Вычислительные технологии, 2021, том 26, › 3, с. 86–106.

© ФИЦ ИВТ, 2021 ISSN 1560-7534

Computational Technologies, 2021, vol. 26, no. 3, pp. 86–106. © FRC ICT, 2021 eISSN 2313-691X

COMPUTATIONAL TECHNOLOGIES

DOI:10.25743/ICT.2021.26.3.006
Stochastic optimization over the Pareto front by the
augmented weighted Tchebychev program
Younsi-Abbaci L.* , Moulaı̈ M.
USTHB University, LaROMaD laboratory, Faculty of Mathematics, BP 32 Bab-Ezzouar, El-Alia
16111, Algiers, Algeria
*
Correspoding author: Leila Younsi-Abbaci, e-mail: [email protected]
Received December 14, 2020, revised April 21, 2021, accepted April 28, 2021.

In this paper, we propose a novel algorithm to deal with multi-objective stochastic

integer linear programming problems (MOSILP). Given a stochastic linear function
𝜑, we will optimize it over the full set of efficient solutions of a MOSILP. We convert
the latter into an equivalent deterministic problem using uncertain aspirations which
are inputs specified by the decision maker. For this purpose, we adopt a 2-stage
recourse approach where an augmented weighted Tchebychev program is progressively
optimized to generate an efficient solution, the value of the utility function 𝜑 is improved
to enumerate all efficient solutions. The approach proposed here defines and solves a
sequence of progressively more constrained integer linear programs, so that a new
efficient solution is generated at each step of the algorithm. A numerical example is
presented for illustration.
Keywords: multiple objective, integer programming, stochastic linear programming,
Tchebychev norm.
Citation: Younsi-Abbaci L., Moulaı̈ M. Stochastic optimization over the Pareto
front by the augmented weighted Tchebychev program. Computational Technologies.
2021; 26(3):86–106. DOI:10.25743/ICT.2021.26.3.006.

Introduction
Let us consider the basic problem
{︂
min Φ(𝑥) = 𝑑(𝜉)𝑥,
(𝑃𝐸 ) (1)
s.t. 𝑥 ∈ E𝑠 ,

where 𝑑 is a random vector of dimension 𝑛 and E𝑠 is the efficient solution set of the multiple
objective stochastic Integer linear programming problem MOSILP,
⎧
⎪
⎪ min 𝑍𝑖 = 𝐶𝑖 (𝜉)𝑥, 𝑖 = 1, ..., 𝑝,
s.t. 𝐴𝑥 = 𝑏,
⎨
(𝑀 𝑂1 ) (2)
⎪
⎪ 𝑇 (𝜉)𝑥 = ℎ(𝜉),
𝑥 ≥ 0, integer,
⎩

where 𝑥 is the decision variable vector of dimension (𝑛 × 1). 𝐶, 𝑇 and ℎ are random matrices
of respective dimensions (𝑝 × 𝑛), (𝑚1 × 𝑛) and (𝑚1 × 1) with a joint probability distribution

86
Stochastic optimization over the Pareto front . . . 87

(independent again of the choice of 𝑥) defined on some probability space (Ξ, 𝐸, 𝑝𝑟𝑜𝑏). 𝐴 and
𝑏 are deterministic matrices of dimensions (𝑚 × 𝑛) and (𝑚 × 1), respectively. Let 𝐶(𝜉) be a
(𝑝 × 𝑛)random matrix with 𝑝 rows 𝐶𝑖 (𝜉) ∈ R.
The main difficulty of problem (𝑃𝐸 ) arises from the nonconvexity of the efficient set E𝑠 ,
indeed (E𝑠 ) the union of several faces of the feasible set of problem (𝑀 𝑂1 ). Consequently,
(𝑃𝐸 ) is a global optimization problem.
(𝑃𝐸 ) have been discussed extensively in the literature and a variety of methods have been
developed for its solution (or resolution); see for example Philip in [1] studied the problem and
described schematically a cutting plane procedure to solve it. Later, Isermann and Steuer [2]
proposed a similar procedure for solving the problem they optimized one criterion among
the multi-objective linear program functions. Necessary and sufficient conditions for this
problem to be unbounded were established by Benson [3]. In [4], Ecker and Song used Philip’s
approach to introduce two implementable algorithms that involve a pivoting technique on
the feasible set a reduced one of a multiple objectives integer linear program. Philip’s method
was implemented by Bolintineanu [5] for the case where the objective function of the problem
is quasiconcave. Sayin in [6] formulated problem (𝑃𝐸 ) as a linear program with an additional
reserve convex constraint and proposed a cutting plane method to solve the latter problem.
In [7], Abbas and Chaabane optimized linear function over an integer efficient set and Jorge
developed in [8] another approach that defines a sequence of progressively more constrained
single-objective integer problems that successively eliminates undesirable points, the most
recent work on this topic was conducted by Chabaane et al. in [9].
The first interactive method for solving MOSILP problems was the STRANGE-MOMIX
developed by Teghem [10]. In [11], Abbas and Belhacen (2006) proposed an algorithm that
combines the cutting plane technique [12] and the L-shaped decomposition method described
in [13]. The authors Amrouche and Moulaı̈ (2012) developed in [14] an approach for detecting
all stochastic integer efficient solutions of problem MOSILP based on solving a deterministic
multiple objective integer linear program. When the decision variables are integers, few
methods exist in the literature and cuts or branch and bound techniques are unavoidable.
In this paper, we propose an exact algorithm for solving (𝑃𝐸 ), it is based on Jorge’s
approach [8] with the concepts L-shaped integer method [15]. We will use the Augmented
Weighted Tchebychev program [16] to generate the set of nondominated objective vectors.
The remainder of the paper is organized as follows: in Section 1 we convert the problem
MOSILP into an equivalent deterministic one; also, definitions and some results concerning
the L-shaped decomposition method are given. Section 2 introduces the concepts of the
utopian vector and the Augmented Weighted Tchebychev program. We describe our algo-
rithm for optimizing a linear function over the efficient set of MOSILP in section 3. Every
step of the method will be illustrated in Section 4 by a numerical example and Section 6
ends the paper with concluding remarks.

1. Construction of equivalent deterministic problem

The basic dual decomposition method for two-stage recourse problems is essentially an ap-
plication of Benders decomposition [15], due to Van Slyke and Wets [13], and is usually
called the L-shaped method in the literature. Assume that we have a joint finite discrete
probability distribution (𝜉 𝑟 , 𝑝𝑟𝑜𝑏𝑟 ), 𝑟 = 1, ..., 𝑅, of the random data.
88 Younsi-Abbaci L., Moulaı̈ M.

1. In the first stage, for each realization 𝜉 𝑟 of 𝜉, we associate a criterion 𝑍𝑖𝑟 = 𝐶𝑖 (𝜉 𝑟 )𝑥, a
matrix 𝑇 (𝜉 𝑟 ) and a vector ℎ(𝜉 𝑟 ) to take into account the different scenarios affecting
the 𝑝 objectives and the stochastic constraints.
2. The second stage is to come back to the same idea of recourse used in single-criterion
stochastic programming [17, 18]. Of course, we assume that the Decision Maker (DM) is
able to specify the penalties 𝑞 𝑟 = 𝑞(𝜉 𝑟 ) of the constraint violations 𝑦 𝑟 , 𝑟 = 1, ..., 𝑅, and
the size of the associated deterministic problem remains reasonable. Then, unlike the
Strange method where a supplementary criterion is created to penalize the constraint
violations, a recourse function 𝑄(𝑥, 𝜉 𝑟 ) is added to each criterion 𝑍𝑖𝑟 . This penalty
(called the recourse function) is given by:

𝑄(𝑥, 𝜉 𝑟 ) = min{(𝑞 𝑟 )𝑇 𝑦|𝑊 (𝜉 𝑟 )𝑦 = ℎ(𝜉 𝑟 ) − 𝑇 (𝜉 𝑟 )𝑥, 𝑦 ≥ 0}. (3)

𝑦

Then the (DM) has to minimize the expected value of the total costs:

𝑍̃︀𝑖 = 𝐸𝑠𝑝[𝑍𝑖 + 𝑄(𝑥, 𝜉)], 𝑖 = 1, . . . , 𝑝,

with 𝐸𝑠𝑝 meaning expected value. It results in the following deterministic MOSILP problem
⎧
⎨ min 𝑍̃︀𝑖 = 𝑍𝑖′ + 𝑄(𝑥)
(𝑀 𝑂2 ) s.t. 𝐴𝑥 = 𝑏, (4)
𝑥 ≥ 0, integer,
⎩

where

𝑅
∑︁ 𝑅
∑︁
𝑟 𝑟
𝑄(𝑥) = 𝐸𝑠𝑝[𝑄(𝑥, 𝜉)] = 𝑝𝑟𝑜𝑏 (𝑄(𝑥, 𝜉 )) = (𝑝𝑟𝑜𝑏𝑟 𝑞 𝑟 )𝑇 𝑦 𝑟 ,
𝑟=1 𝑟=1
𝑝
∑︁
𝑍𝑖′ = 𝐸𝑠𝑝[𝑍𝑖 ] = 𝑝𝑟𝑜𝑏𝑟 𝐶𝑖 (𝜉 𝑟 )𝑥 = 𝐸𝑠𝑝[𝐶𝑖 (𝜉)𝑥], note 𝐶𝑥
̃︀ = 𝐸𝑠𝑝[𝐶𝑖 (𝜉)𝑥]
𝑖=1

are respectively the recourse-function 𝑄(𝑥, 𝜉) and the expected values of 𝑍̃︀𝑖 .
We expect the second-stage program (3) to be feasible for all the realizations 𝜉 𝑟 , 𝑟 =
1, ..., 𝑅, of 𝜉. Depending on the (𝑚1 × 𝑛1 )-recourse matrix 𝑊 (𝜉 𝑟 ), this needs not to be true
for all the first-stage decisions 𝑥 ∈ {𝑥|𝐴𝑥 = 𝑏, 𝑥 ≥ 0}. Then, the first-stage decisions are
restricted to 𝑥 ∈ {𝑥|𝐴𝑥 = 𝑏, 𝑥 ≥ 0} ∩ K ̸= ∅ where, K = {𝑥|𝑇 (𝜉 𝑟 )𝑥 + 𝑊 (𝜉 𝑟 )𝑦 𝑟 = ℎ(𝜉 𝑟 ), 𝑦 𝑟 ≥
0, 𝑟 = 1, . . . , 𝑅}, is the induced first-stage feasibility set.
In the second-stage programs (3), the recourse-matrices 𝑊 (𝜉 𝑟 ), could be replaced by a
fixed recourse-matrix 𝑊 without any changes in the presentation of the proposed algorithm.
Even if 𝑊 is being fixed or not, the problem we face is that {𝑥|𝐴𝑥 = 𝑏, 𝑥 ≥ 0} ∩ K
can be empty. To avoid this problem, complete fixed recourse-matrices that satisfy {𝑡|𝑡 =
𝑊 𝑦, 𝑡 ≥ 0} are recommended. This implies that, whatever the first-stage decisions 𝑥 and
the realizations 𝜉 𝑟 of 𝜉 turn out to be, the second-stage programs (3) are always feasible. A
special case of complete fixed recourse matrix is simple recourse with the identity matrix 𝐼
of order 𝑚1 , 𝑊 = (𝐼, −𝐼).
Stochastic optimization over the Pareto front . . . 89

1.1. Associated relaxed problem and basic definitions

Associated to (𝑃𝐸 ), the deterministic relaxed problem is defined by:
⎧
⎨ min Φ(𝑥)
̃︀ = 𝐸𝑠𝑝(𝑑(𝜉)𝑥) + 𝑄(𝑥)
(𝑃𝑅 ) s.t. 𝐴𝑥 = 𝑏, (5)
𝑥 ≥ 0, integer.
⎩

1.2. Feasibility
The recourse-matrix 𝑊 is being fixed. The question is then: how can we state that a given
decision vector 𝑥0 will yield feasible second-stage problems for all possible realisations of 𝜉.
Therefore, it is a lot advantageous to work with the dual [17]
max{𝜋 𝑇 (ℎ(𝜉) − 𝑇 (𝜉)𝑥0 )| 𝜋𝑊 ≤ 𝑞(𝜉), 𝜋 ∈ R} (6)
on the other hand, the Farkas lemma states that {𝑦|𝑊 𝑦 = ℎ(𝜉) − 𝑇 (𝜉)𝑥0 , 𝑦 ≥ 0} =
̸ ∅ if and
𝑇 𝑇 0
only if 𝜎 𝑊 ≤ 0 implies that 𝜎 [ℎ(𝜉) − 𝑇 (𝜉)𝑥 ] ≤ 0.
We conclude that 𝑄(𝑥0 , 𝜉 𝑟 ) is infeasible if and only if 𝑃 = {𝜋 : 𝜋𝑊 ≤ 𝑞(𝜉)} has an
extreme ray 𝜎 such that 𝜎 𝑇 [ℎ(𝜉) − 𝑇 (𝜉)𝑥0 ] > 0.
Then to check for feasibility of the second stage-problems, we have to find a direction
vector 𝜎 by solving the dual problems:
max{𝜎 𝑇 (ℎ(𝜉) − 𝑇 (𝜉)𝑥0 ) 𝜎 𝑇 𝑊 ≤ 0, ‖𝜎‖1 ≤ 1, 𝜎 ∈ R}, (7)
where the constraint ‖𝜎‖1 ≤ 1 is added to bound 𝜎. In case where for some 𝑟, 𝑟 = 1, ..., 𝑅
with 𝑟 is the optimal solution of dual problem; we have 𝜎𝑟𝑇 [ℎ(𝜉) − 𝑇 (𝜉)𝑥0 ] > 0. Then we add
the feasibility cut:
𝜎𝑟𝑇 [ℎ(𝜉) − 𝑇 (𝜉)𝑥0 ] ≤ 0. (8)

1.3. Optimality
Assuming that all the feasibility cuts are there, we can reformulate the problem (5) by
introducing a new variable 𝜃:
⎧
⎪
⎪ min Φ(𝑥)
̃︀ = 𝐸𝑠𝑝(𝑑(𝜉)𝑥) + 𝜃
s.t. 𝑥 ∈ 𝐷 = 𝐷 ̃︀ ∩ N,
⎨
(9)
⎪
⎪ 𝜃 ≥ 𝑄(𝑥),
⎩
𝑥 integer,
̃︀ = {𝑥 ∈ R𝑛 |𝐴𝑥 = 𝑏, 𝜎 𝑇 (𝑇 (𝜉 𝑟 ) − ℎ(𝜉 𝑟 )) ≥ 0, 𝑟 = 1, ..., 𝑅 } = {𝑥 ∈ R𝑛 |𝐴𝑥
where 𝐷 ̃︀ = ̃︀𝑏}.
𝑟
̃︀ is assumed to be a non-empty, compact polyhedron in R𝑛 .
Throughout this paper, 𝐷
The constraint
𝜃 ≥ 𝑄(𝑥), (10)
is in the optimality cut [17].
We define the notion of optimality for (𝑀 𝑂2 ) according to the Pareto concept.
Definition. A point 𝑥* ∈ 𝐷 is said to be efficient for (4) if and only if there does not exist
another point 𝑥1 ∈ 𝐷 such that 𝑍̃︀𝑖 (𝑥* ) ≥ 𝑍̃︀𝑖 (𝑥1 ), 𝑖 ∈ 1, . . . , 𝑝, and 𝑍̃︀𝑖 (𝑥* ) > 𝑍̃︀𝑖 (𝑥1 ) for at
least one 𝑖 ∈ 1, . . . , 𝑝 and for all the realisations, 𝜉 𝑟 𝑟 = 1, . . . , 𝑅. Otherwise, 𝑥* is not
efficient and the corresponding vector (𝑍̃︀1 (𝑥), 𝑍̃︀2 (𝑥), . . . , 𝑍̃︀𝑝 (𝑥)) is said to be dominated.
90 Younsi-Abbaci L., Moulaı̈ M.

2. The utopian criterion vector and the Augmented Weighted

Tchebychev program
Bowman [19] used a weighted Tchebychev norm for scalarization of multiple objective opti-
mization problems. Based on this approach, Steuer and Choo [20] introduced the augmented
weighted Tchebychev program and the lexicographic weighted Tchebychev program.
Let 𝑍̃︀𝑖𝑑𝑒𝑎𝑙 ∈ R𝑝 be the ideal criterion vector such that 𝑍̃︀𝑖𝑖𝑑𝑒𝑎𝑙 = min{𝑍̃︀𝑖 (𝑥)|𝑥 ∈ 𝐷}. A
vector strictly better than 𝑍̃︀𝑖𝑑𝑒𝑎𝑙 is called an utopian point 𝑍̃︀𝑢𝑡𝑜𝑝 , 𝑍̃︀𝑢𝑡𝑜𝑝 < 𝑍̃︀𝑖𝑑𝑒𝑎𝑙 or 𝑍̃︀𝑢𝑡𝑜𝑝 =
𝑍̃︀𝑖𝑑𝑒𝑎𝑙 − 𝜗 where 𝜗 > 0 and small. The augmented weighted Tchebychev norm of 𝑍̃︀ consist
of measuring the distance between any criteria vector 𝑍̃︀ and the utopian vector 𝑍̃︀𝑢𝑡𝑜𝑝 , is
defined as follows [19]:
𝑝
∑︁
̃︀ 𝜆∞ = max {𝜆𝑖 |𝑍̃︀𝑢𝑡𝑜𝑝 − 𝑍̃︀𝑖 |} + 𝜌
‖𝑍̃︀𝑢𝑡𝑜𝑝 − 𝑍‖ |𝑍̃︀𝑖𝑢𝑡𝑜𝑝 − 𝑍̃︀𝑖 |,
𝑖
𝑖=1,...,𝑝
𝑖=1

where 𝜌 is a sufficiently small positive scalar and 𝜆 is weight vector.

Steuer (1986) has shown that if the 𝜌 is small enough, the augmented weighted Tcheby-
chev program not only guarantees to return a nondominated objective vector but generates
any particular nondominated objective vector for an appropriate 𝜆 ∈ Λ. 𝜆𝑖 is the weight of
𝑝
∑︁
the design objective 𝑖, and satisfies 𝜆𝑖 = 1 and 𝜆𝑖 ≥ 0, 𝑖 = 1, 2, ..., 𝑝. The set
𝑖=1

{︂ 𝑝 }︂
∑︁
𝑝
Λ= 𝜆∈R | 𝜆𝑖 = 1 and 𝜆𝑖 ≥ 0, ∀𝑖 ⊂ R𝑝
𝑖=1

is the weighting vector space and any 𝜆 ∈ Λ is called a weighting vector.

The idea of this approach is to find a vector 𝑍̃︀ in the criteria space which minimizes the
distance to the utopian vector:
min ‖𝑍̃︀𝑢𝑡𝑜𝑝 − 𝑍‖
̃︀ 𝜆 ,
∞
𝑍∈Z

where Z is a feasible region in criteria space.

Theorem. [9] Let 𝜆 ∈ Λ, for a small enough fixed 𝜌 > 0, any optimal solution to (𝑃𝜌 (𝜆))
problem
𝑝
⎧
∑︁
(𝑍̃︀𝑖𝑢𝑡𝑜𝑝 − 𝑍̃︀𝑖 )
⎪
⎪
⎪
⎪
⎪ min 𝛼 + 𝜌
𝑖=1
⎨
(𝑃𝜌 (𝜆)) s.t. 𝛼 ≥ 𝜆𝑖 (𝑍̃︀𝑢𝑡𝑜𝑝 − 𝑍̃︀𝑖 ), (11)
⎪ 𝑖
𝑥 ∈ 𝐷,
⎪
⎪
⎪
⎪
𝛼≥0
⎩

is a nondominated objective vector to problem (𝑀 𝑂2 ). Here

𝑝
[︂ ∑︁ ]︂−1
1 1
𝜆𝑖 = ∀1 ≤ 𝑖 ≤ 𝑝,
𝑍̃︀𝑢𝑡𝑜𝑝 − 𝑍̃︀𝑖* 𝑖=1 𝑍̃︀𝑢𝑡𝑜𝑝 − 𝑍̃︀𝑖*

with 𝑍̃︀𝑖* = 𝐶 𝑖 𝑥* , where 𝑥* is a prefixed vector in 𝐷.

Stochastic optimization over the Pareto front . . . 91

3. Description of the method

Initially, the procedure determines the utopian objective vector 𝑍̃︀𝑢𝑡𝑜𝑝 . And the relaxed
problem (𝑃𝑅 ) associated to problem (𝑃𝐸 ) is being solved, its feasible set is defined by deter-
ministic constraints of problem (𝑀 𝑂1 ) without any feasibility or optimality cut. If for some
realisations 𝜉 𝑟 , 𝑟 ∈ 1, . . . , 𝑅, the second-stage problems yielded by the integer solution 𝑥
found are not feasible, feasibility cuts (8) are introduced, we then get the integer optimal so-
lution 𝑥, compute the recours function 𝑄(𝑥). For a sufficiently small value 𝜌, the augmented
weighted Tchebychev program 𝑝𝜌 (𝜆) is solved in order to find the nondominated vector 𝑍¯
that is the closest to the utopian objective vector 𝑍̃︀𝑢𝑡𝑜𝑝 in the direction determined by 𝑍 𝑢𝑡𝑜𝑝
and 𝑍.¯ We then get the integer optimal solution 𝑥¯ of 𝑝𝜌 (𝜆), feasibility cuts (8) may be added
if infeasibility of second-stage problems appears, and the corresponding value of 𝜃.
Given that, in the decision space it may happen that the obtained solution is not better
than an equivalent efficient solution on the main objective function , therefore, the following
problem has to be solved to find an equivalent efficient solution which improves the main
objective before reducing the current admissible region.

(𝑇 𝑙 ) : min{𝑑𝑥|𝑥
̃︀ ∈ 𝐷, 𝐶𝑥 ¯
̃︀ + 𝜃 = 𝑍}.

The optimal solution 𝑥*𝑙 of this problem is considered as a first efficient solution.
Afterwards, at an iteration 𝑙, using Sylva and Crema’s idea, see [21], we add to (𝑃𝑅𝑙 )
new constraints that eliminate all the solutions dominated by 𝑥*𝑙 . There by, the admissible
domain is reduced. This task is performed by the resolution of the following problem 𝑃𝑅𝑙 . It
is worthnothing that all coefficients of 𝐶
̃︀ are supposed to be integers:

𝑃𝑅𝑙 ≡ min{𝑑𝑥|𝑥
̃︀ ∈ 𝐷 − ∪𝑙 𝐷𝑠 },
𝑠=1 (12)

where 𝐷𝑠 = {𝑥 ∈ Z𝑛 |𝐶𝑥̃︀ ≥ 𝐶𝑥 ̃︀ 𝑠 } and {𝐶𝑥 ̃︀ 𝑠 }𝑙 is a subset of nondominated criteria vectors

𝑠=1
for problem (𝑃𝐸 ). {𝑥𝑠 ; 𝑠 = 1, ..., 𝑙−1} are solutions of (𝑃𝐸 ) obtained at iterations 1, 2, ..., 𝑙−1
respectively
⎧ ⎫
⎪ 𝐶̃︀𝑖 𝑥 ≤ (𝐶̃︀𝑖 𝑥𝑠 + 1)𝑦𝑖𝑠 + 𝑀𝑖 (1 − 𝑦𝑖𝑠 ), ⎪
⎪
⎪ ⎪
⎪
⎪
⎪
⎪ 𝑖 = 1, 2, ...𝑝, 𝑠 = 1, 2, ..., 𝑙, ⎪
⎪
⎪
⎪
⎨ ∑︁𝑠 ⎪
⎬
𝑙 𝑠
𝐻 = 𝐷 − ∪𝑠=1 𝐷𝑠 =
̃︀ 𝑦𝑖 ≥ 1, 𝑠 = 1, 2, ..., 𝑙, ,
⎪ ⎪
⎪
⎪ 𝑖=1 ⎪
⎪
𝑦𝑖𝑠 ∈ {0, 1}, 𝑖 = 1, 2, ...𝑝, 𝑠 = 1, 2, ..., 𝑙, ⎪
⎪
⎪ ⎪
⎪
⎪ ⎪
⎪
𝑥∈𝐷
⎩ ⎭

where 𝑀𝑖 is an upper bound for any feasible value of the 𝑖𝑡ℎ objective function. The asso-
ciated variables 𝑦𝑖𝑠 𝑖 = 1, 2, ...𝑝, of 𝑥*𝑠 and additional constraints are added to impose an
improvement on at least one objective function. Note that when 𝑦𝑖𝑠 = 0, the constraint is
not restrictive, and when 𝑦𝑖𝑠 = 1 a strict improvement is forced in the 𝑖𝑡ℎ objective function
evaluated at 𝑥*𝑠 .

3.1. Algorithm
The technical description of the method provides a new algorithm with an exponential
complexity.
92 Younsi-Abbaci L., Moulaı̈ M.

Algorithm: Stochastic optimizing over the Pareto optimal front

input :
𝐴(𝑚×𝑛) : matrix of deterministic constraints
𝑏(𝑚×1) : RHS vector;
𝐶(𝑝×𝑛) : matrix criterion of stochastic coefficients;
ℎ(𝑚1 × 1): vector of stochastic constraints ;
𝑊(𝑚1 ×𝑛1 ) : matrix of stochastic constraints;
𝑇 (𝑚1 × 𝑛): matrix of stochastic constraints;
output :
𝑥𝑜𝑝𝑡 : optimal solution of the problem (𝑃𝐸 ), Φ ̃︀ 𝑜𝑝𝑡 : optimal value of the main criterion
Φ̃︀
initialization:
for 𝑖 ← 1 to 𝑝 do
Solve 𝑍̃︀𝑖𝑖𝑑𝑒𝑎𝑙 = 𝑚𝑖𝑛{𝐶̃︀𝑖 𝑥,𝑥 ∈ 𝐷} is called the ideal point.
and set the upper bounds 𝑀𝑖 = 𝑚𝑎𝑥{𝐶 ̃︀𝑖 𝑥,𝑥 ∈ 𝐷}.
where 𝜗 = 1 therefore 𝑍̃︀𝑢𝑡𝑜𝑝 = 𝑍̃︀𝑖𝑑𝑒𝑎𝑙 − 1 ;
̃︀ 𝑜𝑝𝑡 := +𝑖𝑛𝑓 , 𝑙 := 1, 𝐸1 := ∅,
Φ
𝑇 𝑒𝑟𝑚𝑖𝑛𝑎𝑡𝑒 := 𝐹 𝑎𝑙𝑠𝑒,, 𝐻 ̃︀ := 𝐷, 𝜃 := −∞;
while Terminate:=False do
Solve 𝑃𝑅 ≡ 𝑚𝑖𝑛{𝑑𝑥|𝑥 ̃︀ ∈ 𝐻}. ̃︀ Let 𝑥𝑙 be an optimal solution of 𝑃𝑅 ;
feasibility and optimality test
for 𝑟:=1 to R do
feasibility:=False;
while feasibility:=False do
ˆ 𝑇 an optimal solution of the problem:
𝜎
𝑚𝑎𝑥{𝜎 𝑇 (ℎ(𝜉) − 𝑇 (𝜉)𝑥𝑙 ) | 𝜎 𝑇 𝑊 ≤ 0, ‖𝜎‖1 ≤ 1,𝜎 ≥ 0}
coup:= 𝜎 ˆ 𝑇 (ℎ(𝜉) − 𝑇 (𝜉)𝑥𝑙 )
if 𝑐𝑜𝑢𝑝 > 0 then
𝐻̃︀ = 𝐻 𝜎 𝑇 [ℎ(𝜉) − 𝑇 (𝜉)𝑥) < 0};
̃︀ ⋃︀{ˆ
Let 𝑥 be an optimal solution of 𝑅2𝑙
𝑙

else
Fallibility:= true

recours function Q:=0;

for 𝑟:=1 to R do
solve problem: max{𝜋 𝑇 (ℎ(𝜉) − 𝑇 (𝜉)𝑥𝑙 ) | 𝜋𝑊 ≤ 𝑞(𝜉)}
𝑄(𝑥) := 𝑄(𝑥) + 𝑝𝑟𝑜𝑏𝑟 × 𝑄(𝑥𝑙 , 𝜉 𝑟 )
𝜃 := 𝑄(𝑥),
(𝑥𝑙 , 𝜃): optimal solution for 𝑃𝑅𝑙 (after feasibility and optimality tests)
if Terminate:=false or Φ(𝑥 ̃︀ 𝑙 ) ≥ Φ ̃︀ 𝑜𝑝𝑡 then
𝑥𝑜𝑝𝑡 an optimal solution of 𝑃𝐸 , Terminate:=True.
else
efficiency test of 𝑍̃︀𝑙 = 𝐶𝑥 ̃︀ 𝑙 ;
𝑇
compute the weighted vector of 𝑃𝜌 (𝜆𝑙 ) ;
Let (¯ 𝑥𝑙 , 𝑍¯ 𝑙 ) be an optimal solution of 𝑃𝜌 (𝜆𝑙 )
feasibility and optimality for 𝑥¯𝑙
̃︀𝑥𝑙 = Φ
if 𝑑¯ ̃︀ 𝑜𝑝𝑡 then
𝑥𝑜𝑝𝑡 = 𝑥¯𝑙 , Φ ̃︀ 𝑥𝑙 ), Terminated:=True
̃︀ 𝑜𝑝𝑡 = Φ(¯
else
solve 𝑇 (¯ 𝑥𝑙 ) = 𝑚𝑖𝑛{𝑑𝑥 ̃︀ + 𝜃|𝑥 ∈ 𝐷, 𝐶𝑥 ̃︀ + 𝜃 = 𝑍¯ 𝑙 };
*𝑙
let 𝑥 be an optimal solution of 𝑇 (¯ 𝑥𝑙 );
feasibility and optimality test if Φ ̃︀ *𝑙 ) then
̃︀ 𝑜𝑝𝑡 ≤ Φ(𝑥
𝑥𝑜𝑝𝑡 an optimal solution of 𝑃𝐸 , Terminate:=True
else
𝑥𝑜𝑝𝑡 := 𝑥*𝑙 , Φ ̃︀ *𝑙 ); let E𝑙+1 = E𝑙 ∪ 𝑥*𝑙 ;
̃︀ 𝑜𝑝𝑡 := Φ(𝑥
𝑠 𝑠
̃︀ := 𝐷∖ ∪ 𝐷𝑠=1 𝑙−1
𝑙 := 𝑙 + 1 and 𝐻 ;

Proposition. The algorithm terminates in a finite number of iterations.

Stochastic optimization over the Pareto front . . . 93

Proof. Since there are finitely many feasible bases coming from the recourse-matrix 𝑊 , there
are only finitely many feasibility and optimality cuts. On other hand, at each iteration of the
algorithm, a new improved efficient solution is generated and the admissible region is being
reduced there until infeasibility. All these additional cuts exclude the points or the edges
once scanned, leading to the convergence of the procedure in a finite number of steps.

4. An illustrative example
The problem of optimization over the efficient set of the MOSILP.
Two scenarios (𝑅 = 2).
ˆ The principal problem:

𝑑(𝜉 1 ) = (4, −10), 𝑑(𝜉 2 ) = (−6, 12),

we calculate the expected value 𝑑̃︀

(︂ )︂ (︂ )︂
1 𝑥 1 1 𝑥 1
𝑑̃︀ = 𝐸𝑠𝑝(𝑑(𝑥, 𝜉)) = 𝑑(𝜉 1 ) + 𝑑(𝜉 2 ) =
2 𝑥(︂2 )︂2 𝑥2 (︂ )︂
1 𝑥1 1 𝑥1
= (4, −10) + (−6, 12) =
2 𝑥2 2 𝑥2
= −𝑥1 + 𝑥2

and then we obtain a linear function optimization problem over an efficient set
{︂
min Φ(𝑥) = −𝑥1 + 𝑥2
(𝑃𝐸 ) (13)
s.t. 𝑥 ∈ E𝑠 .

ˆ Multiobjective stochastic problem: let us consider the following example with a struc-
ture similar to problem (𝑀 𝑂1 ), 𝑝 = 2, 𝑛1 = 4, 𝑚1 = 𝑚 = 𝑛 = 2.
ˆ Matrix C:
𝐶1 (𝜉 1 ) = (4, −9), 𝐶1 (𝜉 2 ) = (−6, 3),
𝐶2 (𝜉 1 ) = (8, 5), 𝐶2 (𝜉 2 ) = (−2, −3).
ˆ Matrix T and vector h:
(︂)︂ (︂ )︂
1 1 2 2 1 0
T(𝜉 ) = , T(𝜉 ) = ,
−2 1 3 4
(︂ )︂ (︂ )︂
1 3 2 6
h(𝜉 ) = , h(𝜉 ) = .
5 1
ˆ The penalties:
)︀𝑇 )︀𝑇
𝑞(𝜉 1 ) = 𝑞(𝜉 2 ) =
(︀ (︀
1 0 6 2 , 5 3 2 1 ,
1 1
𝑝𝑟𝑜𝑏(𝜉 1 ) = , 𝑝𝑟𝑜𝑏(𝜉 2 ) = .
2 2
ˆ Recourse-matrix:
(︂ )︂
−2 −1 2 1
W(𝜉) = W = ,
3 2 −5 −6
94 Younsi-Abbaci L., Moulaı̈ M.

we calculate the expected value 𝑍¯1 and 𝑍¯2 :

(︂ )︂ (︂ )︂
1 𝑥 1 𝑥
𝑍¯1 = 𝐸𝑠𝑝(𝑍1 (𝑥, 𝜉)) = 𝐶1 (𝜉 )
1 1 2
+ 𝐶1 (𝜉 ) 1
=
2 (︂ 𝑥2 )︂ 2 (︂𝑥2 )︂
1 𝑥1 1 𝑥1
= (4, −9) + (−6, 3) =
2 𝑥 2 2 𝑥2
= −𝑥1 − 3𝑥2 ,
(︂ )︂ (︂ )︂
1 𝑥 1 𝑥
𝑍¯2 = 𝐸𝑠𝑝(𝑍2 (𝑥, 𝜉)) = 𝐶2 (𝜉 )
1 1 2
+ 𝐶2 (𝜉 ) 1
=
2 (︂ 𝑥2)︂ 2 (︂𝑥2 )︂
1 𝑥1 1 𝑥1
= (8, 5) + (−2, −3) =
2 𝑥 2 2 𝑥2
= 3𝑥1 + 𝑥2 .
Deterministic constraints
−2𝑥1 + 5𝑥2 ≤ 23,
8𝑥1 + 𝑥2 ≤ 55,
𝑥1 − 𝑥2 ≤ 4,
−𝑥1 − 𝑥2 ≤ −6.

Stochastic constraints
First scenario 𝜉 1 𝑥1 + 2𝑥2 = 3,
−2𝑥1 + 𝑥2 = 5.

Second scenario 𝜉 2 𝑥1 + 0𝑥2 = 6,

3𝑥1 + 4𝑥2 = 1.
We obtain the following deterministic multiple objective integer linear programming
problem: ⎧
⎪
⎪ min 𝑍̃︀1 = −𝑥1 − 3𝑥2 + 𝑄(𝑥),
⎪
⎪
⎪
⎪ min 𝑍̃︀2 =⎧3𝑥1 + 𝑥2 + 𝑄(𝑥)
−2𝑥1 + 5𝑥2 ≤ 23,
⎪
⎪
⎨ ⎪
⎪
𝑀 𝑂1 ⎨ 1 + 𝑥2 ≤ 55,
⎪
⎪ 8𝑥 (14)
𝑥1 − 𝑥2 ≤ 4,
⎪
⎪
⎪
⎪ s.t. 𝐷 =
−𝑥1 − 𝑥2 ≤ −6,
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
𝑥1 , 𝑥2 ≥ 0, integer,
⎩ ⎩

1 1
with 𝑄(𝑥) = 𝑄(𝑥, 𝜉 1 ) + 𝑄(𝑥, 𝜉 2 ), and the second stage problem associated with the
2 2
two scenarios 𝑄(𝑥, 𝜉 1 ) and 𝑄(𝑥, 𝜉 2 ) respectively:
⎧
⎪
⎪ min 𝑦1 + 6𝑦3 + 2𝑦4
s.t. −2𝑦1 − 𝑦2 + 2𝑦3 + 𝑦4 = 3 − 𝑥1 − 2𝑥2 ,
⎨
Q(𝑥, 𝜉 1 ) (15)
⎪
⎪ 3𝑦1 + 2𝑦2 − 5𝑦3 − 6𝑦4 = 5 + 2𝑥1 − 𝑥2 ,
𝑦 ≥ 0,
⎩
⎧
⎪
⎪ min 5𝑦1 + 3𝑦2 + 2𝑦3 + 𝑦4
s.t. −2𝑦1 − 𝑦2 + 2𝑦3 + 𝑦4 = 6 − 𝑥1 ,
⎨
Q(𝑥, 𝜉 2 ) (16)
⎪
⎪ 3𝑦1 + 2𝑦2 − 5𝑦3 − 6𝑦4 = 1 + 3𝑥1 − 4𝑥2 ,
𝑦 ≥ 0.
⎩
Stochastic optimization over the Pareto front . . . 95

Fig. 1. Admissibility domain 𝐷1 with efficient set

We can use one of the algorithms developed in [11] to find the efficient set, it can be
shown that five of them are efficient. Particularly, the efficient set E𝑠 is given by
E𝑠 = {(1, 5); (2, 5); (3, 5); (4, 6); (6, 6)},
as shown in Fig. 1.
For this example, the parameter 𝜌 has been fixed at 0.001.
Initial iteration
ˆ We calculate the upper bound of each objective function 𝑀1 = −7.25 , 𝑀2 = 36.9855,
𝐷1 = 𝐷, 𝜃 = −∞;
̃︀𝑖𝑑𝑒𝑎𝑙 = −15.765, 𝑍̃︀𝑖𝑑𝑒𝑎𝑙 = 14, 𝑍̃︀1𝑢𝑡𝑜𝑝 = −16.765, 𝑍̃︀2𝑢𝑡𝑜𝑝 = 13.
ˆ 𝑍 1 2
ˆ Step 1. With 𝜃 = −∞ and without feasibility and optimality cuts, solve{︂the main de-
terministic relaxed problem 𝑃𝑅 under the deterministic constraints 𝑃𝑅1 ≡ min Φ(𝑥) =
}︂
−𝑥1 + 𝑥2 |𝑥 ∈ 𝐷 , an optimal solution is 𝑥1 = (5, 1). To test the feasibility of the
second-stage problems (15) and (16), we solve the program (7) with:
(︂ )︂ (︂ )︂ (︂ )︂ (︂ )︂
1 1 1 3 1 2 5 −4
ℎ(𝜉 ) − 𝑇 (𝜉 )𝑥 = − = ,
5 −2 1 1 14
(︂ )︂ (︂ )︂ (︂ )︂ (︂ )︂
6 1 0 5 1
ℎ(𝜉 2 ) − 𝑇 (𝜉 2 )𝑥1 = − = .
1 3 4 1 −18

max −4𝜎11 + 14𝜎12

s.t. −2𝜎11 + 3𝜎12 ≤ 0,
maximum is(︂at )︂
−1𝜎11 + 2𝜎12 ≤ 0,
2 1
2𝜎11 − 5𝜎12 ≤ 0, 𝜎1𝑇 = (𝜎11 , 𝜎12 ) = ,
3 3
1𝜎11 − 6𝜎12 ≤ 0,
1𝜎11 + 1𝜎12 ≤ 1
max 1𝜎21 + −18𝜎22
s.t. −2𝜎21 + 3𝜎22 ≤ 0,
−1𝜎21 + 2𝜎22 ≤ 0, maximum is at
2𝜎21 − 5𝜎22 ≤ 0, 𝜎2𝑇 = (𝜎21 , 𝜎22 ) = (0, 0)
1𝜎21 − 6𝜎22 ≤ 0,
1𝜎21 + 1𝜎22 ≤ 1
96 Younsi-Abbaci L., Moulaı̈ M.

Fig. 2. Admissibility domain without stochastic constraint

(︂ )︂ (︂ )︂
2 1 −4
𝜎1𝑇 [ℎ(𝜉 1 ) 1 1
− 𝑇 (𝜉 )𝑥 ] = , = 2 > 0,
3 3 14
(︂ )︂ (︂ )︂
𝑇 2 2 1 5 2 1
𝜎2 [ℎ(𝜉 ) − 𝑇 (𝜉 )𝑥 ] = , = 0.
7 7 −18
𝜎1𝑇 [ℎ(𝜉 1 ) − 𝑇 (𝜉 1 )𝑥1 ] > 0. It means that the second-stage is not feasible for the 𝜉 1 .
Then we create a feasibility cut
(︂ )︂ (︂ )︂ (︂ )︂ (︂ )︂ (︂ )︂
2 1 1 0 𝑥1 2 1 6
≥ ⇐⇒ 5𝑥1 ≥ 11.
3 3 3 4 𝑥2 3 3 1
The cut is added to the first problem 𝑃𝑅1 . We get a new integer point (6, 3) (see Fig. 2).

To test the feasibility of the second-stage problems (15) and (16), we solve the pro-
gram (7) with:
(︂ )︂ (︂ )︂ (︂ )︂ (︂ )︂
1 1 3 1 2 6 −9
ℎ(𝜉 ) − 𝑇 (𝜉 )(6, 3) = − = ,
5 −2 1 3 14
(︂ )︂ (︂ )︂ (︂ )︂ (︂ )︂
2 2 6 1 0 6 0
ℎ(𝜉 ) − 𝑇 (𝜉 )(6, 3) = − = .
1 3 4 3 −29

max −9𝜎11 + 14𝜎12

T.Q. −2𝜎11 + 3𝜎12 ≤ 0,
−1𝜎11 + 2𝜎12 ≤ 0, maximum is at
2𝜎11 − 5𝜎12 ≤ 0, 𝜎1𝑇 = (𝜎11 , 𝜎12 ) = (0, 0)
1𝜎11 − 6𝜎12 ≤ 0,
1𝜎11 + 1𝜎12 ≤ 1
max 0𝜎21 − 29𝜎22
T.Q. −2𝜎21 + 3𝜎22 ≤ 0,
−1𝜎21 + 2𝜎22 ≤ 0, maximum is at
2𝜎21 − 5𝜎22 ≤ 0, 𝜎2𝑇 = (𝜎21 , 𝜎22 ) = (0, 0)
1𝜎21 − 6𝜎22 ≤ 0,
1𝜎21 + 1𝜎22 ≤ 1
Stochastic optimization over the Pareto front . . . 97

(︂ )︂
−9
𝜎1𝑇 [ℎ(𝜉 1 ) 1
− 𝑇 (𝜉 )(6, 3)] = (0, 0) = 0,
14
(︂ )︂
0
𝜎2𝑇 [ℎ(𝜉 2 ) 2
− 𝑇 (𝜉 )(6, 3)] = (0, 0) = 0,
−29
𝜎1 = 𝜎2 = 0, this implies that the solution 𝑥1 = (6, 3) yields feasible second-stage
problems. To test the optimality of 𝑥1 = (6, 3), the dual (6) is solved for 𝜉 1 and 𝜉 2 .
max −9𝜋11 + 14𝜋12
T.Q. −2𝜋11 + 3𝜋12 ≤ 1, maximum (︂
is at )︂
−1𝜋11 + 2𝜋12 ≤ 0, 1
𝜋1𝑇 = (𝜋11 , 𝜋12 ) = −1, −
2𝜋11 − 5𝜋12 ≤ 6, 2
1𝜋11 − 6𝜋12 ≤ 2
max 0𝜋21 − 29𝜋22
T.Q. −2𝜋21 + 3𝜋22 ≤ 5,
maximum is at
−1𝜋21 + 2𝜋22 ≤ 3,
𝜋2𝑇 = (𝜋21 , 𝜋22 ) = (0, −0.17)
2𝜋21 − 5𝜋22 ≤ 2,
1𝜋21 − 6𝜋22 ≤ 1
We calculate the value of the second stage problem associated with the two scenarios
𝑄(𝑥1 , 𝜉 1 ) and 𝑄(𝑥1 , 𝜉 2 ) respectively:
(︂ )︂ (︂ )︂
1 1 𝑇 1 1 1 1 −9
Q(𝑥 , 𝜉 ) = 𝜋1 [ℎ(𝜉 ) − 𝑇 (𝜉 )𝑥 ] = −1, − = 2,
2 14
(︂ )︂
1 1 0
Q(𝑥 , 𝜉 ) = 𝜋2𝑇 [ℎ(𝜉 2 ) 2 1
− 𝑇 (𝜉 )𝑥 ] = (0, −0.17) = 4.93,
−29
1 1 1
Q(𝑥1 ) = Q(𝑥1 , 𝜉 1 ) + Q(𝑥1 , 𝜉 1 ) = (2 + 4.93) = 3.465.
2 2 2
𝜃 = −∞ < Q(𝑥1 ) we introduce the optimality cut of the form
𝑅
∑︁
𝜃> 𝑝𝑟 [ℎ(𝜉𝑟 ) − 𝑇 (𝜉𝑟 )𝑥],
𝑟=1

adding this cut 𝜃 ≥ 2.835 + 0.255𝑥1 + 1.593𝑥2 , and reoptimize the precedent program
𝑃𝑅1 , 𝜃 = 15.315, then 𝑥1 = (6, 3) is the optimal basic feasible solution on the current
region. Let 𝑍̃︀1 (𝑥1 ) = (−15, 21)+3.465 = (−11.535, 24.465), Φ(6,
̃︀ 3) = −3+𝜃 = −0.465,
Φ(6,
̃︀ 3) < Φ ̃︀ 𝑜𝑝𝑡 . We compute the weighted vector 𝜆 of the 𝑍̃︀1 :
1

[︂ ]︂−1
1 1 1
𝜆11 = + ,
−16.765 − (−11.535) (−16.765) − (−11.535) (13) − (24.465)
[︂ ]︂−1
1 1 1
𝜆12 = + ,
(13) − (24.465) (−16.765) − (−11.535) (13) − (24.465)
𝜆1 = (0.686, 0.314).
98 Younsi-Abbaci L., Moulaı̈ M.

ˆ Step 2. We solve the generalized Tchebychev program (𝑃𝜌 (𝜆1 )), which is defined as
follows: ⎧
⎪
⎪ min 𝛼 + 0.001(−3.7765 − 2𝑥1 + 2𝑥2 )
s.t. 𝛼 ≥ 0.686(−16.765 + 𝑥1 + 3𝑥2 ),
⎨
(𝑃𝜌 (𝜆1 )) (17)
⎪
⎪ 𝛼 ≥ 0.313(13 − 3𝑥1 − 𝑥2 ),
𝑥 ∈ 𝐷.
⎩

Then, 𝑥¯1 = (2, 5) is the optimal basic feasible solution of (𝑃𝜌 (𝜆1 )). To test the feasibility
of the second-stage problems (15) and (16), we solve the program (7) with:
(︂ )︂ (︂ )︂ (︂ )︂ (︂ )︂
1 1 1 3 1 2 2 −9
ℎ(𝜉 ) − 𝑇 (𝜉 )¯ 𝑥 = − = ,
5 −2 1 5 4
(︂ )︂ (︂ )︂ (︂ )︂ (︂ )︂
2 2 1 6 1 0 2 4
ℎ(𝜉 ) − 𝑇 (𝜉 )¯ 𝑥 = − = .
1 3 4 5 −25

max −9𝜎11 + 4𝜎12

T.Q. −2𝜎11 + 3𝜎12 ≤ 0,
−1𝜎11 + 2𝜎12 ≤ 0, maximum is at
2𝜎11 − 5𝜎12 ≤ 0, 𝜎1𝑇 = (𝜎11 , 𝜎12 ) = (0, 0)
1𝜎11 − 6𝜎12 ≤ 0,
1𝜎11 + 1𝜎12 ≤ 1
max 4𝜎21 − 25𝜎22
T.Q. −2𝜎21 + 3𝜎22 ≤ 0,
−1𝜎21 + 2𝜎22 ≤ 0, maximum is at
2𝜎21 − 5𝜎22 ≤ 0, 𝜎2𝑇 = (𝜎21 , 𝜎22 ) = (0, 0)
1𝜎21 − 6𝜎22 ≤ 0,
1𝜎21 + 1𝜎22 ≤ 1
The solution is feasible for both first and second scenario.
To test the optimality of 𝑥¯1 = (2, 5), the dual (6) is solved for 𝜉 1 and 𝜉 2 .
max −9𝜋11 + 4𝜋12
T.Q. −2𝜋11 + 3𝜋12 ≤ 1, maximum (︂
is at )︂
−1𝜋11 + 2𝜋12 ≤ 0, 1
𝜋1𝑇 = (𝜋11 , 𝜋12 ) = −1, −
2𝜋11 − 5𝜋12 ≤ 6, 2
1𝜋11 − 6𝜋12 ≤ 2
max 4𝜋21 − 25𝜋22
T.Q. −2𝜋21 + 3𝜋22 ≤ 5,
maximum is at
−1𝜋21 + 2𝜋22 ≤ 3,
𝜋2𝑇 = (𝜋21 , 𝜋22 ) = (0, −0.17)
2𝜋21 − 5𝜋22 ≤ 2,
1𝜋21 − 6𝜋22 ≤ 1
We calculate the value of the second stage problem associated with the two scenarios
𝑥1 , 𝜉 1 ) and 𝑄(¯
𝑄(¯ 𝑥1 , 𝜉 2 ) respectively:
(︂ )︂ (︂ )︂
1 1 𝑇 1 1 1 1 −9
Q(¯𝑥 , 𝜉 ) = 𝜋1 [ℎ(𝜉 ) − 𝑇 (𝜉 )¯
𝑥 ] = −1, − = 7,
2 4
(︂ )︂
1 1 𝑇 2 2 1 4
𝑥 , 𝜉 ) = 𝜋2 [ℎ(𝜉 ) − 𝑇 (𝜉 )¯
Q(¯ 𝑥 ] = (0, −0.17) = 4.25,
−25
Stochastic optimization over the Pareto front . . . 99

1 1
𝑥1 ) = Q(¯
Q(¯ 𝑥1 , 𝜉 1 ) + Q(¯
𝑥1 , 𝜉 1 ) = 5.625.
2 2
The solution 𝑍̃︀1 (¯
𝑥1 ) = (−17, 11) + 5.625 = (−11.375, 16.625) is a nondominated point
with minimal weighted Tchebychev distance, we obtain 𝑥¯1 = (2, 5) and Φ(2, ̃︀ 5) =
3 + 5.625 = 8.625.
ˆ Step 3. Solve the equivalent efficient solutions program:
⎧
⎪
⎪ min Φ(𝑥) = −𝑥1 + 𝑥2 + 𝜃
s.t. 𝑥1 , 𝑥2 ∈ 𝐷,
⎨
(𝑇 1 ) (18)
⎪
⎪ −𝑥1 − 3𝑥2 + 𝜃 = −11.375,
3𝑥1 + 1𝑥2 + 𝜃 = 16.625.
⎩

An optimal solution is 𝑥*1 = 𝑥1 = (2, 5) with 𝜃 = 5.625, Φ(𝑥 ̃︀ *1 ) = 8.625 < Φ̃︀ 𝑜𝑝𝑡 ,
𝑥𝑜𝑝𝑡 := 𝑥*1 , Φ ̃︀ *1 , and let E1 = {(2, 5)}, 𝑙 := 𝑙 + 1 = 2 and we solve problem 𝑃 2 .
̃︀ 𝑜𝑝𝑡 := 𝑑𝑥 𝑠 𝑅
Iteration 2
ˆ Step 1. ⎧
⎪
⎪ min Φ(𝑥) ̃︀ = −𝑥1 + 𝑥2 + 𝜃
⎨ s.t. 𝑥1 , 𝑥2 ∈ 𝐻,
⎪
⎪ ˜
2
(𝑃𝑅 ) −𝑥1 − 3𝑥2 + 𝜃 ≤ (−11.375 + 1)𝑦11 − 7.25(1 − 𝑦11 ), (19)
1 1
3𝑥1 + 𝑥2 + 𝜃 ≤ (16.625 + 1)𝑦2 + 36.985(1 − 𝑦2 ),
⎪
⎪
⎪
⎪
𝑦11 + 𝑦11 ≥ 1, 𝑦11 , 𝑦21 ∈ {0, 1}.
⎩

An optimal solution is 𝑥2 = (6, 4), with 𝑦 1 = (1, 0). To test the feasibility of the
second-stage problems (15) and (16), we solve the program (7) with:
(︂ )︂ (︂ )︂ (︂ )︂ (︂ )︂
1 1 2 3 1 2 6 −11
ℎ(𝜉 ) − 𝑇 (𝜉 )𝑥 = − = ,
5 −2 1 4 13

(︂ )︂ (︂ )︂ (︂ )︂ (︂ )︂
2 2 2 6 1 0 6 0
ℎ(𝜉 ) − 𝑇 (𝜉 )𝑥 = − = .
1 3 4 4 −33

max −11𝜎11 + 13𝜎12

T.Q. −2𝜎11 + 3𝜎12 ≤ 0,
−1𝜎11 + 2𝜎12 ≤ 0, maximum is at
2𝜎11 − 5𝜎12 ≤ 0, 𝜎1𝑇 = (𝜎11 , 𝜎12 ) = (0, 0)
1𝜎11 − 6𝜎12 ≤ 0,
1𝜎11 + 1𝜎12 ≤ 1
max 0𝜎21 − 33𝜎22
T.Q. −2𝜎21 + 3𝜎22 ≤ 0,
−1𝜎21 + 2𝜎22 ≤ 0, maximum is at
2𝜎21 − 5𝜎22 ≤ 0, 𝜎2𝑇 = (𝜎21 , 𝜎22 ) = (0, 0)
1𝜎21 − 6𝜎22 ≤ 0,
1𝜎21 + 1𝜎22 ≤ 1
The solution is feasible for both first and second scenario. To test the optimality of
𝑥2 = (6, 4), the dual (6) is solved for 𝜉 1 and 𝜉 2 .
100 Younsi-Abbaci L., Moulaı̈ M.

max −11𝜋11 + 13𝜋12

T.Q. −2𝜋11 + 3𝜋12 ≤ 1, maximum (︂
is at )︂
−1𝜋11 + 2𝜋12 ≤ 0, 1
𝜋1𝑇 = (𝜋11 , 𝜋12 ) = −1, −
2𝜋11 − 5𝜋12 ≤ 6, 2
1𝜋11 − 6𝜋12 ≤ 2
max 0𝜋21 − 33𝜋22
T.Q. −2𝜋21 + 3𝜋22 ≤ 5,
maximum is at
−1𝜋21 + 2𝜋22 ≤ 3,
𝜋2𝑇 = (𝜋21 , 𝜋22 ) = (0, −0.17)
2𝜋21 − 5𝜋22 ≤ 2,
1𝜋21 − 6𝜋22 ≤ 1
We calculate the value of the second stage problem associated with the two scenarios
𝑄(𝑥2 , 𝜉 1 ) and 𝑄(𝑥2 , 𝜉 2 ) respectively: (︂ )︂ (︂ )︂
2 2 𝑇 1 1 2 1 −11
Q(𝑥 , 𝜉 ) = 𝜋1 [ℎ(𝜉 ) − 𝑇 (𝜉 )𝑥 ] = −1, − = 4.5,
2 13
(︂ )︂
2 1 𝑇 2 2 2 0
Q(𝑥 , 𝜉 ) = 𝜋2 [ℎ(𝜉 ) − 𝑇 (𝜉 )𝑥 ] = (0, −0.17) = 5.2,
−33
1 1
Q(𝑥2 ) = Q(𝑥2 , 𝜉 1 ) + Q(𝑥2 , 𝜉 1 ) = 5.1.
2 2
Then 𝑥2 = (6, 4) is the optimal basic feasible solution in the current region. Let
𝑍̃︀2 (𝑥2 ) = (−18, 22) + 5.1 = (−12.9, 27.3), Φ(6,̃︀ 4) = −2 + 𝜃 = 3.1, Φ(6, ̃︀ 4) < Φ̃︀ 𝑜𝑝𝑡 . We
compute the weighted vector 𝜆2 of the 𝑍̃︀2 : 𝜆2 = (0.782, 0.218).
ˆ Step 2. We solve the generalized Tchebychev program (𝑃𝜌 (𝜆2 )), which is defined as
follows: ⎧
min 𝛼 + 0.001(−3.7765 − 2𝑥1 + 2𝑥2 )
⎪
⎪
s.t. 𝛼 ≥ 0.782(−16.765 + 𝑥1 + 3𝑥2 ),
⎨
(𝑃𝜌 (𝜆2 )) (20)
⎪
⎪ 𝛼 ≥ 0.218(13 − 3𝑥1 − 𝑥2 ),
𝑥 ∈ 𝐷.
⎩

Then, 𝑥¯2 = (3, 5) is the optimal basic feasible solution of (𝑃𝜌 (𝜆2 )). To test the feasibility
of the second-stage problems (15) and (16), we solve the program (7) with:
(︂ )︂ (︂ )︂ (︂ )︂ (︂ )︂
1 1 2 3 1 2 3 −10
ℎ(𝜉 ) − 𝑇 (𝜉 )¯ 𝑥 = − = ,
5 −2 1 5 6
(︂ )︂ (︂ )︂ (︂ )︂ (︂ )︂
2 2 2 6 1 0 3 3
ℎ(𝜉 ) − 𝑇 (𝜉 )¯ 𝑥 = − = .
1 3 4 5 −28
max −10𝜎11 + 6𝜎12
T.Q. −2𝜎11 + 3𝜎12 ≤ 0,
−1𝜎11 + 2𝜎12 ≤ 0, maximum is at
2𝜎11 − 5𝜎12 ≤ 0, 𝜎1𝑇 = (𝜎11 , 𝜎12 ) = (0, 0)
1𝜎11 − 6𝜎12 ≤ 0,
1𝜎11 + 1𝜎12 ≤ 1
max 3𝜎21 − 28𝜎22
T.Q. −2𝜎21 + 3𝜎22 ≤ 0,
−1𝜎21 + 2𝜎22 ≤ 0, maximum is at
2𝜎21 − 5𝜎22 ≤ 0, 𝜎2𝑇 = (𝜎21 , 𝜎22 ) = (0, 0)
1𝜎21 − 6𝜎22 ≤ 0,
1𝜎21 + 1𝜎22 ≤ 1
Stochastic optimization over the Pareto front . . . 101

The solution is feasible for both first and second scenario. To test the optimality of
𝑥¯2 = (3, 5), the dual (6) is solved for 𝜉 1 and 𝜉 2 .
max −10𝜋11 + 6𝜋12
T.Q. −2𝜋11 + 3𝜋12 ≤ 1, maximum (︂
is at )︂
−1𝜋11 + 2𝜋12 ≤ 0, 1
𝜋1𝑇 = (𝜋11 , 𝜋12 ) = −1, −
2𝜋11 − 5𝜋12 ≤ 6, 2
1𝜋11 − 6𝜋12 ≤ 2
max 3𝜋21 − 28𝜋22
T.Q. −2𝜋21 + 3𝜋22 ≤ 5,
maximum is at
−1𝜋21 + 2𝜋22 ≤ 3,
𝜋2𝑇 = (𝜋21 , 𝜋22 ) = (0, −0.17)
2𝜋21 − 5𝜋22 ≤ 2,
1𝜋21 − 6𝜋22 ≤ 1
We calculate the value of the second stage problem associated with the two scenarios
𝑄(¯𝑥2 , 𝜉 1 ) and 𝑄(¯𝑥2 , 𝜉 2 ) respectively:
(︂ )︂ (︂ )︂
2 1 𝑇 1 1 2 1 −10
Q(¯𝑥 , 𝜉 ) = 𝜋1 [ℎ(𝜉 ) − 𝑇 (𝜉 )¯ 𝑥 ] = −1, − = 7,
2 6
(︂ )︂
2 1 𝑇 2 2 2 3
𝑥 , 𝜉 ) = 𝜋2 [ℎ(𝜉 ) − 𝑇 (𝜉 )¯
Q(¯ 𝑥 ] = (0, −0.17) = 4.76,
−28
1 1
Q(¯ 𝑥2 ) = Q(¯ 𝑥2 , 𝜉 1 ) + Q(¯ 𝑥2 , 𝜉 1 ) = 5.88.
2 2
The solution 𝑍̃︀2 (¯ 𝑥2 ) = (−18, 14)+5.88 = (−13.88, 19.88) is a nondominated point with
minimal weighted Tchebychev distance, we obtain 𝑥¯2 = (3, 5) and
Φ(3, 5) = 2 + 5.88 = 7.88.
̃︀
ˆ Step 3. Solve the equivalent efficient solutions program:
⎧
⎪
⎪ min Φ(𝑥) = −𝑥1 + 𝑥2 + 𝜃
s.t. 𝑥1 , 𝑥2 ∈ 𝐷,
⎨
(𝑇 2 ) (21)
⎪
⎪ −𝑥1 − 3𝑥2 + 𝜃 = −13.88,
3𝑥1 + 1𝑥2 + 𝜃 = 19.88.
⎩

An optimal solution is 𝑥*2 = 𝑥2 = (3, 5) with 𝜃 = 5.88, Φ(𝑥 ̃︀ *2 ) = 7.88 < Φ

̃︀ 𝑜𝑝𝑡 ,
*2 ̃︀ *2 2
̃︀ , and let E = {(2, 5), (3, 5)}, 𝑙 := 𝑙 + 2 = 3 and we solve
𝑥𝑜𝑝𝑡 := 𝑥 , Φ𝑜𝑝𝑡 := 𝑑𝑥 𝑠
3
problem (𝑃𝑅 ).
Iteration 3
ˆ Step 1.
⎧
⎪
⎪ min Φ(𝑥)
̃︀ = −𝑥1 + 𝑥2 + 𝜃
⎨ s.t. 𝑥1 , 𝑥2 ∈ 𝐻,
⎪
⎪ ˜
3
(𝑃𝑅 ) −𝑥1 − 3𝑥2 + 𝜃 ≤ (−13.88 + 1)𝑦11 − 7.25(1 − 𝑦11 ), (22)
1 1
3𝑥1 + 𝑥2 + 𝜃 ≤ (19.88 + 1)𝑦2 + 36.985(1 − 𝑦2 ),
⎪
⎪
⎪
⎪
𝑦11 + 𝑦11 ≥ 1, 𝑦11 , 𝑦21 ∈ {0, 1}.
⎩

An optimal solution is 𝑥3 = (6, 5), with 𝑦 1 = (1, 0). To test the feasibility of the
second-stage problems (15) and (16), we solve the program (7) with:
(︂ )︂ (︂ )︂ (︂ )︂ (︂ )︂
1 1 3 3 1 2 6 −13
ℎ(𝜉 ) − 𝑇 (𝜉 )𝑥 = − = ,
5 −2 1 5 12
102 Younsi-Abbaci L., Moulaı̈ M.

(︂ )︂ (︂ )︂ (︂ )︂ (︂ )︂
2 2 3 6 1 0 6 0
ℎ(𝜉 ) − 𝑇 (𝜉 )𝑥 = − = .
1 3 4 5 −37

max −11𝜎11 + 13𝜎12

T.Q. −2𝜎11 + 3𝜎12 ≤ 0,
−1𝜎11 + 2𝜎12 ≤ 0, maximum is at
2𝜎11 − 5𝜎12 ≤ 0, 𝜎1𝑇 = (𝜎11 , 𝜎12 ) = (0, 0)
1𝜎11 − 6𝜎12 ≤ 0,
1𝜎11 + 1𝜎12 ≤ 1
max 0𝜎21 − 33𝜎22
T.Q. −2𝜎21 + 3𝜎22 ≤ 0,
−1𝜎21 + 2𝜎22 ≤ 0, maximum is at
2𝜎21 − 5𝜎22 ≤ 0, 𝜎2𝑇 = (𝜎21 , 𝜎22 ) = (0, 0)
1𝜎21 − 6𝜎22 ≤ 0,
1𝜎21 + 1𝜎22 ≤ 1
The solution is feasible for both first and second scenario. To test the optimality of
𝑥3 = (6, 5), the dual (6) is solved for 𝜉 1 and 𝜉 2 .
max −13𝜋11 + 12𝜋12
T.Q. −2𝜋11 + 3𝜋12 ≤ 1, maximum (︂
is at )︂
−1𝜋11 + 2𝜋12 ≤ 0, 1
𝜋1𝑇 = (𝜋11 , 𝜋12 ) = −1, −
2𝜋11 − 5𝜋12 ≤ 6, 2
1𝜋11 − 6𝜋12 ≤ 2
max 0𝜋21 − 38𝜋22
T.Q. −2𝜋21 + 3𝜋22 ≤ 5,
maximum is at
−1𝜋21 + 2𝜋22 ≤ 3,
𝜋2𝑇 = (𝜋21 , 𝜋22 ) = (0, −0.17)
2𝜋21 − 5𝜋22 ≤ 2,
1𝜋21 − 6𝜋22 ≤ 1
We calculate the value of the second stage problem associated with the two scenarios
𝑄(𝑥3 , 𝜉 1 ) and 𝑄(𝑥3 , 𝜉 2 ) respectively:
(︂ )︂ (︂ )︂
3 2 𝑇 1 1 3 1 −13
Q(𝑥 , 𝜉 ) = 𝜋1 [ℎ(𝜉 ) − 𝑇 (𝜉 )𝑥 ] = −1, − = 7,
2 12
(︂ )︂
3 1 𝑇 2 2 3 0
Q(𝑥 , 𝜉 ) = 𝜋2 [ℎ(𝜉 ) − 𝑇 (𝜉 )𝑥 ] = (0, −0.17) = 6.46,
−38
1 1
Q(𝑥3 ) = Q(𝑥3 , 𝜉 1 ) + Q(𝑥3 , 𝜉 1 ) = 6.73.
2 2
Then, 𝑥3 = (6, 5) is the optimal basic feasible solution on the current region. Let
𝑍̃︀3 (𝑥3 ) = (−21, 23) + 6.73 = (−14.27, 29.73), Φ(6,̃︀ 5) = −2 + 𝜃 = 4.73, Φ(6, ̃︀ 5) < Φ
̃︀ 𝑜𝑝𝑡 .
3 3 3
We compute the weighted vector 𝜆 of the 𝑍̃︀ : 𝜆 = (0.87, 0.13).
ˆ Step 2. We solve the generalized Tchebychev program (𝑃𝜌 (𝜆3 )), which is defined as
follows: ⎧
⎪
⎪ min 𝛼 + 0.001(−3.7765 − 2𝑥1 + 2𝑥2 )
s.t. 𝛼 ≥ 0.87(−16.765 + 𝑥1 + 3𝑥2 ),
⎨
(𝑃𝜌 (𝜆3 )) (23)
⎪
⎪ 𝛼 ≥ 0.13(13 − 3𝑥1 − 𝑥2 ),
𝑥 ∈ 𝐷.
⎩
Stochastic optimization over the Pareto front . . . 103

Then, 𝑥¯3 = (4, 6) is the optimal basic feasible solution of (𝑃𝜌 (𝜆3 )). To test the feasibility
of the second-stage problems (15) and (16), we solve the program (7) with:
(︂ )︂ (︂ )︂ (︂ )︂ (︂ )︂
1 1 3 3 1 2 4 −13
ℎ(𝜉 ) − 𝑇 (𝜉 )¯ 𝑥 = − = ,
5 −2 1 6 7
(︂ )︂ (︂ )︂ (︂ )︂ (︂ )︂
2 2 3 6 1 0 4 2
ℎ(𝜉 ) − 𝑇 (𝜉 )¯ 𝑥 = − = .
1 3 4 6 −35

max −13𝜎11 + 7𝜎12

T.Q. −2𝜎11 + 3𝜎12 ≤ 0,
−1𝜎11 + 2𝜎12 ≤ 0, maximum is at
2𝜎11 − 5𝜎12 ≤ 0, 𝜎1𝑇 = (𝜎11 , 𝜎12 ) = (0, 0)
1𝜎11 − 6𝜎12 ≤ 0,
1𝜎11 + 1𝜎12 ≤ 1
max 2𝜎21 − 35𝜎22
T.Q. −2𝜎21 + 3𝜎22 ≤ 0,
−1𝜎21 + 2𝜎22 ≤ 0, maximum is at
2𝜎21 − 5𝜎22 ≤ 0, 𝜎2𝑇 = (𝜎21 , 𝜎22 ) = (0, 0)
1𝜎21 − 6𝜎22 ≤ 0,
1𝜎21 + 1𝜎22 ≤ 1
The solution is feasible for both first and second scenario.
To test the optimality of 𝑥¯3 = (4, 6), the dual (6) is solved for 𝜉 1 and 𝜉 2 .
max −13𝜋11 + 7𝜋12
T.Q. −2𝜋11 + 3𝜋12 ≤ 1, maximum (︂
is at )︂
−1𝜋11 + 2𝜋12 ≤ 0, 1
𝜋1𝑇 = (𝜋11 , 𝜋12 ) = −1, −
2𝜋11 − 5𝜋12 ≤ 6, 2
1𝜋11 − 6𝜋12 ≤ 2
max 2𝜋21 − 35𝜋22
T.Q. −2𝜋21 + 3𝜋22 ≤ 5,
maximum is at
−1𝜋21 + 2𝜋22 ≤ 3,
𝜋2𝑇 = (𝜋21 , 𝜋22 ) = (0, −0.17)
2𝜋21 − 5𝜋22 ≤ 2,
1𝜋21 − 6𝜋22 ≤ 1
We calculate the value of the second stage problem associated with the two scenarios
𝑥3 , 𝜉 1 ) and 𝑄(¯
𝑄(¯ 𝑥3 , 𝜉 2 ) respectively:
(︂ )︂ (︂ )︂
3 1 𝑇 1 1 3 1 −13
𝑥 , 𝜉 ) = 𝜋1 [ℎ(𝜉 ) − 𝑇 (𝜉 )¯
Q(¯ 𝑥 ] = −1, − = 9.5,
2 7
(︂ )︂
3 1 𝑇 2 2 3 2
𝑥 , 𝜉 ) = 𝜋2 [ℎ(𝜉 ) − 𝑇 (𝜉 )¯
Q(¯ 𝑥 ] = (0, −0.17) = 5.95,
−35
1 1
Q(¯𝑥3 ) = Q(¯ 𝑥3 , 𝜉 1 ) + Q(¯
𝑥3 , 𝜉 1 ) = 7.725.
2 2
3 3
The solution 𝑍 (¯
̃︀ 𝑥 ) = (−22, 18) + 5.88 = (−14.275, 25.725) is a nondominated point
with minimal weighted Tchebychev distance, we obtain 𝑥¯3 = (4, 6) and Φ(4, ̃︀ 6) =
2 + 7.725 = 9.725.
104 Younsi-Abbaci L., Moulaı̈ M.

Fig. 3. The end iteration

ˆ Step 3. Solve the equivalent efficient solutions program:

⎧
⎪
⎪ min Φ(𝑥) = −𝑥1 + 𝑥2 + 𝜃
s.t. 𝑥1 , 𝑥2 ∈ 𝐷,
⎨
(𝑇 3 ) (24)
⎪
⎪ −𝑥1 − 3𝑥2 + 𝜃 = −14.275,
3𝑥1 + 1𝑥2 + 𝜃 = 25.725.
⎩

An optimal solution is 𝑥*3 = 𝑥3 = (4, 6) with 𝜃 = 7.725, Φ(𝑥 ̃︀ *3 ) = 9.725 > Φ

̃︀ 𝑜𝑝𝑡 = 7.88.
Terminated:= True
𝑥𝑜𝑝𝑡 = 𝑥*2 = (3, 5), Φ
̃︀ 𝑜𝑝𝑡 = 7.88 is the optimal solution of the problem (𝑃𝐸 ) (fig. 3).
The set of problems presented above have been solved by the MATLAB 7.0 environment.
However, our algorithm optimizes the linear function Φ(𝑥) = −𝑥1 + 𝑥2 without having to
determine all these solutions but only E3𝑠 = {(2, 5), (3, 5), (4, 6)}.

Conclusion
In this work we have presented a new exact algorithm for optimizing over the integer efficient
set of a stochastic multi-objective program based on the augmented weighted Tchebychev
Program. The algorithm finds an integer optimal solution for problem (𝑃𝐸 ) in a finite number
of steps.
The proposed algorithm is formed on, the stochastic data are treated by recourse ap-
proach to obtain an equivalent deterministic program. We achieve this objective by com-
bining two ideas: one consists of solving the augmented weighted Tchebychev program in
the outcome space criteria to characterize nondominated criterion vector; then adding suc-
cessive Gomory cuts, if necessary, we obtain an integer feasible solution and the feasibility
cuts eliminate some parts of the first-stage decision set. And the second idea is to reduce
progressively the admissible domain by adding more constraints eliminating all the points
dominated by the current solution. A small number of iterations is necessary to obtain the
optimal solution for (𝑃𝐸 ).
In some applications the decision maker does not often have the possibility to use recours
in the future, after the occurrence of a scenario (all 𝐾 induced stresses are empty). In
this case, we have to use another approach to the stochastic programming to convert the
Stochastic optimization over the Pareto front . . . 105

problem to a deterministic stochastic problem. Concerning the complexity, as we are obliged

to transform the stochastic problem into deterministic one, problem (𝑃𝐸 ) remains hard as
was stated in deterministic case by Guyen (1992).

Acknowledgements. This work was supported in part by the Ministry of Higher Educa-
tion and scientific research of Algeria. The authors would also like to thank Professor Farouk
Yalaoui and Dr. Faicel Hnaien from LOSI University of Technology of Troyes, France, for
their help and encouragement during the preparation of this paper.

References
[1] Johan Ph. Algorithms for the vector maximization problem. Mathematical Programming.
1972; 2(1):207–229.
[2] Isermann H., Steuer R.E. Computational experience concerning payoff tables and mini-
mum criterion values over the efficient set. European Journal of Operational Research. 1988;
33(1):91–97.
[3] Benson H.P. Optimization over the efficient set. Journal of Mathematical Analysis and Ap-
plications. 1984; 98(2):562–580.
[4] Ecker J.G., Song J.H. Optimizing a linear function over an efficient set. Journal of Opti-
mization Theory and Applications. 1994; 83(3):541–563.
[5] Bolintineanu S. Minimization of a quasi-concave function over an efficient set. Mathematical
Programming. 1993; 61(1–3):89–110.
[6] Sayin S. Optimizing over the efficient set using a top-down search of faces. Operations Re-
search. 2000; 48(1):65–72.
[7] Abbas M., Chaabane D. Optimizing a linear function over an integer efficient set. European
Journal of Operational Research. 2006; 174(2):1140–1161.
[8] Jorge J.M. An algorithm for optimizing a linear function over an integer efficient set. Euro-
pean Journal of Operational Research. 2009; 195(1):98–103.
[9] Chaabane D., Brahmi B., Ramdani Z. The augmented weighted tchebychev norm for
optimizing a linear function over an integer efficient set of a multicriteria linear program.
International Transactions in Operational Research. 2012; 19(4):531–545.
[10] Teghem J. Strange-momix: An interactive method for mixed integer linear programming.
Stochastic VerKall – Mayersus fuzzy apprroaches to multiobjective mathematical programming
under uncertainty. Dordrecht: Kluwer; 1990: 101–115.
[11] Abbas M., Bellahcene F. Cutting plane method for multiple objective stochastic integer
linear programming. European Journal of Operational Research. 2006: 168(3):967–984.
[12] Abbas M., Moulai M. An algorithm for mixed integer linear fractional programming prob-
lem. Jorbel. 1999; (39):21–30.
[13] Van Slyke R.M., Wets R. L-shaped linear programs with applications to optimal control
and stochastic programming. SIAM Journal on Applied Mathematics. 1969; 17(4):638–663.
[14] Amrouche S., Moulai M. Multi-objective stochastic integer linear programming with fixed
recourse. International Journal of Multicriteria Decision Making. 2012; 2(4):355–378.
[15] Benders J.F. Partitioning procedures for solving mixed-variables programming problems.
Numerische Mathematik. 1962; 4(1):238–252.
[16] Steuer R.E. Multiple criteria optimization: Theory, computation, and applications. Wiley;
1986: 546.
 106 Younsi-Abbaci L., Moulaï M.

[17] Kall P., Mayer J. Stochastic linear programming. Springer, 1976; (21):92.
[18] Kall P., Wallace S.W. Stochastic programming. Journal of the Operational Research Soci-
ety. 1994; 46(9). DOI:10.2307/2584504.
[19] V Joseph Bowman Jr. On the relationship of the Tchebycheff norm and the efficient frontier
of multiple-criteria objectives. In Multiple criteria decision making. Springer; 1976: 76–86.
[20] Steuer R.E., Choo E.-U. An interactive weighted tchebycheff procedure for multiple objec-
tive programming. Mathematical Programming. 1983; 26(3):326–344.
[21] Sylva J., Crema A. A method for finding the set of non-dominated vectors for mul-
tiple objective integer linear programs. European Journal of Operational Research. 2004;
158(1):46–55.

Вычислительные технологии, 2020, том 25, › 6, с. 86–106. © ФИЦ ИВТ, 2020 ISSN 1560-7534
Computational Technologies, 2020, vol. 25, no. 6, pp. 86–106. © FRC ICT, 2020 eISSN 2313-691X

ВЫЧИСЛИТЕЛЬНЫЕ ТЕХНОЛОГИИ

DOI:10.25743/ICT.2021.26.3.006

Стохастическая оптимизация по фронту Парето с помощью расширенной

взвешенной программы Чебышева

Л. Юнси-Аббаси , М. Мула
*

Университет науки и технологии Уари Бумедьен, лаборатория LaROMaD, факультет математики,

16111, Баб-Эззуар, Эль-Алия, Алжир

* Контактный автор: Лейла Юнси-Аббаси, e-mail: [email protected]

Поступила 14 декабря 2020 г., доработана 21 апреля 2021 г., принята в печать 28 апреля 2021 г.

Аннотация

В этой статье мы предлагаем новый алгоритм для решения многоцелевых задач стохасти-
ческого целочисленного линейного программирования (MOSILP). Мы оптимизируем данную
стохастическую линейную функцию 𝜑 по полному набору эффективных решений MOSILP,
которые были преобразованы в эквивалентную детерминированную задачу с использовани-
ем неопределенных предположений, вводимых лицом, принимающим решения. Для этой цели
мы применяем двухэтапный рекурсивный подход, при котором расширенная взвешенная про-
грамма Чебышева постепенно оптимизируется для создания эффективного решения, тем са-
мым улучшая значение вспомогательной функции 𝜑. Предлагаемый здесь подход определяет
и решает последовательность целочисленных линейных программ с нарастающими ограниче-
ниями, так что на каждом этапе алгоритма генерируется новое эффективное решение. Для
иллюстрации представлен числовой пример.
Ключевые слова : многоцелевая задача, целочисленное программирование, стохастическое
линейное программирование, норма Чебышева.
Цитирование : Younsi-Abbaci L., Moulaï M. Stochastic optimization over the Pareto front by
the augmented weighted Tchebychev program. Computational Technologies. 2021; 26(3):86–106.
DOI:10.25743/ICT.2021.26.3.006.
Благодарности. Исследование выполнено при частичной поддержке Министерства высшего
образования и научных исследований Алжира. Авторы также выражают благодарность про-
фессору Фаруку Ялауи и доктору Файселу Хнайену из Технологического университета LOSI
(Труа, Франция) за их помощь и поддержку в подготовке данной статьи.