S.H. Nasseri and Z. Alizadeh/ TJMCS Vol .3 No.

3 (2011) 318 - 328

Solving Linear Programming Problem with Fuzzy Right

Hand Sides: A Penalty Method
S.H. Nasseri∗ and Z. Alizadeh
Department of Mathematics, University of Mazandaran, Babolsar, Iran

Abstract
Linear programming problems with trapezoidal fuzzy variables (FVLP) have recently at-
tracted some interest. Some methods have been developed for solving these problems. Fuzzy
primal and dual simplex algorithms have been recently proposed to solve these problems.
These methods have been developed with the assumption that an initial Basic Feasible So-
lution (BFS) is at hand. In many cases, finding such a BFS is not straightforward and some
works may be needed to get the simplex algorithm started. In this paper, we propose a
penalty method to solve FVLP problems in which the BFS is not readily available.

1 Introduction
Many application problems, modeled as mathematical programming problems, may be for-
mulated with uncertainty. Fuzzy programming approach is useful to treat a problem under
∗
Corresponding author. [email protected] (Hadi Nasseri)

318
 S.H. Nasseri and Z. Alizadeh/ TJMCS Vol .3 No.3 (2011) 318 - 328

uncertainty. The concept of decision making in fuzzy environment was first proposed by
Bellman and Zadeh [2]. Several kinds of fuzzy linear programming problems have appeared
in the literature and different methods have been proposed to solve such problems. The first
formulation of Fuzzy Linear Programming (FLP) is proposed by Zimmermann [12]. Some
methods have been developed for solving these problems by introducing and solving certain
auxiliary problems. Mahdavi-Amiri et al. [10] have developed a new fuzzy primal and dual
simplex algorithm for solving the FVLP problem. As they have been mentioned the fuzzy
primal algorithm needs an initial basic feasible solution (see[10]). In some cases, such a
solution is not straightforward. Hence in this paper we propose a fuzzy Big-M method for
solving these types of problems where directly can find a basic feasible solution if there exist.
Our discussion, here is outlined as follows: In next section we briefly give some necessary
concepts of fuzzy set theory. In Section 3 we first define a Linear Programming with Fuzzy
Variables (FVLP) problem and then formulate a real problem as FVLP problem. Further-
more,the fundamental concepts and theorems are given to complete our discussion in this
section. We also introduce a Big-M method for solving FVLP problems and illustrate it with
an example in Section 4. Finally, we conclude in Section 5.

2 Preliminaries
We review the fundamental notations of fuzzy set theory, initiated by Bellman and Zadeh
[2], to be used throughout this note and which is taken from ([5],[6],[8],[9]and[10]). We give
here, some basic definitions on fuzzy numbers comparison that will be used for illustrating
our approach. In this way, and for methodological reasons, in spite that the approach that
is designed will work correctly for any kind of fuzzy numbers, we prefer in the following to
focus on the usual case of trapezoidal fuzzy numbers.

Definition 1. Let X be the universal set. A e is called a fuzzy set in X if A

e is a set of ordered
{( ) }
pairs Ae = x, µ e (x) |x ∈ X , where µ e (x) is the membership function of x in A. e
A A
remark 1. The membership function of A e specifies the degree of membership of element x
in fuzzy set Ae (in fact, µ e shows the degree that x belongs to A).e
A
{ }
Definition 2. The α-level set of A e is the set Aeα = x ∈ R|µ e (x) ≥ α , where α ∈ (0, 1] .
A
The lower and upper bounds of any α-level set A eα are represented by finite numbers inf e
x∈Aα
and supx∈Aeα .
Definition 3. The support of a fuzzy set A e is a set of elements in X for which µ e (x) is
{ } A
positive, that is, suppAe = x ∈ X|µ e (x) > 0 .
A

319
 S.H. Nasseri and Z. Alizadeh/ TJMCS Vol .3 No.3 (2011) 318 - 328

e is convex if
Definition 4. A fuzzy set A

{ }
µAe (λx + (1 − λ) y) ≥ min µAe (x) , µAe (y) , ∀x, y ∈ Xand λ ∈ [0, 1] .

Definition 5. A convex fuzzy set A e on R is a fuzzy number if the following conditions hold:
(a)Its membership function is piecewise continuous.
(b)There exist three intervals [a, b],[b, c] and [c, d] such that µAe is increasing on [a, b], equal
to 1 on [b, c], decreasing on [c, d] and aqual to 0 elsewhere.
remark 2. In the above definition, we say interval [b, c] is the modal set of fuzzy number A. e
( L U ) [ L ]
Definition 6. Let A e = a , a , α, β denote the trapezoidal fuzzy number, where a − α, aU + β
[ ]
is the support of A e and aL , aU its modal set.
remark 3. We denote the set of all trapezoidal fuzzy numbers by F (R). If a = aL = aU
then we obtain a triangular fuzzy number, and we show it with A e = (a, α, β).

2.1 Arithmetic on fuzzy numbers

( ) ( )
a = aL , aU , α, β and eb = bL , bU , γ, θ be two trapezoidal fuzzy numbers. Define
Let e

( )
x > 0, x ∈ R; x.ea = xaL , xaU , xα, xβ ,
( )
x < 0, x ∈ R; x.ea = xaU , xaL , −xβ, −xα ,
( )
a + eb = aL + bL , aU + bU , α + γ, β + θ ,
e
( )
a − eb = aL − bU , aU − bL , α + θ, β + γ .
e

2.2 Ranking function

An effective approach for ordering the elements of F (R) is also to define a ranking function
R : F (R) → R which maps each fuzzy number into the real line, where a natural order exists.
We define orders on F (R) by:

( )
a ≽ eb if and only if R (e
e a) ≥ R eb
( )
a ≻ eb if and only if R (e
e a) ≥ R eb
( )
e ∼ e
a = b if and only if R (ea) = R eb

a and eb are in F (R) . Also we write e

where e a ≼ eb if and only if eb ≽ e
a.
We restrict our attention to linear ranking functions, that is, a ranking function R such that

320
 S.H. Nasseri and Z. Alizadeh/ TJMCS Vol .3 No.3 (2011) 318 - 328

( ) ( )
a + eb = kR (e
R ke a) + R eb

a and eb belonging to F (R) and any k ∈ R.

for any e
We consider the linear ranking functions on F (R) as:

a) = cL aL + cU aU + cα α + cβ β,
R (e
( )
where ea = aL , aU , α, β , and cL , cU , cα , cβ are constants, at least one of which is nonzero. A
special version of the above linear ranking function was first proposed by Yager[11] as follows:
∫1
R (ea) = 12 0 (inf e aλ + sup e
aλ ) dλ

which reduces to
aL +aU
R (e
a) = + 14 (β − α).
2
( ) ( )
a = aL , aU , α, β and eb = bL , bU , γ, θ , we have
Then, for trapezoidal fuzzy numbers e

a ≽ eb
e if and only if aL + aU + 12 (β − α) ≥ bL + bU + 12 (θ − γ) .

3 Fuzzy linear programme

3.1 Definition of model
Here we first define a general form of the problem which will be discussed in this paper.
Consider a Linear Programming with Fuzzy Variables (FVLP) as follows:

min ze ≈ ce
x (1)

x ≽ eb
s.t. Ae
e≽e
x 0
e ∈ (F (R))n , A ∈ Rm×n , eb ∈ (F (R))m and eb ≽ e
where c ∈ Rn , x 0.
Note that an FVLP problem is a linear programming problem in fuzzy environment in which
the decision making variables and right-hand sides are fuzzy numbers.

321
 S.H. Nasseri and Z. Alizadeh/ TJMCS Vol .3 No.3 (2011) 318 - 328

3.2 Formulation of a problem

An alloy producer, produces 2 types of alloys P1 and P2 . These alloys consists of Zinc and
Tin(F1 , F2 ). The tableau below shows the amount of F1 and F2 used in per unit of alloys.
The average cost of per kilogram of P1 and P2 are 10 and 6 cents, respectively. The minimum
daily supplyes metals F1 and F2 are approximately 49 and 45 units, respectively. The producer
wants to know in order to minimize the cost of Zinc and Tin how many kilograms of alloys
P1 and P2 he must produce daily?

P roducts
M etal P1 P2
F1 1 6
F2 4 2

Now if we assume that the fuzzy variables xe1 and x e2 be the daily supplyes metals F1 and F2
respectively, then this problem can be formulated as follows:

min ze ≈ 10e
x1 + 6e
x2

e1 + 6e
s.t. x x2 ≽ (46, 52, 2, 2)
4e x2 ≽ (42, 48, 4, 4)
x1 + 2e
x e2 ≽ e
e1 , x 0

3.3 Fuzzy basic feasible solution

First of all, we briefly describe Fuzzy Basic Feasible Solution (FBFS) for the FVLP prob-
lem as established by Mahdavi-Amiri and Nasseri [8]. Consider the FVLP problem and
A = [aij ]m×n . Assume rank(A) = m. Partition A as [B N ] where B, m × m, is non-
singular. It is obvious that rank(B) = m. Let yj be the solution to By = aj . It is
apparent that the basic solution x e ≈ (e eBm )T ≈ B −1eb, x
xB1 , ..., x eN ≈ e
0 is a solution of
( T T )TB
Ae e
x = b. In fact, x e = x eB x eN . If x e
eB ≽ 0, then the basic solution is feasible and
the corresponding fuzzy objective value is: ze ≈ cB x eB , where cB = (cB1 , ..., cBm ). Now,
corresponding to every nonbasic variable x ej , 1 ≤ j ≤ n, j ̸= Bi , i = 1, ..., m, define
−1
zj = cB yj = cB B aj . The following result concerns the non-degenerated problems, where
every fuzzy basic variable corresponding to every basis B is positive. Observe that for any
basic index j = Bi , 1 ≤ i ≤ m, we have B −1 aj = ei where ei = (0, ..., 0, 1, 0, ..., 0)T is the ith
unit vector, since Bei = [aB1 , ..., aBi , ..., aBm ] ei = aBi = aj , and so we have:

322
 S.H. Nasseri and Z. Alizadeh/ TJMCS Vol .3 No.3 (2011) 318 - 328

zj − cj = cB B −1 aj − cj = cB ei − cj = cBi − cj = cj − cj = 0.

The following theorem characterizes optimal solutions.

Theorem 1. (optimality conditions) Assume the FVLP problem is non-degenerate. If a

eB = B −1eb, x
basic solution x eN ≈ e0 is feasible to (1)and zj ≥ cj for all j, 1 ≤ j ≤ n then the
fuzzy basic solution is a fuzzy optimal solution to (1).

Proof: See Mahdavi-Amiri and Nasseri [8].

Theorem 2. Let x eB = B −1eb is a fuzzy basic feasible solution of the problem max ze ≈ ce x
such that Ae e
x ≈ b and x e≽e 0. If for any column aj in A which is not in B, the condition
(zj − cj ) < 0 hold and yij > 0 for some i, j ∈ {1, 2, ..., m} then it is possible to obtain a new
fuzzy basic feasible solution by replacing one of the columns in B by aj .

Proof: See[8].

Theorem 3. If in an FVLP problem, there is a column aj in A (not in basis) for which

zj − cj < 0 and yij ≤ 0, i = 1, ..., m, then the FVLP problem is unbounded.

Proof: It is straight forward.

We write the above FVLP problem in the following tableau format:

Basis x̃B x̃N R.H.S.

ze 0 zN − cN = cB B −1 N − cN ye00 = cB B −1eb
eB
x I Y = B −1 N ye0 = B −1eb

Now we are going to give the concept of pivoting and change of basis.
ek enters the basis and x
If x eBr leaves the basis, then pivoting on yrk in the simplex tableau is
stated as follows:

1. Divide row r by yrk

2. For i = 0, 1, ..., m and i ̸= r, update the ith row by adding to it −yik times the new rth
row.

4 Big-M Method
The simplex algorithm was conceived by Dantzig for solving Linear Programming (LP)problems.
This method starts with a Basic Feasible Solution (BFS)and moves to an improved BFS, un-
til the optimal point is reached or else unboundedness of the objective function is verified.

323
 S.H. Nasseri and Z. Alizadeh/ TJMCS Vol .3 No.3 (2011) 318 - 328

In order to initialize this algorithm a BFS must be available. In many cases, finding such
a BFS is not straightforward and some work may be needed to get the simplex algorithm
started.To this end, there are two techniques in linear programming literature: Two-phase
method and Big-M method [1]. But there may be some LP models for which there are not any
BFSs, i.e., the model is infeasible. Both Two-phase method and Big-M method distinguish
the infeasibility. Here we focus on Big-M method.
Consider a generic LP model. After manipulating the constraints and introducing the re-
quired slack variables, the constraints are put in the format AX = b, X ≥ 0 where A is a
m × n matrix and b ≥ 0 an m × 1 vector. Considering C as cost vector, the following LP
model is dealt with:

min CX (P )

s.t. AX = b

X≥0

Furthermore, suppose that we do not have a starting BFS for simplex metod, i.e, A has no
identity submatrix. In this case we shall resort to the artificial variables to get a starting
BFS, and then use the simplex method itself and get rid of these artificial variables. The use
of artificial variables to obtain a starting BFS was first provided by Dantzig [3, 4].
To illustrate, suppose that we change the restrictions by adding an artificial vector R leading
to the system AX + R = b, (X, R) ≥ 0. This forces an identity submatrix corresponding to
the artificial vector and gives an immediate BFS of the new system, namely (X = 0, R = b).
Even though we now have a starting BFS and the simplex method can be applied, we have
in effect changed the problem. In order to get back to our original problem, we must force
these artificial variables to zero, because AX = b ⇔ AX + R = b, R = 0. In Big-M method
we assign a large penalty coefficient to these variables in the original objective function in
such a way as to make their presence in the basis at a positive level very unattractive from
the objective function point of view. More specifically, (P ) is changed to:

min CX + M R P (M )

s.t. AX + R = b

(X, R) ≥ 0

324
 S.H. Nasseri and Z. Alizadeh/ TJMCS Vol .3 No.3 (2011) 318 - 328

where M is a very large positve number. The term M R can be interpreted as a penalty
to be paid by any solution with R ̸= 0. Therefore the simplex method itself will try to get
the artificial variables out of the basis, and then continue to find an optimal solution of the
original problem. Hereafter ‘*’ indicates the optimality, zj − cj is the reduced cost of the jth
variable, and yj = B −1 aj , where B is the basis of the simplex method associated with the
related iteration and aj is the jth column of the technological coefficients matrix (for P (M )
it is [A I]). Four possible cases may arise while solving P(M):
(A1 ) : (X ∗ , R∗ ) is an optimal solution of P(M), in which R∗ = 0
(A2 ) : (X ∗ , R∗ ) is an optimal solution of P(M), in which R∗ ̸= 0
(B1 ) : zk − ck = max (zj − cj ) > 0, yk ≤ 0, and all artificials are equal to zero.
(B2 ) : zk − ck = max (zj − cj ) > 0, yk ≤ 0, and not all artificials are equal to zero.
remark 4. A Two-phase method for solving FVLP problem was proposed later [7]. Here we
are going to solve an FVLP problem using Big-M method.
Consider the following problem:

min Ze = 10e
x1 + 6e
x2

e1 + 6e
s.t. x x ≥ (46, 52, 2, 2)
4e x2 ≥ (42, 48, 4, 4)
x1 + 2e
x e2 ≽ e
e1 , x 0

First of all we add the slack variables:

min Ze = 10e
x1 + 6e
x2
e1 + 6e
s.t. x x−x
e3 = (46, 52, 2, 2)
4e x2 − x
x1 + 2e e4 = (42, 48, 4, 4)
e1 , x
x e2 , x e4 ≽ e
e3 , x 0
As its seen no initial basic variable is available and hence we may use the Big-M method to
solve the problem.
Therefore, the problem changes to the following form:

min Ze = 10e
x1 + 6e e1 + M R
x2 + M R e2

e1 + 6e
s.t. x x−x e1 = (46, 52, 2, 2)
e3 + R

325
 S.H. Nasseri and Z. Alizadeh/ TJMCS Vol .3 No.3 (2011) 318 - 328

4e x2 − x
x1 + 2e e2 = (42, 48, 4, 4)
e4 + R
e1 , x
x e2 , x
e3 , x e1 , R
e4 , R e2 ≽ e
0

The problem is written in the tableau format as:

Basis x̃1 x̃2 x̃3 x̃4 R̃1 R̃2 R.H.S.

ze −10 −6 0 0 −M −M 0
e
R1 1 6 −1 0 1 0 (46, 52, 2, 2)
Re2 4 2 0 −1 0 1 (42, 48, 4, 4)

Because Re1 and R e2 are basic variables, so the cost row must be equal to zero. Hence the next
tableau will be as following:

Basis x̃1 x̃2 x̃3 x̃4 R̃1 R̃2 R.H.S.

ze −10 + 5M −6 + 8M −M −M 0 0 (88M, 100M, 6M, 6M )
Re1 1 6 −1 0 1 0 (46, 52, 2, 2)
e
R2 4 2 0 −1 0 1 (42, 48, 4, 4)

e1 is a leaving fuzzy variable. Then

e2 is an entering fuzzy variable and R
It is obvious that x
after pivoting the next tableau is given as:

Basis x̃1 x̃2 x̃3 x̃4 R̃1 R̃2 R.H.S.

ze −9 + 3 M 0
11 M
3
− 1 −M 1 − 3 M 0 (48 +
4 74M
3
, 52 + 98M 3
, 2 + 14M 3
,2 + 14M
3
)
−1
e2
x 1
6
1 6
0 1
6
0 46 52 2 2
( 6 , 6 , 6, 6)
Re2 11
0 1
−1 −1
1 ( 74 , 98 , 14 , 14 )
3 3 3 3 3 3 3

So xe1 is an entering fuzzy variable and Re2 is a leaving fuzzy variable. The last tableau is
given in the below. Now we can see all artificial variables leaves the basis. So we can remove
their columns

Basis x̃1 x̃2 x̃3 x̃4 R̃1 R̃2 R.H.S.

ze 0 0 −2 11
−27
11
2
11
−M 27
11
− M ( 11 , 11 , 11 , 11 )
1148 1478 176 176

−2 −1
e2
x 0 1 11 1
22
2
11 22
( 68 , 83 , 6 , 6 )
11 11 11 11
−3 −1
e1
x 1 0 11 1
11 11
3
11
( 74 , 98 , 14 , 14 )
11 11 11 11

Therefore, the fuzzy optimal solution of the FVLP problem which is obtained by the Big-M
( 98 14 14 ) ( 83 6 6 )
e1 = 74
method is x e2 = 68
, 11 , 11 , 11 , x , , , and the fuzzy optimal value of its objective
( 1148 1478 176 176 ) 11 11 11 11
11
function is ze = 11 , 11 , 11 , 11 .

326
 S.H. Nasseri and Z. Alizadeh/ TJMCS Vol .3 No.3 (2011) 318 - 328

5 Conclusions
The Big-M method is a known auxiliary technique to start the simplex algorithm. In particu-
lar, we have illustrated the mentiond method by formulating and solving a linear programming
problem with fuzzy variables.

6 Acknowledgment
The authors would like to thank from the anonymous referees for their valuable comments to
improve the earlier version of this paper.

References
[1] M.S. Bazarra, J.J. Jarvis and H.D. Sherali, Linear Programming and Network Flows,
John Wiley and Sons, 1990.

[2] R.E. Bellman, L.A. Zadeh, Decision making in a fuzzy environment, Manag. Sci.
17(1970) 141-164.

[3] G.B. Dantzig, Maximization of a linear function of variables subject to linear inequali-
ties, T.C. Koopmans(Ed.), Activity Analysis of Production and Allocation, John Wiley
and Sons, New York, 1951, pp. 339-347.

[4] G.B. Dantzig, Programming in a linear structure, Comptroller, United Air Force, Wash-
ington, D.C., February, 1948.

[5] D. Dubois, H. Prade, Fuzzy Sets and Systems, Theory and Applications, Academic
Press, New York, 1980.

[6] A. Ebrahimnejad, S.H. Nasseri and S.M. Mansourzadeh, Bounded Primal Simplex Al-
gorithm for Bounded Linear Programming with Fuzzy Cost Coefficients, International
Journal of Operations Research and Information Systems, 2(1)January-March 2011,
96-120.

[7] B. Khabiri, S.H. Nasseri and Z. Alizadeh, Starting fuzzy solution for the fuzzy primal
simplex algorithm using a fuzzy two-phase method, The 5th International Conference
on fuzzy Information and Engineering (ICFIE), Oct.14-16, 2011, Chengdu, China.

327
 S.H. Nasseri and Z. Alizadeh/ TJMCS Vol .3 No.3 (2011) 318 - 328

[8] N. Mahdavi-Amiri and S.H. Nasseri, Duality results and a dual simplex method for
linear programming problems with trapezoidal fuzzy variables, Fuzzy Sets and Systems
158(2007)1961-1978.

[9] N. Mahdavi-Amiri and S.H. Nasseri, Duality in fuzzy number linear programming by
use of a certain linear ranking function, Applied Mathematics and Computation 180
(2006)206-216.

[10] N. Mahdavi-Amiri, S.H. Nasseri and A. Yazdani, Fuzzy primal simplex algorithms for
solving fuzzy linear programming problems, Iranian Journal of Operational Research
1(2),(2009)68-84.

[11] R.R.Yager, A procedure for ordering fuzzy subsets of the unit interval, Information
Sciences 24, (1981)143-161.

[12] H.J. Zimmermann, Fuzzy programming and linear programming with several objective
functions, Fuzzy Sets and Systems 1, (1978) 45-55.

328