Int. J. Industrial and Systems Engineering, Vol. 12, No.

2, 2012 119

A primal-dual method for solving linear programming

problems with fuzzy cost coefficients based on linear
ranking functions and its applications

Ali Ebrahimnejad
Department of Mathematics,
Qaemshahr Branch,
Islamic Azad University,
P.O. Box 163,
Qaemshahr, Iran
Fax: +98 1232240091
E-mail: [email protected]
E-mail: [email protected]

Abstract: There are two important approaches based on linear ranking

functions for solving linear programming problems with cost coefficients as an
auxiliary problem to obtain a fuzzy solution of fuzzy variable linear
programming problem. The first approach uses the primal simplex method that
assumes an initial primal feasible basic solution is at hand. The second
approach is based on dual simplex method that begins with a basic dual feasible
basic solution and proceeds by pivoting through a series of dual basic solutions
until the associated complementary primal basic fuzzy solution is feasible. In
this paper, we propose a new method called the primal-dual algorithm, which is
similar to the dual simplex method and begins with dual feasibility and
proceeds to obtain primal feasibility while maintaining complementary
slackness. An important difference between the dual simplex method and the
primal-dual method is that the primal-dual algorithm does not require a dual
feasible solution to be basic. This algorithm is useful specially for solving
minimum fuzzy cost flow problem in which finding an initial dual feasible
solution turns out to be a trivial task.

Keywords: FNLP; fuzzy number linear programming; primal-dual simplex

method; ranking functions; TRFN; trapezoidal fuzzy number.

Reference to this paper should be made as follows: Ebrahimnejad, A. (2012)

‘A primal-dual method for solving linear programming problems with fuzzy
cost coefficients based on linear ranking functions and its applications’, Int. J.
Industrial and Systems Engineering, Vol. 12, No. 2, pp.119–140.

Biographical notes: Ali Ebrahimnejad serves as an Assistance Professor at

Mathematics Department, Qaemshahr Branch, Islamic Azad University. He
obtained his PhD degree in Applied Mathematics in the Department of
Mathematics at the Islamic Azad University (Sciences and Research Branch),
Tehran, Iran. He received his BS from Mazandaran University, Iran and his
MSc from Tehran University, Iran. He also serves as a Reviewer for some
journals such as Journal of Computational and Applied Mathematics, Int. J.
Uncertainty, Fuzziness and Knowledge-Based Systems, Neural Computing and
Applications, Int. J. Advanced Operations Management and Int. J. Fuzzy

Copyright © 2012 Inderscience Enterprises Ltd.

120 A. Ebrahimnejad

Systems. His research interests include operations research, network flows, data
envelopment analysis and fuzzy linear programming.

1 Introduction

In many practical applications of mathematical programming, the decision-maker may

not be in a position to specify the objective and/or constraint functions precisely but
rather can specify them in a fuzzy sense. In such situations, it is desirable to use some
fuzzy linear programming (FLP) type of modelling so as to provide more flexibility to the
decision-maker, see e.g. Smimou (2007), Kumar et al. (2008), Tang and Beynon (2009),
Vrana and Aly (2010) and Azadeh et al. (2010, 2011). The first formulation of FLP is
proposed by Zimmermann (1978). Afterwards, many authors considered various kinds of
the FLP problems and proposed several approaches for solving these problems. Ganesan
and Veeramani (2006) formulated FLPs with trapezoidal fuzzy numbers (TRFNs) and
introduced a primal simplex method for solving these problems based on a new type of
fuzzy arithmetic for symmetric TRFNs. In addition, Ebrahimnejad et al. (2010a) extended
their method for solving bounded linear programmes with symmetric TRFNs and then
Ebrahimnejad (2011a) obtained some new results for this kind of fuzzy problem. Chen
(2007) proposed a mathematical programming method to construct the membership
functions of the fuzzy objective value of the cost-based queuing decision problem with
the cost coefficients and the arrival rate being fuzzy numbers. On the basis of Zadeh’s
extension principle, he formulated three pairs of mixed integer non-linear programmes
parameterised by the possibility level Į to calculate the lower and upper bounds of the
minimal expected total cost per unit time at Į, through which the membership function of
the minimal expected total cost per unit time of the fuzzy objective value is constructed.
Then, to provide a suitable optimal service rate for designing queuing systems, he used
the Yager’s ranking index method.
This paper is organised as follows: in Section 2, we review the current methods for
solving FLP by focusing on some major contributions to the field to motivate the readers
in the background and justification for the research. In Section 3, we first give some
necessary concepts of fuzzy set theory and then define the linear programming problems
with fuzzy number. The review of the fuzzy number linear programming (FNLP)
problem and the definition of the corresponding dual problem are given in Section 4. We
develop and present the primal-dual algorithm for solving the FNLP problems in
Section 5 and explain it by an illustrative example. Section 6 gives the tableau format of
primal-dual method. As an application of primal-dual algorithm in solving FNLP
problem, the minimum cost flow problem is sketched in Section 7. Finally, we conclude
in Section 8.

2 Literature review

Some authors used the concept of comparison of fuzzy numbers based on ranking
functions for solving FLP problems (Ebrahimnejad and Nasseri, 2010; Ebrahimnejad
et al., 2010b, 2011; Fortemps and Roubens, 1996; Mahdavi-Amiri and Nasseri, 2006,
 A primal-dual method for solving linear programming problems 121

2007; Mahdavi-Amiri et al., 2009; Maleki et al., 2000; Nasseri and Ebrahimnejad,
2010a,b, 2011). Usually in such methods, authors define a crisp model which is
equivalent to the FLP problem and then use optimal solution of the model as the optimal
solution of the FLP problem. Maleki et al. (2000) used linear ranking functions for
solving FNLP problem as an auxiliary problem to obtain a fuzzy solution of fuzzy
variable linear programming (FVLP) problem. A review of the literature concerning
fuzzy mathematical programming and comparisons of fuzzy numbers have been given in
Klir and Yuan (1995) and Lai and Hwang (1992). However, in contrast with the vast
literature on modelling and solution procedures for a linear programme in a fuzzy
environment, the studies in duality are rather scarce. The most basic results on duality in
FLP are due to Rodder and Zimmermann (1980) and Hamacher et al. (1978). Rodder and
Zimmermann (1980) presented a generalisation of maximum and minmax problems in a
fuzzy environment and thereby constructed a pair of fuzzy dual linear programming
problems. The paper by Hamacher et al. (1978) is mostly devoted to the study of
sensitivity analysis in FLP.
Ebrahimnejad et al. (2010b) proposed a primal-dual method for solving linear
programming with fuzzy variables by use of a certain linear ranking function. Moreover,
Ebrahimnejad and Nasseri (2009) used the complementary slackness to solve FNLP and
FVLP problems without the need of a simplex tableau. Then Nasseri and Ebrahimnejad
(2010a) suggested the fuzzy primal simplex method to solve the flexible linear
programming problems directly without solving any auxiliary problem. Also, Hashemi
et al. (2006) have introduced the weak duality theorem based on an alphabetic order
function for fully fuzzified linear programming. Furthermore, Inuiguchi et al. (2003) have
proved some important dual theorems on linear programming using satisfying concepts.
In addition, Zhong and Shi (2002) gave duality concepts on fuzzy multi-criteria and
multi-constraint linear programming applying a parametric approach. Also, Ghatee and
Hashemi (2007) and Ghatee et al. (2008) used Hukuhara’s difference and approximated
multiplication to represent fully fuzzified minimum cost flow problem. In addition, to
give duality theorems, they utilised a total order on fuzzy numbers due to level of risk
and realised optimality conditions for providing some efficient combinatorial algorithms.
Hosseinzadeh Lotfi et al. (2009) discussed full FLP problems of which all parameters and
variable are triangular fuzzy numbers. After that Kumar et al. (2011) proposed a new
method to find the fuzzy optimal solution of same type of FLP problems. Also, Mahdavi-
Amiri and Nasseri (2006) proposed another approach to define dual of an FNLP problem
as a same kind problem leading to the dual simplex algorithm for solving FNLP and
bounded FNLP problems (Ebrahimnejad and Nasseri, 2010; Nasseri and Ebrahimnejad,
2010a,b). After that Ebrahimnejad (2011b) developed the concept of sensitivity analysis
in FNLP problems based on fuzzy primal and dual simplex algorithms. The dual simplex
method begins with a basic dual feasible basic solution and proceeds by pivoting through
a series of dual basic solutions until the associated complementary primal basic fuzzy
solution is feasible. In this paper, we propose a new method called the primal-dual
algorithm, which is similar to the dual simplex method and begins with dual feasibility
and proceeds to obtain primal feasibility while maintaining complementary slackness. An
important difference between the dual simplex method and the primal-dual method is that
the primal-dual algorithm does not require a dual feasible solution to be basic. This
algorithm is useful specially for solving minimum fuzzy cost flow problem (MFCFP) in
which finding an initial dual feasible solution turns out to be a trivial task. We emphasise
that for the sake of illustrating the performance of the algorithm in this paper, it has been
122 A. Ebrahimnejad

here developed using the TRFNs and the linear ranking functions, but it is no restrict at
all to use these fuzzy numbers and ranking function.

3 Preliminaries

In this section, we give some basic definitions on fuzzy numbers comparison that will be
used for illustrating our approach.
Fuzzy set is defined as a% of universal set X ˆ  by its membership function Na% (¸)
which assigns to each element x ‰ X , a real number Na% ( x) in the interval [0, 1]. The
Į-cut or Į-level of a fuzzy set a% , which plays an essential role in fuzzy optimisation, is
defined as an ordinary set [a]B for which the degree of its membership function exceeds
the level Į. A fuzzy number is a convex normalised fuzzy set of the real line  ; whose
membership function is piecewise continuous. The set of fuzzy numbers on  is denoted
by F ( ) .
An left-right (LR) type flat fuzzy number (Okada and Soper, 2000) is denoted as
a (a L , aU , D , E )LR , if
­ § aL  x ·
° L ¨© D ¸¹ for a L  D d x d a L
°
°1 for a L d x d aU
Pa ( x) ® (1)
U
°R § x  a · for aU d x d aU  E
° ¨© E ¸¹
°
¯0 else
where the symmetric non-increasing function L :[0, f) o [0,1] is the left shape function,
that L(0) = 1. Also, a right shape function R(·) is similarly defined as L(·)
TRFNs are special cases of LR fuzzy numbers with L( x) R( x) 1  x and the
following membership function:
­§ x   aL  D  ·
°¨ ¸ for a L  D d x d a L
°© D ¹
°
°1 for a L d x d aU
Pa ( x) ® (2)
° §  aU  E   x ·
°¨ ¸ for aU d x d aU  E
E
°© ¹
°0 else
¯
We denote all TRFNs by F () . Now define arithmetic on TRFNs. Let a (a L , aU , D , E)
and b (b L , bU , J , T) be two TRFNs. Define,

x ! 0, x  ; xa  xa L , xaU , xD , x E 
x  0, x  ; xa  xaU , xa L ,  x E ,  xD 
a  b  a L  b L , aU  bU , D  J , E  T 
 A primal-dual method for solving linear programming problems 123

To define the inequality relation between two fuzzy numbers, many methods have been
proposed in the literature. But perhaps the most convenient and directive method in this
area is based on the concept of comparison of fuzzy numbers by the use of ranking
functions, in which a ranking function R : F () o  that maps each fuzzy number into
the real line is defined for ordering the elements of F () . Thus, we define orders on
F () by:

a % b if and only if R(a) p R(a) (3)

a ; b if and only if R(a) ! R(b) (4)

a% ; b% if and only if R(a%)  R(b%) (5)

In this paper, we restrict our attention to linear ranking functions, i.e. a ranking function
R such that R(ka  b)  kR(a)  R(b) for any a and b belonging to F () and any
k ‰ .

Remark 1: For any TRFN a, the relation a% % %0 holds, if there exist F p 0 and B p 0
such that a % (F, F, B, B) . We realise that R(F, F, B, B)  0 (we also consider a  0
if and only if R(a)  0 ). Thus, without loss of generality, throughout this paper we let
0  (0, 0, 0, 0) as the zero TRFN.

A special version of the linear ranking functions was first proposed by Yager (1981) as
follows:
1   C B¯
Y2 a
 1 ¨ inf < a>  sup < a >B
dB  1 ¡ a L  aU  ° (6)
2 0
B 2 ¡¢ 2 °±

Then, for TRFNs a  (a L , aU , B, C ) and b  (b L , bU , H, R) , we have a % b if and

only if

a L  aU  1 (C  B) p b L  bU  1 (R  H ) (7)
2 2
We pursue this linear ranking function in this paper to illustrate our approach. But there is
no restrict at all to use this linear ranking function.

4 Linear programming with fuzzy cost coefficients

In this section, we first review the concept of solution for FNLP problems proposed by
Maleki et al. (2000) as well as a fuzzy simplex method for solving FNLP problems. Then,
we review the definition of the dual of FNLP problems proposed by Mahdavi-Amiri and
Nasseri (2006) that leads to a dual simplex method for solving such problems (Nasseri
and Ebrahimnejad, 2010a,b).
124 A. Ebrahimnejad

Definition 1: A FNLP problem is defined as (Mahdavi-Amiri and Nasseri, 2006; Maleki

et al., 2000):
min z  cx
 (8)
s.t. Ax % b

xp0

where b  F ( )m , c  F () n and A  F ( )mun (rank( A) m) are given and x   n is

to be determined.
Maleki et al. (2000) developed the concepts of feasible and optimal solutions for the
FNLP problems. Then they gave the following auxiliary problem, having only fuzzy cost
coefficients, for solving linear programming problems which in the decision variables are
fuzzy:
min z  cx

s.t. Ax p b (9)
xp0

where b   m , x   n , c  ( F ())n and A   mun .

We name this problem as reduced fuzzy number linear programming (RFNLP)
problem.

Definition 2: We say that a fuzzy vector x*   n is an optimal solution to (9) if for all
 * - cx
fuzzy feasible solution x, we have cx  .
Now, consider the RFNLP problem in standard form:
min z  cx

s.t. Ax  b (10)
xp0

where the parameters of the problem are as defined in Equation (9).

Let A (aij )m u n and assume rank ( A) m . Partition A as [B N] where B, m u m , is
non-singular. It is obvious that rank (B) = m. Let yj be the solution to By j a j . It is
1
apparent that the basic solution x ( xB , xN ) ( B b,0) is a solution of Ax b . If
xB t 0 then the basic solution is feasible and the corresponding fuzzy objective value is:
z  cB xB where cB (cB1 , cB2 ,!, cBm ) . Now, corresponding to every non-basic variable
x j , 1 d j d n, j z Bi , i 1,!, m , define z j cB y j cB B 1a j . Maleki et al. (2000) proved
fuzzy analogous of some important theorems of linear programming that lead to a new
method for solving RFNLP problems. Mahdavi-Amiri et al. (2009) proposed the tableau
format of this method as follows:
Algorithm 1 A primal simplex method for RFNLP problems
Initialisation step
Choose a starting basic feasible solution with basic B and form the initial simplex tableau
as shown in Table 1.
 A primal-dual method for solving linear programming problems 125

Table 1 The initial FNLP simplex tableau

Basis xB xN Right-hand-side (RHS)

R ( z ) 0 R( z N  cN ) R(cB B 1N  cN ) R( z ) R(cB B 1b)
z 0 z N  cN cB B 1N  cN z cB B 1b
xB I B 1N b B 1b

Main step
Let zk  ck max{z j  c j , j  R} in which R is the index set of the current non-basic
 then stop; the current solution is optimal. Otherwise examine yk.
variables. If zk  ck - 0,
If yk d 0 , then stop; the problem is unbounded. If yk d/ 0 , determine the index of
variable xBr leaving the basic as follows:

br ­b ½
min ® i yik ! 0 ¾
yrk 1di d m ¯ yik ¿
Pivot on yrk and update the simplex tableau. Update the basic and non-basic variables
where xk enters the basis and xBr leaves the basis and repeat the main step.
Maleki et al. (2000) used the RFNLP problem (9) as an auxiliary problem for solving
FVLP problem defined as follows:
max y  wb

s.t.  - c
wA (11)
w p 0

where b   m , w  F ( ) m , c  F ( )n and A   mun .

In fact, based on the following theorem and using the optimal crisp solution obtained
by Algorithm 1 for RFNLP problem, Maleki et al. (2000) found an optimal fuzzy solution
for FVLP problem.
Theorem 1: (Strong duality) If any one of the RFNLP problem (9) or FVLP problem (11)
has an optimal solution, then both problems have optimal solutions and the two optimal
objective fuzzy values are equal. (In fact, if xB B 1b is optimal solution of the RFNLP
problem then the vector w cB B 1 , where B is the optimal basis, is an optimal solution
of the FVLP problem.)
Proof: See Maleki et al. (2000) and Mahdavi-Amiri and Nasseri (2007).
Corollary 1: If x0 and w 0 are fuzzy feasible solutions to RFNLP (9) and FVLP (11)
problems, respectively, and cx
 0 w0 b , then x0 and w 0 are fuzzy optimal solutions to
their respective problems.
After that, Mahavi-Amiri and Nasseri (2007) extended the usual definition of a dual
problem in the crisp linear programming to the problem with fuzzy variables based on
126 A. Ebrahimnejad

ranking functions. They proved the RFNLP problem is indeed the dual of FVLP problem
and hence developed some duality results. Using the results, they developed a new dual
algorithm for solving the FVLP problem directly, making use of the primal simplex
tableau. But the solution space of the problem (11) is a fuzzy space, thus solving this
problem by the dual simplex method proposed by Mahdavi-Amiri and Nasseri (2007)
needs more computations comparing with the Maleki et al.’s (2000) method which solved
the problem (9) in a crisp space.
On the other hand, in certain instances it is difficult to find an initial primal feasible
basis without adding artificial variables. In this same instances, it is often possible to
obtain an initial basis (not necessarily primal feasible) solution that is dual feasible
(i.e. z j  cj - 0 in a minimisation problem). Thus, Nasseri and Ebrahimnejad (2010a,b)
developed a new method called the fuzzy dual simplex method that starts with a dual
feasible, but primal infeasible basis and walks to an optimal basis by moving among
adjacent dual feasible basis. At each pivot step, this algorithm tries to reduce primal
infeasibility while retaining dual feasibility. If a feasible basis is reached, the fuzzy dual
simplex algorithm terminates by declaring it as an optimum basis. Below, we state a
summary of the fuzzy dual simplex algorithm for solving the RFNLP problems:
Algorithm 2 A dual simplex method for RFNLP problems
Initialisation step

Find a basic B such that z j  c j - 0 for all j. Form the initial dual simplex similar to
Table 1.
Main step

Let br min1didm {bi } . If br t 0 , then stop; the current solution is optimal. Otherwise
examine yrj for all j. If yrj d 0 , then stop; the RFNLP problem is infeasible. Else,
determine the pivot column k by the following minimum ratio test:

zk  ck °­ z j  c j °½
min ® yik  0 ¾
yrk 1d j d n ° yrj
¯ °¿

Pivot on yrk and update the simplex tableau. Update the basic and non-basic variables
where xk enters the basis and xBr leaves the basis and repeat the main step.
Also, Ebrahimnejad and Nassei (2009) used the following complementary slackness
theorem to solve FVLP problem based on the crisp solution obtained by the dual simplex
method.
Theorem 2: (Complementary slackness theorem) Suppose u and v be the slack variables
to the RFNLP (9) and FVLP (11), respectively. Let ( x*, u*) and ( w *, v*) be any feasible
solutions to RFNLP problem and FVLP problem. Then x * and w * are, respectively,
optimal if and only if

v * x*  0, w
 * u*  0 (12)
 A primal-dual method for solving linear programming problems 127

Remark 2: Complementary slackness theorem indicates at optimality if a variable in

RFNLP (FVLP) problem is positive (fuzzy positive) then corresponding constraint in the
other problem must be tight. If a constraint in one problem is not tight, then the
corresponding variable (fuzzy variable) in the other problem must be zero (fuzzy zero).
Hence, condition (13) is equivalent to (below, ai refers to the ith row and aj refers to the
jth column of A):

* a j ; c j º x*j  0 or
w x*j  0 º w
*a j  cj j  1, !, n
*
ai x  bi º i*
w  0 or i*
w ; 0 º ai x*  bi i  1,!, m

5 A primal-dual method for RFNLP problems

Note that the dual simplex method, which is proposed by Nasseri and Ebrahimnejad
(2010b), begins with a basic (not necessarily feasible) solution and a basic feasible dual
solution to the primal RFNLP problem. The dual simplex method proceeds by pivoting
through a series of dual basic fuzzy solution until the associated complementary primal
basic solution is feasible. In this section, we describe a new method called the primal-
dual algorithm, similar to the dual simplex method, which begins with dual feasibility
and proceeds to obtain primal feasibility while maintaining complementary slackness. An
important advantage of this method is that the optimal solution of RFNLP problem and
its dual problem (FVLP problem) is obtained simultaneously and does not require more
computation comparing with Maleki et al.’s (2000) and Nassei and Ebrahimnejad’s
(2010b) methods which obtained the fuzzy solution of FVLP problem based on
Theorems 1 and 2, respectively.
Corollary 2: The optimality criteria z j  cj - 0 , for all j, for the RFNLP problem is
equivalent to the feasibility condition for the FVLP problem.
In the fuzzy primal-dual method, given a dual feasibility solution, the primal variables
that correspond to the tight dual constraints (so that complementary slackness is satisfied)
are determined. It attempts to attain primal feasibility using only these variables. If it is
unable to obtain primal feasibility, it changes the dual feasible solution in such a way as
to admit at least one new variable. This is continued until either the primal becomes
feasible or the dual becomes unbounded. Consider the following RFNLP problem:
min z  cx
s.t. Ax  b (13)
xp0
The dual of FLP (13) is defined as follows:
max u  wb
(14)
s.t.  - c
wA

% j:
Let w be an initial dual feasible solution, i.e. wa  c%j , for all j. By complementary
slackness, if wa
 j  cj , then xj is allowed to be positive and we attempt to attain primal
 j  cj } , i.e. the set of the indices
feasibility from among these variables. Let -  { j; wa
128 A. Ebrahimnejad

of the primal variables allowed to be positive. Now consider the following Phase I
problem that attempts to find a feasible solution to the primal RFNLP problem among
variables in the set ȁ:
min ¦
0 x  1 x
j/
j a

s.t. ¦a x
j:
j j  Ixa b
(15)

x j t 0 for all j  /
xa t 0

where xa ( x1a , ! , xma )   m and 1 ((1,1,0,0),!,(1,1,0,0))  F ( )m .

We utilise the artificial vector xa to obtain a starting basic feasible solution to the
Phase I problem. It is apparent that the Phase I problem is a RFNLP problem and so can
be solved by Algorithm 1.
Denote the primal objective fuzzy value of foregoing problem by x0 . At optimality of
this problem either x  0 or 
0 x % 0 . When x  0 , we have a feasibility solution to the
0 0
primal RFNLP problem since all artificial variables are zeros. Furthermore, we have a
dual feasible solution, and the complementary slackness condition (wa  j  c j ) x j  0
 j  cj  0 , or else j  / in which case
holds because either j / in which case wa
xj 0 . Therefore, we have an optimal solution of the overall problem whenever x0  0 .
If we have x%0 ; %0 , then primal feasibility is not achieved and we must construct a new
dual solution that would admit a new variable to the Phase I problem in such a way that
x0 might be decreased. We shall modify the dual fuzzy vector w such that all the basic
primal variables in the Phase I problem remain in the new Phase I problem, and in
addition, at least one primal variable that did not belong to the set ȁ would get passed to
the Phase I problem. Furthermore, this variable would reduce x0 if introduced in the
basic. To construct such a dual fuzzy vector, consider the following dual of the Phase I
problem:
max vb
s.t.  j - c j
va j / (16)
v -1

Let v * be an optimal solution to the foregoing problem. Then if a real variable xj is a

member of the optimal basis for the Phase I problem, the associated dual constraint must
 j  0 . Also the criterion for basis entry in the Phase I problem is that the
be tight, i.e. va
associated dual constraint be violated, i.e. va  j ; 0 then if xj could be passed to
 j ; 0 . If va
the Phase I problem it would be a candidate to enter the basis with the potential of a
further decrease in x0 . Therefore, we must find a way to force some variable xj with
v * a ; 0 into the set ȁ.
j
 A primal-dual method for solving linear programming problems 129

Construct the following dual vector w  w

  Rv * , where T ! 0 . Then
wa j  c%j ; w
%  Rv% *
a j  c%j ; wa
% j  c% j
 Rv% * a j (17)

 j  cj  0 and v * a j d 0 , for j  / . Thus, Equation (17) implies that

Note that wa
wa j  cj - 0 for j  / . In particular, if xj, j  / is a basic variable in the Phase I
problem, then v * a j  0 and wa j  cj  0 , permitting j in the new Phase I problem. If
% j  c%j - %0 , we have
j / and v * a j d 0 , then from Equation (17) and noting that wa
wa j  cj - 0 . Finally, consider j, j / with v * a j ; 0 . Examining Equation (17) and
nothing that wa  j  c j E 
~ 0 for j  / , it is evident that we can choose T ! 0 such that
wa j  cj - 0 for j  / with at least one component equal to fuzzy zero. In particular,
define ș as follows:

R  ck  wa
 k ° 
­ R c  wa
j  j  ½
°
T min ®  
R v * a j ! 0 ¾
R  v * ak  j
°¯ R v *a j  °¿

By definition of ș and from Equation (17), we see that wak  ck  0 and wa j  c j - 0 for
each j with v * a j ; 0 .
To summarise, modifying the dual fuzzy vector leads to a new feasible dual fuzzy
solution. Also, all the variables that belonged to the Phase I problem basis are passed to
the new basis. In addition, a new variable xk, i.e. a candidate to enter the basis is passed to
the Phase I problem. Hence, we continue from the present Phase I problem basis by
entering xk, which leads to a potential reduction in x0 .

Theorem 3: At the end of the Phase I problem, if x0 ; 0 and v * a j - 0 , for j  / , then
the RFNLP problem has no solution and the FVLP problem is unbounded.
Proof: In this case, consider w  w  Rv * . Since wa  for all j and by assumption
 j  c j - 0,
v * a j - 0 for all j  / , then from Equation (17), w is a dual feasible fuzzy solution for
all T ! 0 . In addition, the dual objective fuzzy value is as wb  ( w   Rv * b .
  Rv*)b  wb
Since x0 is optimal objective value for the Phase I problem and v * b is the optimal value
for its dual problem, we have x0  v * b . Also by assumption x0 is positive, so wb can
be increased indefinitely by choosing ș arbitrarily large. Therefore, the dual is unbounded
and hence the primal is infeasible.
Algorithm 3 A fuzzy primal-dual simplex method
Initialisation step
 j  c j - 0 , for all j.
Choose a fuzzy vector w such that wa
Main step
Let ’  { j : wa
 j  c j  0} and solve the Phase I problem. If x0  0, then stop; the current
solution is optimal. Else suppose v be the optimal solution of dual of Phase I problem. If
130 A. Ebrahimnejad

 j - 0 , then stop; the RFNLP problem is infeasible. Else let

va

R  ck  wa
 k ° 
­ R c  wa
j  j  ½
°
T min ®  
R v * a j ! 0 ¾
R  v * ak  j 
°¯ R v * a j
  °¿

Replace w by w  T v * and go to main step.

Example 1: For an illustration of the above method, consider the following RFNLP
problem:
5 7 1 1¬ 7 9 1 1¬
min %z ; žž , , , ­­­ x1  žž , , , ­­­ x2  (2, 4,1,1) x3
Ÿž 2 2 2 2 ® Ÿž 2 2 2 2 ®
s.t. 2 x1  x2  x3  6 (18)
x1  x2  x4  3
x1 , x2 , x3 , x4 p 0
The dual problem is given by the following:
min z  6 w
1  3w2
 5 7 1 1¬
s.t. 2w 2 - žž , , , ­­­
1  w
žŸ 2 2 2 2®
7 9 1 1¬ (19)
w 2 - žž , , , ­­­
1  w
žŸ 2 2 2 2 ®
1 - (0, 0, 0, 0)
w
2 - (0, 0, 0, 0)
w
An initial dual feasible fuzzy solution is given by w (0,1)   , where 0 (0,0,0,0) and
1 (1,1,0,0) . Substituting w
 in each dual constraint, we find that only third constraint is
tight so that / {3} . Denoting the artificial variables by x5 and x6, the Phase I problem as
follows:
min x  1 x  1x
0 5 6
s.t. x3  x5  6
(20)
x6  3
x3 , x5 , x6 p 0
The optimal solution to this problem is clearly ( x3 , x5 , x6 ) (0,6,3) and the optimal
objective fuzzy value is x0 (9,9,0,0) with R( x0 ) 9 . The dual of the Phase I problem is
the following:
max y  6v1  3v2
s.t. v1 - (0, 0, 0, 0)
(21)
v1 - (1,1, 0, 0)
v2 - (1,1,0, 0)
 A primal-dual method for solving linear programming problems 131

  and its corresponding objective

Utilising complementary slackness, we have v* (1,1)
fuzzy value is (9,9,0,0). Computing v * a j for j  / conclude that v * a1 (3,3,0,0) ,
v * a2 (0,0,0,0) and v * a4 (1, 1,0,0) with R(v * a1 ) 3 , R(v * a2 ) 0 and R(v * a4 ) 1 ,
respectively. Then we can determine the parameter ș as T min{(2) / 3} 2 / 3 and
   2 / 3(1,1)
w  (0,1)    (2  / 3) where 2 / 3 (2 / 3, 2 / 3, 0, 0) and 5 / 3 (5 / 3,5 / 3, 0, 0) . By
 / 3,5
the new dual fuzzy solution, we compute ȁ and obtain / {1} , giving the following
Phase I problem:
max x  1x  1x
0 5 6
s.t. 2 x1  x5  6
(22)
x1  x6  3
x1 , x5 , x6 p 0
This time an optimal solution to the Phase I problem is given by ( x1, x5 , x6 ) (3,0,0) with
x0  0 . Thus, we have an optimal solution to the original problem with the optimal
primal and dual being ( x1 , x2 , x3 , x4 ) (3,0, 0, 0) and w  (2 / 3,5 / 3) .

6 Tableau form of primal-dual method

In this section, we are going to state the primal-dual method in tableau format. To this
end, let %z j  c%j be the row zero coefficients for the original primal RFNLP problem and
let z j  c j be the row zero coefficient for the Phase I problem. Then for each real variable
xj, we have z j  c j wa
 j  c j and z j  c j  j  0
va  j.
va We also have
( wa
 j  c j )  T va
 j ( z j  c j )  T ( z j  c j ) and R ( wa
 j  c j ) / (va
 j) [ R ( z j  c j )] / [ R ( z j  c j )] .
We can carry out all of the necessary operations directly in one tableau. In this tableau,
we have four objective rows: the first gives the z j  c j ’s, the second gives the z j  c j ’s,
the third and fourth give R ( z j  c j ) ’s and R( z j  c j ) ’s, respectively. We shall apply this
tableau method to the numerical Example 1. The initial tableau is displayed in Table 2. In
  , so that the row zero coefficients for the original problem in initial
this example w (0,1)
tableau have been computed by this fuzzy vector.
Since we begin with x5 and x6 in the basis for the Phase I problem, we must perform
some preliminary pivoting to make their coefficients fuzzy zero in the Phase I objective.
We do this by multiplying the first and second constraints rows by 1 and adding to the
Phase I objective row (also we add the first and second constraints rows to R( x0 ) ). Then
z  c  0 and R( z  c ) 0 for the two basic variables x5 and x6. Let symbol
j j j j

indicates the variables in the Phase I problem, i.e. those for which z j  c j  0 (see
Table 3). As the Phase I problem is solved, only the variables signalled with symbol
are allowed to enter the basis.
132 A. Ebrahimnejad

Since z j  c j - 0 and R( z j  c j ) ื 0 for all variables in the Phase I problem, we

have an optimal solution for Phase I. Then ș is given by

R  zk  ck  ­ 
° R z j  c j  °
½
­ ( 2) ½ 2
T min ® 
R z j  c j ! 0¾  min ® ¾

R zk  ck  j

°¯ R z j  c j  °¿ ¯ 3 ¿ 3

Thus, we multiply the Phase I objective row by 2/3 and add to the original objective row.
Also we multiply the R( x0 ) row by 2/3 and add to the R ( z ) row. This leads to Table 4.
The Phase I problem is solved by only utilising the variables in the set ȁ, i.e. those with
z j  c j  0 or R( z j  c j ) 0.
In this case, x1 is an entering variable and x6 is a leaving variable. Therefore, the new
tableau is given in Table 5.
Now, x4 is entering variable and x5 is leaving variable. Therefore, the new tableau is
given in Table 6.
In this case, we have x0  0 , then the optimal solution is found, namely
( x1 , x2 , x3 , x4 ) (3, 0, 0, 0) .
Table 2 The initial tableau

Basis x1 x2 x3 x4 x5 x6 RHS R
§ 5 3 3 3· § 7 5 1 1·
z ¨©  ,  ,  ,  ¸¹ ¨©  ,  , , ¸¹ 0 1 0 0 3 3
2 2 2 2 2 2 2 2
x0 0 0 0 0 1 1 0 0
R ( z ) 2 3 0 1 0 0 3 –
R ( x0 ) 0 0 0 0 1 1 0 –
x5 2 1 1 0 1 0 6 –
x6 1 1 0 1 0 1 3 –

Table 3 The second tableau

Basis x1 x2 x3¥ x4 x5¥ x6¥ RHS R

§ 5 3 3 3· § 7 5 1 1·
z ¨©  ,  ,  ,  ¸¹ ¨©  ,  , , ¸¹ 0 1 0 0 3 3
2 2 2 2 2 2 2 2
x0 3 0 1 1 0 0 9 9
R ( z ) 2 3 0 1 0 0 3 –
R ( x0 ) 3 0 1 1 0 0 9 –
x5 2 1 1 0 1 0 6 –
x6 1 1 0 1 0 1 3 –
 A primal-dual method for solving linear programming problems 133

Table 4 The third tableau

Basis x1¥ x2 x3 x4 x5¥ x6¥ RHS R

§ 1 1 3 3· § 7 5 1 1· § 2 2 · § 5 5 ·
z ¨©  , , , ¸¹ ¨©  , , , ¸¹ ¨©  ,  ,0,0¸¹ ¨©  ,  ,0,0¸¹ 0 0 9 9
2 2 2 2 2 2 2 2 3 3 3 3
x0 3 0 1 1 0 0 9 9
2 5
R ( z ) 0 3   0 0 9 –
3 3
R ( x0 ) 3 0 1 1 0 0 9 –
x5 2 1 1 0 1 0 6 –
x6 1 1 0 1 0 1 3 –

Table 5 The fourth tableau

Basis x1¥ x2 x3 x4 x5¥ x6¥ RHS R

§ 1 1 3 3· § 7 5 1 1· § 2 2 · § 5 5 ·
z ¨©  , , , ¸¹ ¨ , , , ¸ ¨©  ,  ,0,0¸¹ ¨©  ,  ,0,0¸¹ 0 0 9 9
2 2 2 2 © 2 2 2 2¹ 3 3 3 3
x0 0 3 1 2 0 3 0 0
2 5
R ( z ) 0 3   0 0 9 –
3 3
R ( x0 ) 0 3 1 2 0 3 0 –
x5 0 3 1 2 1 2 0 –
x1 1 1 0 1 0 1 3 –

Table 6 The optimal tableau

Basis x1¥ x2 x3 x4 x5¥ x6¥ RHS R

§ 1 1 3 3· § 7 5 1 1· § 2 2 · § 5 5 ·
z ¨©  , , , ¸¹ ¨  ,  , , ¸ ¨©  ,  ,0,0¸¹ ¨©  ,  ,0,0¸¹ 0 0 9 9
2 2 2 2 © 2 2 2 2¹ 3 3 3 3
x0 0 0 0 0 1 1 0 0
2 5
R ( z ) 0 3   0 0 9 –
3 3
R ( x0 ) 0 0 0 0 1 1 0 –
3 1 1
x4 0   1 1 0 –
2 2 2
1 1 1
x1 1   0 0 3 –
2 2 2

7 Application in MFCFP

In this section, as an application of primal-dual algorithm in solving RFNLP problem, the

minimum cost flow problem is sketched which arises naturally in engineering and
134 A. Ebrahimnejad

economics contexts; it appears in problems involving equilibrium models, resistive

electrical networks, production distribution problems and machine scheduling (Ahuja
et al., 1993). The central concept in all of these problems is to find the least transportation
cost of a commodity through a capacitated network to satisfy demands at certain nodes
using available supplies at other nodes. However, in many real problems the traverse
costs are known imprecisely. For instance, when the costs are related to safety parameters
or environmental phenomena, the cost may be represented as fuzzy numbers. Keeping
fuzzy costs in mind, MFCFP should be solved.
Let G = (V, E) be a directed network, V and E denote sets of nodes and arcs,
respectively. For each arc (i, j )  E , a fuzzy cost cij can be considered dependent on
complex human, social, economic and political interactions. Two real numbers for arc
(i, j) are supposed as the lower and upper capacity, llj and ulj, respectively. The RFNLP
formulation of the MFCFP is as follows:
min z  œ
cij xij
(i , j )‰ E

s.t. œ xij  œ xij  bi for all i ‰ V (23)

{ j : (i , j )‰E } { j: ( j ,i )‰ E}

lij b xij b uij for all (i, j ) ‰ E

In the above formulation, xij is the amount of flow on arc (i, j). Moreover, if
bi  0 (bi ! 0) , the corresponding node is said demander (supplier), otherwise transient
node. Suppose ¦ b
i V i
0.

Example 2: Consider the network depicted in Figure 1 as a large numerical example.

Define a MFCFP with presented parameters in Tables 7–9.

Figure 1 A network with 16 nodes and 33 arcs

Table 7 The supplier nodes

b1 b2 b3 b4 b6 b9 b15
345 100 45 200 150 135 100
 A primal-dual method for solving linear programming problems 135

Table 8 The demander nodes

b5 b7 b10 b11 b13 b14 b16

223 87 100 120 150 45 350

Table 9 The capacity and cost of network arcs

Arc index Tail node Head node Lower bound Upper bound Fuzzy cost
1 1 2 50 115 (100, 150, 20, 30)
2 1 3 60 90 (250, 300, 10, 20)
3 1 5 55 325 (25, 75, 5, 15)
4 2 4 47 75 (150, 200, 15, 25)
5 2 5 25 50 (100, 400, 10, 40)
6 2 8 45 180 (180, 220, 25, 45)
7 3 5 45 115 (125, 225, 25, 35)
8 3 6 25 70 (100, 200, 5, 25)
9 3 9 35 80 (110, 240, 10, 20)
10 4 7 65 220 (300, 400, 30, 90)
11 4 8 50 140 (120, 280, 40, 30)
12 4 11 70 150 (150, 200, 10, 20)
13 5 8 40 65 (0,0, 0, 0)
14 5 9 35 75 (200, 400, 35, 15)
15 5 12 40 90 (125, 225, 10, 45)
16 6 9 25 45 (215, 235, 20, 30)
17 6 10 85 275 (150, 250, 10, 25)
18 6 13 30 45 (100, 300, 15, 35)
19 7 11 40 70 (75, 125, 10, 25)
20 8 11 50 85 (25, 75, 5, 15)
21 8 12 25 80 (50, 150, 20, 30)
22 8 14 30 75 (125, 175, 15, 25)
23 9 12 30 55 (115, 185, 25, 15)
24 9 13 55 375 (25, 75, 15, 10
25 9 15 25 125 (25, 75, 15, 10)
26 10 13 20 75 (110, 190, 35, 15)
27 11 14 40 90 (185, 215, 30, 20)
28 12 14 25 60 (75, 125, 10, 25)
29 12 15 35 85 (150, 250, 20, 40)
30 12 16 30 100 (100, 200, 10, 25)
31 13 15 40 115 (55, 125, 15, 25)
32 14 16 50 105 (275, 325, 30, 10)
33 15 16 55 275 (100, 150, 20, 10)
136 A. Ebrahimnejad

In this example, the total supply is equal to total demand within this network. Table 7
shows that the nodes 1–4, 6, 9 and 15 are the supplier nodes. For example, the available
supply of node 1 is equal to 345 and of node 15 is equal to 100. In a similar way, Table 8
shows that the nodes 5, 7, 10, 11, 13, 14 and 16 are the demander nodes. As instance, the
required demand of node 5 is 223 units and the required demand of node 16 is 350 units.
Moreover, the fuzzy cost per unit of each arc is given in Table 9. For example, the unit
shipping fuzzy cost of arc (1, 2) is equal to (100, 150, 20, 100). Also, in this table lower
and upper capacity of each arc has been given. For example, the at least and at most
shipping flows along with arc (1, 2) are equal to 50 and 150 units, respectively.
We solved this RFNLP problem by Algorithm 3. The optimal solution is given in
Table 10. For example, it shows that to satisfy the demand of node 5, it needs to send
235, 58 and 45 units from nodes 1–3, to node 5, respectively, and also must send 40, 35
and 40 units from node 5 to nodes 8, 9 and 12, respectively. It will be possible to give a
similar discussion for other nodes. Also the optimal objective function value is (244,800,
410,550, 33,645, 53,400). In fact the least transportation fuzzy cost of a commodity
through this capacitated network to satisfy demands at certain nodes using available
supplies to other nodes is equal to (244,800, 410,550, 33,645, 53,400).

Table 10 The optimal flow

Arc index Tail node Head node Optimal flow

1 1 2 50
2 1 3 60
3 1 5 235
4 2 4 47
5 2 5 58
6 2 8 45
7 3 5 45
8 3 6 25
9 3 9 35
10 4 7 127
11 4 8 50
12 4 11 70
13 5 8 40
14 5 9 35
15 5 12 40
16 6 9 25
17 6 10 120
18 6 13 30
19 7 11 40
20 8 11 50
21 8 12 55
22 8 14 30
23 9 12 30
24 9 13 175
 A primal-dual method for solving linear programming problems 137

Table 10 The optimal flow (continued)

Arc index Tail node Head node Optimal flow

25 9 15 25
26 10 13 20
27 11 14 40
28 12 14 24
29 12 15 35
30 12 16 65
31 13 15 75
32 14 16 50
33 15 16 235

8 Conclusions and future work

In this paper, we considered linear programming problems which involve fuzzy numbers
only in cost coefficients of objective function. Then, by use of a linear ranking function,
we gave a fuzzy primal-dual algorithm to solve the FNLP problems. The fuzzy primal-
dual algorithm does not require a dual feasible fuzzy solution to be basic. It begins with
dual feasibility and proceeds to obtain primal feasibility while maintaining
complementary slackness and thus satisfying all of the conditions for optimality. In
addition, in the absence of degeneracy, this algorithm terminates in a finite number of
steps due to adding an improving variable to the Phase I problem at each iteration.
Therefore, in the absence of degeneracy in the Phase I problem, at each step the optimal
objective x0 strictly decreased. This means that the set ȁ generated at any iteration is
distinct from all those generated at previous iteration. Since there is only a finite number
of sets of ȁ (recall /  {1, 2,!, n} ) and none of them can be repeated, then the algorithm
terminates in a finite number of steps. In addition, as an application of primal-dual
algorithm in solving RFNLP problems, the MFCFP sketched which arises naturally in
engineering and economics contexts. The fuzzy primal-dual algorithm proposed here is
indeed the generalisation of the primal-dual algorithm in deterministic environment
(Bazarra et al., 2005).
In our opinion, we feel that there are many other points of research and should be
studied later on. Some of these points are discussed below.
x In this paper, we use the fuzzy primal-dual simplex algorithm for solving MFCFPs
as a special kind of reduced FNLP problems. Development of the network simplex
algorithms and combinational algorithms for solving these problems in fuzzy
environment may also produce interesting results.
x It needs to point out that we used the linear ranking function for solving reduced
FNLP problems. In fact, a ranking function R : F ( ) o  that maps each fuzzy
number
138 A. Ebrahimnejad

into the real line is defined for ordering the elements of F ( ) , i.e. using the natural
orders on real numbers, we compared fuzzy numbers as follows:

a % b if and only if R a
p R b

a ; b if and only if R a
 R b

a  b if and only if R a
 R b

We pursued this approach from decision-maker’s point of view. It is clear that if we have
R(a ) R(b ) , we cannot guarantee the equality of a  b . Thus, research on the topic for
introducing a new ranking function in solving reduced FNLP such that it can satisfy the
third property of linear ranking functions will be an interesting research work in the
future.
x On the other hand, in this paper those kinds of FLP problems have been studied in
which not all part of problems were assumed to be fuzzy. Introducing a new method
for solving these kinds of FLP problems are left to the next work.

Acknowledgements

The author thanks the Editor-in-Chief of the journal of IJISE, Prof Angappa Gunasekaran
and the anonymous reviewers for their helpful comments which contributed to the
improvement of this paper. Finally, the author greatly appreciates to the office of Vice
Chancellor for research of Islamic Azad University-Qamshahr Branch for financial
support.

References
Ahuja, R.K., Magnanti, T.L. and Orlin, J.B. (1993) Network Flows: Theory, Algorithms, and
Applications. Englewood Cliffs, NJ: Prentice-Hall.
Azadeh, A., Seifoory, M. and Abbasi, M. (2010) ‘Integration of simulation and fuzzy multi-
attribute decision making for modelling and assessment of fuzzy parameters’, Int. J. Industrial
and Systems Engineering, Vol. 6, No. 4, pp.483–502.
Azadeh, A., Seraj, O. and Saberi, M. (2011) ‘A comparative assessment of fuzzy regression
models: the case of oil consumption estimation’, Int. J. Industrial and Systems Engineering,
Vol. 7, No. 2, pp.195–223.
Bazarra, M.S., Jarvis, J.J. and Sherali, H.D. (2005) Linear Programming and Network Flows
(3rd ed.). New York: John Wiley and Sons.
Chen, S. (2007) ‘Solving fuzzy queueing decision problems via a parametric mixed integer
nonlinear programming method’, European Journal of Operational Research, Vol. 177,
pp.445–457.
Ebrahimnejad, A. (2011a) ‘Some new results in linear programs with trapezoidal fuzzy numbers:
finite convergence of the Ganesan and Veeramani’s method and a fuzzy revised simplex
method’, Applied Mathematical Modelling, Vol. 35, No. 9, pp.4526–4540.
Ebrahimnejad, A. (2011b) ‘Sensitivity analysis in fuzzy numbers linear programming problems’,
Mathematical and Computer Modelling, Vol. 53, Nos. 8–9, pp.1878–1888.
 A primal-dual method for solving linear programming problems 139

Ebrahimnejad, A. and Nasseri, S.H. (2009) ‘Using complementary slackness property to solve
linear programming with fuzzy parameters’, Fuzzy Information and Engineering, Vol. 3,
pp.233–245.
Ebrahimnejad, A. and Nasseri, S.H. (2010) ‘A dual simplex method for bounded linear
programmes with fuzzy numbers’, Int. J. Mathematics in Operational Research, Vol. 2, No. 6,
pp.762–779.
Ebrahimnejad, A., Nasseri, S.H. and Hosseinzadeh Lotfi, F. (2010a) ‘Bounded linear programs
with trapezoidal fuzzy numbers’, Int. J. Uncertainty, Fuzziness and Knowledge-Based
Systems, Vol. 18, No. 3, pp.269–286.
Ebrahimnejad, A., Nasseri, S.H., Hosseinzadeh Lotfi, F. and Soltanifar, M. (2010b) ‘A primal-dual
method for linear programming problems with fuzzy variables’, European Journal of
Industrial Engineering, Vol. 4, No. 2, pp.189–209.
Ebrahimnejad, A., Nasseri, S.H. and Mansourzadeh, S.M. (2011) ‘Bounded primal simplex
algorithm for bounded linear programming with fuzzy cost coefficients’, Int. J. Operations
Research and Information Systems, Vol. 2, No. 1, pp.96–120.
Fortemps, P. and Roubens, M. (1996) ‘Ranking and defuzzification methods based on area
compensation’, Fuzzy Sets and Systems, Vol. 82, pp.319–330.
Ganesan, K. and Veeramani, P. (2006) ‘Fuzzy linear programming with trapezoidal fuzzy
numbers’, Annals of Operational Research, Vol. 143, pp.305–315.
Ghatee, M. and Hashemi, S.M. (2007) ‘Ranking function-based solutions of fully fuzzified minimal
cost flow problem’, Information Sciences, Vol. 177, pp.4271–4294.
Ghatee, M., Hashemi, S.M., Hashemi, B. and Dehghan, M. (2008) ‘The solution and duality of
imprecise network problem’, Computers and Mathematics with Applications, Vol. 55,
pp.2767–2790.
Hamacher, H., Liberling, H. and Zimmermann, H.J. (1978) ‘Sensitivity analysis in fuzzy linear
programming’, Fuzzy Sets and Systems, Vol. 1, pp.269–281.
Hashemi, S.M., Modarres, M., Nasrabadi, E. and Nasrabadi, N.N. (2006) ‘Fully fuzzified linear
programming, solution and duality’, Journal of Intelligent and Fuzzy Systems, Vol. 17,
pp.253–261.
Hosseinzadeh Lotfi, F., Allahviranloo, T., Alimardani Jondabeh, M. and Alizadeh, L. (2009)
‘Solving a full fuzzy linear programming using lexicography method and fuzzy approximate
solution’, Applied Mathematical Modelling, Vol. 33, pp.3151–3156.
Inuiguchi, M., Ramik, J., Tanio, T. and Vlach, M. (2003) ‘Satisfying solutions and duality in
interval and fuzzy linear programming’, Fuzzy Sets and Systems, Vol. 135, pp.151–177.
Klir, G.J. and Yuan, B. (1995) Fuzzy Sets and Fuzzy Logic: Theory and Applications. New Jersey:
Prentice-Hall, PTR.
Kumar, A., Kaur, J. and Singh, P. (2011) ‘A new method for solving fully fuzzy linear
programming problems’, Applied Mathematical Modelling, Vol. 35, pp.817–823.
Kumar, P., Surendra, R. and Yadav, S. (2008) ‘An integrated approach of analytic hierarchy
process and fuzzy linear programming for supplier selection’, Int. J. Operational Research,
Vol. 3, No. 6, pp.614–631.
Lai, Y.J. and Hwang, C.L. (1992) Fuzzy Mathematical Programming Methods and Applications.
Berlin: Springer.
Mahdavi-Amiri, N. and Nasseri, S.H. (2006) ‘Duality in fuzzy number linear programming by use
of a certain linear ranking function’, Applied Mathematics and Computation, Vol. 180,
pp.206–216.
Mahdavi-Amiri, N. and Nasseri, S.H. (2007) ‘Duality results and a dual simplex method for linear
programming problems with trapezoidal fuzzy variables’, Fuzzy Sets and Systems, Vol. 158,
pp.1961–1978.
Mahdavi-Amiri, N., Nasseri, S.H. and Yazdani, A. (2009) ‘Fuzzy primal simplex algorithms for
solving fuzzy linear programming problems’, Iranian Journal of Operational Research,
Vol. 1, pp.68–84.
140 A. Ebrahimnejad

Maleki, H.R., Tata, M. and Mashinchi, M. (2000) ‘Linear programming with fuzzy variables’,
Fuzzy Sets and Systems, Vol. 109, pp.21–33.
Nasseri, S.H. and Ebrahimnejad, A. (2010a) ‘A fuzzy primal simplex algorithm and its application
for solving flexible linear programming problems’, European Journal of Industrial
Engineering, Vol. 4, No. 3, pp.372–389.
Nasseri, S.H. and Ebrahimnejad, A. (2010b) ‘A fuzzy dual simplex method for fuzzy number linear
programming problem’, Advances in Fuzzy Sets and Systems, Vol. 5, No. 2, pp.81–95.
Nasseri, S.H. and Ebrahimnejad, A. (2011) ‘Sensitivity analysis on linear programming with
trapezoidal fuzzy numbers’, Int. J. Operations Research and Information Systems, Vol. 2,
No. 2, pp.96–120.
Nasseri, S.H. and Mahdavi-Amiri, N. (2009) ‘Duality in fuzzy linear programming with symmetric
trapezoidal numbers’, Fuzzy Information and Engineering, Vol. 1, pp.59–66.
Okada, S. and Soper, T. (2000) ‘A shortest path problem on a network with fuzzy arc lengths’,
Fuzzy Sets and Systems, Vol. 109, pp.22–39.
Rodder, W. and Zimmermann, H.J. (1980) ‘Duality in fuzzy linear programming’, in A.V. Fiacoo
and K.O. Kortanek (Eds.), Proceedings of the International Symposium on External
Methods and System Analysis. Berlin, Heidelberg, New York: University of Texas at Austin,
pp.415–427.
Smimou, K. (2007) ‘A fuzzy transhipment model for allocating foreign currencies’, Int. J.
Operational Research, Vol. 2, No. 3, pp.284–307.
Tang, Y-C. and Beynon, M.J. (2009) ‘Group decision-making within capital investment: a fuzzy
analytic hierarchy process approach with developments’, Int. J. Operational Research, Vol. 4,
No. 1, pp.75–96.
Vrana, I. and Aly, S. (2010) ‘Assessing candidate industrial technologies utilising hierarchical
fuzzy decision making systems’, Int. J. Industrial and Systems Engineering, Vol. 6, No. 2,
pp.187–206.
Yager, R.R. (1981) ‘A procedure for ordering fuzzy subsets of the unit interval’, Information
Sciences, Vol. 24, pp.143–161.
Zhong, Y. and Shi, Y. (2002) ‘Duality in fuzzy multi-criteria and multi-constraint level linear
programming: a parametric approach’, Fuzzy Sets and Systems, Vol. 132, pp.335–346.
Zimmermann, H.J. (1978) ‘Fuzzy programming and linear programming with several objective
functions’, Fuzzy Sets and Systems, Vol. 1, pp.45–55.