World Academy of Science, Engineering and Technology 8 2005

Simplex Method for Fuzzy Variable Linear

Programming Problems
S.H. Nasseri, and E. Ardil

μa% ( x) is called the membership function of x in a% which

Abstract—Fuzzy linear programming is an application of fuzzy
set theory in linear decision making problems and most of these maps R to a subset of the nonnegative real numbers whose
problems are related to linear programming with fuzzy variables. A
convenient method for solving these problems is based on using of
supremum is finite. If sup x μa% ( x) = 1 the fuzzy set a% is
auxiliary problem. In this paper a new method for solving fuzzy called normal.
variable linear programming problems directly using linear ranking Definition 2.2. The support of a fuzzy set a% on R is the
functions is proposed. This method uses simplex tableau which is
used for solving linear programming problems in crisp environment crisp set of all x ∈ R such that μa% ( x) > 0 .
before.
Definition 2.3. The set of elements that belong to the fuzzy set
Keywords—Fuzzy variable linear programming, fuzzy number, a% on R at least to the degree α is called the α − cut set:
ranking function, simplex method. aα = {x ∈ R | μa% ( x) ≥ α } .
I. INTRODUCTION Definition 2.4. A fuzzy set a% on R is convex if

Z IMMERMANN [10] proposed the first formulation of fuzzy

linear programming. Fang and Hu [4] considered linear
programming with fuzzy constraint coefficients. Vasant and et
μa% (λ x + (1 − λ ) y ) ≥ min{μ a% ( x), μa% ( y )} ,
x, y ∈ R, and λ ∈ [0,1] .
al [9] applied linear programming with fuzzy parameters for Note that, a fuzzy set is convex if all α − cuts are convex.
decision making in industrial production planning. Maleki and Definition 2.5. A fuzzy number a% is a convex normalized
et al [6, 7] introduced a linear programming problem with fuzzy set on the real line R such that
fuzzy variables and proposed a new method for solving these 1) It exists at least one x0 ∈ R with μ a% ( x0 ) = 1 .
problems using an auxiliary problem. Mahdavi-Amiri and
Nasseri [5] described duality theory for the fuzzy variable 2) μa% ( x) is piecewise continuous.
linear programming (FVLP) problems. This study focuses on A fuzzy number a% is a trapezoidal fuzzy number if the
FVLP problems. Hence, first some important concepts of
membership function of it be in the following form:
fuzzy theory are reviewed and concept of the comparison of
fuzzy numbers by introducing a linear ranking function is
described. Moreover, fuzzy basic feasible solution for the
FVLP problems and also optimality conditions along with
fuzzy simplex algorithm for solving the fuzzy variable linear
programming problems is proposed.

II. DEFINITIONS AND NOTATIONS

In this section, some of the fundamental definitions and
concepts of fuzzy sets theory initiated by Zadeh [2] (taken aL −α aL aU aU + β
from Bezdek [3]) are reviewed.
Fig. 1 Trapezoidal Fuzzy Number
Definition 2.1. A fuzzy set a% in R is a set of ordered pairs:
a% = {( x, μa% ( x)) | x ∈ R} We may show any trapezoidal fuzzy number by
a% = (a L , aU , α , β ) , where the support of a% is
(a − α , a + β ) , and the core of a% is [a , a ] . Let
L U L U

S.H. Nasseri, Department of Mathematical and Computer Sciences, F (R ) be the set of trapezoidal fuzzy numbers. Note that, we
Sharif University of Technology, Tehran, Iran (corresponding author: e-mail:
[email protected]). consider F ( R ) throughout this paper.
E. Ardil is with Department of Computer Engineering, Trakya University,
Edirne Turkey (e-mail: [email protected]).

198
 World Academy of Science, Engineering and Technology 8 2005

A. Arithmetic on Fuzzy Numbers IV. FUZZY LINEAR PROGRAMMING

Since in this paper we only consider the trapezoidal fuzzy In this section, we introduce fuzzy linear programming
numbers therefore we define arithmetic on the elements of (FLP) problems. In order to the definition of fuzzy linear
~
F (R ) . Let a~ = (a L , a U , α , β ) and b = (b L , bU , γ , θ ) programming problems it is necessary to introduce linear
programming problems.
be two trapezoidal fuzzy numbers and x ∈ R . Then, we
define A. Linear Programming
A linear programming (LP) problem is defined as:
x > 0, x a% = ( x a L , x a U , x α , x β ) Max z = cx
s.t. Ax = b (6)
x < 0, x a% = ( x a U , x a L , − x β , − x α )
~ x≥0
a~ + b = (a L + b L , a U + bU , α + γ , β + θ ) where c = (c1,..., cn ), b = (b1,..., bm ) , and A = [ aij ]m×n .
T
~
a~ − b = (a L − bU , a U − b L , α + θ , β + γ ) In the above problem the all of parameters are crisp [1].
Now, if the some of parameters be fuzzy numbers we obtain a
III. RANKING FUNCTIONS fuzzy linear programming which is defined in the next
A convenient method for comparing of the fuzzy numbers subsection.
is by use of ranking functions. We define a ranking
function ℜ : F ( R ) → R , which maps each fuzzy number B. Fuzzy Linear Programming
into the real line. Now, suppose that a% and b% be two Suppose that in the linear programming problem some
trapezoidal fuzzy numbers. Therefore, we define orders on parameters be fuzzy number. Then, we have a fuzzy linear
programming problem. Hence, it is possible the some
F (R ) as following:
~ coefficients of the problem in the objective function, technical
a~ ≥ b if and only if ℜ(a% ) ≥ ℜ(b% ) (1) coefficients, the right-hand side coefficients or decision
ℜ
making variables be fuzzy number [5], [6], [7], [8]. Here, we
~
a~ > b if and only if ℜ(a% ) > ℜ(b% ) (2) focus on the linear programming problems with fuzzy
ℜ
variables which is defined in the next section
~
a~ = b if and only if ℜ(a% ) = ℜ(b% ) (3)
ℜ
V. FUZZY VARIABLE LINEAR PROGRAMMING
where a~ and b~ are in F ( R) . Also we write a~ ≤ b~ if and
ℜ
A fuzzy variable linear programming (FVLP) problem is
~ ~ defined as follows:
only if b ≥ a .
ℜ Max ~ z = c~
x
ℜ
Lemma 3.1. Let ℜ be any linear ranking function. Then ~
~ ≥ b~ if and only if a~ − b~ ≥ 0 if and only if − b~ ≥− a~ s.t. A~
x =b (7)
i) a ℜ
ℜ ℜ ℜ ~
x ≥0
~ ≥ b~ and c~ ≥ d~ , then a~ + c~ ≥ b~ + d~ .
ii) If a
ℜ

where b% ∈ (F (R)) , x% ∈ (F (R)) , A∈ R

m×n
ℜ ℜ ℜ m n
, cT ∈ Rn , and ℜ
These are many numbers ranking function for comparing
is a linear ranking function.
fuzzy numbers. Here, we use from linear ranking functions,
Definition5.1. We say that fuzzy vector x% ∈ ( F ( R )) is a
n
that is, a ranking function ℜ such that
fuzzy feasible solution to (7) if and only if ~
x satisfies the
ℜ(ka% + b% ) = k ℜ(a% ) + ℜ(b% ) . (4) constraints of the problem.
One suggestion for a linear ranking function as following: Definition 5.2. A fuzzy feasible solution ~
x* is a fuzzy
1 optimal solution for (7), if for all fuzzy feasible solution ~
ℜ(a~ ) = a L + a U + ( β − α ). (5) x for
2 ~ ~
(7), we have cx* ≥ cx .
where a% = (a , a , α , β ) ∈ F ( R ) , and adopted by
L U ℜ

Maleki and et al [6, 7]. A. Fuzzy Basic Feasible Solution

~ = (a L , a U , α , β )
Then, for trapezoidal fuzzy numbers a Here, we describe fuzzy basic feasible solution (FBFS) for
~ the FVLP problems which established by Mahdavi-Amiri and
and b = (b L , bU , γ , θ ) , we have Nasseri [5].
~ 1 1 For the FVLP problem is defined in (7), consider the
a~ ≥ b if and only if aL + aU + (β −α) ≥ bL + bU + (θ −γ ).
ℜ 2 2 system Ax% = b% and x% ≥ 0 . Let A = [aij ]m ×n . Assume
ℜ ℜ

199
 World Academy of Science, Engineering and Technology 8 2005

rank (A ) = m . Partition A as [ B N ] where B , x%B + B −1 Nx% N = B −1b% , and also for objective function
ℜ
m × m , is nonsingular. It is obvious that rank ( B ) = m .
z% + (cB B N − cN ) x% N = cB B −1b%.
−1
Let y j be the solution to By =a j . It is apparent that the ℜ

Currently x% N = 0 , and then x% B = B b% , and z% = cB B b% .

−1 −1
basic solution
~ ~
xB = (~
x B1 ,..., ~
x Bm ) T = B −1b , ~
ℜ ℜ ℜ
xN = 0 (8) Then, we may rewrite the above FVLP problem in the
ℜ ℜ
~ following tableau format:
is a solution of A~
x = b . We call x% , accordingly partitioned
ℜ
TABLE I
as ( x%
T~
x NT ) T , a fuzzy basic solution corresponding to the FUZZY SIMPLEX TABLEAU
B

basis B . If ~ x B ≥ 0 , then the fuzzy basic solution is feasible z% x%B x% N R.H.S.

ℜ

z = cB ~
and the corresponding fuzzy objective value is ~ xB , z% 1 0 cB B −1 N − cN cB B −1b%
ℜ

where c B = (c B1 ,..., c Bm ) . Now, corresponding to every x%B

fuzzy nonbasic variable ~
x j , 1 ≤ j ≤ n, j ≠ Bi , and 0 I B −1 N B −1b%
i = 1,..., m, define
The above tableau gives us all the information we need to
z j = c B y j = c B B −1 a j . (9) proceed with the simplex method. The cost row in the above
If x% B > 0 , then is x% called a nondegenerate fuzzy basic tableau is (γ j ) j ≠ Bi = (cB B −1a j − c j ) j ≠ Bi = ( z j − c j ) j ≠ Bi .
ℜ
According to the optimality condition for these problems we
feasible solution, and if at least one component of x% B is zero,
are at the optimal solution if γ j ≥ 0 for all j ≠ Bi . On the
then x% is called a degenerate fuzzy basic feasible solution.
The following theorem characterizes optimal solutions. The other hand, if γl < 0 , for a l ≠ Bi then we may
result corresponds to the so-called nondegenerate problems, exchange x% Br with x%l . Then we compute the vector
where all fuzzy basic variables corresponding to every basis B
are nonzero (and hence positive) [5]. yl = B −1al . If yl ≤ 0 , then x%l can be increase
indefinitely, and then the optimal objective is unbounded. On
Theorem 5.1. Assume the FVLP problem is nondegenerate. A
~ the other hand, if yl has at least one positive component, then
fuzzy basic feasible solution ~
x B = B −1b , ~
x N = 0 is optimal the increase in will be blocked by one of the current basic
ℜ ℜ
variables, which drops to zero.
to (7) if and only if z j ≥ c j for all 1 ≤ j ≤ n.
B. Pivoting
Maleki and et al [6, 7] was proposed a method for solving
FVLP problems by use of solving an auxiliary problem. They If x%l enters the basis and x% Br leaves the basis, then
discuss on the some relations between the FVLP problem and
the auxiliary problem. Then, they used from these results for pivoting on yrl can be stated as follows:
solving the FVLP problems. Here, we propose simplex 1) Divide row r by yrl .
method for solving FVLP problems.
2) For i = 1,..., m and i ≠ r , update the i th row by adding
VI. SIMPLEX METHOD FOR THE FVLP PROBLEMS to it − yil times the new r th row.

A. Fuzzy Simplex Method in Tableau Format 3) Update row zero by adding to it γ l times the new r th row.
Consider the FVLP problem as is defined in (7).
Max z% = cB x% B + c N x% N Theorem 6.1. If in a fuzzy simplex tableau, an l exists such
ℜ that zl − cl < 0 and there exists a basic index i such that
s.t. Bx% B + Nx% N = b% (10) yil > 0 , then a pivoting row r can be found so that pivoting
ℜ

x%B , x% N ≥ 0 on yrl will yield a fuzzy feasible tableau with a corresponding

ℜ
nondecreasing objective value.
Then, it is possible we write x% B = B b% − B Nx% N and
−1 −1
ℜ Proof. We need a criterion for choosing a fuzzy basic variable
to leave the basis so that the new simplex tableau will remain
z% = cB ( B b% + B −1 Nx% N ) + cN x% N . Also, we may rewrite
−1
feasible and the new objective value is nondecreasing.
ℜ
Assume column l is the pivot column. Also, suppose that

200
 World Academy of Science, Engineering and Technology 8 2005

x% =( x%BT , x% N T )T is a fuzzy basic feasible solution to the Now, if we enter x%l into the basis, then we have x%l > 0 , and
ℜ ℜ

FVLP problem, where x% B = B b% , and

−1
x% N = 0 . Then, the x% j = 0 , for all j ≠ Bi ∪ l . Since yil ≤ 0 , i = 1,..., m , hence
ℜ ℜ ℜ

corresponding fuzzy objective value is z% = cB B b% = cB y%0 . −1 y%i 0 − yil x%l ≥ 0 (20)

ℜ
ℜ ℜ
On the other hand, for any fuzzy basic feasible solution to the Therefore, the current fuzzy basic solution will remain
FVLP problem, we have feasible. Now, the value of ẑ for the above fuzzy feasible
solution as following:
x%B + ∑ y j x% j = y% 0 (11) m
ℜ
j ≠ Bi
zˆ = cB x% B + cN x% N = ∑ cBi ( y%i 0 − yil x%l ) + cl x%l
−1 ℜ ℜ
where y j = B a j . i =1
m m
So, if x%l enters into the basis we may write = ∑ cBi y%i 0 − (∑ cBi yil − cl ) x%l
ℜ
x%B = y% 0 − yl x%l (12) i =1 i =1
ℜ = cB y% 0 − (cB yl − cl ) x%l = z% − ( zl − cl ) x%l
ℜ ℜ
Since, we want x% B be feasible, hence
So,
y%i 0 − yil x%l ≥ 0 , for all i = 1,..., m. zˆ = z% − ( zl − cl ) x%l . (21)
ℜ
ℜ
If yil ≤ 0 , then it is obvious that the above condition is hold. Hence, we can enter x%l into the basis with arbitrarily large
Hence, for all yil > 0 , we need to have fuzzy value. Then, from (21) we have unbounded solution.
y%i 0
x%l ≤ (13) VII. A FUZZY SIMPLEX METHOD
ℜ yil
Suppose that we are given a basic feasible solution with
To satisfy (5) it is sufficient to let basis B . Then:
y% r 0 y%
1.The basic feasible solution is given by x% B = B b% = y% 0 and
−1
= min{ i 0 | yil > 0} (14)
yrl ℜ yil ℜ ℜ

Also, for any fuzzy basic feasible solution to the FVLP x% N = 0 . The fuzzy objective z% = cB B b% = cB y% 0 . −1
ℜ ℜ ℜ
problem, we have
−1
z% = cB y% 0 − ∑ ( z j − c j )x% j (15) 2.Calculate w = cB B , and y 0 = ℜ( y% 0 ) . For each
ℜ
j ≠ Bi nonbasic variable, calculate γ j = z j − c j = cB B −1a j − c j
So, if we enter x%l into the basis we have
= wa j − c j . Let γ l = min j {γ j } . If γ l ≥ 0 , then stop; the
z% = cB y% 0 − ( zl − cl ) x%l (16) current solution is optimal. Otherwise go to step 3.
ℜ
−1
We note that the new objective value is nondecreasing, since 3.Calculate yl = B al . If yl ≤ 0 , then stop; the optimal
z% = cB y% 0 − ( zl − cl ) x%l ≥ cB y% 0 (17) solution is unbounded. Otherwise determine the index of the
ℜ ℜ
variable x% Br leaving the basis as follows:
Using the fact that ( zl − cl ) x%l ≤ 0 . yr 0 y
ℜ = min{ i 0 | yil > 0}
Theorem 6.2. If for any fuzzy basic feasible solution to the yrl 1≤i ≤ m yil
FVLP problem there is some column not in basis for which y%
zl − cl < 0 and yil ≤ 0 , i = 1,..., m , then the FVLP problem Update y% i 0 by replacing y% i 0 − r 0 yil for i ≠ r and y% r 0 by
yrl
has an unbounded solution.
y% r 0
Proof. Suppose that x% B is a fuzzy basic solution to the FVLP replacing . Also, update z% by replacing
problem, so
yrl
x%Bi + ∑ yij x% j = y%i 0 , i = 1,..., m, j = 1,..., n, (18) y%
z% − r 0 ( zl − cl ) . Then, update B by replacing aBr with al
ℜ
j ≠ Bi yrl
or and go to step 2.
x%Bi = y%i 0 − ∑ yij x% j , i = 1,..., m, j = 1,..., n. (19)
ℜ
j ≠ Bi VIII. A NUMERICAL EXAMPLE
For an illustration of the above method we solve a FVLP
problem by use of fuzzy simplex method.

201
 World Academy of Science, Engineering and Technology 8 2005

Example 8.1. and introduced the fuzzy basic feasible solution for these
max z% = 3 x%1 + 4 x%2 problems. Finally, we proposed a new algorithm for solving
ℜ these problems directly, by use of linear ranking function.
s.t. 3 x%1 + x%2 ≤(2, 4,1,3)
ℜ
REFERENCES
2 x%1 − 3 x%2 ≤(3,5, 2,1) [1] M.S. Bazaraa, J.J. Jarvis and H.D. Sherali, Linear Programming and
ℜ
Network Flows, John Wiley, New York, Second Edition, 1990.
x%1 , x%2 ≥ 0 [2] R.E. Bellman and L.A. Zadeh, “Decision making in a fuzzy
ℜ
environment”, Management Sci. 17 (1970) 141--164.
Now, we may rewrite the above problem in form (10): [3] J.C. Bezdek, “Fuzzy models - What are they, and Why?”, IEEE
Transactions on Fuzzy Systems 1 (1993) 1--9.
[4] S.C. Fang and C.F. Hu, “Linear programming with fuzzy coefficients in
3 x%1 + x%2 + x%3 =(2, 4,1,3) constraint”, Comput. Math. Appl. 37 (1999) 63--76.
ℜ
[5] N. Mahdavi-Amiri and S.H. Nasseri, “Duality in fuzzy variable linear
2 x%1 − 3 x%2 + x%4 =(3,5, 2,1) programming”, 4th World Enformatika Conference, WEC'05, June 24-
ℜ 26, 2005, Istanbul, Turkey.
x%1 , x%2 , x%3 , x%4 ≥ 0 [6] H.R. Maleki, “Ranking functions and their applications to fuzzy linear
ℜ programming”, Far East J. Math. Sci. 4 (2002) 283--301.
[7] H.R. Maleki, M. Tata and M. Mashinchi, “Linear programming with
Therefore, using fuzzy simplex tableau (Table I), we obtain
fuzzy variables”, Fuzzy Sets and Systems 109 (2000) 21--33.
first tableau as follow: [8] H. Rommelfanger, R. Hanuscheck and J. Wolf, “Linear programming
with fuzzy objective”, Fuzzy Sets and Systems 29 (1989) 31--48.
[9] P. Vasant, R. Nagarajan, and S. Yaacab, ‘‘Decision making in industrial
basis x%1 x%2 x%3 x%4 R.H.S. ℜ( R.H .S .) production planning using fuzzy linear programming”, IMA, Journal of
Management Mathematics 15 (2004) 53--65.
z% -3 -4 0 0 0% 0 [10] H. J. Zimmermann, “Fuzzy programming and linear programming with
several objective functions”, Fuzzy Sets and Systems 1 (1978) 45--55.
x%3 3 1 1 0 (2,4,1,3) 7
x%4 2 -3 0 1 (3,5,2,1) 7.5

From the above tableau, we obtain

γ1 = z 1 − c1 = −3 < 0 γ 2 = z 2 − c 2 = −4 < 0 .
and
Then, γ 2 < γ 1 . Hence, related fuzzy nonbasic variable to γ 2 ,
that is x%2 is an entering variable. Therefore, according to the
minimum ration test is given in the step 3 of the fuzzy simplex
algorithm, x%3 is a leaving variable. Now, after pivoting (as
given in the part B of section 6) the new tableau is:

basis x%1 x%2 x%3 x%4 R.H.S. ℜ( R.H .S .)

z% 9 0 4 0 (8,16, 4,12) 28
x%2 3 1 1 0 (2,4,1,3) 7
x%4 11 0 3 1 (9,17,5,10) 28.5

According to above tableau, for fuzzy nonbasic variables

x%1 , x% 3 we have γ 1 = 9 > 0, γ 3 = 4 > 0 . Hence, using the
optimality condition for the FVLP problems is given in
Theorem 5.1, the optimal fuzzy solution is obtained
x% 1* = (0, 0, 0, 0) , x% *2 = (2, 4,1, 3) , x% *3 = (0, 0, 0, 0) ,
ℜ ℜ ℜ

x% *4 =(9,17,5,10) and z% =(8,16, 4,12) with ℜ( z% ) = 28 .

ℜ ℜ

IX. CONCLUSION
We considered fuzzy variable linear programming problems

202