World Academy of Science, Engineering and Technology

International Journal of Mathematical and Computational Sciences

Vol:5, No:3, 2011

Block homotopy perturbation method

for solving fuzzy linear systems
Shu-Xin Miao

Abstract—In this paper, we present an efficient numerical al- simple problem which is easily solved. In most cases, using
gorithm, namely block homotopy perturbation method, for solving HPM, gives a very rapid convergence of the solution series,
fuzzy linear systems based on homotopy perturbation method. Some and usually only a few iterations leading to very accurate
numerical examples are given to show the efficiency of the algorithm.
solutions. The HPM has been used to solve various types of
nonlinear problems, see [23], [5], [4], [10], [9] and references
Keywords—Homotopy perturbation method, Fuzzy linear systems,
therein.
Block linear system, Fuzzy solution, Embedding parameter.
International Science Index, Mathematical and Computational Sciences Vol:5, No:3, 2011 [Link]/Publication/428

In [15] and [22], the HPM has been used to solve the linear
systems. Especially, the point HMP method for solving n × n
I. I NTRODUCTION nonsingular fuzzy linear system have been studied in [8] while

M ANY engineering problems, such as equilibrium and

steady-state problems, a mechanism using the kineto-
static approach, require the solution of simultaneous algebraic
the block HMP method for solving full fuzzy linear system
have been considered in [16].
In this paper, we consider the block HMP method for
linear equations. However, many real-world engineering sys- solving n × n fuzzy linear system, which is efficient and
tems are too complex to be defined in precise terms, therefore, practical because the procedure only require the nonsingularity
imprecision is often involved. Fuzzy linear systems, which can of the coefficient matrix of n × n fuzzy linear system while
formulate uncertainty in actual environment, play an essential the point HMP method require the diagonal entries of the
role in such cases [12], [18], [11] since the concept of fuzzy coefficient matrix are nonzero (see [8]).
number and arithmetic operations with these numbers are The structure of this paper is organized as follows: In Sect.
first introduced by Zadeh [24]. It is immensely important II, we introduce the notation, the definitions, and preliminary
to development mathematical model and numerical procedure results that will be used throughout the paper. In Sect. III,
that would appropriately treat general fuzzy linear systems and the block HMP method for solving fuzzy linear system is
solve them. proposed. The proposed model is illustrated by solving some
Friedman et al. [12] proposed a general model for solving examples in Sect. IV and conclusions are drawn in Sect. V.
an n × n fuzzy linear system whose coefficient matrix is
a crisp matrix and the right hand column is an arbitrary II. P RELIMINARIES
fuzzy number vector. Using the embedding method given in Zadeh [24] defined a fuzzy number as follows:
[21], Friedman et al. [12] replace the original fuzzy linear Definition 1. A fuzzy number is a fuzzy set like ũ : R →
system by an 2n × 2n crisp linear system. Then solving I = [0, 1], which satisfies
n × n fuzzy linear system is equal to solving 2n × 2n 1. ũ is upper semi-continuous,
crisp linear system. Followed Friedman et al. [12], the point 2. ũ(x) = 0 outside some interval [c, d],
Jacobi, Gauss-Seidel, SOR and steepest descent methods and 3. There are real numbers a, b such that c ≤ a ≤ b ≤ d and
conjugate gradient methods have been presented for solving 3.1 ũ(x) is monotonic increasing on [c, a],
2n × 2n crisp fuzzy linear system, see for example [6], [7], 3.2 ũ(x) is monotonic decreasing on [b, d],
[19], [3], [1] and references therein. The direct methods based 3.3 ũ(x) = 1 when a ≤ x ≤ b.
on LU decomposition have been proposed and analyzed by The set of all these fuzzy numbers is denoted by E. An
Abbasbandy, R. Ezzati and A. Jafarian [2]. equivalent parametric of a fuzzy number is given in [13] as
However, if there is an diagonal element of the coefficient Definition 2. A fuzzy number in parametric form is an
matrix being zero, then classic point iterative methods are not ordered pair of functions (u(r), u(r)), 0 ≤ r ≤ 1, which
working. The block method has been studied in [20], [17], satisfy the following requirements:
including block Jacobi, block Gauss-Seidel and block SOR 1. u(r) is a bounded left continuous nondecreasing function
methods. over [0, 1],
Recently, an analytic approach based on the basic ideas 2. u(r) is a bounded left continuous nonincreasing function
of homotopy, which is called homotopy perturbation method over [0, 1],
(HPM), is provided by He [14] for nonlinear problems. The 3. u(r) ≤ u(r), 0 ≤ r ≤ 1.
HPM, which is a coupling of the traditional perturbation A crisp number α is simply represented by u(r) = u(r) =
method and homotopy in topology, deforms continuously to a α, 0 ≤ r ≤ 1.
The addition and scalar multiplication of fuzzy numbers
Shu-Xin Miao is with the College of Mathematics and Information Science,
Northwest Normal University, Lanzhou, 730070, People’s Republic of China, can be described as follows, for arbitrary u = (u(r), u(r)),
e-mail: shuxinmiao@[Link]. v = (v(r), v(r)), 0 ≤ r ≤ 1, and real number k,

International Scholarly and Scientific Research & Innovation 5(3) 2011 347 [Link]/1307-6892/428
 World Academy of Science, Engineering and Technology
International Journal of Mathematical and Computational Sciences
Vol:5, No:3, 2011

1. u + v 
= (u(r) + v(r), u(r) + v(r)), only if the coefficient matrix S is nonsingular. The following
(ku(r), ku(r)), k ≥ 0, theorem tell us when S is nonsingular.
2. ku =
(ku(r), ku(r)), k < 0. Theorem 1. [12] The matrix S is nonsingular if and only
Definition 3. The n × n linear system if A = B − C and B + C are both nonsingular.
⎧ If the matrix S is nonsingular, then the solution vector X
⎪
⎪ a11 x1 + a12 x2 + · · · + a1n xn = y1 ,
⎪
⎨ a21 x1 + a22 x2 + · · · + a2n xn = y2 , represent a solution fuzzy vector to the fuzzy system (1) if
.. (1) and only if (xj (r), xj (r)) is 
a fuzzy numberfor all j.
⎪
⎪ . Definition 6. Let X = (xj (r), xj (r)) , 1 ≤ j ≤ n
⎪
⎩
an1 x1 + an2 x2 + · · · + ann xn = yn , denote
 the solution of (3).The fuzzy number vector U =
(uj (r), uj (r)), 1 ≤ j ≤ n defined by
or briefly
 
Ax = y, uj (r) = min xj (r), xj (r), xj (1), xj (1) ,
 
where the coefficient matrix A = (aij ), 1 ≤ i, j ≤ n is a uj (r) = max xj (r), xj (r), xj (1), xj (1)
crisp matrix, y = (y1 , y2 , · · · , yn )T is known with yi ∈ E and is called the fuzzy solution of (3). If (xj (r), xj (r)), 1 ≤ j ≤ n
x = (x1 , x2 , · · · , xn )T is unknown with xi ∈ E, 1 ≤ i ≤ n, are all fuzzy numbers then uj (r) = xj (r), uj (r) = xj (r),
International Science Index, Mathematical and Computational Sciences Vol:5, No:3, 2011 [Link]/Publication/428

is called a fuzzy linear system (FLS). 1 ≤ j ≤ n and U is called a strong fuzzy solution. Otherwise,
Definition 4. A fuzzy number vector X = U is called a weak fuzzy solution.
(x1 , x2 , · · · , xn )T given by xj = (xj (r), xj (r)), 1 ≤
j ≤ n, 0 ≤ r ≤ 1, is called a solution of the fuzzy linear III. A NALYSIS OF THE HPM
system (1) if
⎧ n Consider the crisp linear system (3) and let
⎪  n
⎪
⎪ aij xj = aij xj = y i , L(U) = SU − Y, F (U) = QU − Y,
⎪
⎪
⎨ j=1 j=1
i = 1, · · · , n. (2) where Q is nonsingular. We define homotopy H(U, p) by
⎪
⎪
⎪
⎪ n n
⎪
⎩ aij xj = aij xj = y i , H(U, 0) = F (U), H(U, 1) = L(U).
j=1 j=1
We may choose a convex homotopy
By (2) and the operation of fuzzy numbers, Friedman et
al. [12] replace the original fuzzy linear systems (1) by an H(U, p) = (1 − p)F (U) + pL(U) = 0, (4)
2n × 2n crisp function linear system
   and continuously trace an implicitly defined curve from a
B C X Y starting point H(U, 0) to a solution H(U, 1). The embedding
SX = Y or = , (3)
C B −X −Y parameter p monotonically increases from zero to one as
the trivial problem F (U) = 0 is continuously deformed to
where S = (skl ), 1 ≤ k, l ≤ 2n, skl are determined as follows
the original problem L(U) = 0. The embedding parameter
aij ≥ 0 ⇒ sij = aij , si+m, j+n = aij , p ∈ [0, 1] can be considered as an expanding parameter [14]
aij < 0 ⇒ si, j+n = −aij , si+m, j = aij ,
U = U0 + pU1 + p2 U2 + · · · , (5)
and any skl which is not determined by the above items is
when p → 1, (4) corresponds to L(U) = 0 and (5) becomes
zero and ⎡ ⎤ the approximate solution of (3), i.e.,
x1
⎢ .. ⎥ ∞

⎢ . ⎥
 ⎢ ⎥ X = lim (U0 + pU1 + p2 U2 + · · ·) = Uk .
X ⎢ xn ⎥ p→1
X= =⎢⎢ −x1
⎥,
⎥
k=0
−X ⎢ ⎥
⎢ . ⎥ Substituting (5) into (4) and equating the terms with identical
⎣ .. ⎦ powers of p, we have
−xn
p0 : QU0 − Y = 0,
⎡ ⎤ pk : QUk + (S − Q)Uk−1 = 0, k = 1, 2, · · · .
y1
⎢ .. ⎥ This implies that
⎢ . ⎥
 ⎢ ⎥
Y ⎢ y ⎥ U0 = Q−1 Y,
Y= =⎢
⎢ −y
n ⎥,
⎥
−Y ⎢ 1 ⎥ Uk = (I − Q−1 S)Uk−1 , k = 1, 2, · · · ,
⎢ . ⎥
⎣ .. ⎦ where I is an indent matrix with order 2n. Moreover, we can
−y n rewrite Uk in terms of the vector Y as

B contains the positive entries of A, C the absolute of the Uk = (I − Q−1 S)k Q−1 Y, k = 1, 2, · · · .
negative entries of A and A = B − C. Then solving the fuzzy
Hence, the solution of (3) can be of the form
linear system (1) is equal to solving crisp linear system (3).
The crisp linear system (3) can be uniquely solved for X if and X = U0 + U1 + U2 + · · ·

International Scholarly and Scientific Research & Innovation 5(3) 2011 348 [Link]/1307-6892/428
 World Academy of Science, Engineering and Technology
International Journal of Mathematical and Computational Sciences
Vol:5, No:3, 2011

or The extended linear system is the form of (3), where

⎛ ⎞
X = −1
[Q + (I − Q S)Q + (I − Q −1 −1 −1 2
S) Q −1
+ · · ·]Y 2 3 0 0 0 1
∞ ⎜ 3 0 2 0 1 0 ⎟
 ⎜ ⎟
= (I − Q−1 S)k Q−1 Y. (6) ⎜ 1 2 3 0 0 0 ⎟
S=⎜ ⎜ 0 0 1 2 3 0 ⎟,
⎟
k=0 ⎜ ⎟
⎝ 0 1 0 3 0 2 ⎠
In practice, all terms of series (6) cannot be determined and
so we use an approximation of the solution by the following 0 0 0 1 2 3
⎛ ⎞
truncated series: −1 + 4r

m−1 ⎜ 4 + 2r ⎟
⎜ ⎟
X= (I − Q−1 S)k Q−1 Y. ⎜ 4 + 5r ⎟
Y=⎜ ⎜ −6 + 3r ⎟
⎟
k=0
⎜ ⎟
⎝ −12 + 6r ⎠
The following theorem gives the convergent result of the
above series. −13 + 4r
Theorem 2. The sequence and ⎛ ⎞
International Science Index, Mathematical and Computational Sciences Vol:5, No:3, 2011 [Link]/Publication/428

x1 (r)

m−1
⎜ x2 (r) ⎟
U=[ (I − Q−1 S)k Q−1 Y ⎜ ⎟
⎜ x3 (r) ⎟
X=⎜ ⎟.
k=0
⎜ −x1 (r) ⎟
⎜ ⎟
is convergent if ⎝ −x2 (r) ⎠
−x3 (r)
(I − Q−1 S)k  < 1,
By HPM and using nine iterations, we obtain the approxi-
where  ·  denotes any norm of a matrix. mation to the solution of this extended system as
To find the solution of linear system (3), we should choose ⎛ ⎞
a nonsingular matrix Q. From Theorem 2, the matrix Q can 1.00000 + 0.00000r
⎜ 0.00000 + 1.00000r ⎟
be selected as ⎜ ⎟
 ⎜ 1.00000 + 1.00000r ⎟
B−C 0 X=⎜ ⎜ ⎟.
⎟
Q=
0 B−C ⎜ −2.00000 + 1.0000r ⎟
⎝ −1.00000 + 0.00000r ⎠
or  −3.00000 + 1.00000r
B+C 0
Q= Therefore, the approximation fuzzy solution of this example
0 B+C
is ⎧
or other block forms, see for example [17]. ⎨ x1 = (x1 (r), x1 (r)) = (1, 2 − r),
If the matrix Q is selected as x2 = (x2 (r), x2 (r)) = (r, 1),
⎩
 x3 = (x3 (r), x3 (r)) = (1 + r, 3 − r).
B−C 0
Q= , Example 2. Consider the 3 × 3 fuzzy system
0 B−C
⎧
then we have ⎨ −2x2 + 5x3 = (−3, − 2 − r),
 x1 + 2x2 = (r, 2 − r),
I − (B − C)−1 B −(B − C)−1 C ⎩
I−Q −1
S= , 3x1 − x3 = (1 + r, 3 − r).
−(B − C)−1 C I − (B − C)−1 B
The exact solution is
where I is an indent matrix with order n. ⎧
⎪
⎪ x1 = (x1 (r), x1 (r))
⎪
⎪
⎪
⎪ = (−0.18750 + 0.31250r, 0.87500 − 0.37500r),
IV. N UMERICAL EXAMPLES ⎨
x2 = (x2 (r), x2 (r))
For a given parameter r ∈ [0, 1], we consider the following ⎪
⎪ = (−0.09375 + 0.34375r, 0.56250 − 0.31250r),
⎪
⎪
two examples. ⎪
⎪ x 3 = (x3 (r), x3 (r))
⎩
Example 1. Consider the 3 × 3 fuzzy system = (−0.37500 − 0.12500r, 0.43750 − 0.06250r).
⎧
⎨ 2x1 + 3x2 − x3 = (−1 + 4r, 6 − 3r), In fact x3 is not a fuzzy number, therefore the fuzzy solution
3x1 − x2 + 2x3 = (4 + 2r, 12 − 6r), is
⎩ ⎧
x1 + 2x2 + 3x3 = (4 + 5r, 13 − 4r). ⎪
⎪ u1 = (u1 (r), u1 (r))
⎪
⎪
⎪
⎪ = (−0.18750 + 0.31250r, 0.87500 − 0.37500r),
The exact solution is ⎨
⎧ u2 = (u2 (r), u2 (r))
⎨ x1 = (x1 (r), x1 (r)) = (1, 2 − r), ⎪
⎪ = (−0.09375 + 0.34375r, 0.56250 − 0.31250r),
x2 = (x2 (r), x2 (r)) = (r, 1), ⎪
⎪
⎪
⎪ u3 = (u3 (r), u3 (r))
⎩ ⎩
x3 = (x3 (r), x3 (r)) = (1 + r, 3 − r), = (−0.50000, − 0.37500 − 0.12500r),
which is a strong fuzzy solution. which is a weak fuzzy solution.

International Scholarly and Scientific Research & Innovation 5(3) 2011 349 [Link]/1307-6892/428
 World Academy of Science, Engineering and Technology
International Journal of Mathematical and Computational Sciences
Vol:5, No:3, 2011

The extended linear system is the form of (3), where R EFERENCES

⎛ ⎞ [1] S. Abbasbandy, A. Jafarian, and R. Ezzati, ”Conjugate gradient method
0 0 5 0 2 0 for fuzzy symmetric positive-definite system of linear equations”, Appl.
⎜ 1 2 0 0 0 0 ⎟ Math. Comput., 171 (2005) 1184-1191.
⎜ ⎟
⎜ 3 0 0 0 0 1 ⎟ [2] S. Abbasbandy, R. Ezzati, and A. Jafarian, ”LU decomposition method
S=⎜ ⎜ 0 2 0
⎟, for solving fuzzy system of linear equations”, Appl. Math. Comput., 172
⎜ 0 0 5 ⎟ ⎟ (2006) 633-643.
⎝ 0 0 0 1 2 0 ⎠ [3] S. Abbasbandy and A. Jafarian, ”Steepest descent method for system of
fuzzy linear equations”, Appl. Math. Comput., 175 (2006) 823-833.
0 0 1 3 0 0 [4] S. Abbasbandy, ”Application of He’s homotopy perturbation method for
Laplace transform”, Chaos Solitons Fractals 30 (2006) 1206-1212.
⎛ ⎞ [5] S. Abbasbandy, ”Application of He’s homotopy perturbation method to
−3 functional integral equations”, Chaos Solitons Fractals 31 (2007) 1243-
⎜ r ⎟
⎜ ⎟ 1247.
⎜ 1+r ⎟ [6] T. Allahviranloo, ”Numerical methods for fuzzy system of linear equa-
Y=⎜
⎜
⎟
⎟ tions”, Appl. Math. Comput., 155 (2004) 493-502.
⎜ 2+r ⎟ [7] T. Allahviranloo, ”Successive over relaxation iterative method for fuzzy
⎝ −2 + r ⎠ system of linear equations”, Appl. Math. Comput., 162 (2005) 189-196.
−3 + r [8] T. Allahviranloo, M. Ghanbari, ”Solving fuzzy linear systems by ho-
motopy perturbation method”, International Journal of Computational
International Science Index, Mathematical and Computational Sciences Vol:5, No:3, 2011 [Link]/Publication/428

Cognition, 8 (2010) 27-30.

and ⎛ ⎞ [9] A. Belendez, A. Hernandez, T. Belendez, ”Application of He’s homotopy
x1 (r) perturbation method to the Duffing-harmonic oscillator”, Int. J. Nonlinear
⎜ x2 (r) ⎟ Sci. Numer. Simul. 8 (2007) 79-88.
⎜ ⎟
⎜ x3 (r) ⎟ [10] J. Biazar, M. Eslami, H. Ghazvini, ”Homotopy perturbation method for
X=⎜
⎜
⎟.
⎟
systems of partial differential equations”, Int. J. Nonlinear Sci. Numer.
⎜ −x1 (r) ⎟ Simul. 8 (2007) 413-418.
⎝ −x2 (r) ⎠ [11] J.J. Buckley, ”Solving fuzzy equations in economics and finance”, Fuzzy
Sets and Systems, 48 (1992) 289-296.
−x3 (r) [12] M. Friedman, M. Ma, A. Kandel, ”Fuzzy linear systems”, Fuzzy Sets
Syst., 96 (1998) 201-209.
By HPM and using six iterations, we obtain the approxima- [13] R. Goetschel, W. Voxman, ”Elementary calculus”, Fuzzy Sets and
tion to the solution of this extended system as Systems, 18 (1986) 31-43.
[14] J.H. He, ”Homotopy perturbation technique”, Comput. Methods. Appl.
⎛ ⎞ Mech. Engrg. 178 (1999) 257-262.
−0.18750 + 0.31250r [15] B. Keramati, ”An approach to the solution of linear system of equations
⎜ −0.09375 + 0.34375r ⎟ by He’s homotopy perturbation method”, Chaos Solitons Fractals, 41
⎜ ⎟
⎜ −0.37500 − 0.12500r ⎟ (2009) 152-156.
X=⎜ ⎜ ⎟. [16] H.K. Liu, ”On the solution of fully fuzzy linear systems”, International
⎟
⎜ −0.87500 + 0.37500r ⎟ Journal of Computational and Mathematical Sciences, 4 (2010) 29-33.
⎝ −0.56250 + 0.31250r ⎠ [17] S.-X. Miao, B. Zheng, K. Wang, ”Block SOR methods for fuzzy linear
−0.43750 + 0.06250r systems”, J. Appl. Math. Comput., 26 (2008) 201-218.
[18] S.S. Rao and L. Chen, ”Numerical solution of fuzzy linear equations
in engineering analysis”, Internat. J. Numer. Methods Engrg., 42 (1998)
Therefore, the approximation weak fuzzy solution of this 829-846.
example is [19] K. Wang, B. Zheng, ”Symmetric successive overrelaxation methods for
fuzzy linear systems”, Appl. Math. Comput., 175 (2006) 891-901.
⎧ [20] K. Wang, B. Zheng, ”Block iterative methods for fuzzy linear systems”,
⎪
⎪ x1 = (x1 (r), x1 (r))
⎪
⎪
J. Appl. Math. Comput., 25 (2007) 119-136.
⎪
⎪ = (−0.18750 + 0.31250r, 0.87500 − 0.37500r), [21] C.X. Wu, M. Ma, ”Embedding problem of fuzzy number space: Part I”,
⎨
x2 = (x2 (r), x2 (r)) Fuzzy Sets and Systems, 44 (1991) 33-38.
[22] E. Yusufoğlu, ”An improvement to homotopy perturbation method for
⎪
⎪ = (−0.09375 + 0.34375r, 0.56250 − 0.31250r),
⎪
⎪
solving system of linear equations”, Computers and Mathematics with
⎪
⎪ x3 = (x3 (r), x3 (r)) Applications, 58 (2009) 2231-2235.
⎩
= (−0.37500 − 0.12500r, 0.43750 − 0.06250r). [23] E. Yusufoğlu, ”Improved homotopy perturbation method for solving
Fredholm type integro-differential equations”, Chaos Solitons Fractals,
41 (2009) 28-37.
[24] L.A. Zadeh, ”Fuzzy sets”, Information and Control, 8 (1965) 338-353.
V. C ONCLUSION

In this paper, we consider the block HPM for finding an

approximation solution of fuzzy linear systems. The block
HMP method is efficient and practical because the procedure
only require the nonsingularity of the coefficient matrix of
n×n fuzzy linear system while the point HMP method require
the diagonal entries of the coefficient matrix are nonzero (see
[8]). The numerical results show that the block HPM converges
to the exact solution of fuzzy linear systems.

ACKNOWLEDGEMENTS

This work was supported by NWNU-KJCXGC-3-47.

International Scholarly and Scientific Research & Innovation 5(3) 2011 350 [Link]/1307-6892/428