4OR (2020) 18:177–196

https://doi.org/10.1007/s10288-019-00410-4

RESEARCH PAPER

Single-machine scheduling of proportional-linearly

deteriorating jobs with positional due indices

Rubing Chen1 · Jinjiang Yuan1

Received: 18 January 2018 / Revised: 8 June 2019 / Published online: 22 June 2019
© Springer-Verlag GmbH Germany, part of Springer Nature 2019

Abstract
In this paper, we study the scheduling of proportional-linearly deteriorating jobs with
positional due indices, release dates, deadlines and precedence relations on a single
machine. The scheduling criteria studied in this paper include the makespan, maximum
lateness, maximum tardiness, maximum flow time, maximum weighted completion
time, maximum scheduling cost, total completion time, and the number of tardy jobs.
By applying Lawler’s rule and Smith’s rule, polynomially solvable problems are pro-
cessed by using two unified methods. We also present some new NP-hardness results
when processing times of the jobs have no deterioration. Our results generalize a series
of known achievements in the literature.

Keywords Single-machine scheduling · Deteriorating jobs · Due indices · NP-hard ·

Polynomial-time algorithms

Mathematics Subject Classification 90B35 (Scheduling theory, deterministic) · 90C27

(Combinatorial optimization)

1 Introduction

We study the scheduling of proportional-linearly deteriorating jobs with positional

due indices, release dates, deadlines and precedence relations on a single machine.
Our research is motivated by the strong application background of deteriorating jobs
and positional due indices.
In the traditional scheduling research, it is always assumed that the processing time
of every job is a fixed value and independent of its starting time and the position of
each job has no restriction in any feasible schedule. However, in a variety of real life

B Jinjiang Yuan
[email protected]

1 School of Mathematics and Statistics, Zhengzhou University, Zhengzhou 450001,

Henan, People’s Republic of China

123
178 R. Chen, J. Yuan

applications, the actual processing time of a job depends on its starting time. It is
particularly an increasing function of its starting time, which is the case of deteriorat-
ing jobs. Such kind of deterioration may appear in the steel industry. The producers
have to process molten steel from steel furnace into steel products needed for produc-
tion and living. Before rolling the ingots, the producers need to reheat the ingots to a
pretty high temperature. The later the producers reheat ingots to begin, the more time
and energy will be consumed. The deterioration also arises in cleaning assignments,
scheduling maintenance, etc. On the other side, in many practical make-to-order pro-
duction systems, the customers want their needs to be met as early as possible. In their
opinion, the most advanced position the order is, the sooner it will be completed. Thus,
they not only require the completion times of their orders, but also have requirements
on the processing positions of their orders. With the deterioration and position con-
straints, the starting times of jobs and their processing positions should be considered
simultaneously to balance the completion times and production positions of jobs. This
motivates us to study scheduling problems by putting deterioration and processing
position constraints together.
Problem formulation: Suppose that we are given a set of n proportional-linearly
deteriorating jobs J = {J1 , J2 , . . . , Jn } with precedence relations to be scheduled on
a single machine. Each job J j has a normal processing time α j > 0, a release date
r j ≥ 0, a due date d j ≥ 0, a deadline d̄ j ≥ d j , a weight w j > 0, and a positional due
index k j ∈ {1, 2, . . . , n}. In this paper, “proportional-linearly deteriorating” means
that the actual processing time of each job J j starting at time t is given by p j (t) =
α j (a + bt), where a ≥ 0, b ≥ 0, t > 0 and a + b > 0. Note that proportional-linearly
deterioration is a special case of linear deterioration in which the actual processing
time of a job J j starting at time t is given by p j (t) = a j + b j t. The term “positional
due index k j ” requires that, in a feasible schedule, the position number of every job
J j is at most k j . Moreover, we use “prec” to denote the precedence relations among
the n jobs. If a scheduling criterion f is to be minimized, then the problem is denoted
by
1|r j , k j , p j (t) = α j (a + bt), d̄ j , prec| f , (1)
 
where f ∈ {Cmax , L max , Tmax , W Cmax , Fmax , f max , C j , U j }. Note that, when
b = 0, the problem degenerates into the traditional version 1|r j , k j , d̄ j , prec| f without
the job deterioration assumption.
Our contribution: We study the complexity classification
  of the scheduling model in
Eq. (1) for f ∈ {Cmax , L max , Tmax , Fmax , f max , C j , U j }. By applying Lawler’s
rule and Smith’s rule, polynomially solvable problems are processed by using two uni-
fied methods, where Lawler’s rule is an extended version of the well-known algorithm
by Lawler (1973) for solving problem 1|prec| f max , and Smith’s rule is an extended

version of the well-known algorithm by Smith (1956) for solving problem 1|d̄ j | C j .
We also presented some new NP-hardness results when processing times of the jobs
have no deterioration. Our results generalize a series of known results in the literature.
The results obtained in this paper are summarized in Table 1.

Remark Without deadlines d̄ j and positional  due , all the problems 1| p j (t) =
indices k j 
α j (a+bt)| f for f ∈ {L max , Tmax , f max , C j , w j C j , U j } have been addressed

123
Single-machine scheduling of proportional-linearly… 179

Table 1 Summary of complexity results obtained in this paper

Indices Scheduling problems Time complexity References

P1 1|k j , p j (t) = α j (a + bt), d̄ j , prec| f max O(n 2 ) Theorem 4.2

P2 1|r j , k j , p j (t) = α j (a + bt), prec|Cmax O(n 2 ) Theorem 4.2
P3 1|r j , k j , p j (t) = α j (a + bt), prec|Fmax O(n 2 ) Theorem 4.2
(under Assumption 1 in Sect. 4)

P4 1|k j , p j (t) = α j (a + bt), d̄ j | C j O(n 2 ) Theorem 4.2

P5 1|k j , p j (t) = α j (a + bt)| C j O(n log n) Theorem 5.2

P6 1| p j (t) = α j (a + bt), d̄ j | C j O(n log n) Theorem 5.2
P7 1|k j , p j (t) = α j (a + bt)|L max O(n log n) Theorem 5.2
P8 1|k j , p j (t) = α j (a + bt)|Tmax O(n log n) Theorem 5.2
P9 1|k j , p j (t) = α j (a + bt)|W Cmax O(n log n) Theorem 5.2
P10 1| p j (t) = α j (a + bt), d̄ j |L max O(n log n) Theorem 5.2
P11 1| p j (t) = α j (a + bt), d̄ j |Tmax O(n log n) Theorem 5.2
P12 1| p j (t) = α j (a + bt), d̄ j |W Cmax O(n log n) Theorem 5.2
P13 1|r j , k j , p j (t) = α j (a + bt)|Cmax O(n log n) Theorem 5.2
P14 1|r j , k j , p j (t) = α j (a + bt)|Fmax O(n log n) Theorem 5.2
P15 1|r j , p j = α j |W Cmax Unary NP-hard Theorem 6.1
P16 1|r j , p j = α j , chains|Fmax Unary NP-hard Theorem 6.2

P17 1|k j , p j = α j | U j Unary NP-hard Theorem 6.3

in the literature. These problems are solved by extending the techniques in  Smith
(1956), Moore (1968) or Lawler (1973). For the criteria L max , Tmax , f max and Cj,
when deadlines d̄ j or positional due indices k j are taken into consideration, the research
in this paper develops the techniques existing in the literature  and leads to similar

results. But such extensions do not work  for the two criteria  w j C j and Uj.
In fact,
 for the four special problems 1| d̄ j | w j C j , 1| d̄ j | U j , 1|k j | w j C j and
1|k j | U j , the first two problems are unary NP-hard, the complexity of the third
problem is still open, and the last problem will be proved to be unary NP-hard in this
paper. Our research also covers the two criteria Cmax and Fmax . In some cases, the
release dates r j and the precedence relations “prec” are taken into consideration.

Organization of this paper: This paper is organized as follows. In Sect. 2, we present

a review of relevant research. In Sect. 3, we present some notations and terminologies
which will be used in the followed sections. In Sect. 4, we give the definitions about
the Lawler Property and Lawler’s rule, and described the problems how to be solved
by Lawler’s rule. In Sect. 5, we give the definitions about the Smith Property and
Smith’s rule, and described the problems how to be solved by Smith’s rule. In Sect. 6,
we presentthe unary NP-hardness proofs for problems 1|r j |W Cmax , 1|r j , chains|Fmax
and 1|k j | U j , respectively. Conclusions are presented in the last section.

123
180 R. Chen, J. Yuan

2 Relevant research

Scheduling with deteriorating jobs was first studied by Melnikov and Shafransky
(1979). Gawiejnowicz (2008) and Gawiejnowicz et al. (2009a, b) presented a compre-
hensive study of time complexity results of scheduling problems with deteriorating
jobs under single-, parallel- and dedicated-machine environments. Agnetis et al. (2014)
summarized the time complexity of single-machine two-agent scheduling problems
with deteriorating jobs. Strusevich and Rustogi (2017) presented a comprehensive
study of time complexity results of scheduling problems with deteriorating jobs to min-
imize either the maximum completion time or the total completion time under single-
and parallel-machine environments. The deterioration model considered in Strusevich
and Rustogi (2017) includes not only start-time dependent but also positional, cumula-
tive and their combinations. Although there are many forms of job deterioration in the
literature, we mainly review some relevant results on scheduling with proportional-
linearly deteriorating jobs. This means that each job J j which starts at time t has an
actual processing time p j (t) = α j (a + bt). Note that proportional-linearly deteriora-
tion is a special version of linearly deterioration in which p j (t) = a j +b j t. In the case
that p j (t) = α j t, the setting is called proportional deterioration, which is a special
version of proportional-linear deterioration.
For the setting of proportional deterioration, earlier work can be found in Mosheiov
(1994). Qi et al. (2009) further extended this deterioration model to some unbounded
parallel-batch scheduling problems on a single machine. Cheng and Sun (2009)
provided NP-hardness results of the single machine scheduling problems in which
p j (t) = α j t and jobs can be rejected by paying corresponding penalty costs. Ng et al.
(2010) provided polynomial-time algorithms for problems 1|r j , p j (t) = α j t|Cmax ,
1|r j , p j (t) = α j t, pmtn|F (F is an arbitrary symmetric function), 1|r j , p j (t) =
α j t, pmtn| f max and 1|r j , p j (t)
= α j t, pmtn, prec| f max . They also showed the prob-
lem 1|r j , p j (t) = α j t, pmtn| w j C j is NP-hard. Miao et al. (2012) showed that
problem 1|r j , p j (t) = α j t|L max is NP-hard. In fact, this result in Miao et al. (2012)
follows immediately from the NP-hardness of problem 1|r j , p j (t) = α j t, d̄ j |Cmax
showed early in Gawiejnowicz (2007). Miao et al. (2012) also showed that problem
1|p-batch, r j , p j (t) = α j t, b = ∞|L max is NP-hard. Yu and Wong (2013) presented
 lcompetitive ratio (1 + αmax ) for the online problem
a best-possible algorithm with l

1|online, r j , p j (t) = α j t| C j , where l > 0. Li et al. (2014) presented an approxi-

mation algorithm for problem Pm| p j (t) = α j t|Cmax . They tackled the problem with
a game theoretic approach, in which each job was regarded as player and tried to
make a decision to optimize its own payoff while ignoring the others. Chen et al.
(2017) also studied parallel-machine scheduling problems where the actual process-
ing time is p j (t) = α j t. They mainly considered the criteria of the minimization of
makespan, the minimization of total completion time, and the minimization of total
machine load. Jobs are selfish players, and each job can choose one of the m identical
parallel machines to be processed. They considered three coordination mechanisms
and proved the existence of Nash equilibrium under these coordination mechanisms.
They also presented the Price of Anarchy (PoA) and the Price of Stability (PoS)
with social costs of minimizing the makespan and total machine load. Miao (2018)

123
Single-machine scheduling of proportional-linearly… 181

 
showed that problems 1|r j , p j (t) = α j t| w j C j , 1|r j , p j (t) = α j t| U j and
1|r j , p j (t) = α j t, d̄ j | C j are NP-hard.
For the setting of proportional-linear deterioration, some basic results can be
found in Kononov (1998) (in Russian) and Zhao et al. (2003). Liu et al. (2011)
further extended this deterioration model to some two-agent scheduling problems
on a singlemachine. Li et al. (2009) considered the problem 1| p j (t) = α j (a +
bt)| nj=1 i= n
j |C j − Ci |. They showed that the optimal schedule is V-shaped with
respect to the  j -values. Wu et al. (2013) considered the problem 1|r j , p j (t) =
α
α j (a + bt)| T j , and proposed a branch-and-bound algorithm incorporating with
several dominance properties and two lower bounds to search for the optimal solu-
tion. They also proposed a marriage in honey-bees optimization algorithm (MBO)
to provide a near-optimal solution. Gawiejnowicz and Kononov (2014) considered
isomorphic scheduling problems that constituted a new class of mutually related
scheduling problems. The criteria mainly considered in their paper were makespan,
maximum lateness and the weighted number of tardy jobs. They defined the class
applying a one-to-one transformation of instances of a generic scheduling problem
with fixed processing times into instances of scheduling problems with p j (t) =
α j (a + bt). Moreover, they showed how to convert polynomial-time algorithms for
scheduling problems with fixed processing times into polynomial-time algorithms for
scheduling problems with p j (t) = α j (a + bt). Ma et al. (2016) presented the best-
possible algorithm with competitive  ratio 1 + λ(a) + αmax b for the online problem
1|online, r j , p j = α j (a + bt)| w j C j , where αmax = max1≤ j≤n α j and λ(a) = 0
or λ(a) = 1 depending on whether a = 0 or a > 0.
In the scheduling with positional due indices, each job J j has a positional due index
k j . The feasibility requirement is that, in a feasible schedule, the position number of
every job is at most its positional due index, that is, if J j is the x-th job in a feasible
schedule, then x ≤ k j . Scheduling with positional due indices was first introduced
in Zhao and Yuan (2017). In their paper, they presented an O(n 2 )-time algorithm for
problem 1|k j , prec| f max and an O(n log n)-time  algorithm for problem 1|k j |L max . An
O(n log n)-time algorithm for problem 1|k j | C j was presented in Zhao et al. (2016).
 presented an O(n )-time algorithm for the Pareto optimization
Gao and Yuan (2015) 4

problem 1|k j | ( C j , f max ). Gao and Yuan (2017) further presented polynomial-
#

time algorithms for various bi-criteria Pareto-scheduling problems under positional

due indices.
 It was posed in Gao and Yuan (2017) that the complexity of  problem
1|k j | w j C j is still open. Lenstra et al. (1977) showed that problem 1|d̄ j | w j C j
is unary NP-hard. Since the positional due indices k  j is similar to the deadlines d̄ j to
a certain extent, we tend to think that problem 1|k j | w j C j is also unary NP-hard.

Remark It should be noticed that the concept of “positional due indices” has some dual-
similarity with the concept of “generalized due dates” introduced by Hall (1986). Both
of the two concepts are related to positions. In the scheduling with generalized due
dates, there are n due dates d1 , d2 , . . . , dn with d1 ≤ d2 ≤ · · · ≤ dn . In a schedule π ,
let Jπ( j) be the job scheduled in the j-th position. Then we must assign the due date d j
to the job Jπ( j) , 1 ≤ j ≤ n. But in the scheduling with positional due indices, for each
job J j , 1 ≤ j ≤ n, the due index k j is a characteristic of job J j , and is independent
of the schedule of the jobs. Thus, the difference between these two concepts can be

123
182 R. Chen, J. Yuan

briefly described as follows: Generalized due dates are defined on positions and are
related to jobs by specific schedules, and positional due indices are defined on jobs
directly and are related to positions by specific schedules in such a way that, in every
feasible, each job J j must be scheduled in a position no larger than k j .

For the classical problem 1|| U j  , Moore (1968) presented an O(n log n)-time
algorithm. For the weighted version 1|| w j U j , Karp (1972) proved that this problem
is NP-hard, even if all the jobs had a common due date. For this problem, Lawler
and Moore (1969) presented
 a pseudo-polynomial dynamic programming
 algorithm
that takes O(n min{ p j , max j d j })) time. For problem 1|d̄ j | U j , Lawler (1983)
showed that it is binary NP-hard. Recently, Yuan (2017) showed that  this problem
is unary NP-hard. By our knowledge, the complexity of problem 1|k j | U j is still
unaddressed.
Lawler (1973) presented an O(n 2 )-time algorithm for problem 1|prec| f max .
Debczynski and Gawiejnowicz (2013) presented polynomial-time algorithms for sev-
eral single-machine scheduling problems with same type job processing time or mixed
job processing times and arbitrary precedence constraints to minimize the maximum
cost criterion. Brauner et al. (2016) further presented necessary and sufficient condi-
tions for the optimality of a schedule for a similar problem with strictly increasing cost
functions. Based on the structure of optimal schedules, they presented an O(n 2 )-time
algorithm for the problem with uncertainty under the decomposable assumption. For
the problem 1|prec| f max , Brauner et al. (2017) developed a fast updating algorithm
which depends on the well-known algorithm by Lawler (1973). Their new algorithm is
used as a subroutine for solving several problems with uncertainty in O(n 3 ) or O(n 2 )
time.

3 Notation and terminology

Consider a set J = {J1 , J2 , . . . , Jn } of proportional-linearly deteriorating jobs. Then

the actual processing time of a job J j starting at time t is given by p j (t) = α j (a + bt).
From this description, it is easy to verify that, if a set of jobs K ⊆ J is consecutively
processed (in an arbitrary order) starting at time t, then the last job is completed at
time  a  a
C(K, t) = t + (1 + bα j ) − . (2)
b b
J j ∈K

The equation in (2) will be repeatedly used in this paper. Let τ0 = min{r j : j =
1, 2, . . . , n}. In the case that a = 0, the actual processing time of a job J j starting at
time t is given by p j (t) = α j bt. In order to avoid the trivial situation that all jobs
are released, started and completed at time 0, we take the convention that τ0 > 0. Let
f j (·) be the cost function of job J j ∈ J , which is assumed to be non-decreasing in
[0, +∞), and, for each time t ∈ [0, +∞), the function value f j (t) can be computed
in a constant time. In a feasible schedule σ , we use the following notation.
• σ [J j ] is the position number of job J j in σ . This means that J j is the σ [J j ]-th job
in σ . Then σ [J j ] ≤ k j for every job J j .

123
Single-machine scheduling of proportional-linearly… 183

• S j (σ ) is the starting time of job J j in σ .

• p j (S j (σ )) = α j (a + bS j (σ )) is the actual processing time of job J j in σ .
• C j (σ ) = S j (σ ) + p j (S j (σ )) = S j (σ ) + α j (a + bS j (σ )) is the completion time
of job J j in σ .
• L j (σ ) = C j (σ ) − d j is the lateness of job J j in σ .
• T j (σ ) = max{0, L j (σ )} is the tardiness of job J j in σ .
• F j (σ ) = C j (σ ) − r j is the flow time of job J j in σ .
• U j (σ ) = 1 if C j (σ ) > d j and U j (σ ) = 0 if C j (σ ) ≤ d j . We call U j (σ ) the
indicator of tardy job J j in σ .
• w j C j (σ ) is the weighted completion time of job J j in σ .
• Cmax (σ ) = max{C j (σ ) : 1 ≤ j ≤ n} is the makespan of σ .
• W Cmax (σ ) = max{w j C j (σ ) : 1 ≤ j ≤ n} is the maximum weighted completion
time of σ .
• Fmax (σ ) = max{F j (σ ) : 1 ≤ j ≤ n} is the maximum flow time of σ .
• L max (σ ) = max{L j (σ ) : 1 ≤ j ≤ n} is the maximum lateness of σ .
• Tmax (σ ) = max{T j (σ ) : 1 ≤ j ≤ n} is the maximum tardiness of σ .
• f max (σ ) = max{ f j (σ ) : 1 ≤ j ≤ n} is the maximum scheduling cost of σ .
n
• w C (σ ) is the total weighted completion time of σ .
nj=1 j j
• j=1 w j U j (σ ) is the total weighted number of tardy jobs of σ .
A feasible schedule σ is called active if there is no other feasible schedule σ  such
that C j (σ  ) ≤ C j (σ ) for 1 ≤ j ≤ n and at least one of the n inequalities is strict.
In this paper, all the criteria are non-decreasing in [0, +∞). Thus, for each problem,
there is an optimal schedule which is also active. Therefore, we consider only active
schedules.

4 Problems solvable by Lawler’s rule

We use ≺ to denote the precedence relation among the jobs in J . Then Ji ≺ J j means
that job Ji must be scheduled before job J j in any feasible schedule. A topological
order of the jobs in (J , ≺), denoted by O(J ) = (Jo(1) , Jo(2) , . . . , Jo(n) ), is a per-
mutation of the jobs in J such that Jo(i) ≺ Jo( j) implies i < j. Note that, given the
partially ordered set (J , ≺), a topological order of the jobs in (J , ≺) can be obtained
in O(n 2 ) time. Let J˜ be the set of jobs in J each of which has no successors under the
precedence relation ≺. Then, for each J j ∈ J , in a constant time, we can determine
either J j ∈ J˜ or not by considering its out degree in the corresponding directed graph.
It is helpful to notice that for a subset K ⊆ J , the restriction of the topological order
O(J ) on K, denoted O(J )|K , is also a topological order on (K, ≺).
For the classical scheduling problem 1|prec| f max , Lawler (1973) showed that, for
every job instance J , there is an optimal schedule such thatthe last job in the schedule
can be chosen as a job J j ∈ J˜ such that f j (τ ), where τ = Ji ∈J pi is the completion
time of the last job in every active schedule, is as small as possible. Based on this
property, Lawler (1973) presented an O(n 2 )-time algorithm for problem 1|prec| f max
by the rule that, for k from n to 1, always assign a job as the k-th job by the established
optimality property.

123
184 R. Chen, J. Yuan

Remark Debczynski and Gawiejnowicz (2013) formulated two properties, called

makespan invariantness and schedule non-idleness, which must be satisfied by a
scheduling problem with variable job processing times to be solvable by Lawler’s
algorithm. “Makespan invariantness” states that all feasible schedules have the same
makespan, and “schedule non-idleness” states that there exists no idle time for any fea-
sible schedule. Note that we only focus on active schedules, and so, the two properties
are also satisfied in our discussions.
We notice that Lawler’s approach can be efficiently used in the problems studied
in this paper after suitable extension. Let us consider a single-machine scheduling
problem 1| • | f , where f is nondecreasing in the completion times of the n jobs. In
the following, we use A(J ) to denote the set of jobs in J which can be processed in
the last position in an active schedule. We call A(J ) the set of available jobs at the
last position. For our purpose, we provide the following definition.
Definition 4.1 A scheduling problem 1| • | f has the Lawler Property if, on every
feasible job instance J of n jobs, the problem has the following properties:
(i) A(J ) can be obtained in O(n) time.
(ii) There is a Selection Rule on J , which in O(n) time finds a job J j ∈ A(J ) such
that there is an optimal active schedule σ of the jobs in J in which J j acts as the
last job.
Remark The Lawler Property established in Definition 4.1 will be used for solving
some problems of scheduling proportional-linearly deteriorating jobs with positional
due indices by using the idea of the algorithm for solving problem 1|prec| f max in
Lawler (1973). Such a method is not original in the application of the idea of Lawler’s
algorithm. In fact, the idea of Lawler’s algorithm has been widely used in the literature.
A successful example can be found in Debczynski and Gawiejnowicz (2013), in which
the authors addressed several single-machine scheduling problems with mixed job
processing times, arbitrary precedence constraints and maximum cost criterion.
Given the Lawler Property, problem 1|•| f can be solved by the following algorithm,
denoted by Lawler’s rule.
Lawler’s rule: For problem 1| • | f with the Lawler Property.
1: σ ← ∅
2: J ← {J1 , J2 , . . . , Jn }
3: l ← n
4: while l ≥ 1, do
5: Find a job J j ∈ A(J ) by applying Selection Rule on J
6: J ← J \{J j }
7: σ ← (J j , σ )
8: l ← l − 1
9: end while
10: return σ
Theorem 4.1 Suppose that problem 1| • | f has the Lawler Property. Then Lawler’s
rule solves problem 1| • | f in O(n 2 ) time.

123
Single-machine scheduling of proportional-linearly… 185

Table 2 Available jobs and the Selection Rule for problems P1–P4

A(J ) of P1 Jobs J j ∈ J˜ with k j ≥ |J | and C(J , τ0 ) ≤ d̄ j

Selection Rule of P1 Pick a job J j ∈ A(J ) with the minimum value f j (C(J , τ0 ))
A(J ) of P2 Jobs J j ∈ J˜ with k j ≥ |J |
Selection Rule of P2 Pick a job J j ∈ A(J ) with the maximum release date r j
A(J ) of P3 Jobs J j ∈ J˜ with k j ≥ |J |
Selection Rule of P3 Pick a job J j ∈ A(J ) with the maximum release date r j
A(J ) of P4 Jobs J j ∈ J with k j ≥ |J | and C(J , τ0 ) ≤ d̄ j
Selection Rule of P4 Pick a job J j ∈ A(J ) with the largest value α j

Proof Correctness of Lawler’s rule follows from the definitions of the available job
set A(J ) and the Selection Rule.
As for the time complexity, the algorithm has n iterations. In each iteration, from
items (i) and (ii) of the Lawler Property in Definition 4.1, the available job set A(J )
can be obtained in O(n) time and the Selection Rule find the last job J j ∈ A(J ) in an
optimal schedule in O(n) time. It follows that the total time complexity of Lawler’s
rule is given by O(n 2 ). 


We now apply Theorem 4.1 to the problems P1–P4 in Table 1 to obtain O(n 2 )-
time algorithms. In an instance of problem P2 (i.e., 1|r j , k j , p j (t) = α j (a +
bt), prec|Cmax ), for two jobs Ji and J j with Ji ≺ J j , Ji must be processed before J j
in every feasible schedule. Then we can modify the release dates of each job J j by
setting

r j := max{ri : i = j or Ji ≺ J j }.

This modification can be done in O(n 2 ) time and does not change the optimal schedules
and the optimal value. Then we take the following assumption for problem P2.

Assumption 1 For each pair of jobs Ji and J j , if Ji ≺ J j , then ri ≤ r j .

This modification for release dates is not suitable for problem 1|r j , k j , p j (t) =
α j (a + bt), prec|Fmax . In fact, we will show in Sect. 6 that the special version
1|r j , chains|Fmax is already unary NP-hard. Thus, problem P3 is defined to be the
problem 1|r j , k j , p j (t) = α j (a + bt), prec|Fmax under Assumption 1.
For the four problems P1–P4 on the job instance J , the available job set A(J ) and
the Selection Rule are presented in Table 2.
It can be observed that, in Table 2, the available jobs A(J ) for the four problems are
correctly chosen and can be determined in O(n) time. Moreover, each of the Selection
Rule for the four problems runs in O(n) time. It remains to show the correctness of
these decision methods.

Lemma 4.1 For each of the problems P1–P4, the job J j determined by the correspond-
ing Selection Rule is the last job in an optimal active schedule.

123
186 R. Chen, J. Yuan

Proof We first show the correctness of the decision method for problem P1. Let σ
be an optimal active schedule and suppose that the job J j determined by the corre-
sponding Selection Rule is not the last job in σ . Let Ji be the last job in σ . Since
both Ji and J j are available at the last position in σ , we have that Ji , J j ∈ J˜ ,
|J | ≤ min{ki , k j }, and C(J , τ0 ) ≤ min{d̄i , d̄ j }. The Selection Rule implies that
f j (C(J , τ0 )) ≤ f i (C(J , τ0 )). Let σ  be the active schedule obtained from σ by
shifting job J j to the last position and keeping the order of other jobs unchanged. It
is clear that σ  satisfies the precedence condition and σ  [J j ] = |J | ≤ k j . Moreover,
for every job Jl with l = j, we have σ  [Jl ] ≤ σ [Jl ] ≤ kl and Cl (σ  ) ≤ Cl (σ ) ≤ d̄l .
So fl (Cl (σ  )) ≤ fl (Cl (σ )) for any job Jl with l = j. From (2), we know that
C j (σ  ) = Ci (σ ) = C(J , τ0 ) ≤ d̄ j , and so, f j (C j (σ  )) ≤ f i (Ci (σ )). It follows that
σ  is a feasible schedule for P1 on instance J and f max (σ  ) ≤ f max (σ ). Consequently,
σ  is a required optimal active schedule.
For problems P2 and P3, their A(J ) and Selection Rule are completely the same.
Then the proof can be put together. Let σ be an optimal active schedule for P2 or P3
and suppose that the job J j determined by the corresponding Selection Rule is not the
last job in σ . Let Ji be the last job in σ . Since both Ji and J j are available at the last
position in σ , we have that Ji , J j ∈ J˜ and |J | ≤ min{ki , k j }. The Selection Rule
implies ri ≤ r j . Let σ  be the active schedule obtained from σ by shifting J j to the last
position and keeping the order of other jobs unchanged. It is clear that σ  satisfies the
precedence condition and σ  [J j ] = |J | ≤ k j . Moreover, for every job Jl with l = j,
we have σ  [Jl ] ≤ σ [Jl ] ≤ kl . This means that σ  is a feasible schedule for both P2
and P3 on instance J . Obviously, we have Cl (σ  ) ≤ Cl (σ ), and so, Fl (σ  ) ≤ Fl (σ )
for any job Jl with l = j. The remaining issue is to prove that C j (σ  ) ≤ Ci (σ ) and
F j (σ  ) ≤ Fi (σ ), which implies that Cmax (σ  ) ≤ Cmax (σ ) and Fmax (σ  ) ≤ Fmax (σ ).
Let t = S j (σ ) and let K = {Jl ∈ J : σ [Jl ] ≥ σ [J j ]}. We claim that rl ≤ t for all
Jl ∈ K.
To prove the claim, we pick Jl ∈ K arbitrarily and let Jl  ∈ J˜ such that either
Jl  = Jl or Jl ≺ Jl  . By the choice of J j , we have r j ≥ rl  . From Assumption 1, we
have rl ≤ rl  . Thus, we have rl ≤ rl  ≤ r j ≤ S j (σ ) = t. The claim follows.
From the above claim, the jobs in K are consecutively processed starting at time t.
From Eq. (2), we know that Cmax (σ  ) = Cmax (σ ) = C(K, t). The fact r j ≥ ri further
implies that F j (σ  ) = C(K, t) − r j ≤ C(K, t) − ri = Fi (σ ). It follows that σ  is a
required optimal active schedule.
Now we consider problem P4. Again, let σ be an optimal active schedule and
suppose that the job J j determined by the corresponding Selection Rule is not the
last job in σ . Let Ji be the last job in σ . Since both Ji and J j are available at the
last position in σ , we have that |J | ≤ min{ki , k j } and C(J , τ0 ) ≤ min{d̄i , d̄ j }. The
Selection Rule implies that α j ≥ αi . Let σ  be the active schedule obtained from σ
by exchanging the position of jobs Ji and J j . Then J j is the last job in σ  . Clearly, for
every job Jl , we have σ  [Jl ] ≤ kl . Moreover, from the fact α j ≥ αi and the Eq. (2),
we have Cl (σ  ) ≤ Cl (σ ) ≤ d̄l for l ∈ / {Ji , J j }, Ci (σ  ) ≤ C j (σ ) ≤ C(J , τ0 ) ≤ d̄i
and C j (σ ) = Ci (σ ) = C(J , τ0 ) ≤ d̄ j . This means that σ  is a feasible schedule


for P4 on instance J not worse than σ . Consequently, σ  is a required optimal active

schedule. 


123
Single-machine scheduling of proportional-linearly… 187

The above discussion means that the four problems P1–P4 have the Lawler Property.
From Theorem 4.1, we conclude the following result.

Theorem 4.2 Problems P1–P4 are solvable in O(n 2 ) time.

It is interesting to notice that, for problems P2 and P3 on a common job instance

J , the available job set A(J ) and the Selection Rule are completely the same. Then
we have the following corollary.

Corollary 4.1 Under Assumption 1, the two problems 1|r j , k j , p j = α j (a +

bt), prec|Cmax and 1|r j , k j , p j = α j (a + bt), prec|Fmax on a common job instance
have a common optimal active schedule.

It can be verify that the problems P5–P14 in Table 1 also have the Lawler Property,
and so, can be solved by the Lawler’s rule in O(n 2 ) time. But we will show in the
next section that these problems can be more efficiently solved by Smith’s rule in
O(n log n) time.

5 Problems solvable by Smith’s rule


For another classical scheduling problem 1|d̄ j | C j , Smith (1956) showed that, for
every job instance J , there is an optimal schedule such that the last job in the schedule
can be chosen as a job J j ∈ J with τ ≤ d̄ j , where τ = Ji ∈J pi is the completion
time of the last job in every active schedule, such that p j is as large as possible. Based
on this property, Smith (1956) presented an O(n log n)-time algorithm for problem
1|d̄ j | C j by the rule that, for k from n to 1, always assign a job as the k-th job by
the above optimality property.
Let us again consider a single-machine scheduling problem 1| • | f in which the
actual processing time of job J j is given by p j (t) = α j (a + bt). As in the last
section, we use A(J ) to denote the set of jobs in J which can be processed in the last
position in an active schedule. Moreover, we assume that, associated with problem
1| • | f there are two permutations: β-permutation Jβ = (Jβ(1) , Jβ(2) , . . . , Jβ(n) ) and
γ -permutation Jγ = (Jγ (1) , Jγ (2) , . . . , Jγ (n) ). Roughly speaking, the β-permutation
indicates that, if there are no restrictions, the active schedule obtained by processing
the jobs in the order Jβ = (Jβ(1) , Jβ(2) , . . . , Jβ(n) ) is optimal and the γ -permutation
indicates that the active schedule (if any) obtained by processing the jobs in the order
Jγ = (Jγ (1) , Jγ (2) , . . . , Jγ (n) ) always guarantees the feasibility.

Definition 5.1 A scheduling problem 1| • | f has the Smith Property if, on every job
instance J of n jobs, it has the following properties:
(i) The two associated permutations (Jβ(1) , Jβ(2) , . . . , Jβ(n) ) and (Jγ (1) , Jγ (2) , . . . ,
Jγ (n) ) can be generated in O(n log n) time.
(ii) There is a certain index x such that A(J ) = {Jγ (x) , Jγ (x+1) , . . . , Jγ (n) } and each
relationship J j ∈ A(J ) can be checked in a constant time.
(iii) There is an optimal active schedule σ such that the job Jβ( j) ∈ A(J ) with the
maximum index j is the last job in σ .

123
188 R. Chen, J. Yuan

(iv) For every job J j ∈ A(J ), the relation A(J )\{J j } ⊆ A(J \{J j }) holds.
Given Smith Property, problem 1| • | f can be solved by the following Smith’s rule.
Smith’s rule: For problem 1| • | f with the Smith Property.
– Initially, generate the two associated permutations Jβ = (Jβ(1) , Jβ(2) , . . . , Jβ(n) )
and Jγ = (Jγ (1) , Jγ (2) , . . . , Jγ (n) ). Set J (n) := {J1 , J2 , . . . , Jn }, R(n) :=
{J1 , J2 , . . . , Jn } and l := n. Here, R(l) denotes the set of jobs which are unscheduled
and unavailable up to iteration l.
– Calculate A(J (l) ) by finding the minimum index x with Jγ (x) ∈ A(J (l) ). Then
A(J (l) ) = {Jγ (x) , Jγ (x+1) , . . . , Jγ (l) }.
– If A(J (l) ) = ∅ and l ≥ 1, terminate the algorithm and output infeasibility.
– Find a job Jβ( j) ∈ A(J (l) ) such that j is as large as possible and assign Jβ( j) as
the l-th job in an optimal schedule.
– Reset J (l−1) := J (l) \{J j } and l := l − 1.
– Repeat the above procedure until either l = 0 or the algorithm claims the infea-
sibility.
By adding a simple date structure, we will show that Smith’s rule runs correctly in
O(n log n) time.
Theorem 5.1 Suppose that problem 1| • | f has the Smith Property. Then Smith’s rule
solves problem 1| • | f in O(n log n) time.
Proof Correctness of Smith’s rule follows from the definitions of the available job set
A(J ) and the two associated permutations Jβ and Jγ , and the item (iii) in Definition
5.1.
As for the time complexity, the algorithm has n iterations. Initially, the two associ-
ated permutations Jβ and Jγ are generated in O(n log n) time. To guarantee that the
Smith’s rule runs in O(n log n) time, we introduce a simple data structure. The jobs
in A(J (l) ) at each iteration l = n, n − 1, . . . , 1 are stored in a linked list L(l) which
sorts the jobs in A(J (l) ) in the same order as that in permutation Jβ . Moreover, the
(l)
unscheduled and unavailable jobs in R(l) are stored in the permutation Rγ , which is
(l)
the restriction of Jγ on R(l) , i.e., Rγ = Jγ |R(l) .
Initially, L(n) is set to be empty. At each iteration l, if the last job in R(l) γ , say J y ,
is available at position l, we delete Jy from R(l) γ and insert J y into the list L(l) until
this procedure cannot be implemented further. From item (ii) in Definition 5.1, the
resulted list L(l) is just the required sorting of the jobs in A(J (l) ). Then the last job
in L(l), say J j , is deleted from the list and is scheduled at the l-th position. From the
item (iv) in Definition 5.1, L(l)\{J j } takes as a sublist in iteration l − 1.
Thus, the Smith’s rule in fact consists of the following operations: n jobs are deleted
from the permutation Jγ from the last one and one by one, n jobs are inserted into
a dynamic linked list L(l) one by one, and n jobs are scheduled one by one. Each
insertion or deletion needs a time at most O(log n) by the binary search, and the
scheduling of each job needs a constant time. Then the total time complexity of the
Smith’s rule is given by O(n log n). 

In the following we show that problems P5–P14 can be solved by the Smith’s
rule. The β-permutations and γ -permutations of these problems are given in Table 3,

123
Single-machine scheduling of proportional-linearly… 189

Table 3 The two associated permutations for problems P5–P14

PX Scheduling problems β-permutations γ -permutations


P5 1|k j , p j (t)| C j αβ(1) ≤ αβ(2) ≤ · · · ≤ αβ(n) kγ (1) ≤ kγ (2) ≤ · · · ≤ kγ (n)

P6 1| p j (t), d̄ j | C j αβ(1) ≤ αβ(2) ≤ · · · ≤ αβ(n) d̄γ (1) ≤ d̄γ (2) ≤ · · · ≤ d̄γ (n)
P7 1|k j , p j (t)|L max dβ(1) ≤ dβ(2) ≤ · · · ≤ dβ(n) kγ (1) ≤ kγ (2) ≤ · · · ≤ kγ (n)
P8 1|k j , p j (t)|Tmax dβ(1) ≤ dβ(2) ≤ · · · ≤ dβ(n) kγ (1) ≤ kγ (2) ≤ · · · ≤ kγ (n)
P9 1|k j , p j (t)|W Cmax wβ(1) ≥ wβ(2) ≥ · · · ≥ wβ(n) kγ (1) ≤ kγ (2) ≤ · · · ≤ kγ (n)
P10 1| p j (t), d̄ j |L max dβ(1) ≤ dβ(2) ≤ · · · ≤ dβ(n) d̄γ (1) ≤ d̄γ (2) ≤ · · · ≤ d̄γ (n)
P11 1| p j (t), d̄ j |Tmax dβ(1) ≤ dβ(2) ≤ · · · ≤ dβ(n) d̄γ (1) ≤ d̄γ (2) ≤ · · · ≤ d̄γ (n)
P12 1| p j (t), d̄ j |W Cmax wβ(1) ≥ wβ(2) ≥ · · · ≥ wβ(n) d̄γ (1) ≤ d̄γ (2) ≤ · · · ≤ d̄γ (n)
P13 1|r j , k j , p j (t)|Cmax rβ(1) ≤ rβ(2) ≤ · · · ≤ rβ(n) kγ (1) ≤ kγ (2) ≤ · · · ≤ kγ (n)
P14 1|r j , k j , p j (t)|Fmax rβ(1) ≤ rβ(2) ≤ · · · ≤ rβ(n) kγ (1) ≤ kγ (2) ≤ · · · ≤ kγ (n)

where p j (t) = α j (a + bt). It is routine to see that, with these β-permutations and γ -
permutations in hand, problems P5–P14 satisfy the items (i), (ii) and (iv) in Definition
5.1. Here, it should be noticed that, for problems P5–P12, all jobs are released at time
τ0 , and so, the completion time of the last job C(J , τ0 ) can be calculated in O(n) time
in the preprocessing procedure. At each iteration to perform the Smith’s rule, when
the last job J j is determined, the updated value C(J \{J j }, τ0 ) can be calculated by
using the following formula in a constant time:

C(J , τ0 ) + a
b a
C(J \{J j }, τ0 ) = − .
1 + bα j b

The correctness of this formula is guaranteed by Eq. (2). It remains to show that the
item (iii) in Definition 5.1 also holds for each problem.
Lemma 5.1 For each of the problems P5–P14, the job Jβ( j) ∈ A(J ) with the maximum
index j is the last job in an optimal active schedule.
Proof Let X ∈ {5, 6, . . . , 14}. For  X ∈ {5, 6}, PX is a subproblem of problem P4,
i.e., 1|k j , p j (t) = α j (a + bt), d̄ j | C j . According to the β-permutation of PX in
Table 3, Jβ( j) is the job in A(J ) so that the normal processing time αβ( j) is of the
maximum. This coincides with the Selection Rule for problem P4 in Sect. 4. Hence,
Jβ( j) is the last job in an optimal active schedule.
For X ∈ {7, 8, . . . , 12}, PX is a subproblem of problem P1, i.e., 1|k j , p j (t) =
α j (a + bt), d̄ j , prec| f max . According to the β-permutation of PX in Table 3, Jβ( j) is
the job in A(J ) so that the f β( j) (C(J , τ0 )) is as small as possible. This coincides
with the Selection Rule for problem P1 in Sect. 4. Hence, Jβ( j) is the last job in an
optimal active schedule.
Problems P13 and P14 are the subproblems of P2 and P3, respectively. According
to the β-permutation of PX in Table 3, Jβ( j) is the job in A(J ) so that the rβ( j) is as
large as possible. This coincides with the Selection Rule for problems P2 and P3 in
Sect. 4. Hence, Jβ( j) is the last job in an optimal active schedule. 


123
190 R. Chen, J. Yuan

The above discussion means that the problems P5–P14 have the Smith Property.
From Theorem 5.1, we conclude the following result.
Theorem 5.2 Problems P5–P14 are solvable in O(n log n) time.

6 NP-hardness proof

By using the unary NP-complete 3-Partition problem in Garey and Johnson

(1979) for reduction, we first show in this section that problems 1|r j |W Cmax and
1|r j , chains|Fmax are unary NP-hard.
3-Partition: Given a set of 3t positive integers a1 , a2 , . . . , a3t and a positive integer
B such that 3tj=1 a j = t B and B/4 < a j < B/2 for each j with 1 ≤ j ≤ 3t, does
thereexist a partition (I1 , I2 , . . . , It ) of the index set {1, 2, . . . , 3t}, such that |Ii | = 3
and j∈Ii a j = B for each i with 1 ≤ i ≤ t?
Theorem 6.1 Problem 1|r j |W Cmax is unary NP-hard.
Proof The decision version of problem 1|r j |W Cmax is clearly in NP. For a given 3-
Partition instance (a1 , a2 , . . . , a3t ; B), we construct an instance of the decision version
of problem 1|r j |W Cmax as follows.
• We have n = 4t jobs J1 , J2 , . . . , Jt and J1 , J2 , . . . , J3t of two types.
• J1 , J2 , . . . , Jt are t separation jobs with r j = j B + j − 1, p j = 1, and w j =
j B+ j , for j = 1, 2, . . . , t.
t B+t−1

• J1 , J2 , . . . , J3t are 3t normal jobs with r j = 0, p j = a j , and w j = 1, for

j = 1, 2, . . . , 3t. Note that the 3t normal jobs correspond to the 3t numbers
a1 , a2 , . . . , a3t in the 3-Partition instance.
• The threshold value is defined by Y = t B + t − 1.
• The decision asks whether there is a feasible schedule π such that W Cmax (π ) ≤ Y .
It can be observed that the above construction can be done in polynomial time
under unary encoding. From the above construction, each separation job Ji with 1 ≤
i ≤ t satisfies wi (ri + pi ) = t B + t − 1 = Y . Thus, in a feasible schedule π with
W Cmax (π ) ≤ Y , all the t separation jobs are started at their release dates. This leaves
t time slots, each of length B, to schedule the 3t normal jobs corresponding to the
3t numbers a1 , a2 , . . . , a3t in the 3-Partition instance. Therefore, it can be observed
that the 3-Partition instance has a solution if and only if there is a feasible schedule π
of the job instance such that W Cmax (π ) ≤ Y . It follows that problem 1|r j |W Cmax is
unary NP-hard. 

Theorem 6.2 Problem 1|r j , chains|Fmax is unary NP-hard.
Proof The decision version of problem 1|r j , chains|Fmax is clearly in NP. For a given 3-
Partition instance (a1 , a2 , . . . , a3t ; B), we construct an instance of the decision version
of problem 1|r j , chains|Fmax as follows.
• We have n = 5t jobs J1 , J2 , . . . , Jt , J1 , J2 , . . . , Jt and J1 , J2 , . . . , J3t of two
types.

123
Single-machine scheduling of proportional-linearly… 191

• J1 , J2 , . . . , Jt and J1 , J2 , . . . , Jt are 2t separation jobs with chain relations Ji ≺
Ji for i = 1, 2, . . . , t. Their release dates and processing times are given by
ri = (t + i − 1)B, ri = (i − 1)B, and pi = pi = 0, i = 1, 2, . . . , t.
• J1 , J2 , . . . , J3t are 3t normal jobs with r j = (t − 1)B and p j = a j ,
j = 1, 2, . . . , 3t. Note that the 3t normal jobs correspond to the 3t numbers
a1 , a2 , . . . , a3t in the 3-Partition instance.
• The threshold value is defined by Y = t B.
• The decision asks whether there is a feasible schedule π such that Fmax (π ) ≤ Y .
It can be observed that the above construction can be done in polynomial time under
unary encoding. From the above construction, for each pair of separation jobs Ji and
Ji with 1 ≤ i ≤ t, we have ri = (t + i − 1)B = ri + Y . Thus, from the chain relation
Ji ≺ Ji , in a feasible schedule π with Fmax (π ) ≤ Y , all the t pairs separation jobs
Ji and Ji are started and completed at time ri = (t + i − 1)B, i = 1, 2, . . . , t. This
leaves t time slots, each of length B, from the common release dates (t − 1)B of the
normal jobs to time (2t − 1)B to schedule the 3t normal jobs corresponding to the 3t
numbers a1 , a2 , . . . , a3t in the 3-Partition instance. Therefore, it can be observed that
the 3-Partition instance has a solution if and only if there is a feasible schedule π of
the job instance such that Fmax (π ) ≤ Y . It follows that problem 1|r j , chains|Fmax is
unary NP-hard. 


Theorem 6.3 Problem 1|k j | U j is unary NP-hard.
Proof  By using 3-Partition for the reduction, Yuan (2017) showed that problem
1|d̄ j | U j is unary NP-hard. The proof in Yuan (2017) in fact implies that, on
special instances,
 called ( , D)-dominated instances, the decision version of prob-
lem 1|d̄ j | U j is still unary NP-complete. We first describe the ( , D)-dominated
instances. 
For a job instance J = {J j = ( p j , d j , d̄ j ) : 1 ≤ j ≤n} of problem 1|d̄ j | U j ,
we use U ∗ (J ) to denote the optimal value of problem 1|| U j on instance J without
considering the deadlines. By Moore (1968), the value U ∗ (J ) can be calculated in
O(n log n) time.

Definition 6.1 An instance of the decision version of problem 1|d̄ j | U j is called
( , D)-dominated, if the job instance J = {J j = ( p j , d j , d̄ j ) : 1 ≤ j ≤ n} and its
decision satisfy the following conditions:

is a sufficiently large positive integer so that p j > for all jobs J j ∈ J
(C1) There
and nj=1 p j < (n + 1) .
(C2) max{d j : 1 ≤ j ≤ n} < min{ d̄ j : 1 ≤ j ≤ n}.
(C3) max{d̄ j : 1 ≤ j ≤ n} = nj=1 p j and for two distinct deadlines d̄i < d̄ j , we
have d̄ j > d̄i + 2 . 
 is a feasible∗ schedule π of problem 1|d̄ j | U j
(C4) The decision asks whether there
on the instance J such that U j (π ) ≤ U (J ).
From the instance construction in Yuan (2017), we have the following observation.

Observation 1 The decision version of problem 1|d̄ j | U j is unary NP-complete
even for ( , D)-dominated instances.

123
192 R. Chen, J. Yuan

Given a ( ,D)-dominated instance J = {J j = ( p j , d j , d̄ j ) : 1 ≤ j ≤ n} of

problem 1|d̄ j | U j , let D (1) < D (2) < · · · < D (m) be the distinct deadlines and let
J (i) = {J j ∈ J : d̄ j = D (i) }, i = 1, 2, . . . , m. We use λ(J j ) to denote the deadline
index of job J j . Then λ(J j ) = i means that  d̄ j = D (i) , or equivalently, J j ∈ J (i) .
We construct an instance K of problem 1|k j | U j in the following way.
• We have n + m jobs of two types: n normal jobs J j , 1 ≤ j ≤ n, and m restriction
jobs J (i) , 1 ≤ i ≤ m.
• Each normal job J j , which corresponds to the job J j in the ( , D)-dominated
instance, has a processing time p j = p j , a due date d j = d j and a due index
d̄
k j =  j  + λ(J j ) − 1.
• Each restriction job J (i) , which corresponds to the i-th deadline D (i) in the ( , D)-
dominated instance, has a processing time p (i) = 1, a due date d (i) = D (i) + i
(i)
and a due index k (i) =  D  + i. 
 schedule ∗σ of problem 1|k j | U j
• The decision asks whether there is a feasible
on the constructed instance K such that U j (σ ) ≤ U (J ).
It can be observed that the above construction can be done in polynomial time
under unary encoding. In a schedule σ of the jobs in K, we will use C j (σ ) to denote
the completion time of a normal job J j , and use C (i) (σ ) to denote the completion
time of a restriction job J (i) . Moreover, U j (σ ) and U (i) (σ ) are used to denote the
tardy indicator numbers of J j and J (i) , respectively. We show in the following that

the decision version of problem 1|d̄ j | U j  on the instance J has a solution if and
only if the decision version of problem 1|k j | U j on the constructed instance K has
a solution. 
Suppose first that there is a feasible schedule π of problem 1|d̄ j | U j on the
instance J such that U j (π ) ≤ U ∗ (J ). From condition (C2) in Definition 6.1, we
may assume that all the early jobs are scheduled before all the tardy jobs. Moreover,
from condition (C1) and the deadline restriction C j (π ) ≤ d̄ j = D (λ(J j )) , we have
   
d̄ j D (λ(J j ))
π [J j ] ≤ = for all J j ∈ J . (3)

We obtain a schedule σ of instance K by the following way:

(i) Beginning from schedule π , replace each job J j in J by its corresponding job
J j in K to obtain a partial schedule σ  of instance K;
(ii) Beginning from schedule σ  , find the m jobs J j1 , J j2 , . . . , J jm which are the last
jobs completed before the deadlines D (1) , D (2) , . . . , D (m) , respectively, in σ  .
From condition (C3) in Definition 6.1, the m jobs J j1 , J j2 , . . . , J jm are distinct and
well-defined. Next, for i from 1 to m, insert the restriction job J (i) corresponding
to the deadline D (i) after job J ji . Then we obtain our schedule σ of instance K.

Now we prove that σ ∗is a feasible
schedule of problem
 1|k j | U j on instance K
such that U j (σ ) ≤ U (J ), i.e., j=1 U j (σ ) + i=1 U (i) (σ ) ≤ U ∗ (J ).
n m

123
Single-machine scheduling of proportional-linearly… 193

For each normal job J j ∈ K, from Eq. (3) and the construction of σ  , we have
  (λ(J j ))
d̄
σ  [J j ] ≤ j = D . From the construction of σ , at most λ(J j ) − 1 restriction

d̄
jobs are inserted before J j . Thus, σ [J j ] ≤ j + λ(J j ) − 1 = k j , and so, J j satisfies
its position constraint in σ .
For each restriction job J (i) ∈ K, the last job processed before J (i) in σ is given
by J ji . Since C ji (π ) ≤ D (i) , from condition (C1) and the construction of σ  again,
 (i)
we have σ  [J ji ] = π [J ji ] ≤ D . From the construction of σ , we have σ [J (i) ] =
 (i)
σ  [J ji ] + i ≤ D + i = k (i) , and so, J (i) satisfies its position constraint in σ .
The above discussion implies that σ is a feasible schedule of instance K. To estimate
the value U j (σ ) = nj=1 U j (σ ) + i=1 m
U (i) (σ ), we notice that J j is early in π if

and only J j is early in σ , and so, in σ . Moreover, every restriction job J (i) is early in σ ,


since the completion time of J j(i) in σ is up bounded by C ji (π ) + i ≤ D (i) + i = d (i) .

 
It follows that U j (σ ) = U j (π ) ≤ U ∗ (J ), as required. 
that there is a∗ feasible schedule σ for problem 1|k j | U j on
Conversely, suppose
instance K such that U j (σ ) ≤ U (J ). We obtain a schedule π of instance J by
the following way: Beginning from schedule σ , delete all the restriction jobs from σ
and replace each job J j in K by its corresponding job J j in J to obtain our schedule
π of instance J . Note that, for j = 1, 2, . . . , n, the normal job J j ∈ K has the same
processing time p j and the same due date d j as the job J j ∈ J . Thus, from the
construction of π , we have C j (π ) ≤ C j (σ ) and U j (π ) ≤ U j (σ ) for j = 1, 2, . . . , n.
This means that

U j (π ) = U j (π ) ≤ U j (σ ) ≤ U j (σ ) ≤ U ∗ (J ). (4)
j=1 j=1


Since U ∗ (J ) is the optimal value of problem
 1|| U j ∗on instance J without con-
sidering the deadlines, we further
 have U j (π ) ≥ U (J ). By putting Eq. (4) in
consideration, we conclude that U j (π ) = U ∗ (J ) and all the inequalities in Eq. (4)
hold with equalities. Consequently, all the restriction jobs are early in σ , that is

C (i) (σ ) ≤ d (i) = D (i) + i for 1 ≤ i ≤ m. (5)

The remaining issue is to show that every job in J meets its deadline in π .
Suppose on the contrary that Jx is a job in J so that C x (π ) > d̄x = D (y) , where
y = λ(Jx ). From condition (C1), we have

 
D (y)
π [Jx ] ≥ . (6)

123
194 R. Chen, J. Yuan

Since σ is a feasible schedule of K, we have

 
D (y)
σ [Jx ] ≤ k x = + y − 1. (7)

From Eqs. (6) and (7) and putting the relation of π and σ in consideration, we conclude
that at most y − 1 restriction jobs in K are processed before job Jx in σ . Let R(y) =
{J (i) : 1 ≤ i ≤ y}. Since there are totally y restriction jobs in R(y) , at least one
job in R(y) are scheduled after Jx in σ . Let J (i) be the last jobs of R(y) in σ . Then
d (i) = D (i) + i ≤ D (y) + y and C (i) (σ ) ≥ C x (π ) + y > D (y) + y. This leads to a
contradiction. Thus, every job in J meets its deadline in π . 
From the above discussion,
 π is a feasible schedule for problem 1|d̄ j | U j on
instance J such that U j (π ) ≤ U ∗ (J ). Then we conclude that problem 1|k j | U j
is unary NP-hard. 

Note that, the above discussions are taken for the processing times p j (t) = α j (a +
bt) with b = 0. Then we have the following corollary.

Corollary 6.1 The three problems 1|r j , p j (t) = α j (a +bt)|W Cmax , 1|r j , p j (t) =
α j (a + bt), chains|Fmax and 1|k j , p j (t) = α j (a + bt)| U j are unary NP-hard.

7 Conclusions

In this paper, we studied the scheduling of proportional-linearly deteriorating jobs

with positional due indices, release dates, deadlines and precedence relations on a
single machine. The scheduling criteria studied in this paper include the makespan
Cmax , maximum lateness L max , maximum tardiness Tmax , maximum flow time Fmax ,
maximum weighted  complete time W Cmax , maximum  scheduling cost f max , total
completion time C j , and number of tardy jobs U j . By applying Lawler’s rule
and Smith’s rule, polynomially solvable problems are processed by using two unified
methods. We also present some new NP-hardness results when processing times of
the jobs have no deterioration. Our results generalize a series of known achievements
in the literature.
For further research, it is worthy of study to present polynomial-time algorithms  for
the following problems with time complexity better than O(n 2 ): 1|k j , p j (t), d̄ j | C j ,
1|k j , p j (t), d̄ j |L max , 1|k j , p j (t), d̄ j |Tmax , and 1|k j , p j (t), d̄ j |W Cmax . In our opin-
ion, O(n(log n)2 )-time algorithms are expected for these problems. Moreover, the
corresponding research on linearly deteriorating jobs with positional due indices is
worthy of a deep study. It is also interesting and valuable to study the scheduling
models with generalized due dates (GDD) and positional due indices, where the GDD
assumption was first introduced in Hall (1986) and has been widely studied in the
literature (without the positional due indices).

Acknowledgements The authors would like to thank the associate editor and two anonymous referees for
their constructive comments and helpful suggestions. This research was supported by NSFC (11671368,
11771406).

123
Single-machine scheduling of proportional-linearly… 195

Compliance with ethical standards

Conflict of interest The authors declare that there is no conflict of interests regarding the publication of this
paper.

References
Agnetis A, Billaut JC, Gawiejnowicz S, Pacciarelli D, Soukhal A (2014) Multiagent scheduling: models
and algorithms. Springer, Berlin
Brauner N, Finke G, Shafransky Y, Sledneu D (2016) Lawler’s minmax cost algorithm: optimality conditions
and uncertainty. J Sched 19:401–408
Brauner N, Finke G, Shafransky Y (2017) Lawler’s minmax cost problem under uncertainty. J Comb Optim
34:31–46
Chen QQ, Lin L, Tan ZY, Yan YJ (2017) Coordination mechanisms for scheduling games with proportional
deterioration. Eur J Oper Res 263:380–389
Cheng YS, Sun SJ (2009) Scheduling linear deteriorating jobs with rejection on a single machine. Eur J
Oper Res 194:18–27
Debczynski M, Gawiejnowicz S (2013) Scheduling jobs with mixed processing times, arbitrary precedence
constraints and maximum cost criterion. Comput Ind Eng 64:273–279
Gao Y, Yuan JJ (2015) Pareto minimizing total completion time and maximum cost with positional due
indices. J Oper Res Soc China 3:381–387
Gao Y, Yuan JJ (2017) Bi-criteria Pareto-scheduling on a single machine with due indices and precedence
constraints. Discrete Optim 25:105–119
Garey MR, Johnson DS (1979) Computers and intractability: a guide to the theory of NP-completeness.
Freeman, San Francisco
Gawiejnowicz S (2007) Scheduling deteriorating jobs subject to job or machine availability constraints. Eur
J Oper Res 180:472–478
Gawiejnowicz S (2008) Time-dependent scheduling. Springer, Berlin
Gawiejnowicz S, Kononov A (2014) Isomorphic scheduling problems. Ann Oper Res 213:131–145
Gawiejnowicz S, Kurc W, Pankowska L (2009a) Equivalent time-dependent scheduling problems. Eur J
Oper Res 196:919–929
Gawiejnowicz S, Kurc W, Pankowska L (2009b) Conjugate problems in time-dependent scheduling. J Sched
12:543–553
Hall NG (1986) Scheduling problems with generalized due dates. IIE Trans 18:220–222
Karp RM (1972) Reducibility among combinatorial problems. In: Miller RE, Thatcher JW, Bohlinger JD
(eds) Complexity of computer computations. Springer, Boston, MA, pp 85–103
Kononov A (1998) Single machine scheduling problems with processing times proportional to an arbitrary
function. Discrete Anal Oper Res 5:17–37 (in Russian)
Lawler EL (1983) Scheduling a single machine to minimize the number of late jobs. Report No. UCB/CSD
83/139, Computer Science Division, University of California, Berkeley, USA
Lawler EL (1973) Optimal sequencing of a single machine subject to precedence constraints. Manag Sci
19:544–546
Lawler EL, Moore JM (1969) A functional equation and its applications to resource allocation and sequenc-
ing problems. Manag Sci 16:77–84
Lenstra JK, Rinnooy AHG, Kan AHG, Brucker P (1977) Complexity of machine scheduling problems. Ann
Oper Res 1:343–362
Li YQ, Li G, Sun LY, Xu ZY (2009) Single machine scheduling of deteriorating jobs to minimize total
absolute differences in completion times. Int J Prod Econ 118:424–429
Li KL, Liu CB, Li KQ (2014) An approximation algorithm based on game theory for scheduling simple
linear deteriorating jobs. Theor Comput Sci 543:46–51
Liu P, Yi N, Zhou XY (2011) Two-agent single-machine scheduling problems under increasing linear
deterioration. Appl Math Model 35:2290–2296
Ma R, Tao JP, Yuan JJ (2016) Online scheduling with linear deteriorating jobs to minimize the total weighted
completion time. Appl Math Comput 273:570–583
Melnikov OI, Shafransky YM (1979) Parametric problem in scheduling theory. Cybernetics 15:352–357

123
196 R. Chen, J. Yuan

Miao CX (2018) Complexity of scheduling with proportional deterioration and release dates. Iran J Sci
Technol Trans Sci 42:1337–1342
Miao CX, Zhang YZ, Wu CL (2012) Scheduling of deteriorating jobs with release dates to minimize the
maximum lateness. Theor Comput Sci 462:80–87
Moore JM (1968) An n job, one machine sequencing algorithm for minimizing the number of late jobs.
Manag Sci 15:102–109
Mosheiov G (1994) Scheduling deteriorating jobs under simple linear deterioration. Comput Oper Res
21:653–659
Ng CT, Li SS, Cheng TCE, Yuan JJ (2010) Preemptive scheduling with simple linear deterioration on a
single machine. Theor Comput Sci 411:3578–3586
Qi XL, Zhou SG, Yuan JJ (2009) Single machine parallel-batch scheduling with deteriorating jobs. Theor
Comput Sci 410:830–836
Smith WE (1956) Various optimizers for single-stage production. Naval Res Logist 3:59–66
Strusevich V, Rustogi K (2017) Scheduling with times-changing effects and rate-modifying activities.
Springer, Berlin
Wu CC, Cheng SR, Wu WH, Yin YQ, Wu WH (2013) The single-machine total tardiness problem with
unequal release times and a linear deterioration. Appl Math Comput 219:10401–10415
Yu S, Wong PWH (2013) Online scheduling of simple linear deteriorating jobs to minimize the total general
completion time. Theor Comput Sci 487:95–102
Yuan JJ (2017) Unary NP-hardness of minimizing the number of tardy jobs with deadlines. J Sched 20:211–
218
Zhao QL, Yuan JJ (2017) Rescheduling to minimize the maximum lateness under the sequence disruptions
of original jobs. Asia Pac J Oper Res 34:12, Article ID: 1750024
Zhao CL, Zhang QL, Tang HY (2003) Scheduling problems under linear deterioration. Acta Autom Sin
29:531–535
Zhao QL, Lu LF, Yuan JJ (2016) Rescheduling with new orders and general maximum allowable time
disruptions. 4OR Q J Oper Res 14:261–280

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations.

123
 Reproduced with permission of copyright owner.
Further reproduction prohibited without permission.