Computers & Operations Research 39 (2012) 3244–3246

Contents lists available at SciVerse ScienceDirect

Computers & Operations Research

journal homepage: www.elsevier.com/locate/caor

A note on ‘‘Two-machine flow-shop scheduling with rejection’’ and its link

with flow-shop scheduling and common due date assignment
V. T’kindt a,n, F. Della Croce b
a
LI., ERL CNRS OC 6305, Université Franc- ois Rabelais, Tours, France
b
D.A.I., Politecnico di Torino, Torino, Italy

a r t i c l e i n f o a b s t r a c t

Available online 19 April 2012 In a recent paper by Shabtay and Gasper ‘‘Two-machine flow-shop scheduling with rejection,
Keywords: Computers and Operations Research’’, forthcoming, doi:10.1016/j.cor.2011.05.023, several complexity
Two-machine flow-shop and approximation results are proposed for a two-criteria two-machine flow-shop scheduling problem
Scheduling with rejection with rejection. The two criteria to be minimized are the makespan the total rejection cost. This note
Common due date assignment positions the contribution of such results with respect to the contributions of the literature on common
due date assignment and flow-shop scheduling not considered in the work of Shabtay and Gasper.
& 2012 Elsevier Ltd. All rights reserved.

1. Introduction However, as all the jobs in A will be completed within the

makespan C max , such makespan can be seen as a common due
In [5] Shabtay and Gasper have recently tackled the two-machine date d to be respected by the jobs in A (hence all jobs A A will be
flow-shop problem with rejection where two objectives are con- early with respect to d), while the jobs in A can be assumed
sidered, namely the makespan and the total rejection cost, and four w.l.o.g. to be completed after d (hence all jobs A A will be tardy
combinations of such objectives are analyzed. Such combinations with respect to d).
are the problem of minimizing the weighted sum of such objectives But then, using the extended three-field classification of [7]
(denoted by P1), the E constraint problem with respect to the total (which is more common for multi-objective scheduling), problem
P
rejection cost (denoted by P2), the E constraint problem with respect P4 can be denoted by F29dj ¼ d, unknown d9d, j wj U j . Correspond-
P
to the makespan (denoted by P3) and the more general problem ingly, P1 can be denoted by F29dj ¼ d, unknown d9d þ j wj U j , P2
P
devoted to the search for the Pareto optimal solutions with respect can be denoted by F29dj ¼ d, unknown d9Eðd= j wj U j Þ and P3 can
P
to both objectives (denoted by P4). Several complexity and approx- be denoted by F29dj ¼ d, unknown d9Eð j wj U j =dÞ.
imation results are provided on problems P1, y, P4 but some related It turns out that P4 is actually the weighted generalization of
P
literature [3,8] on common due date assignment and flow-shop problem F29dj ¼ d, unknown d9d, j U j considered in [8] (denoted
scheduling is missing. Purpose of this note is to position the hereafter by P5) and is also strictly related to problem F29dj ¼
P
contribution of [5] with respect to such literature. D9 j wj U j (minimization of the weighted sum of tardy jobs in a
Consider the more general problem P4. We have n jobs two-machine flow-shop with common due date D—denoted here-
available for processing at time zero to be scheduled on m¼2 after P6) considered in [3]. This has several implications with respect
machines in a flow-shop scheduling system. Each job j has a to the overall contribution of [5].
processing time aj on the first machine, a processing time bj on
the second machine and a rejection cost wj. The jobs can be either
accepted (belonging then to the set A of accepted jobs) or rejected 2. Main remarks
(belonging then to the set A of rejected jobs). The objectives are
2.1. NP-hardness of problem P1—Section 2.1. in [5]
the makespan of the accepted jobs and the total rejection cost
P
w and the aim is to search for the Pareto optimal solutions
jAA j It can be directly derived from the NP-hardness of problem P6
with respect to both objectives. proved in [3] by showing that P6 reduces to P1. Here is a sketch of
the proof. Let us denote by D0 the makespan obtained by Johnson’s
algorithm [2] when computing the optimal schedule for the
n
Corresponding author. Tel.: þ33 247 361 427; fax: þ 33 247 361 422.
F2JC max problem. Given an instance of P6 (where we assume
E-mail addresses: [email protected] (V. T’kindt), w.l.o.g. wi Z 1 integer, i¼1, y, n), consider solving several
[email protected] (F. Della Croce). instances of P1 that keep the same processing times but multiplies

0305-0548/$ - see front matter & 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.cor.2012.04.009
 V. T’kindt, F. Della Croce / Computers & Operations Research 39 (2012) 3244–3246 3245

the weights by a coefficient g with g A ½0,D0 . Notice that both for Table 1
g ¼ 0 and for g ¼ D0 , P1 is polynomial as the best you can do in the CPU time required by the ILP solver on problem P1 for different values of n.
first case is to reject all jobs and in the second case is to accept all
Size n Average CPU time (s) Maximum CPU time (s)
jobs. Notice also that, for g increasing, the value of d in the
optimal solution of P1 is obviously nondecreasing. But then, to 100 0.29 3.70
find the optimal solution of problem P6, it is sufficient to apply a 300 0.76 4.52
binary search on g until the value of d reaches the largest value 500 1.40 4.14
1000 4.83 7.12
z r D that corresponds to the optimal solution of problem P6. But 1500 11.31 17.00
then, the number of instances to be solved is OðlogfD0 gÞ inducing a 2000 22.51 57.08
polynomial time reduction from P6 to P1.

2.2. Exact pseudo-polynomials algorithms for problems P1 and

P4—Sections 2.2 and 3.1 in [5]
We have considered the same plan of the computational
Consider the pseudo-polynomial algorithm presented in [3] for experiment proposed in [5] and tested the CPU time required
problem P6. That procedure computes the minimum weighted by a pure MIP solver (CPLEX 12.1) to compute the optimal
number of tardy jobs in a two machine flow-shop with common solution for instances with up to 2000 jobs. The tests have been
due date D with complexity OðnD2 Þ. The optimal solution value is conducted on a PC Intel Core 2 Duo 2.4 GHz 3Go of RAM. The
d d
given by f 1 ð0; 0Þ where f k ðt 1 ,t 2 Þ denotes the minimum weighted results are depicted in Table 1. We point out that the optimal
number of late jobs for the jobs in fi=k ri r ng, provided that the solution can be computed in the worst-case within approxi-
first early job in fi=kr i rng starts processing at time t1 on the first mately 57 s on an instance with 2000 jobs (while the FPTAS
machine and no earlier than time t2 on the second machine. It is with E ¼ 1 proposed in [5] and denoted H3 required in the
shown in [8] that, by applying such procedure with common due average case on a 100-job instance more than 160 s on a
D
date D0 , the values f 1 ðj,jÞ for j ¼ 0, . . . ,D provide the optimal comparable machine).
P
solution to the F29dj ¼ d9 wj U j problem for all values of d with Notice also that a similar model can be devised for problem P2
0 r d r D (in [8], the unweighted problem was considered, but the
0
and hence by means of an E-constraint approach also problem P4
procedure works exactly the same if weights are added as the can be solved to optimality. We have then applied the same
D
algorithm in [3] indeed). Hence, by comparing all such f 1 ðj,jÞ computational campaign also for instances with just n ¼50 jobs
values with respect to the cost function of problem P1 we get an for problem P4 due to the huge number of Pareto-optima on each
optimal pseudo-polynomial procedure for such problem that instance. The average CPU time was in this case approximately
requires OðnD02 Þ time. Same computational time is shown in [8] 380 s for computing on the average approximately 1900 paretian
for problem P5 that obviously works also for P4, hence a pseudo- solutions (hence, on the average about 0.20 s for each paretian
polynomial exact approach for problems P1 , . . . ,P 4 could be easily solution).
derived from the results in [3,8]. On the other hand, notice, that in
[5], the complexity of the proposed pseudo-polynomial algorithm
for problem P4 that clearly works also for problem P5 is
P P 3. Final remarks
Oðn ni¼ 1 bi ni¼ 1 wi Þ. For problem P4, none of these two complex-
ities is dominant. However, as for problem P5 we have wi ¼1, i¼ 1,
In this note we positioned the contribution of [5] with respect
y, n, the complexity of the algorithm proposed in [5] becomes
P to some relevant literature not considered in that paper. We
Oðn2 ni¼ 1 bi Þ and hence improves upon the result of [8]. Also, as
notice, however, that, apparently, in the scheduling literature,
far as problem P6 is concerned, the same binary search sketched
there has been little attention in ascertaining the (dis)-similarities
above for the NP-hardness proof, can be used to solve such
P between scheduling with rejection and bi-objective scheduling
problem. The related complexity becomes then Oðn ni¼ 1 bi
0 with common due date assignment and this is particularly
logfD gÞ that improves upon the result of [3].
unfortunate in [5] as one of the authors is the author of several
other publications on due date assignment such as, for
2.3. Experimental testing—Section 2.5. in [5]
instance, [6]. Indeed, in a recent paper [9], several complexity
results are provided on single machine scheduling with rejection.
The contribution of the approximation algorithms of [5] is
By keeping the notation of [7], one of the results provided was the
obviously mainly theoretical. Nonetheless, as an experimental
simple (four rows of length) NP-hardness proof of problem
testing subsection on problem P1 is considered, it is worthy to see P
19dj ¼ d, unknown d9Eð j wj U j =dÞ. However, the NP-hardness proof
how an exact approach performs on the same problem. Consider P
of the related 19dj ¼ d, unknown d9 j wj U j þ d problem given in [1]
the ILP below which is a straightforward adaptation to model P1
and reported also on [4] was not even mentioned in [9]. Main
of the ILP model presented in [8]. Notice that the jobs are sorted
purpose of this note was indeed to point out such discrepancy in
and indexed according to Johnson’s schedule for the F2JC max
order to avoid in the future incomplete contributions like the one
problem. Let xi be a 0=1 variable where xi ¼1 if job i is on-time and
of [5].
let d be the unknown common due date.
X
n
min d þ wi ð1xi Þ ð1Þ
i¼1
References

X
u X
n
[1] De P, Ghosh JB, Wells CE. Optimal delivery time quotation and order
ai x i þ bi xi r d, u ¼ 1, . . . ,n ð2Þ sequencing. Decision Sciences 1991;22:379–390.
i¼1 i¼u [2] Johnson SM. Optimal two- and three-stage production schedules with setup
times included. Naval Research Logistics Quarterly 1954;1:61–67.
xi A f0; 1g, i,j ¼ 1, . . . ,n ð3Þ [3] Józefowska J, Jurish B, Kubiak W. Scheduling shops to minimize the weighted
number of late jobs. Operations Research Letters 1994;16:277–283.
[4] Leung JYT. Handbook of scheduling: algorithms, models, and performance
dZ0 ð4Þ
analysis. Boca Raton, FL, USA: Chapman & Hall/CRC; 2004.
3246 V. T’kindt, F. Della Croce / Computers & Operations Research 39 (2012) 3244–3246

[5] Shabtay D, Gasper N. Two-machine flow-shop scheduling with rejection. [8] T’Kindt V, Della Croce F, Bouquard JL. Enumeration of Pareto optima for a
Computers and Operations Research 2011;39(5):1087–1096, http://dx.doi.org/ flowshop scheduling problem with two criteria. INFORMS Journal on Comput-
10.1016/j.cor.2011.05.023. ing 2007;19:64–72.
[6] Shabtay D, Steiner G. Two due date assignment problems in scheduling a [9] Zhang L, Lu L, Yuan J. Single-machine scheduling under the job rejection
single machine. Operations Research Letters 2006;34:683–691. constraint. Theoretical Computer Science 2010;411:1877–1882.
[7] T’kindt V, Billaut JC. Multicriteria scheduling: theory, models and algorithms.
Heidelberg, Germany: Springer-Verlag; 2006.