A note on optimal policies for a periodic inventory system with emergency orders

* Tel.: #886-35-712121x5201.

E-mail address: [email protected] (C. Chiang)

A note on optimal policies for a periodic inventory system

with emergency orders

Chi Chiang*

Department of Management Science, National Chiao Tung University, 1001 Ta Hsueh Road, Hsinchu, Taiwan, Republic of China

Received 24 December 1998; received in revised form 23 July 1999

Abstract

In this article, we develop a dynamic programming model for a periodic review inventory system in which emergency orders can be placed at the start of each period, while regular orders are placed at the beginning of an order cycle (which consists of a number of periods). We assume that the regular and emergency supply lead times di!er by one period. We devise a simple algorithm of computing the optimal policy parameters. Thus, the ordering policy developed is easy to implement.

Scope and purpose

Alternative resupply modes are commonly used in inventory systems. For example, a materials manager could choose to replenish the inventory of an item by air if its inventory position gets dangerously low. In this article, we study a periodic review inventory system in which there is an emergency supply mode in addition to a regular supply mode. We develop optimal ordering policies that minimize the total expected discounted cost of procurement, holding, and shortage over a "nite planning horizon. These optimal policies are next shown to converge as the planning horizon is extended, if some conditions that are easy to hold are satis"ed. Finally, we derive an algorithm which involves solving only a one-order-cycle dynamic program for the optimal policy parameters.  2000 Elsevier Science Ltd. All rights reserved.

Keywords: Inventory model; Periodic review; Dynamic programming; Optimization

1. Introduction

Alternative resupply modes are commonly used in practice. For example, a retailer could choose to replenish the inventory of an item by a fast resupply mode (e.g., by air) if its inventory position is

0305-0548/00/$ - see front matter  2000 Elsevier Science Ltd. All rights reserved. PII: S 0 3 0 5 - 0 5 4 8 ( 9 9 ) 0 0 0 9 2 - 1

dangerously low. In this article, we study an inventory system in which there are two resupply modes: namely a regular mode and an emergency mode. Orders placed via the emergency mode, compared to orders placed via the regular mode, have a shorter lead time but are subject to higher ordering costs.

Many studies in the literature address this problem. Some studies assume a particular policy form and devise methods for evaluating it, while other studies develop the true optimal policy and solve speci"c instances of the problem under consideration. While the policy-evaluation studies use broader assumptions and simpler policies, the policy-optimization studies typically have stronger results. This paper treats the problem in a periodic review setting and contributes to the optimiza-tion literature.

The earlier research in this area are policy-optimization studies. See Barankin [1], Daniel [2], Neuts [3], Bulinskaya [4], Fukuda [5], Veinott [6], and Wright [7]. They assume that the lead times for regular and emergency replenishment di!er by exactly one period (whose length is one or few working days). Whittmore and Saunders [8] extend the analysis by allowing the emergency and regular lead times to be of arbitrary lengths. Unfortunately, the form of the optimal policy they derive is extremely complex. They are able to obtain explicit results only for the case in which the two lead times di!er by one period. Later works in this area are policy-evaluation studies. See, e.g., Moinzadeh and Nahmias [9], Moinzadeh and Schmidt [10], and Moinzadeh and Aggarwal [11] for continuous review models.

All of the policy-optimization studies cited above assume that both regular and emergency orders can be placed at the start of each period. In this paper, we also assume that an emergency order can be placed at the start of a period if the inventory position of an item is dangerously low. However, we assume that regular orders are placed at the beginning of an order cycle which consists of a number of periods. Possible reasons for this include avoiding large "xed order costs and achieving economies in the coordination and consolidation of orders for di!erent items. The latter is particularly true if many items are purchased from the same source. For example, a retailer may order hundreds of items from a distribution center every two weeks (which then is the length of an order cycle). Also, companies in the import auto industry in Taiwan typically establish their weekly or biweekly ordering of auto parts from abroad. In addition, they place an emergency order if the inventory level of an item falls below a warning point at the start of a working day. The periodic inventory system considered in this article is similar to that depicted in Chiang and Gutierrez [12]. However, Chiang and Gutierrez assume that the regular and emergency supply lead times di!er by more than one period but less than the order-cycle length. In this article, we assume that the emergency supply mode has a lead time one period shorter than that of the regular mode. This assumption is used in most of the policy-optimization studies cited above, and may really be true in some real-world situations in which the supplier's warehouse is not too far away from the buyer and emergency orders delivered by a faster transportation mode will arrive earlier than regular orders by one or few working days.

We analyze the problem within the framework of a stochastic dynamic program. We assume that emergency orders have larger variable item costs. It is possible that emergency orders also have a "xed order cost. This paper, like Chiang and Gutierrez [12], treats only a special case, i.e., the cost of placing an emergency order is assumed to be negligible. We will develop optimal regular and emergency ordering policies that minimize the total expected discounted cost of procurement, holding, and shortage over a "nite planning horizon. These optimal policies are next shown to

converge as the planning horizon is extended, if some conditions that are easy to hold are satis"ed. Hence, the ordering rules to which these policies respectively converge are optimal for the in"nite horizon model.

The contribution of this paper is twofold. Firstly, we present a dynamic programming model for the "nite horizon problem in which there is only one state variable, as opposed to two state variables in the model of Chiang and Gutierrez [12]. We also derive the convergence conditions (for the optimal policies) which are much simpler than those in Chiang and Gutierrez [12]. Secondly, we develop a simple algorithm, which is not available in Chiang and Gutierrez [12], of computing the optimal policy parameters. Thus, the ordering policy developed is easy to implement.

2. A dynamic programming model

Assume that there are two resupply modes available: namely a regular mode and an emergency mode. The unit item costs for the regular and emergency supply modes are c and c, respectively, where c(c. Assume that an order cycle, whose length is exogenously determined (as in Chiang and Gutierrez [12]), consists of m periods. For notational simplicity, the lead times for the regular and emergency modes are assumed to be one and zero periods, respectively. It can be shown that this case can be generalized to situations where the two lead times di!er by exactly one period. See Chiang [13] for details. Assume that all demand which is not "lled immediately is backlogged. There is a holding cost h( ) ) based on inventory on hand and a shortage cost p( ) ) based on backlogged demand. Both the holding cost and shortage cost are charged at the end of each period. Let(t) be the probability density function for demand t during a period with mean k. Demand is assumed to be non-negative and independently distributed in disjoint time intervals.

Suppose the net inventory (i.e., inventory on hand minus backorder) at the beginning of a period is x; then the expected holding and shortage costs incurred in that period are given by

¸(x)"



h(x!t)(t) dt#



V>p(t!x)(t) dt, (1)

where ( ) )> denotes max+ ) , 0,. Other functional forms of ¸(x) are allowed; however, for our analysis we need ¸(x) to be a convex and di!erentiable function. Denote <GH(x) as the expected discounted cost with i order cycles and j periods remaining (where 0)j)m!1) until the end of the planning horizon when the starting net inventory is x and an optimal ordering policy is used at every review epoch. Then, <GH(x) satis"es the functional equation

<G(x)" min VWPW0+cr#c(R!r)#¸(r)#aER<G\K\(R!t),!cx, (2) <GH(x)"min VWP +cr#¸(r)#aER<GH\(r!t),!cx, j"m!1,2, 1, (3) where <(x),0, a (0(a(1) is the discount factor, r is the net inventory after a possible emergency order is placed at a review epoch, and R is the inventory position (i.e., net inventory plus inventory on order) after a possible emergency order and a regular order are placed at the

beginning of an order cycle (r and R are decision variables). As we see above, <G(x), for example, consists of the purchase cost c(r!x)#c(R!r), one-period expected holding and shortage cost ¸(r), and the expected discounted costaER<G\K\(R!t) from the next review epoch until the end of the planning horizon.

De"ne the function GGH(r) as

GGH(r)"cr#¸(r)#aER<GH\(r!t), j"m!1,2, 1. (4) Assume that G(r)"cr#¸(r) attains its minimum (this assumption is satis"ed if k(R, limV dh(x)/dx'0, i.e., there is a positive holding cost, and limV"dp(x)/dx"'c, i.e., the shortage cost per unit is greater than c. However, limV"dp(x)/dx"'c can be relaxed to limV"dp(x)/dx"'(1!a)c if <(x),!cx for x(0, i.e., any unsatis"ed demand at the end of the planning horizon has to be "lled at the unit cost of c). Also, de"ne H(r) as

H(r)"cr!cr#¸(r). (5)

Denote by Df the "rst derivative of the function f. Let rH_ be the value of r that minimizes H( ) ). If

rH_ is not an unique minimum, we choose the smallest such value (i.e., DH(r)(0 for r(rH_ and DH(rH_{)"0). Then <G(x) can be expressed as}

<G(x)" min VWPW0









The result below due to Karush [14] (see also Veinott [6]) is used to simplify (6). Let

aRb,max+a, b,.

Lemma 1 (Karush [14]). Let f (y) be a convex function which is minimized by yH. Then min

*WWW3

f (y)"f(¸)#f (;),

where f(¸)"f (¸RyH) is convex decreasing in L and f (;)"f (;)!f (;RyH) is convex non-increasing in U.

It follows by Lemma 1 that min

VWPW0 H(r)"H(x)#H3(R),

(7) where H(x)"H(xRrH

) is convex non-decreasing in x and H3(R)"H(R)!H(RRrH) is convex non-increasing in R. Note that (i) if R)rH_{, H(x)#H3(R)"H(R), (ii) if x*r}H

, H(x)#H3(R)"

H(x), and (iii) if x(rH

(

R, H(x)#H3(R)"H(rH

). Substituting (7) into (6) yields <G(x)"min

VW0+cR#aER<G\K\(R!t)#H3(R),!cx#H(x).

De"ne the function GG(R) as

GG(R)"cR#aER<G\K\(R!t)#H3(R). (9)

We next show in Lemma 2 that the cost function <GH(x) is convex. Lemma 2. <GH(x) for each (i, j) is a convex function.

Proof. <(x) is convex. Assume that <H\(x) is convex. It follows from (4) that GH(r) is a convex function. Thus, <H(x) is convex by Proposition B-4 of Heyman and Sobel [15]. This implies from (9) that G(R) is convex. Hence, <(x) is convex again by Proposition B-4 of [15]. Convexity is established by induction for the remaining <GH(x). 䊐

Let rGH be the (smallest) value minimizing GGH(r). Then it follows from (3) and (4) that the optimal policy at a review epoch with i order cycles and j ( jO0) periods remaining is (i) order up to rGH if

x(rGH and (ii) do not order if x*rGH. Let RG minimize GG(R) and rG"min(rH

, RG). Then it can be seen from (8) that the optimal policy at a review epoch with i order cycles remaining is (i) if

x(rG, order amounts rG!x and RG!rG at unit costs c and c, respectively, (ii)

if rG)x(RG, order an amount RG!x at unit cost c, and (iii) if x*RG, do not order. Notice that if rH_*

RG (thus rG"RG), then the regular supply mode is never used at that review epoch with i cycles remaining. We will elaborate on this in the next section.

3. Properties of the optimal policy

In this section, we present important properties about the regular order-up-to levels RG and emergency order-up-to levels rGH, j"0, 1,2, m!1._{We "rst introduce some preliminary observations that will be useful to establish the results of} this section. It follows from (3) and (4) that <GH(x), jO0, can be expressed as

<GH(x)"GGH(xRrGH)!cx, j"m!1,2, 1. (10) Similarly, from (8) and (9), <G(x) can be rewritten as

<G(x)"GG(xRRG)!cx#H(x). (11)

Thus, D<GH(x)"!c for x)rGH, j"m!1,2, 1, and D<G(x)"!c#DH(x) for x)RG (which implies, due to rG"min(rH

, RG), that D<G(x)"!c for x)rG). Also, D<G(x)* !

c#DH(x)*!c and D<GH(x)*!c, j"m!1,2, 1, for all x.

We show in the following theorem that emergency order-up-to levels are non-decreasing within an order cycle as the planning horizon is extended.

Theorem 1. rGK\*rGK\*2*rG.

Proof. We show that rG*rG. The remaining inequalities can be established similarly. To show

aERD<G(r!t)(0 for r(rG. Also, DGG(r)"c#D¸(r)#aERD<G(r!t)"c#D¸(r)!ac for r(rG. Hence, DGG(r)(!a[c#ERD<G(r!t)])0, for r(rG. 䊐

We next show that if c(ac, the regular order-up-to level at the beginning of an order cycle is greater than or equal to the emergency order-up-to level at the next review epoch.

Lemma 3. If c(ac, then RG>*rGK\ for all i.

Proof. We show that if c(ac, DGG>(R)(0 for R(rGK\, implying that RG>*rGK\. It follows from (9) that for _{R(rGK\, DGG>(R)"DH3(R)#c#aERD<GK\(R!t)"} DH3(R)#c!ac. As DH3()))0, if c(ac, DGG>(R)(0 for R(rGK\. 䊐

The condition c(ac is usually true; otherwise, the regular supply mode never will be used. This is shown in the following theorem.

Theorem 2. If c*ac, then rH *

RG for all i.

Proof. We show that if c*ac, DGG(R)*0 for R*rH

and thus rH*

RG. It follows from (9) that

DGG(R)"DH3(R)#c#aERD<G\K\(R!t). As DH3(R)"0 for R*rH

, DGG(R)"

c#aERD<G\K\(R!t)*c!ac*0 for R*rH . 䊐

The most important result of this section is stated in Theorem 3, which shows that if two consecutive regular order-up-to levels are equal to each other, and greater than or equal to all intermediate emergency order-up-to levels, then both the regular and emergency order-up-to levels have converged.

Theorem 3. If RG>"RG*rGK\, then rL>"rG, rL>"rG,2, rL>K\"rGK\, and

RL>"RG>, for n*i.

Proof. We show that if RG>"RG*rGK\, then rG>"rG, rG>"rG,2, rG>K\"rGK\, and _{RG>"RG>, and thus similarly, the equalities hold for all n*i#1. If}

RG>"RG*rGK\, RG>"RG*rG by Theorem 1. Notice that D<G>(x)"D<G(x) for x)RG>"RG. Also, DGG(r)"c#D¸(r)#aERD<G(r!t) and DGG>(r)"c#D¸(r)#

aERD<G>(r!t). Hence, DGG(r)"DGG>(r) for r)RG>"RG. As rG minimizes GG(r), it follows that it also minimizes GG>(r), i.e., rG>"rG. In addition, for x'rG>"rG, D<G>(x)"D¸(x)#aERD<G>(x!t) and D<G(x)"D¸(x)#aERD<G(x!t). Thus D<G>(x)"D<G(x) for x3(rG,RG]. Also, D<G>(x)"D<G(x)"!c for x)rG. As a result, D<G>(x)"D<G(x) for x)RG. By the same logic, we can show that rG>H"rGH for j"2,2, m!1 and _{D<G>H(x)"D<GH(x), j"2,2, m!1, for x)RG"RG>.} Moreover, DGG>(R)"DH3(R)#c#aERD<GK\(R!t) and DGG>(R)"DH3(R)#c# aERD<G>K\(R!t). As D<G>K\(x)"D<GK\(x) for x)RG"RG>, DGG>(R)" DGG>(R) for R)RG>. As RG> minimizes GG>(R), it follows that it also minimizes GG>(R), i.e., RG>"RG>. 䊐

We see from Theorem 3 that if for some i, RG>"RG*rGK\, then the sequences +RL, and +rLH,, j"1, 2,2,m!1, converge, respectively, to RH"RG> and rHH"rGH, j"1,2,m!1. The condition of RG>"RG*rGK\ for some i usually holds if c(ac, as implied by Lemma 3. Consequently, RH, rH_H for j"1,2, m!1, and min+rH,RH, are, respectively, the optimal regular and emergency order-up-to levels for the in"nite horizon model.

To illustrate, consider the example (referred to as the base case thereafter):

m"10, c"$10,c"$15, k"2 (with Poisson demand), h(x)"0.01x and p(x)"20x (both for x*0) (this choice of the holding and shortage cost functions implies that holding and shortage are

charged at $0.01 and $20 per unit, respectively), and a"0.999. After solving, we "nd that

rH_"

3, r"1, r"3, r"4, r"5, r"6, r"2"r"7, R"19; r"4, r "

r"6, r"2"r"7, R"32; and rG"4, rG"rG"6, rG"2"rG"7 for i*2, and RG"32 for i*3. Thus, we see that the sequences +RL,, +rL,, +rL,, +rL,, and

+rLH,, j"4, 5,2, 9, converge, respectively, to RH"32,rH"4,rH"rH"6, and

rH_"2"rH

"7 after just two order cycles. 4. Discussion

We brie#y compare the results derived in Section 3 to those in Chiang and Gutierrez [12]. As mentioned in Section 1, both this paper and Chiang and Gutierrez [12] study the same periodic inventory system, except that this paper assumes that the regular and emergency lead times di!er by one period, while Chiang and Gutierrez [12] assumes that the two lead times di!er by more than one period (but less than the order-cycle length).

The non-decreasingness property of emergency order-up-to levels within an order cycle in Theorem 1 and the convergence conditions for the optimal policy parameters in Theorem 3 are similar to those in Chiang and Gutierrez [12]. However, the convergence conditions in Theorem 3 are simpler and easier to hold than their counterpart in Chiang and Gutierrez [12].

In addition, the emergency order-up-to levels derived in Chiang and Gutierrez [12] are a function of inventory on order (if there is any at a review epoch), while they are not in this paper. Moreover, the optimal regular order-up-to level RH derived in this paper is a "xed level, while it is a variable level in Chiang and Gutierrez [12] (depending on the inventory position after a possible emergency order is placed at the beginning of an order cycle). As a result, the optimal policy developed in this paper consists of only a few parameters (i.e., rH_H, j"0,2, m!1, and RH), while it contains a lot many in Chiang and Gutierrez [12].

Take the base case in Section 3 for example: m"10, c"$10,c"$15, k"2 (Poisson demand), h(x)"0.01x and p(x)"20x (for x*0), and a"0.999. If the emergency and regular supply lead times are one and two periods, respectively (note in this case that expressions (2) and (3) would have ¸(r) replaced by aER¸(r!t) [13]), we obtain after solving that rH"5,rH"7,

rH_"8, rH

"9, rH

"10, rH

"2"rH

"11, and RH"35. On the other hand, if the two lead times are one and six periods, respectively, it is found in Chiang and Gutierrez [12] that

rH_"rH

"2"rH

"11 (rH

H for j"5,2, 9 are a function of inventory on order), and the optimal

regular order-up-to level is between 41 and 45. As we see, the emergency order-up-to levels in the former situation are smaller than their counterpart levels in the latter. This is because the more we are close to the time of the arrival of a regular order, the less inventory we need to carry on hand

(against possible stockouts) and thus the smaller the emergency order-up-to levels. For example, if we are at the beginning of an order cycle, there are only two periods till the time of the arrival of a regular order in the former situation, while there are six periods in the latter; hence, rH_"5 in the former while rH_"11 in the latter. Also, the optimal regular order-up-to level is lower when the di!erence between the two lead times is only one period. The reason for this is simple. As we know, the regular order-up-to level should be large enough to cover demand over an order cycle plus the regular lead time since it will take such time for the next regular order to arrive. Hence, a larger regular lead time (other things being equal) will yield a higher order-up-to level.

5. A simple method

In this section, we develop a simple method for computing RH and rH_H, j"1,2, m!1. While solving <GH(x) until RG>"RG*rGK\ is satis"ed may take little time, this simple method obtains

RH and rH_{H '}s by solving only a m-stage dynamic program.

Assume that there exists some i such that RG>"RG*rGK\, and thus RH"RG> and

rH_H"

rGH, j"1,2, m!1. Let QG(x)"GG(xRRG)"GG(xRRH), which is convex

non-decreas-ing and equal to GG(RH) for x)RH. Thus, it follows from (11) that <G(x) can be expressed as

<G(x)"QG(x)!cx#H(x). (12) Then, <G(x)"min VWP +cr#¸(r)#aER<G(r!t),!cx "min VWP +cr#¸(r)#aERQG(r!t)#aERH(r!t)!acr#ack,!cx. Let J(r)"(1!a)cr#¸(r)#aERH(r!t). (13)

Thus GG(r)"J(r)#aERQG(r!t)#ack. For r)RH, DGG(r)"DJ(r), since DQG(x)"0 for r)RH. As rH_"

rG)rGK\)RG"RH, rH

 can be obtained by solving DJ(r)"0. Let

H(x)"J(xRrH

) and QG(x)"aERQG((xRrH

)!t), which are both convex and non-decreasing. Then,

<G(x)"GG(xRrH

)!cx"QG(x)#ack!cx#H(x). Note that for x)RH, QG(x)"aERQG((xRrH

)!t)"aERGG(((xRrH)!t)RRH)"aGG(RH). Similarly, we can repeat the above logic and show that if we let

JH(r)"(1!a)cr#¸(r)#aERHH\(r!t), j"2,2, m!1 (14) which is minimized by rH_H" rGH and de"ne HH(x)"JH(xRrH H), (15) QGH(x)"aERQGH\((xRrH H)!t) (16)

which are both convex and non-decreasing, then <GH(x)"QGH(x)# H

IaIck!cx#HH(x), j"1,2,m!1,

(17) where QGH(x)"aHGG(RH) for x)RH.

We "nally show in this section that RH can be obtained by solving DJ(R)"0, where J(R) is de"ned as

J(R)"(c!ac)R#H3(R)#aERHK\(R!t). (18)

It follows from (16) and (17) that <G>(x) can be written as <G>(x)"min VW0+H3(R)#cR#aER<GK\(R!t),!cx#H(x) "min VW0





where GG>(R)"J(R)#aERQGK\(R!t)# KIaIck is minimized by the value RG>"RH, i.e., DGG>(RH)"DJ(RH)#aERDQGK\(RH!t)"0. However, aERDQGK\(RH!t)"0 since

QGK\(x)"aK\GG(RH) for x)RH. Thus, it follows that DJ(RH)"0.

6. Algorithm

In this section, we present an algorithm which summarizes the simple method described in Section 5 for computing the optimal policy parameters rH_H, j"0,2, m!1, and RH. We also investigate the sensitivity of several important parameters towards the optimal solution.

Note that all the functions involved are convex when computing rH_H, j"0,2, m!1, and RH in Section 5. Hence, the following algorithm can be easily implemented.

Step 1: Compute rH_{ by using (5), and let H(x)"H(xRr}H ).

Step 2: Compute rH_{ by di!erentiating J(r) in (13) and solving DJ(r)"0, and let} H(x)"J(xRrH

). Similarly, compute rHH for j"2,2, m!1 by alternatively using JH(r) and HH(x), j"2,2, m!1, which are de"ned in (14) and (15), respectively.

Step 3: Finally, compute RH by di!erentiating J(R) in (18) and solving DJ(R)"0.

This algorithm involves solving only a one-order-cycle dynamic program with m stages.

Consider the base case in Section 3: m"10, c"$10, c"$15, k"2 (with Poisson demand),

h(x)"0.01x and p(x)"20x (for x*0) (holding and shortage are charged at $.01 and $20 per unit,

respectively), anda"0.999. Using the above algorithm, we obtain the same emergency and regular order-up-to levels rH_{H '}s and RH (i.e., rH_"3, rH

"4, rH "rH

"6, rH

"2"rH

"7, and RH"32). We next investigate the sensitivity of several important parameters towards the optimal solution. First, as the emergency unit cost c increases (other things being equal), the emergency order-up-to levels rH_{H '}s are likely to decrease and the regular order-up-to level RH is likely to increase, re#ecting the fact that we employ the emergency supply mode less often and tend to order more quantity via the regular mode. For example, compared to the base case, if c"$12.5, then rH

"4, rH "5,

rH_"6, rH

"2"rH "

7, and RH"31; and if c"$20.0, then rH

"2, rH "4, rH "5, rH "rH "6, rH_"2"rH "7, and RH"33.

Second, as the shortage cost per unit increases (other things being equal), rH_{H '}s are likely to increase (and thus RH may also increase), implying that we need to provide more inventory against possible stockouts. For example, compared to the base case, if p(x)"10x, then

rH_"2, rH "4, rH "5, rH "rH "6, rH "2"rH

"7, and RH"32; and if p(x)"40x, then

rH_"4, rH "5, rH "6, rH "rH "7, rH "2"rH "8, and RH"33.

Finally, as the holding cost per unit increases (other things being equal), RH is likely to decrease (rH_{H '}s may also decrease), since it is economical to carry less inventory on hand. For example, compared to the base case, if h(x)"0.005x, then rH_"3, rH

"4, rH "rH "6, rH "rH "7, rH_"2"rH

"8, and RH"33; and if h(x)"0.02x, then rH

"3, rH "4, rH "5, rH "6, rH "2"rH "7, and RH"31. 7. Conclusion

In this paper, we develop a dynamic programming model for an inventory system in which while emergency orders can be placed at the beginning of each period, regular orders are placed at the beginning of order cycles. Such inventory systems are found in the import auto industry in Taiwan. We assume that the regular and emergency channel lead times di!er by one period. We develop optimal ordering policies as well as a simple algorithm for computing the optimal policy para-meters. We hope that these results will help material managers decide how to use the emergency supply mode when this alternative option is available.

References

[1] Barankin RW. A delivery-lag inventory model with an emergency provision. Naval Research Logistics Quarterly 1961;8:285}311.

[2] Daniel KH. A delivery-lag inventory model with emergency. In: Scarf HE, Gilford DM, Shelly MW, editors. Multistage inventory models and techniques. Stanford University Press, Stanford, Ca., 1962 [Chapter 2]. [3] Neuts MF. An inventory model with optional time lag. SIAM Journal of Applied Mathematics 1964;12:179}85. [4] Bulinskaya EV. Some results concerning optimum inventory policies. Theory of Probability Applications

1964;9:389}403.

[5] Fukuda Y. Optimal policies for the inventory problem with negotiable leadtime. Management Science 1964;10:690}708.

[6] Veinott Jr. AF. The status of mathematical inventory theory. Management Science 1966;12:745}77.

[7] Wright GP. Optimal policies for a multi-product inventory system with negotiable lead times. Naval Research Logistics Quarterly 1968;15:375}401.

[8] Whittmore AS, Saunders S. Optimal inventory under stochastic demand with two supply options. SIAM Journal of Applied Mathematics 1977;32:293}305.

[9] Moinzadeh K, Nahmias S. A continuous review model for an inventory system with two supply modes. Manage-ment Science 1988;34:761}73.

[10] Moinzadeh K, Schmidt CP. An (S-1, S) inventory system with emergency orders. Operations Research 1991;39:308}21.

[11] Moinzadeh K, Aggarwal PK. An information based multiechelon inventory system with emergency orders. Operations Research 1997;45:694}701.

[12] Chiang C, Gutierrez GJ. Optimal control policies for a periodic review inventory system with emergency orders. Naval Research Logistics 1998;45:187}204.

[13] Chiang C. Inventory management with two supply modes. Doctoral dissertation, The University of Texas at Austin, 1991.

[14] Karush W. A theorem in convex programming. Naval Research Logistics Quarterly 1959;6:245}60. [15] Heyman DP, Sobel MJ. Stochastic models in operations research. New York: McGraw-Hill, 1984.

Chi Chiang is an Associate Professor of Management Science at National Chiao Tung University in Taiwan. He received his M.A. in Operations Management from the Ohio State University and Ph.D. in Management from the University of Texas at Austin. His interests are in the areas of materials management, JIT planning, and facility layout. He has research papers published in journals such as Naval Research Logistics, European Journal of Operational research, and Journal of the Operational Research Society.