3a); in other words,

X. Liu, Y. Tu / Int. J. Production Economics 111 (2008) 180–191
the zero inventory comes earlier than the period t.
With respect to this second situation (Fig. 3a), we
need to determine the starting period (where
production takes place), which is before the period
t. We denote the starting period by nt. As such, the
definition of nt is given as follows:
( )
Xt
nt ¼ min t0 jst Xpt0 þ hk ; 1pt0 otpT .
k¼t0
With nt, we then know that the first zero inventory
can happen in any period from nt to t1. The
algorithm thus needs to evaluate all these possibi-
lities for the optimal solution. Therefore, we can
formulate the following backward recursive equa-
tion:
f ðtÞ ¼ minff ðt  1Þ þ minfK t þ pt d t ; st d t g,
min ff ðt0  1Þ þ gðt0 ; tÞgg.
nt pt pt1
0
Considering (12)–(14), a dynamic programming
algorithm can be devised. Note that the above
recursive Eq. (14) is composed of two parts: (i) the
first part (called type 1), where the production
period is not the first zero inventory period (see
Fig. 3a), and (ii) the second part (called type 2),
where the production period is the first zero
inventory period (see Fig. 3b). It is further noted
that the minimization of g(t0 ,t) follows Lemma 2;
in other words, it is subject to the constraint
Eqs. (2)–(8) for the period of t’ to t.
The computational complexity of the algorithm is
analyzed as follows. In the proposed algorithm,
namely Eq. (14), we need to first come up with f(t)
and then we need to solve f(t). To come up with f(t),
we need O(T) computation. Since we have the T
number of f(t) and Eq. (14) is a recursive one, we
therefore need O(T2) computations. Solving
Eq. (14) using the dynamic programming approach
is known to need O(T2) computations. As such, the
total computations of our algorithm are OðT 2 þ
T 2 Þ ¼ OðT 2 Þ.
… …
Zero inventory Production
period period
Fig. 3. Two situations when production takes place.
ARTICLE IN PRESS
187
6. A case study and numerical analysis of the
algorithms
6.1. Problem description
The problem addressed in this case study can be
described as follows. We concern the production
planning in a refinery. A typical refinery production
process is illustrated in Fig. 4. In general, the
process consists of three processing units: (1) the
crude oil distillation unit (CDU) which transforms
crude oil into bitumen, naphtha, distillates, etc., (2)
the light hydro-treatment (LT), and (3) the heavy
hydro-treatment (HT). The HT and LT transform
distillates into various naphthenic special oils. The
production planning makes decisions on when and
how production should take place in order to
ð14Þ minimize the total cost.
The production planning in a refinery has the
following characteristics: (1) the production process
can not be stopped, and the products and sub-
products should flow into the special tanks, (2)
production process is restricted by both production
capacities and storage or tank capacities, (3) in
general, production capacities are larger than
storage or tank capacities, (4) in general, the total
demands for petrochemical products are more than
the capacities, (5) setup cost is higher enough to
reject a new production, (6) inventory levels and an
increased need for storage capacities (tanks) are
associated with larger holding and capital costs, (7)
products and sub-products should be put in tanks
for while, and withdrawn at the end of period, (8)
demands should be less than the inventory (tank)
capacity at a period.
Furthermore, a particular refinery company we
dealt with has the following profile. First, the
company produces the product of aviation kero-
sene. Second, the price of the product is primarily
regulated by the government. Third, the company
has a relatively good management of proprietary
… … … …
The first zero
The first zero
inventory period
inventory period
Non-zero Zero inventory Production Non-zero
inventory period period period inventory period
 ARTICLE IN PRESS
188 X. Liu, Y. Tu / Int. J. Production Economics 111 (2008) 180–191

Distillate 1
Naphtha

Distillate 2

Crude oil
LT
Distillate 3 Naphthenic
CDU
Special Oil
Distillate 4 HT

Bitumen

Fig. 4. A refinery production process.

Inventory/tanks
Aviation
Crude oil Distillate kerosene

CDU LT

Fig. 5. A refinery production line.

knowledge, and therefore knowledge and skills for
production are well kept in the company. This
profile of the company is translated into the fact,
associated with our problem, that the production
cost is non-increasing. Accordingly, the lost sale
cost is non-increasing as well (note that in reality,
the sale price is not frequently changed in the length
of the period considered in this study).
Based on the production process shown in Fig. 4,
we can decompose the aviation kerosene production
process as one of the production line as shown in
Fig. 5. Further in the business practice of the
company, customer’s demands may not be satisfied,
which simply means that the sale loss situation
occurs, but the company will meet the demand
whenever the inventory is positive. Apparently, the
company has taken the stockout strategy.
6.2. Optimal solutions to the company’s problem
During our interaction with the company, we
extracted a real problem from the company. In this
problem, there are four periods and the production
costs associated with these periods are non-increas-
ing (see Table 1). The results of production planning
by the company based on our approach are: X mint ¼
5000 for all t, and X maxt ¼ {20 000, 20 000, 6000,
4000}. Note that in the first period the feasible
production quantities are X1 ¼ {0, 20 000}, X2 ¼ {0,
17 750, 19 750}, X3 ¼ {0, 6000}, X4 ¼ {0}, respec-
tively.
Table 2 shows the corresponding optimal solution
using our approach. It is interesting to notice that
we lose the demands in periods 1 and 4 if we cannot
get profit in spite of having enough capacity to meet
them; this means that the stockout occurs. The
demand in period 2 is satisfied by production, while
demand in period 3 is satisfied by inventory. From
Tables 1 and 2, the production quantity in each
period will be {0, 19 750, 0, 0}, and we can find the
minimum total cost to be $353 250 ( ¼ 3000(15)
+25 000+19 750(10+1)+2000(3)+4000(15)) using
our approach. If we take the traditional inventory
strategy (i.e., all demands are satisfied by produc-
tion or inventory), production will take place in
periods 1–3. The production quantity in each period
will be {3000, 19 750, 0, 4000}, and the mini-
mum cost is $403 250 ( ¼ 25 000+3000(10+5)+
25 000+19 750(10+1)+2000(3) +25 000+4000(10+
5)). This example has clearly shown the effectiveness of
 ARTICLE IN PRESS
X. Liu, Y. Tu / Int. J. Production Economics 111 (2008) 180–191 189

the stockout strategy and our approach to such a
stockout problem.
6.3. Numerical analysis of our algorithm
We have implemented our algorithm using
Microsoft Visual C++ 6.0, and the computer code
called Inventory Capacity Limiting with Stockout
(ICLS) runs on the Intel Pentium III 700 MHz
processor. In the following, we illustrate the
efficiency of our algorithm using a numerical study.
Table 1
Information of production of a refinery company
Period t Kt
1 25 000
2 25 000
3 25 000
4 25 000
Inventory capacity Ct ¼ 20000.
Table 2
The optimal solution for the refinery company
Test problems used for the study varied in sizes of
25–500 periods. The inventory capacity Ct was
chosen as 1000 and 10 000. The maximum demand
was chosen to be smaller than the capacity to meet
the assumption of our problem. Demands were
generated randomly from the uniform distributions,
U(50,500), and holding costs varied randomly from
uniform distributions, U(1,5). The setup cost was
chosen as Kt ¼ 25 000, 1000, and 100. The variable
production and the stockout costs were kept
between $10 and $15. As such, the different
minimum production quantities and maximum
holding periods could be generated using our
approach.
pt ht st dt
Fig. 6 shows the average CPU time versus the size
of problems. In Table 3, we show more detailed
10 5 15 3000 results of test problems. In Table 3, the first and
10 1 15 17 750 second columns represent the number of periods
10 3 15 2000
and the maximum inventory capacities correspond-
10 5 15 4000
ing to these periods, respectively. The third and
fourth columns show the setup cost and the average
running time, respectively. From Table 3, we can see
that for the sizes of 500 periods the ICLS problem
only takes 0.059 s, which is very efficient.

Period t Xt It Lt It1
7. Conclusion

1 0 0 3000 0 In this paper, we consider the production-plan-

2 19 750 2000 0 0 ning problem in which inventory capacity is a
3 0 0 0 2000
limiting factor. The problem is further complicated
4 0 0 4000 0
by the considerations: (1) the inventory capacity is
The total cost is $353 250. constant bounded, (2) backlogging is not allowed,

CPU Time VS the size of problem

0.07

0.06

0.05
CPU time (seconds)

0.04

0.03

0.02

0.01

0
25 75 150 250 350 450
Size of problem (T)

Fig. 6. CPU time versus the size of problem.

 ARTICLE IN PRESS
190 X. Liu, Y. Tu / Int. J. Production Economics 111 (2008) 180–191

Table 3
Test problems, results and CPU times

Length of Capacity of Setup Time of Length of Capacity of Setup Time of

period inventory cost CPU (s) period inventory cost CPU (s)

25 000 0.00013 25 000 0.01435

T ¼ 25 1000 1000 0.00014 T ¼ 250 1000 1000 0.01419
100 0.00019 100 0.01449
25 000 0.00013 25 000 0.01422
10 000 1000 0.00013 1000 1000 0.0142
100 0.00014 100 0.01406

25 000 0.00053 25 000 0.02268

T ¼ 50 1000 1000 0.00054 T ¼ 300 1000 1000 0.02201
100 0.00054 100 0.02199
25 000 0.00055 25 000 0.02216
10 000 1000 0.00054 10 000 1000 0.02211
100 0.00055 100 0.02213
25 000 0.00124 25 000 0.03093
T ¼ 75 1000 1000 0.00121 T ¼ 350 1000 1000 0.03038
100 0.0013 100 0.03038
25 000 0.00119 25 000 0.03118
10 000 1000 0.00118 10 000 1000 0.03026
100 0.00122 100 0.03018
25 000 0.00229 25 000 0.04110
T ¼ 100 1000 1000 0.00209 T ¼ 400 1000 1000 0.03921
100 0.00212 100 0.03927
25000 0.00212 25 000 0.03904
10 000 1000 0.00227 10 000 1000 0.03935
100 0.00223 100 0.03919
25 000 0.00486 25 000 0.05058
T ¼ 150 1000 1000 0.00487 T ¼ 450 1000 1000 0.04998
100 0.00476 100 0.0499
25 000 0.00473 25 000 0.04972
10 000 1000 0.00474 10 000 1000 0.04984
100 0.00488 100 0.04987

25 000 0.00889 25 000 0.06009

T ¼ 200 1000 1000 0.00871 T ¼ 500 1000 1000 0.05923
100 0.00868 100 0.05917
25 000 0.00871 25 000 0.0590
10000 1000 0.00877 10000 1000 0.05921
100 0.00874 100 0.05914

(3) an unsatisfied demand is lost with the stockout
strategy, and (4) both production and stockout
costs are non-increasing. This problem has not been
discussed in the current literature, while the problem
is encountered in practice especially with those so-
called process companies.
We formulated the problem model and developed
the algorithm for the model. As a result, we can
conclude that the complexity of the algorithm is
O(T2). This complexity is significant compared with
the complexity of O(T3) for the problem which is
studied by Sandbothe and Thompson (1990).
Further, the problem formulated by Sandbothe
and Thompson (1990) considers a constant produc-
tion capacity with allowed stockout.
Last, the problem featured with the stockout and
limited inventory, discussed in this study, has a
practical root especially in the area of product
 ARTICLE IN PRESS
X. Liu, Y. Tu / Int. J. Production Economics 111 (2008) 180–191
manufacturing. However, in some general planning
applications, such as product-assortment, batch-
queuing, investment-consumption, and reservoir-
control, a similar problem may be presented. We
believe that our model and algorithm can be readily
applied.
Acknowledgment
This research has been partially supported by the
National Science Foundation of China (Grant no.
70571077), Key Laboratory of Process Industry
Automation, Ministry of Education, China and
Natural Sciences and Engineering Research Council
of Canada (Grant no. JIIRP312747-04).
References
Aggarwal, A., Park, J.K., 1993. Improved algorithms for
economic lot-size problems. Operations Research 41 (3),
549–571.
Aksen, D., Altinkemer, K., Chand, S., 2003. The single-item lot
sizing production with immediate lost sales. European
Journal of Operational Research 147, 558–566.
Baker, K.B., Dixon, P., Magazine, M.J., Silver, E.A., 1978. An
algorithm for the dynamic lot-size problem with time-varying
production capacity constraints. Management Science 24 (16),
1710–1720.
Bitran, G.R., Yanasse, H.H., 1982. Computational complexity of
the capaciated lot size problem. Management Science 28 (10),
1174–1186.
Chen, H.D., Hearn, D.W., Lee, C.Y., 1994. A new dynamic
programming algorithm for the single item capacitated
dynamic lot size model. Journal of Global Optimization 4,
285–300.
Chung, C.S., Flynn, J., Lin, C.H.M., 1994. An effective algorithm
for the capacitated single item lot size problem. European
Journal of Operational Research 75, 427–440.
Federgruen, A., Tzur, M., 1991. A single forward algorithm to
solve general dynamic lot sizing models with n periods in
O(n log n) or O(n) time. Management Science 37, 909–925.
Florian, M., Klein, M., 1971. Deterministic production planning
with concave costs and capacity constraints. Management
Science 18 (1), 12–20.
Florian, M., Lenstra, J.K., Rinnooy Kan, A.H.G., 1980.
Deterministic production planning: algorithm and complex-
ity. Management Science 26 (7), 669–679.
191
Gutierrez, J., Sedeno-Noda, A., Colebrook, M., Sicilia, J., 2002.
A new characterization for the dynamic lot size problem with
bounded inventory. Computers & Operation Research 30,
383–395.
Gutierrez, J., Sedeno-Noda, A., Colebrook, M., Sicilia, J., 2005.
A polynomial algorithm for the production/ordering planning
problem with limited storage. Computers & Operation
Research; available online /http://www.Sciencedirect.comS.
Kirca, O., 1990. An efficient algorithm for the capacitated single
item dynamic lot size problem. European Journal of Opera-
tional Research 45, 15–24.
Liu, X., 2004. Models and algorithms for capacitated lot sizing
problems. Ph.D. Thesis. Université de Technologie de Troyes,
France.
Liu, X., Wang, C., Chu, F., Chu, C., 2004. A forward algorithm
for capacitated lot-sizing problem with lost sales. In:
Proceeding of the IEEE Fifth World Congress on Intelligent
Control and Automation, 15–19 June, China, vol. 4,
pp. 3192–3196.
Love, S.F., 1973. Bounded production and inventory models with
piecewise concave costs. Management Science 20 (3), 313–318.
Sandbothe, R.A., Thompson, G.L., 1990. A forward algorithm
for the capacitated lot size model with stockouts. Operations
Research 38 (3), 474–486.
Sandbothe, R.A., Thompson, G.L., 1993. Decision horizons for
the capacitated lot size model with inventory bounds and
stockouts. Computers & Operations Research 20 (5),
455–465.
Shaw, D.X., Wagelmans, A.P.M., 1998. An algorithm for single
item capacitated economic lot-sizing with piecewise linear
production costs and general holding costs. Management
Science 44, 831–838.
Swoveland, C.A., 1975. Deterministic multi-period production
planning model with piecewise concave production and
holding-backorder costs. Management Science 21 (9),
1007–1013.
van Hoesel, C.P.M., Wagelmans, A.P.M., 1996. An O(T3)
algorithm for the economic lot-sizing problem with constant
capacities. Management Science 42, 142–150.
Wagelmans, A., Van Hoesel, S., Kolen, A., 1992. Economic
lot-sizing: an O(n log n) algorithm that runs in linear time
in the Wagner–Whitin case. Operations Research 40,
145–156.
Wagner, H.M., Whitin, T., 1958. Dynamic version of the
economic lot size model. Management Science 5, 89–96.
Zangwill, W.I., 1968. Minimum concave cost flows in certain
networks. Management Science 14 (9), 429–450.
Zangwill, W.I., 1969. A backlogging model and a multi-
echelon model of a dynamic economic lot size pro-
duction system—a network approach. Management Science
15 (9), 506–527.