A Single Machine Model for Determination of Optimal Due Dates and Sequence

Author(s): Suresh Chand and Dilip Chhajed

Source: Operations Research, Vol. 40, No. 3 (May - Jun., 1992), pp. 596-602
Published by: INFORMS
Stable URL: http://www.jstor.org/stable/171420
Accessed: 13-11-2015 23:08 UTC

REFERENCES
Linked references are available on JSTOR for this article:
http://www.jstor.org/stable/171420?seq=1&cid=pdf-reference#references_tab_contents

You may need to log in to JSTOR to access the linked references.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.

INFORMS is collaborating with JSTOR to digitize, preserve and extend access to Operations Research.

http://www.jstor.org

This content downloaded from 128.184.220.23 on Fri, 13 Nov 2015 23:08:02 UTC
All use subject to JSTOR Terms and Conditions
 OF
A SINGLEMACHINEMODELFOR DETERMINATION
OPTIMALDUEDATESAND SEQUENCE
SURESH CHAND
Purdue University, West Lafayette, Indiana

DILIPCHHAJED
University of Illinois, Champaign, Illinois
(Received March 1989; revisions received July 1989, June, December 1990, April 1991; accepted April 1991)

The problem of simultaneous determination of optimal due dates and optimal sequence for N-job single machine
problem with multiple due dates is considered in this paper. The penalty for a job is assumed to be a linear function of
the due date and the earliness/tardiness for the job. The objective is to minimize the total penalty for all jobs. An efficient
optimal algorithm to solve the problem is developed and several results are provided.

T his paperconsidersthe problemof the simulta-
neous determination of optimal due dates and
optimal sequence for N independent jobs to be pro-
cessed on a single machine. All jobs are available at
time zero. Processing times are known and determin-
istic. The penalty for a job in a sequence is assumed
to be a linear function of the due date assigned to the
job and the earliness/tardiness for the job in the
sequence. The same linear penalty function is used
for all jobs. The objective is to minimize the total
penalty for all jobs. The number of distinct due dates,
m, to be assigned to the jobs is assumed to be prespe-
cified and known. We allow m to take any value in
the set 1, 2, .. ., N}. Job preemption is not allowed.
The problem of the simultaneous determination of
optimal due dates and optimal sequence has been
motivated and considered by several authors in the
recent literature (Bagchi 1989, Baker and Bertrand
1981, Panwalkar, Smith and Seidmann 1982, Eilon
and Chowdhary 1976, Weeks 1979, Weeks and Fryer
1977); the reader is directed to Baker and Scudder
(1990), and Bector, Gupta and Gupta (1988) for a
brief review of this literature.The problem considered
in this paperis most directly relatedwith the problems
considered in Panwalkar, Smith and Seidmann, and
Seidmann, Panwalkar and Smith (1981). Panwalkar,
Smith and Seidmann assume a common due date for
all jobs; that is, m = 1. The penalty function used in
this paper is the same as in Panwalkar, Smith and
Seidmann. Several results from that paper are used
here. As in Panwalkar,Smith and Seidmann, we show
that the penalty function can be expressedas a product
Subject classification: Production/scheduling: single machine scheduling, deterministic.
Area of review: MANUFACTURING, PRODUCTION AND SCHEDULING.
of two sequences, which is minimized by arranging
them in opposite order. Seidmann, Panwalkar and
Smith consider the problem when each job is allowed
a different due date. They show that an SPT (shortest
processingtime) sequence is optimal for this problem.
With the number of distinct due dates, m, permitted
to be any number in the set 1, 2, . . ., N}, our paper,
in essence, bridges the gap between the sched-
uling models with a common due date and the
scheduling models with possibly N distinct due dates
for the N-job problem.
Our paper considers two different cases. In the first
case, we assume that the decision maker specifies the
number of jobs to be assigned to different due dates.
In the second case, we assume that the number ofjobs
to be assignedto differentdue dates is unknown. Note
that the second case turned out to be very difficult to
analyze.
Section 1 of the paper defines the notation and
formulates the problem. Section 2 of the paper con-
siders the case when the vector (ni, n2, . .. ., nf), where
nj is the number of jobs assigned to the jth due date,
is externally specified. Note that while nj is assumed
to be known, the specificjobs to be assigned to the Jth
due date are unknown and to be determined. We
show that the objective function can be expressed as
a sum of the pairwise product of two equal sets of
numbers, where one set consists of processing times
and the other set consists of positional weights. The
positional weights are independent of job characteris-
tics. An efficient procedure is presented to find opti-
mal due dates and an optimal sequence. The results

OperationsResearch 0030-364X/92/4003-0596 $01.25

Vol. 40, No. 3, May-June1992 596 ?) 1992OperationsResearchSocietyof America

This content downloaded from 128.184.220.23 on Fri, 13 Nov 2015 23:08:02 UTC
All use subject to JSTOR Terms and Conditions

in this section are used in the next section in devel-
oping an optimal procedure when the vector (n1, n2,
... . nn) is also a decision variable. Section 3 of the
paper develops some dominance results to find an
optimal vector (n1, n2, no)
..., when it is not exter-
nally specified. These dominance results are used in
developing an optimal procedure to find (no, n2,
n,").Even though these results are intuitive, proofs for
these turned out to be very difficult. The paper closes
in Section 4 by providing some concluding remarks
and ideas for further research.
1. PROBLEMFORMULATION
Throughout the paper, we assume that the number of
jobs to be scheduled and the number of due dates to
be assigned to the jobs are known. The following
notation is used to formulate the problems:
N = the number of jobs to be scheduled;
m = the number of due dates;
(n, n + k) = {n, n + 1, n + 2, ..., n + k};
di = the due date for job i, i E ( 1,N);
Ci = the completion time for job i,
i E ( 1, N);
Ei = max(O, di - Ci) = earliness for job i,
i E (1, A);
TF= max(O, Ci - di) = tardiness for job i,
i E (1, N);
t4= the processing time for job i,
iE (lN);
P1 = per unit time due date penalty for each
job;
P2 = per unit time earliness penalty for each
job; and
P3 = per unit time tardinesspenalty for each
job.
In addition, we let Di < D2<1 .. D denote the m
due dates and let Ij denote the set of jobs assigned to
due date Dj for j E ( 1, m). As in Panwalkar, Smith,
and Seidmann, the constants PF, P2 and P3 are as-
sumed to be nonnegative and known.
The objective is to determine D = ID,, D2, .
Dm1, I= {II, I2, ... I Im, and a schedule u to minimize
the total penalty TP(D, I, a), where
m Property 1. For any given D and a, there is an optimal
TP(D, I, a) =X X , IPI *Dj + P2*E, + P3.T1}
j= I ieIj
and d, = Dj for i E I,.
It is easy to see that the problem has an optimal
solution with zero machine idle time. (See Lemma 3
This content downloaded from 128.184.220.23 on Fri, 13 Nov 2015 23:08:02 UTC
All use subject to JSTOR Terms and Conditions
A SingleMachineModel / 597
in the Appendix.) Therefore,it is sufficient to consider
permutation schedules to find optimal solutions. Let
[p1 denote the job in position p in a permutation
schedule, then
p
C[p]= E t[k]. (2)
. . .
k= I
We will use a to denote a permutation schedule in the
rest of the paper. Also, jobs are assumed to be indexed
such that
0 <' toI _ t2 s . .. <1 tN (3)
The next section considers the problem when
I, I = n, is known for all j E ( 1, m). Note that while
Ij I is assumed to be known, the specific jobs in Ij are
unknown and to be determined.
2. RESULTSAND PROCEDUREWITHIijI = n
KNOWN
This section assumes that the vector (n, n2, ...,*nm)
of m positive integers such that Zj':I nj = N and
IIj I = nj for j = 1, 2, . . ., m is externally specified.
D, I (with IIj nj known) and a are decision vari-
ables. The cases when m = 1 and m = Nare considered
in Panwalkar, Smith and Seidmann, and Seidmann,
Panwalkarand Smith, respectively.
For PI > P3, an SPT sequence is optimal with di =
o for i E (1, N) (see Theorems 1 and 2 in Seidmann,
Panwalkar and Smith). We assume that PI < P3 in
developing the propertiesand the algorithm below.
Properties of Optimal Solutions
We now state several properties of the optimal solu-
tions which will be used in developing an algorithm.
Lemmas 4 and 5 needed to develop these properties
are stated in the Appendix. We define Nj = ID-r, nk for
j E ( 1, m) with No = 0; Nj is the total number of jobs
assigned to the firstj due dates.
From Lemma 4, it is easy to see that for any given
D and a, there is an optimal I such that a batch of nj
consecutive jobs in a is assigned to D>.The following
propertygives an optimal I for any given D and a.
(1) I such that
Ii = {aNj_1+1, aNj_1+2, .,
aNj_1+nfj
for j E ( 1, m), where a, is the job in position n in
sequence a.
598 / CHAND AND CHHAJED

This property essentially says that there is an optimal
solution such that nj consecutive jobs (in positions
Nj-I + 1 to NjfI + nj) in a are assigned to D>.
Lemma 5 shows that for any given a, there is an
optimal D such that Dj coincides with the completion
time of a job in a. The following property,which gives
an optimal D for a given a, follows from Panwalkar,
Smith and Seidmann.
Property 2. For a given a, there is an optimal D such
that Dj = C[kj], where
/P3-Pi
k~~=N>
P3+ P2
(x)+ here stands for the smallest integer 3 x.
Property 2 implies that the first (kj - Nj-t) jobs in Ij
are early and the remaining (Nj - kj) jobs are tardy.
We now develop an algorithm to find (D, I, a) to
minimize (1).
Algorithm
Using Properties 1 and 2, we first reduce (1) to the
formEp'=I wpt[p].Knowingthe positionalweightswp's
Note that (6) is a product of two sequences,
which is minimized by arranging them in opposite
order. The following steps give an optimal solution
for a given (n1, n2, ..., nm). We call this algo-
rithm the Given-(n1, n2, ..., nm)-Algorithmor the
Gn-Algorithm.
Steps of the Gn-Algorithm
Step0. Set No = 0 andNj = Ek= nk forj E (1, m).
Step 1. For j E (1, m), compute
+
+P3 P2 )
Step 2. For p E ( 1, N)', compute w, using (7).
Step 3. Rank the positional weights wp'sin a de-
scending order of magnitude such that the largest wp
is rankedfirstand the smallest wpis rankedNth. Break
the ties arbitrarily.
Step 4. Find a sequence a by assigning job i to
position p, where p is such that wphas rank i. Set
D>= Zkp=l t[p],and
d= Djfor i E Ij=I{UN%-+i, 7Nj% +2, .*,oNjl

for positions p E (1, N), an optimal sequence can be forj E (1, m).
found as in Panwalkar,Smith and Seidmann.
Using Property 1, we get: Stop, the solution found in Step 4 is an optimal
solution.
TP(D, a) = Min TP(D, a, I)
I It is easy to see that the first (kj - NjA%)
jobs in Ij
m N are in an LPT order, while the last (Nj - kj) jobs
E E (PI *Dj + P2*E[p]+ P3*T[p]). are in an SPT order. The Gn-Algorithm requires
j=I p=Nj- +l
O(N log N) computations. We now give a numerical
Furthermore,using Property 2, we can write: example to illustrate the procedure.

TP(o) = Min TP(D, a) Numerical Example. Given ten jobs with (t1, t2, t3,
D
. Ito)=(3, 10, 11, 13, 13, 16, 17, 20, 22, 25), PI =
in kj
2, P2 = 11 and P3 = 18. We are given m = 2, ni = 4
= nj[J.Pi*C[k + E P2-(C[kj- C[p]) and n2 = 6.
j=, p=Nji + I
From Steps 0 and 1 of the algorithm, we get N1 =
Nj_
4, N2 = 10, k, = 3 and k2 = 8. The ten positional
+ E P3 (C[p] - C[ki])1 (5)
weights and their ranks are given in Table I.
p=kj+ 1
The optimal sequence is 8-5-2-6-10-7-4-1-3-9; DI =
with kj defined by (4). Using C[k] - k= I t[p1for a given t8 + t5 + t2 = 43, and D2 = t8 + t5 + t2 + t6 +
a, (5) can be simplified to: t0o + t7 + t4 + tI = 117; I, = 18, 5, 2, 61 and I2 =
N {10, 7, 4, 1, 3, 91;penalty = 3,763.
TP(Ci)= E Wpt[pj, (6) The next section considers the problem when the
p=I
vector (nI, n2, . . ., n,,) is unknown.
where
P2.(P- 1 -ANj) + PI(N-ANj1) 3. RESULTSAND PROCEDUREWITHIIj = n
UNKNOWN
for p E (NI-I + 1, kj
Wp= I This section considers the problem when the vector
P3 .(NJ -p + 1) + PI-(N -Nj)
(n1, n2, . . ., nm) is unknown. D, I, and a are decision
l n E (k; + 1, Ni)
~~~~~for (7) variables in this case. Dominance properties are

This content downloaded from 128.184.220.23 on Fri, 13 Nov 2015 23:08:02 UTC
All use subject to JSTOR Terms and Conditions
 A SingleMachineModel / 599
Table I
PositionalWeightsand Ranks
Position p 1 2 3 4 5 6 7 8 9 10
Weight wp 20 31 42 30 12 23 34 45 36 18
Rank 8 5 2 6 10 7 4 1 3 9

developed to find an optimal n,, n2, nm. A key

...), Proof. From (8) we have
result in the section shows that there is an optimal
COST(l) - COST(n) = t(VWT1(l) - WTI(n))
solution with n1 3 n2 3 ... 3 n,. This decreasing
N
property is used to develop an efficient search algo-
rithm to find an optimal n1, n2, ..., n,. Computa- + A \ ii(WT(l) - WT(n)). (10)
i=2
tional requirements for this search algorithm are
discussed below. If (9) holds for given n and 1, then COST(l) -
Recall that w, denotes the positional weight associ- COST(n) > 0 from (10) for any given (t, t2, * * *, W,
ated with position p for a known vector (n,, n2, . . . satisfying (3). If (9) does not hold for all i E ( 1, N),
nin). Step 3 of the Gn-Algorithm gives ranks for these then there is a k E ( 1, N) such that WTk(n)> WTk(l)
weights. We need the following additional notation or WTk(l) - WTk(n)< 0. Then, we can select a vector
associated with these weights: (tI, t2, . . . , tN) satisfying (3) with Ak = tk - tk-1 high
enough to make COST(l) - COST(n) < 0 in (10).
ai = the positional weight with the rank of i; This completes the proof.
N
WT = E ak Lemma 2. For m = 2 and n E (0, (N/2)+), we have
k=i
WT,(n) > WT((N/2)+)for all i E (1, N). (Recall that
Clearly, ae 3c a2 ... a N Note that the weights ai's
Ofv. (x)+ is the smallest integer 3 x.)
and WT1's,while they depend on the vector (n,, n2,
... , nn) and the penalties P1, P2 and P3, do not depend Proof. A proof for this is given in the Appendix.
on the processing times for the jobs.
To develop the results in this section, we first as- Theorem 1. There is an optimal solution (D*, I*, a*)
sume that m = 2. We let WTi(n) denote the weight such that n* > n* ... nM.
WT,when n1 = n and n2 = N - n; aij(n)is similarly
defined. Let COST(n) denote the corresponding op- Proof. First consider the case m = 2. If n1 < (N/2)+,
timum objective function value, then we have then WTi(n,) > WTi((N/2)+) for all i E (1, N) from
Lemma 2, and COST(n,) > COST((N/2)+) from
COST(n) Lemma 1. Thus, there is an optimal solution with
= tial(n) +t2 a2(n) +..+tN.aN(n) ni' > (N/2)+ > (n* + n*)/2, or n* > n*.
The proof for m = 2 can be extended for the
= t* ( WTI(n) - WT2(n)) + t2( WT2(n) m > 2 case by considering two adjacent batches.
- WT3(n)) + . . . + tN (WTN(n)) Thus, there is an optimal solution with n0 > nj*>lfor
j-1, 2, . .., m - 1, or n* 3, n* 3 n*.
n
= t* WTI(n) + (t2- t)) JWT2(n)
The following search algorithm gives an optimal
+ (t3 - t2) 'WT3(n)+ . + (tN-tNI). WTN(n) (n*, n*, . . ., n,*).We call this algorithm the Efficient
N Search Algorithm (or the ES Algorithm).
= t. WT,(n)+ E Ai. WTi(n), (8)
i=2
Efficient Search Algorithm
where Ai = t - t- I. For nI = (N/m)+, (N/m)+ + 1, . . ., N - (m - 1).
Lemma 1. For m = 2, n E (1, N - 1) and I E For n2 = (N n,/m - 1)+,(N - n,/m - 1)+ +
-

(1, N - 1), we have COST(n) < COST(I) for every I, .. , N- (m - 2).

vector(tI t2, . . ., tN) satisfying (3) if and only if
For nk = (N - Zj=-J' nj/m - (k - 1))+, (N -
WT (n) < WT (l) for all i E ( 1, N). (9) ,yt--A nj /m - (k -1))+ + 1, . . ., N - (m - k).

This content downloaded from 128.184.220.23 on Fri, 13 Nov 2015 23:08:02 UTC
All use subject to JSTOR Terms and Conditions
600 / CHAND AND CHHAJED

Table II Theorem 2. For a positive integern 3 2 and r E

Costs Using the Gn-Algorithm for m = 2 (1, n - ,we have
n, 5 6 7 8 9 Pl = n = 1, and
n2 5 4 3 2 1
pn pn-r + pn-r +
Cost 3,624 3,586 3,729 4,121 4,738 = + ... Pmi(r~n-r). (1 1)

See Chand and Chhajed (1990) for a proof of this

For n,,, = N - "Y_,'nj.
Find the minimum total penalty for (n1, n2,
n,1) using the Gn-Algorithm.
Find (n*, n*, . . ., n*) corresponding to the lowest
total penalty.
The following example illustratesthe procedure.
Numerical Example. Consider the 10-job problem
with processing time and penalty data used in the
numerical example in Section 3. We solve it for
m = 2 and m = 3.
Solution for m = 2. To find an optimal (nc, n*),
we need to find the minimum cost using the
Gn-Algorithm for n, = 5, 6, 7, 8 and 9. Note that
n2 = 10 n,. Table II gives the costs using the
- itive, the proof turned out to be difficult. An algorithm
Gn-Algorithm. We get n* = 6 and n* = 4 corre-
sponding to the minimum cost of $3,586.
Solution for m = 3. The costs computed using the
Gn-Algorithmfor variousvaluesof n,, n2 and n3 in
the ES Algorithm are given in Table III. We get
W'= 4, nt = 3 and n* = 3.
We now discuss the computational requirements
for the ES Algorithm. Note that the ES Algorithm
generates vectors (n,, n2, nfm) by partitioning N
theorem.
4. CONCLUDING
REMARKS
This paper makes two major contributions. First, an
optimal algorithm of the order O(N log N) was devel-
oped to determine optimal due dates and a sequence
for an N-job problem when the number of jobs to be
assigned to different due dates are known. We show
that the objective function can be expressedas a sum
of the pairwise product of two equal sets of numbers,
where one set consists of processing times and the
other set consists of positional weights.
Second, an important dominance property was de-
veloped to determine the number of jobs to be as-
signed to different due dates. While the result is intu-
of the order 0(pN Nf log N) was developed to find
optimal due dates, sequence and number of jobs to be
assigned to different due dates. Theorem 2 in Section
4 gives a computational procedure to
find Pi.
APPENDIX
This appendix states and proves several important
results.
...,

into m integer numbers, such that Eye=,nj = N, and
n,3 n2 3 ... 3 n,n 3 1. Let PZ denote the total
number of these vectors, then the ES Algorithm re-
quires PZ applications of the Gn-Algorithm to find
(n*, n*, n*). The following theorem gives a re-
...,
cursive procedure to compute PM.
Table III
Costs Using the Gn-Algorithmfor m = 3
nf n2
4 3
4 4
5 3
5 4
6 2
6 3
7 2
8 1
n3 Cost
3 2,757
2 2,852
2 2,845
1 3,023
2 3,021
1 3,068
1 3,367
1 3,829
Lemma3. The TP(D, I, a) problemhas an optimal
schedulewithzeromachineidle time.
See Chand and Chhajed for a proof.
Lemma4. Let D, a and I be known.Let I be such
that there are three successivejobs, a, b, and c
(C, < Cf < Ce.)in a witha, c E Ij and b 4 Ip. Thenit
is possibletofind an alternateI (withoutchangingD
and a) withthe same or a lowercost such that either
a, b e Ijand c M Iorb, c e Iand a Ij.
See Chand and Chhajed for a proof.
Lemma5. Let D, I, and a be known.Let D be such
that there is a Dj that does not coincide with the
completiontime of anyjob in Ij. Thenit is possibleto
find an alternateDi thatcoincideswiththe completion
time of a job in Ij withoutincreasingthe totalcost.

This content downloaded from 128.184.220.23 on Fri, 13 Nov 2015 23:08:02 UTC
All use subject to JSTOR Terms and Conditions
 A SingleMachineModel / 601
See Chand and Chhajed for a proof. and
Proof of Lemma 2 COST(n)
We prove the lemma for i = 1 and 2. The proof can = (E, + T.) *(E' *t)*P
be extended easily for i = 3, 4, .. . , N.
El.(E' - 1).t.P2
Proof for WT1(n)3 WT1((N/2)+) + 2

Consider the problem with m = 2 and t1 = = ... =

tN = t. (Since WTj(n)'sare independent of processing
T;.(T' + 1).t.P3
times, it is sufficient to prove this result for t = + 22
t2 - . = tN =t.) From (8), we get
+ (E2 + T2*).(E'+ T. + E2)*t*P1
WT,(n) = COST(n) E; *(E21- >t.P2 T2 *(T + 1)0t.P3
t
2 2
and
Now consider another feasible solution constructed
V N+0_COST((N/2)+)
from the COST(n)-solution by changing the assign-
V2J t ment for the firstjob in Batch 2. We assume that the
To prove WT,(n) > WT,((N/2)+), it is sufficient to assignment for the first job in Batch 2 is changed to
show that Batch 1. Thus, the first (n + 1) jobs are now assigned
to Batch 1 and the last N - (n + 1) jobs are assigned
COST(n) 3COST( 2-) ) (A. 1) to Batch 2. The due dates Di and D2 are determined
as follows.
for 1 < n S (N/2)+. Note that COST(l) here denotes Case 1. D= = (El + 1) t and D2= (El + E2 + T)t.
the minimum total cost for the N-job problem with Let COST, denote the cost for this case, so we have
m = 2 due dates, ti = t2= ... = tNand /jobs assigned
to D,. We provide a proof by construction. COST,
Consider a COST(n)-solution with n < N - n.
=(E' + T + 1).(E + 1)-t.Pi
Let Ef( T.) denote the number of early (tardy) jobs
in Batch 1 and E'(T2) denote the number of early (E + 1).E*t.P2 T' *(Tf + 1) *t P3
(tardy)jobs in Batch 2 in the COST(n)-solution. It is 2 2
easy to see that n = E, + T., D1 = E'-t, D2 =
+(E2 + T2- 1).(E + T +E').t*Pl
(El + E2 + Tf)t,
(E- 1)(E2- 2).t.P2 T2 *(T + 1) t P3
Due date penalty for the first n jobs: + 2+ 2
2 2
= nD* -P1 = (El + T ).(E*.t).P,.
and
Earlinesstardiness cost for the first n jobs:
COST(n) - COST,
=P2 .((E' - l)t t+(E2' - 2).t+ .. . + I -t)
= t.(PI + P2) (E - E- 1). (A.2)
+ P3(t + 2t + ... + Tit)
Case 2. Di = E' t and D2 = (El + T' + E2 + 1).t.
P2-EV-(El - -t + P3 TV(T' + 1).t Let COST2denote the cost for this case, so we have
2 2
2 2
COST2
Due date penalty for the last N - n jobs
= (El + Tf + 1).Ef.t.P1
= (N- n).D2.P1,
E(I -1)tP2 (TMlo-1).(T' + 2).t.P3
= (E2 + I ) *(E; + E2 + T;) *t.Pi. 2 2
Earliness/tardinesscost for the last N - n jobs: + (E2 + T2 - 1).(E + T' + E2 + 1).t.P

=
P2 E'(E - 1).t P3.T2'(TM + 1)t E2'(E - 1).t.P2 (TM - O)T2'.t.P3
2 2 2 2

This content downloaded from 128.184.220.23 on Fri, 13 Nov 2015 23:08:02 UTC
All use subject to JSTOR Terms and Conditions
602 / CHAND AND CHHAJED
and
COST(n) - COST2
= (T2 - T -1) -
(P3 .t.
Since n < N - n, we have either E, < E2 and
T. < T2, or E, < E2 and T. < T2. If El < E2, then
COST(n) > COST,, and if T. < T2, then COST(n) >
COST 2. Thus, for n < N - n, an alternate feasible
solution with an equal or lower cost can be constructed
by reassigning a job from Batch 2 to Batch 1. This
implies that WT,(n) > WT,(n + 1) for n < N - n.
This completes the proof for WTI(n)> WT,((N/2)+).
Proof for WT2(n)> WT2((N/2)+)
Consider the problem with m = 2, t1 = 0 and t2 =
t3. . . = tN -t. From (8), we get
WT2(1) = COST(l) for I E (1, N-1).
t Opns.Res. 38, 22-36.
To prove WT2(n)> WT2((N/2)+)for n E (1, (N/2)+),
it is sufficient to prove that
COST(n)> COST((j[)T) for n E (I E
Again, a proof by construction is provided. See Chand
and Chhajed for the details.
The above proof can be extended to show that
WT1(n)> WTi(n + 1) for n < N- n and i E (3, N).
To prove it for i = k, assume that the two due date N-
job problem has (k - 1) jobs with zero processing
times, and N-(k - 1) jobs with processing time of
t > 0 each. All the zero processing time jobs are as-
sumed to complete at D1. For a given COST(n) solu-
tion, an alternate feasible solution is constructed by
increasing the number of jobs assigned to Batch 1
from n to n + 1. Two alternatives for due dates are
considered. Compared to the due dates in COST(n),
the due date for Batch 1 is increased by t while the
due date for Batch 2 is kept the same as in Case 1. In
This content downloaded from 128.184.220.23 on Fri, 13 Nov 2015 23:08:02 UTC
All use subject to JSTOR Terms and Conditions
Case 2, the due date for Batch 1 is kept the same while
the due date for Batch 2 is increased by t.
(A.3) ACKNOWLEDGMENT
This researchwas supported in part by the Center for
the Management of Manufacturing Enterprises at
Purdue University.
REFERENCES
BAGCHI,U. 1989. Due Date or Deadline Assignmentto
Multi-JobOrdersto Minimize Total Penaltyin the
One MachineSchedulingProblem.WorkingPaper,
The Universityof Texas at Austin.
BAKER, K. R., AND J. BERTRAND. 1981. A Comparison
of Due-Date Selection Rules. AIIE Trans. 13,
123-131.
BAKER, K. R., AND G. D. SCUDDER. 1990. Sequencing
With Earlinessand TardinessPenalties:A Review.
BECTOR,C. R., Y. P. GUPTA AND M. C. GUPTA. 1988.
Determinationof an Optimal Common Due Date
and Optimal Sequence in a Single Machine Job
Shop. Int. J. Prod.Res. 26, 613-628.
CHAND,S., AND D. CHHAJED.1990. Determination of
OptimalDue Dates and Sequence.BEBR Working
Paper No. 90-1714, College of Commerce and
Business Administration,University of Illinois at
Urbana-Champaign.
PANWALKAR,
S. S., M. L. SMITHAND A. SEIDMANN.
1982. Common Due Date Assignmentto Minimize
TotalPenaltyforthe One MachineSchedulingProb-
lem. Opns.Res. 30, 391-399.
EILON,S., AND I. J. CHOWDHARY.
1976. Due Date in
Job Shop Scheduling.Int. J. Prod.Res. 14, 223-237.
SEIDMANN, A., S. S. PANWALKAR AND M. L. SMITH.
1981.OptimalAssignmentof Due Datesfor a Single
ProcessorSchedulingProblem.Int. J. Prod.Res. 19,
393-399.
WEEKS,J. K. 1979. A SimulationStudy of Predictable
Due Dates. Mgmt. Sci. 25, 363-373.
WEEKS,J. K., AND J. S. FRYER.1977. A Methodology
for Assigning Minimum Cost Due Dates. Mgmt. Sci.
23, 872-881.