Some guidelines to solve multicriteria scheduling problems

V. Tkindt & J.-C. Billaut

Laboratoire dInformatique, Ecole dIng6nieurs en Informatique pour lIndustrie, 64 avenue Jean Portalis, 37200 TOURS, FRANCE tkindtQe3i.univ-tours.fr ABSTRACT
In this paper we present a state-of-the-art survey about multicriteria scheduling. We introduce a definition of a multicriteria scheduling problem. This problem is decomposed in a modelization problem, a taking into consideration of the criteria problem and a final scheduling problem. A notation for multicriteria scheduling problems is proposed. Then, some basic results of multicriteria optimization literature are presented. These results are used to build the final scheduling problem. Then a survey is presented for one-machine, parallel machines and flowshop multicriteria scheduling problems. section 4 we present a brief survey on multicriteria scheduling problems tackled in the literature.

2.TYPOLOGY AND NOTATION

The minimization of several conflictuous criteria change the way to handle scheduling problems. Generally, a solution that minimizes optimally all the criteria does not exist. This implies that a new definition of optimality must be used, namely the definition of Pareto optimality. Two main definitions are encountered in multicriteria optimization literature: the definitions of weak and strict Pareto optimality. Definition 1 Let I C conflictuous criteria Z i t o minimize. S is the set of solutions and Z its image in the criteria space. x E S is a weak Pareto optimum (or a weak efficient solution) iff $ y E S,x # y, such that &(y) < Zi(x), V i = 1.X. W E is the set of all weak Pareto optima and ZWE its image in the criteria space (i.e. the set of weakly non dominated criteria vectors). Definition 2 x E S is a stract Pareto optimum (or an efficient solution) iff $ y E S,x # y, such that Zj(y) 5 Zi(x), V i = l..K, with at least one strict inequality. E is the set of all strict Pareto optania and ZE its image in the criteria space (i.e. the set of non dominated criteria vectors). One has E C W E . For the decision maker, only the strict Pareto optima are interesting, but unfortunately most of the results available computes a subset (or the entire set) of W E . A multicriteria scheduling problem consists in computing for the decision maker a schedule that minimizes several criteria. This schedule is necessarily a strict Pareto optimum. The study man is the person who designs an algorithm for the multicriteria scheduling problem. For him, this problem is decomposed in at least three problems:
1. a modelisation problem where the study man interacts with the decision maker to defiiie the criteria, the set of solutions considered, ... (see for instance ~511,

1.INTRODUCTION
Scheduling theory appeared in years 1950. Since this time, problems become more and more complex due to the numerous practical constraints they want to take into account. Surprisingly the most important part of literature on scheduling problems is dedicated to monocriterion problems whereas, in practice, the use of multiple criteria often allows to compute a more realistic solution for the decision maker. A survey on multicriteria one-machine scheduling problems can be found in [7, 9, 131. They show that three kinds of problems have been tackled. The first one deals with problems in which a lexicographical order of criteria is minimized. The second class of problems considers a convex combination of criteria. The third class of problems concerns the determination of all strict Pareto optima. Nagar, Haddock and Heragu ([23]) present a similar survey for multimachine scheduling problems. Some results concerning the complexity of multicriteria scheduling problems are presented in [4, 18, 51. It appears that the consideration of more than one criterion does not simplify the resolution of the scheduling problem. In terms of complexity, it means that multicriteria scheduling problems are at least as difficult as monocriterion scheduling problems. In this paper we try to put in evidence the interest of multicriteria optimization theory to solve multicriteria scheduling problems. In section 2 we introduce a typology and a notation for the multicriteria scheduling problems. In section 3 some basic results of multicriteria optimization field are introduced. In

2. the taking into consideration of criteria problem, where the study man interacts with the decision maker to fix the resolution context (i.e. the type of the resolution method) and to define the shape of the objective function of the scheduling problem,

3 . a sch.pduling problem where the study man has to

0-7803-5731-0/99/$10.00 01999 IEEE

VI-463

determine a schedule that optimizes the determined objective function. The obtained solution is a strict Pareto optimum. In the second problem the study man has to determine the resolution context. This is due to the fact that two strict Pareto optima are not comparable, so to choose the final solution the decision maker must help the resolution algorithm. Basically, Evans ([8]) distinguishes three kinds of resolution context. In the first one, the decision maker gives all the necessary informations to the resolution algorithm (for instance the criteria weights if a linear combination of criteria is minimized). This algorithm allows to compute a unique strict Pareto optimum. Such an algorithm is called an a priori algorithm. The algorithms of the second context are interactive algorithms that allow the decision maker to determine interactively its most prefered solution. In the third context, the resolution algorithm determines all the elements of the set E and let the decision maker choose the strict Pareto optimum he prefers. In this case, we talk about an a posteriori algorithm. In practice these three kinds of algorithms are made of two modules: the first one is the interface with the decision maker. This module realizes the settings of the parameters of the objective function. The second module is dedicated to the resolution of the scheduling problem. For instance if a convex combination of criteria is minimized by an a posteriori algorithm, its first module computes all possible weights for the criteria. For each value it runs the second module that solves the corresponding scheduling problem. After the choice of the resolution context, one has to use a multicriteria optimization result to build an objective function for the scheduling problem. Before presenting an overview of such results we propose a notation for multicriteria scheduling problems based on the one presented in [2]. This notation is decomposed into three fields ~ / / 3 / ywhere the field y is dedicated to the criteria. The existing notations does not allow t o take account of all the possible ways to minimize some criteria. The different values we propose for the field y are:
0

FT~(Z ..., ~Z , K ) , if the aim is to minimize the distance to a known ideal solution using a weighted Tchebycheff metric,

FT~~( ..., Z Z~ K) ,,if the aim is to minimize the distance to a known ideal solution using an augmented weighted Tchebycheff metric,
F,(Z1, ..., Z K ) ,if the aim is to maximize a special function that takes into account an ideal solution (satisfiability objective approach),
L e x ( Z 1 , ..., ZK), if the aim is to minimize a lexicographical order of all the criteria, i.e. to minimize criterion 2 1 , then to minimize criterion 2 2 subject to the optiniality of criterion 2 1 , etc. The order in the Lex operator is the lexicographical order considered,
2 1 , ..., Z K , denotes the general multicriteria problem. Generally this approach is associated with an a posteriori resolution and consists in enumerating the set E .

All these values refer to the different multicriteria optimization results presented in section 3. The possible values for Q and /3 fields are the classical ones

(PI).
3.AN OVERVIEW O F MULTICRITERIA OPTIMIZATION RESULTS
Many results are available in multicriteria optimization literature. The aim of this section is to provide general guidelines about the most commonly used results. Steuer ([as]) presents a detailed review on some of these results (mainly for linear problems). Different approaches can be considered to take into account the criteria, depending on some difierent factors: Are the tradeoff between criteria allowed ? Is it possible to associate a weight to each criterion ? Is it possible to associate a particular goal value to each criterion ? Does an upper bound exist for each criterion ? Depending on the approach one factor or more can be taken into account. Several results are then available to compute Pareto optima. One can note that generally the different approaches transform the original multicriteria problem into a new monocriterion problem. In this one, the objective function is an aggregation of the different criteria with new parameters. Geoffrion ([ll]) proposes a result when a convex combination of criteria is minimized. This result allows t o compute the set of proper Pareto optima (set PRE), that is a subset of E. Theorem 1 [ll] Let S be the convex set of solutions and I< criteria Zi that are convex functions on S . T O is a proper Pareto ai = 1 optimum i 8 3 a E IRK, with ai.E]O;1[ and such that x o is an optimal solution of problem (Pa):

2, if the aim is to minimize only criterion 2,

Fe(Z1, ..., Z K ) , if the aim is to minimize a linear combination of I< criteria,

P(Z1, ..., Z K ) , is a non decreasing function of the criteria, that are subjected to bound constraints (parametiic approach),
E ( Z , / Z ~ ,. .. , Zu-l, Zu+l,...,Z K ) ,if the criterion to minimize is Z,, and the other criteria are subjected to bound constraints (-constraint approach),
F ~ ( z 1..., , Z K ) , if the aim is to minimize the distance to a known ideal solution using a Tchebycheff metric,

xEl

V I - 464

Min g ( Z ( x ) ) with g ( Z ( x ) )= ai&(.) subject to xES P R E is the set o f proper Pareto optima and one has P R E g E. For problems that can be modeled by a mixte integer or linear program with a finite number of constraints one has P R E = E ([as]). If the weights ai can be equal to 0 or 1 then the set W E can be computed by theorem 1. At last if the convex hypothesis do not hold then only the sufficient condition of the theorem is true. It means that every solution of problem ( P o ) is a strict or weak Pareto optima (depending if the ai can be equal to 0 and 1 or not). These computable solutions are called the supported Pareto optima and belong to the border of the convex hull of 2. The other Pareto optima are non supported and belong to the interior of the convex hull of 2.
0 The parametric approach consists in minimizing an objective function subject to bound constraints on the criteria. Theorem 2 [28] Let G y be the set o f functions from R" to R that are strictly increasing functions and lower bounded on 2. Let g E G y . xo E S is a strict Pareto optimum iff 3b E R K such that xo is an optimal solution of problem (p(g,6)): Min g ( Z ( x ) ) subject t o x E S and Z ( x ) b The parametric approach is useful because theorem 2 ensures that only the set E can be computed by solving all possible problems ( P ( g , b ) when ) g is fixed. A similar result holds for weak Pareto optima if one consider an increasing function unstead of a strictly increasing function.

consists in minimizing a metric over the set 2. These approches consider generally an utopian point zUt demin(&(x)), V i = l..I<,with at least fined by z:t

<

X-ES

one strict inequality. Here, we consider the augmented weighted Tchebycheff metric. Theorem 4 (see f o r instance [3U]) xo E S is a strict Pareto optimum iff 3~ E zut E IRK an utopian point and 3 p E JR; a small value, such that xo is an optimal solution of problem (P( P ) ): Man llZ(X) - Z " t ( ( T P U subject to x ES with l l Z ( ~ ) ~ ~ = max ~ (XilZ:(x) l l- zY'I) ~
I

1SzSK

K i=l

The main difficulty for the use of theorem 4 lies in the setting of the value p. For instance, Steuer ([as]) proposes to set this valuc. xround lod6. Some particular cases are the weighted Tchebycheff metric ( p = 0) and the Tchebycheff metric ( p = 0 and X i = 1, V i = l . . I i ) for which Bowman ([3]) proposes some results, but weaker than theorem 4. Some other metrics 2s the weighted L, metric can be used. The goal satisfiability approach has been proposed by Gembicki ([lo]). Theorem 5 [IO] (321 xo E S is a weak Pareto optimum iff 3 z r e f E IRK a reference point and w E a weights vector such that xo is an optimal solution of problem ( P ( z r e f , w ) ) : Max g ( Z ( x ) ) 1 with g ( Z ( x ) )= min (-(zTef - Z i ( x ) )
i=l..K W ,

<

0 A similar approach is the -constraint approach and has been first proposed in [12]. A basic result is given in theorem 3. Theorem 3 (301 Let k E [l;I<] be a number of criterion. If 3 ck = ( c t ; . . . ; ~ k - ~ ; : + ~ ;. .. ; ~ kE ) such that xo E S is an optimal solution of problem (PE.): Min Z k ( x ) subject to x E S and Zi(X) 5 ; Y i [l;I<],i # IC then xo is a weak Pareto optimum. It is possible to show that the set of all the solutions determined by theorem 3 is a subset of W E and that this subset contains set E . A more general form of this theorem considers all the problems (PE.), V k = l..K, V c k . It states that the set of all the solutions of all problems (PE.) is a subset of W E ([22]). If the convex hypothesis for S and Zi,V i = l..I<, hold then the whole set W E can be determined by the general form of theorem 3.

subject to xEP The advantage of this theorem compared to theorem 4 lies in the reference criteria vector: in the goal satisfiability approach the reference point can be any point in the criteria space. One disadvantage is that for fixed weights w , the weak Pareto optimum computed depends on the relative position of the goal z P e f compared to the set 2 (see [30]). On another hand, one can show that an optimal solution to the problem
/

is a strict Pareto optimum.

0 Other approaches are possible to minimize multiple criteria. One of the most used for scheduling problems is the minimization of a lexicographical 2 + . . . -+ Z K . order of criteria, for instance Z1 + 2 The aim is to find a solution x o E S K with si = { x E S"'/Z;(X) = min ( Z i ( x ' j ) } , vi = l . . ~ ,

X'ES'-'

Other well known way to compute Pareto optima

and So = S. A solution oi this problem is a strict Pareto opiimum and each solution x 1 E 5'2, V i = l..Ii, is a weak Pareto optimum. Notice that with this

VI -465

approach no tradeoff between criteria are allowed, so it can be only used in an a priori resolution algorithm. At last an obvious way to solve the multicriteria problem is to propose an algorithm that enumerates the whole set E . 4.AN OVERVIEW OF MULTICRITERIA SCHEDULING PROBLEMS Survey on multicriteria scheduling literature only focus on the scheduling problems solved, without presenting the multicriteria optimization results associated. For some multicrit.eria scheduling problems of the literature, authors choose an objective function that is not relevant, faced to the scheduling problem they consider. A knowledge of some basical results of multicriteria optimization field can prevent such mistakes. For instance, an optimal solution of the problem of minimizing a convex combination of criteria is a supported Pareto optimum for non convex problems. It means that some solutions (the non supported Pareto optima) that are potentially interesting for the decision maker can never be determined by using such an approach. On another hand, the minimization of a criterion subjected to bound constraints on the other criteria (approach -constraint) leads to a weak Pareto optimum and not necessarily to a strict Pareto optimum. Unfortunately, we note that most of the multicriteria scheduling problems tackled in the literature deal with one of these two ways to handle the criteria. It implies that either some interesting solutions can not be obtained or some non interesting solutions are determined. Knowing these theoretical results of the multicriteria Optimization field, these problems would have been avoided. The problems encountered in the multicriteria scheduling literature can be separated into different classes of problems. The first class contains problems for which one minimizes a t least one completion times based criterion (such as the makespan or the average flowtime time of jobs). For some multicriteria problems the exact processing times of jobs belong to some intervals and must be computed. These problems are based on the minimization of the total weighled crashing costs time of jobs. Another kind of multicriteria problems involves the minimization of the changing tools costs. For other problems the only criteria considered are due d a t e based criteria. A special case of this kind of problems is when at least a mesure of the earliness of jobs and a mesure of the tardiness of jobs are involved. One talk about Just-in-Time (JiT) scheduling problems and we consider these problems as an entire class of problems. Most of the literature on scheduling problems deal with one machine problems. This kind of problem has interested a lot of researchers since the 50's. Few works on multicriteria scheduling problems with

parallel machines have been done. Some works on multicriteria flowshop scheduling problems can be encountered in the literature. hfwt of them deal with the two machines flowshop problem and involve the minimization of the makespan criterion. Few problems deal with multicriteria Jobshop scheduling. None problems about multicriteria Openshop scheduling exist in the literature to our knowledge. A detailed survey is proposed in [30]. Only a small number of problems consider the minimization of more than two criteria. Moreover most of the literature do not talk about the computation of some Pareto optima. Generally] the authors only try to minimize a convex combination of criteria or to minimize the criteria using a lexicographical order. They also often consider the econstraint approach. In this paper we try to put into the light the scheduling problems that explicitcly tackle the problem of computing some Pareto opiima. Hoogeveen ([13]) studies a one machine scheduling .f , " , , ) problem that can be noted l / / c ( f ~ u x / f ~ , x ,.. with fi,, and increasing function of the completion times of jobs. In a first time Hoogeveen proposes a modification of the algorithm of [17] for problem l/prec/ f , , , , to solve optimally the problem with K = 2. The complexity of this algorithm is in O ( n 2 ) . As presented in section 3 the solutions of the problem tackled by Hoogeveen are weak Pareto optima. Despite this fact, the author shows that his algorithm only computes strict Pareto optima. An algorithm to compute the whole set E is therefore proposed. The number of strict Pareto optima is at. most equal to ( n ( n- 1 ) / 2 + l)K-l. Sen and Gupta ([as])study a one machine problem with a completion time based criterion.- The problem tackled can be noted l/d;/fi (T,,, , C) and is polynomially solvable when the weights are fixed. The authors present a branch-and-bound algorithx that enumerates the set of solutions that minimizes all possible convex combination of the two criteria. Unfortunately as seen in section 3 the set of .il.ese solutions corresponds to the set of supported weak P a r e t o p t i m a , namely WE,. For the problem 1 / / f m a x , C John ([15]) presents an algorithm that enumerates the set E . The complexity of this algorithm is in O ( n 2 X ) with X the cardinality of set E . John shows that 1 2 X < ;in (pmax- pmin) with p,,,, = r.max(pi) and =l..n pmin = i%i,nn(p;). Later Hoogeveen and Van de Velde

([14]) study the problem l / / ~ ( f , ~ , / and ~ )propose a polynomial algorithm to solve it. In a second time, an algorithm that computes the whole set E is proposed. The authors show that the cardinality of set E is at most e p a l to n ( n - 1)/2 1. The , for which the insertion problem l / & nrnit/E(C/E,,,) of idle times before jobs are not allox-:ed (constraint nrnit) is strongly NP-hard. Azizoglu, Kondakci

VI - 466

and Koksalan ( [ l ]propose ) a heuristic that appoxjmates a subset -of W E . The problems l / d i / c ( C / U ) and l / d j / c ( C / U , T,,,) are NP-hard and Nelson, Sarin and Daniels ([24]) propose branch-and-bound algorithms and some heuristics. These algorithms compute or approximate a subset of set W E . The problem l / d i / F , c , i.e. the enumeration of the set E has been solved by Lin ([20])who proposes a dynamic programming algorithm. This problem is NP-hard. The problem 1 / d i , n m i t / f i ( c , Lm,, - Lmin) for which the complexity is unknown has been tackled by Sen, Raizadeh and Dileepan (1271). They propose a branch-and-bound that enumerates the set E,. When the processing times of jobs are not fixed and the total weighted crashing costs time of jobs, noted cc", must be minimized, Van Wassenhove and Baker ([31]) study the problem l / p i E bi;jZ], di/c(Tmax,CCU). They propose an algorithm That computes the set of supported strict Pareto optima, namely E,. Its complexity is in O ( n 2 ) .In a second time, the authors extend this algorithm to solve the problem with criteria cc" and fmax where for f m n x some particular assumptions hold. Concerning Just-in- Time ( J i T ) scheduling problems, Hoogeveen ([13])deal with the problems l / s i , d;, d; - si 5 p i , n m i t / c ( L m a x / P m a x ) and l / s i , di, di - s i 5 pj/c(Lmax/Pmax). si is a desired starting time of job Ji and the constraint nmit means that no machine idle times are allowed. The criterion P,, is the maximum promptness criterion and is defined by P,, = max Pi with Pi = si - t i and i=l. n ti the real starting time of job Ji. To solve the first problem the author propose an O(nlog(n)) optimal algorithm whereas for the second problem the optimal algorithm is in O(n2). When the constraint nmit holds Hoogeveen presents an algorithm that computes the set E and shows -that \ E ( 5 n. For the problems l/pi-= 1 , d i , n m i t / c ( E / U ) , l / p i = 1,d i , nmit/E(E,,,/U) and l / p i = 1,d i , nmit/c(Ft(z,T)/U) with Fl(E,T) = E T , Kondakci, Emre and Koksalan ([16]) propose in a first time some linear programs. Then, for all these problems they propose algorithms that compute a subset of the set W E (some non strict Pareto optima are computed).

Chambers ([e]) present a branch-and-bnund algorithm that computes the set E . A heuristic is also proposed to approximate this set. In the case where the number of machines is m, the authors propose a heuristic to m a x / C m a x ) . For solve the problem F l p e r m u , d i / E ( T the problem F 2 / p e r m u , di/Cma:, , U , Liao, Yu and Joe ([19])propose a branch-and-bound algorithm to compute the set E . The authors - also consider the problem F 2 / p e r m u , di/Cmax, T for which they propose a branch-and-bound algorithm that enumerates the set E .

4.CONCLUSION
In this paper we present some guidelines to solve multicriteria scheduling problems. First we introduce the problem and show that one must take into consideration the multicriteria optimization literature to define correctly the scheduling problem to solve. We also introduce a notation for multicriteria scheduling problems. We put into the light some works in scheduling literature that explicitely try to compute some Pareto optima. A lot of works do not take account of multicriteria optimization results. It confirms the interest of bringing together multicriteria optimization and multicriteria scheduling literatures.

Acknowledgements: The authors wish to thank Professor Christian PROUST for his help and useful1 advices.
5.REFERENCES
[l] M. Azizoglu, S. K. Kondakci, and M. Koksalan. Bi-

criteria scheduling : minimizing ftowtime and maxi m u m earliness on a single machine, pages 279288. In J . Climaco (Ed.) : Multicriteria Analysis, Springer-Verlag, 1997.
[2] J . Blazewicz, K.H. Ecker, E. Pescl;, G. Schmidt, and J . Weglarz. Scheduling computer and manufacturing processes. Springer-Verlag, 1996.

Concerning parallel machines scheduling problems one can cite McCormick and Pinedo ([21])that study the problem & / p m t n / E ( ~ / C m a x ) . The preemption of jobs occur at real times. They propose an algorithm that computes all the strict Pareto optima that correspond to extreme points of the set of solutions. The complexity of this algorithm is in O ( m 3 n ) . Concerning Flowshop scheduling problems one can cite the works of Sayin and Karabati that solve the problem F 2 / p e r r n u / ~ ( ~ / C , ~ , )This . one is strongly NF-hard. The authors propose a branch-and-bound algorithm that computes the set E . The problem F 2 / p e r m u , d;/Cmax, T,,, is "P-hard. Daniels and

[3] V.J. Bowman. On the relationship of the tchebycheff norm and the efficienb fcontier of multiplecriteria objectives. I n H. Thiriez et S. Zionts : Multiple Criteria Decision Making, (Eds.) Springer- Verlag, pages 76-85, 1976.

[4] C.-L. Chen and R.L. Bulfin. Complexity of single machine, multi-criteria scheduling problems. Eur. J. Oper. Res., 70:115-125, 1993.
[5] C.-L. Chen and R.L. Bulfin. Complexity of multiple machines, multi-criteria scheduling problems. 3rd Industrial Engineering Research Conference (IERC'Sd), Atlanta (USA), pages 662-665, 1994.

VI - 467

[6] R.L. Daniels and R.J. Chambers. Multiobjective flow-shop scheduling. Nav. Res. Log., 37:981-995, 1990. [7] P. Dileepan and T . Sen. Bicriterion static scheduling research for a single machine. Omega, 16(1):5359, 1988. [8] G.W. Evans. An overview of techniques for solving multiobjective mathematical programs. Mngnt. Sci., 30(11):1268-1282, 1984.

[19] C.-J. Liao, W.-C. Yu, and C.-B. Joe. Bicriterion scheduling in the two-machine flowshop. J. Oper. Res. Soc., 48:929-935, 1997. [20] K.S. Lin. Hybrid algorithm for sequencing with bicriteria. J. Opt. Theory and Apps., 39(1):105124, 1983. [all S.T. McCormick and M.L. Pinedo. Scheduling n independant jobs on m uniform machines with both flowtime and makespan objectives : a parametric analysis. 0.R.S.A. J. Computing, 7(1):6377, 1995. [22] K. Miettinen. On the Methodology of Multiobjective Optimization with Applications. Ph. d. thesis, University of Jyvaskyla, department of mathematics, 1994. [23] A. Nagar, J . Haddock, and S.S. Heragu. Multiple and bicriteria scheduling : A literature survey. Eur. J. Oper. Res., 81:88-104, 1995. [24] R.T. Nelson, R.K. Sarin, and R.L. Daniels. Scheduling with multiple performance measures : The one-machine case. Mngnt. Scie., 32(4):464479, 1986. [25] B. Roy. Methodologie multicritire daide h la decision. Economica, 1985. [26] T. Sen and S.K. Gupta. A branch-and-bound procedure to solve a bicriterion scheduling problem. IIE Trans., 15(1):84-88, 1983. [27] T. Sen, F.M.E. Raiszadeh, and P. Dileepan. A brach-and-bound approach to the bicriterion scheduling problem involving total flowtime and range of lateness. Mngt. Sci., 34(2):255-260,1988,
[28] R.M. Soland. Multicriteria optimization : A gen-

[9] T.D. Fry, R.D. Armstrong, and H. Lewis. A framework for single machine multiple objective sequencing research. Omega, 17(6):595-607, 1989. Vector Optimization for Control [lo] F. Gembicki. with Performance and Parameter Sensitivity I n dices. Ph. d. thesis, Case Western Reserve University, Cleveland (USA), 1973.

[ll] A.M. Geoffrion. Proper efficiency and the theory of vector maximization. J. Math. Analysis and Apps., 22:618-630, 1968.
[12] Y. Haimes, L. Ladson, and D. Wismer. On a bicriterion formulation of the problems of integrated system identification and system optimization. IIIE Transactions on Systems, Man and Cybernetics, 1:296-297, 1971.

[13] J.A. Hoogeveen. Single-Machine Bicriteria Scheduling. Ph. d. thesis, CWI Amsterdam, 1992.
[14] J.A. Hoogeveen and S.L. VandeVelde. Minimizing total completion time and maximum cost simultaneously is solvable in polynomial time. Oper. Res. Let., 17:205-208, 1995. [15] T.C. John. Tradeoff solutions in single machine production scheduling for minimizing flow time and maximum penalty. Comput. and Oper. Res., 16(5):471-479, 1984.
[la] S.K. Kondakci, E. Emre, and M. Koksalan. Scheduling of unit processing time jobs on a single machine. In G. Fandel and T. Gal (Eds.) : Multiple Criteria Decision Making, Lecture Notes in Economics and Mathematical Systems, SpringerVerlag, pages 654-660, 1997.

eral characterization of efficient solutions. Decision Sci., 10:27-38, 1979. [29] R. Steuer. Multiple criteria optimization : theory, computation and application. Wiley, 1986. [30] V. Tkindt and J.-C. Billaut.. Les problhmes dordonnancernent date!iers multicritbres , research report 206, li/e3i/university of tours. page 190 p., 1999. [31] L. Vanwassenhove and K.K. Baker. A bicriterion approach to time/cost trade-offs in sequencing. Eur. J . Oper. Res., 11(1):48-54, 1982. 1321 A. Wierzbicki. The use of reference objectives in multiahjective optimization, in g. fandel and t. gal (eds.): Multiple criteria decision making, theory and application, springer-verlag, pp. 468-486, 1980.

[17] E.L. Lawler. Optimal sequencing of a single machine subject to precedence constraints. Sci., 19:544-546, 1973. Mngnt.

[18] C.-Y. Lee and G.L. Vairaktarakis. Complexity of single machine hierarchical scheduling: A survey. In P.M. Pardalos (Ed.): Complexity in Numeracal Optimization, World Scientijic Publishing CO, pages 269-298, 1993.

V I -- 468