Chapter 7

Defuzzification

The practical problems occurring in various engineering disciplines often include

imprecision as mentioned in Introduction section of Frontmatter of this book. Due
to the involvement of such uncertainties, the parameters considered in real-world
problems may be expressed as fuzzy sets instead of crisp sets. The methodology of
transformation of a crisp set to appropriate fuzzy set is referred as fuzzification. On the
other hand, defuzzification may be considered as an inverse approach of fuzzification
that converts fuzzy set to appropriate crisp set. Fuzzification and defuzzification are
often encountered in handling fuzzy logic controllers. Detailed discussion on fuzzy
controllers based on fuzzification and/or defuzzification modules may be found in
various literature (Klir and Yuan 2008; Zimmermann 1996; Hellendoorn and Thomas
1993).
In this regard, we define defuzzification as the transformation T,

T : X × [0, 1] → X,
  
where A = (x, µ A(x))µ A(x) ∈ [0, 1], x ∈ X is the fuzzy set and the transformed
crisp set is A. In case of defuzzification of a fuzzy number to corresponding crisp
value, the transformation is considered as

T :  × [0, 1] → 
ã → T (ã) = x ∗

While handling the practical problems with impreciseness, we often encounter

situations where the control command may be considered as a crisp value instead of
the concerned fuzzy number ã. In such cases, defuzzification helps in computation
of a crisp value x ∗ associated with the fuzzy number ã as given in Fig. 7.1.

© Springer Nature Singapore Pte Ltd. 2019 117

S. Chakraverty et al., Concepts of Soft Computing,
https://doi.org/10.1007/978-981-13-7430-2_7
118 7 Defuzzification

Fig. 7.1 Defuzzification of

fuzzy set ã

But there exists no specific approach for the choice of defuzzified crisp value;
instead, the defuzzification method needs to be selected with respect to the system
of the practical problem. Various defuzzification methods exist and may be found in
Zimmermann (2001), Lee (2006) and Sivanandam and Deepa (2012). In this chapter,
we have mainly discussed four types of defuzzification, viz., max-membership, cen-
troid methods, weighted-average, and mean–max methods. Further, MATLAB pro-
grams with respect to the defuzzification methods have also been included in the
present context.
In this regard, the defuzzification methods have been discussed in the next section.

7.1 Defuzzification Methods

The max-membership, centroid, weighted-average, and mean–max methods of

defuzzification have been incorporated in Sects. 7.1.2–7.1.5. However, before we
start with the different defuzzification procedures, we briefly discuss the concept of
aggregation (Lee 2006; Sivanandam and Deepa 2012) of fuzzy sets (required while
performing defuzzification for multiple fuzzy sets).

7.1.1 Aggregation

Aggregation helps in computation of a single fuzzy set with respect to multiple

fuzzy sets. Sivanandam and Deepa (2012) discussed two aggregation rules, viz.,
conjunctive and disjunctive based on fuzzy intersection and union (given in Chap.
2), respectively, that may be useful in handling fuzzy reasoning. In this chapter, we
have focused on the aggregation in terms of fuzzy union as shown in Fig. 7.2.
A MATLAB code has been included in Program 7.1 to perform the aggregation
of multiple fuzzy sets using a fuzzy union.
7.1 Defuzzification Methods 119

Fig. 7.2 Aggregated fuzzy


set C

Program 7.1: Write a MATLAB function “aggregate” to perform aggregation of

fuzzy sets.

%Aggregation of fuzzy sets

function [x,mf] = aggregate(n)
%n is the number of fuzzy sets
% Enter values of the fuzzy numbers TFN,TrFN and GFN
x =[-10:0.1:10]'; %x values
for i=1:n
fprintf('Enter the fuzzy set %d\n',i);
j = input('Select 1 for triangular or 2 for trapezoidal or 3 for Gaussian fuzzy
sets\n');
if(j==1)
tfn = input('Enter a TFN (a,b,c)\n');
yt(i,:) = trimf(x,[tfn(1) tfn(2) tfn(3)]);
a(i) = input('Enter membership scaling factor in interval [0,1] for triangular
type fuzzy set\n');
yt(i,:) = a(i)*yt(i,:);
elseif(j==2)
trfn = input('Enter a TrFN (a,b,c,d)\n');
yt(i,:) = trapmf(x,[trfn(1) trfn(2) trfn(3) trfn(4)]);
b(i) = input('Enter membership scaling factor in interval [0,1] for trapezoial
type fuzzy set\n');
yt(i,:) = b(i)*yt(i,:);
elseif(j==3)
gfn = input('Enter a GFN (a,b,c)\n');
yt(i,:) = gauss2mf(x,[gfn(2) gfn(1) gfn(3) gfn(1)]);
c(i) = input('Enter membership scaling factor in interval [0,1] for Gaussian type
fuzzy set\n');
yt(i,:) = c(i)*yt(i,:);
end
end
120 7 Defuzzification

%Aggregate the fuzzy sets

for i=1:n
yt(1,:) = max(yt(1,:),yt(i,:));
mf = yt(1,:);
end
plot(x,mf,'-r','LineWidth',2)
xlabel('x')
ylabel('\mu(x)')
ylim([-0.05 1.05])
end

Using Program 7.1, the aggregated fuzzy set with respect to four fuzzy sets is
obtained and the corresponding MATLAB plot is depicted in Fig. 7.3.

Input:
>> aggregate(4)
Enter the fuzzy set 1
Select 1 for triangular or 2 for trapezoidal or 3 for Gaussian fuzzy sets
1
Enter a TFN (a,b,c)
[-7, -5, 1]
Enter membership scaling factor between 0 to 1 for triangular type fuzzy set
0.5
Enter the fuzzy set 2
Select 1 for triangular or 2 for trapezoidal or 3 for Gaussian fuzzy sets
2
Enter a TrFN (a,b,c,d)
[-3, 1, 5, 6]
Enter membership scaling factor between 0 to 1 for trapezoial type fuzzy set
0.9
Enter the fuzzy set 3
Select 1 for triangular or 2 for trapezoidal or 3 for Gaussian fuzzy sets
2
Enter a TrFN (a,b,c,d)
[5, 6, 8, 9]
Enter membership scaling factor between 0 to 1 for trapezoial type fuzzy set
1
Enter the fuzzy set 4
Select 1 for triangular or 2 for trapezoidal or 3 for Gaussian fuzzy sets
3
Enter a GFN (a,b,c)
[-2, 3, 4]
Enter membership scaling factor between 0 to 1 for Gaussian type fuzzy set
0.7

Output:
7.1 Defuzzification Methods 121

Fig. 7.3 Aggregated fuzzy

output

Using the aggregation rule as illustrated in Program 7.1, the max-membership

defuzzification of fuzzy sets is discussed in the next section.

7.1.2 Max-Membership Method

The max-membership method gives the defuzzified point as the crisp value x ∗ ∈ X
having maximum membership value ∀x ∈ X . In this regard, the membership or
characteristic value corresponding to defuzzified point x ∗ ∈ X satisfies

µC(x ∗ ) ≥ µC(x) ∀x ∈ X (7.1)

or µC(x ∗ ) = sup (µC(x)). (7.2)

x∈X

Sometimes, this method is also referred as height method as the defuzzification is

performed based on the maximum height as given in Eq. (7.2). An example problem
has been considered in Example 7.1 to illustrate the max-membership method.
Example 7.1 Compute defuzzification of GFN ã = g f n(2, 1, 1) using max-
membership method.
Solution: The GFN ã = g f n(2, 1, 1) as given in Fig. 7.4 has the membership
function
 (x−2)2
e− 2 ,x <2
µã (x) = 2 .
− (x−2)
e 2 ,x ≥2
122 7 Defuzzification

Fig. 7.4 Gaussian fuzzy

number

Using Eq. (7.2), the GFN attends its maximum membership value 1 at x = 2.
Accordingly, the defuzzified crisp value x ∗ is 2.
In order to make max-membership method effective, we have written a MATLAB
code for computation of defuzzified crisp value in Program 7.2.

Program 7.2: Write a “maxmem” MATLAB function to perform defuzzification

using max-membership method.

%Max-membership defuzzification
function maxmem(n)
%n is the number of fuzzy sets
%mf is the membership function of aggregated fuzzy set
[x,mf] = aggregate(n);
maxm = max(mf);
p = length(mf);
z = 0;
for i = 1:p
if(mf(i) == maxm)
z = x(i);
end
end
disp('The defuzzified crisp value is:');
disp(z);
end

Using Program 7.2, defuzzification of fuzzy set using max-membership function

for three fuzzy set is obtained.
7.1 Defuzzification Methods 123

Input:
>> maxmem(2)
Enter the fuzzy set 1
Select 1 for triangular or 2 for trapezoidal or 3 for Gaussian fuzzy sets
3
Enter a GFN (a,b,c)
[-5, 2, 3]
Enter membership scaling factor in interval [0,1] for Gaussian type fuzzy set
0.8
Enter the fuzzy set 2
Select 1 for triangular or 2 for trapezoidal or 3 for Gaussian fuzzy sets
1
Enter a TFN (a,b,c)
[-2, 2, 7]
Enter membership scaling factor in interval [0,1] for triangular type fuzzy set
1

Output:
The defuzzified crisp value is:
2

The corresponding plot depicting defuzzification using max-membership method

has been given in Fig. 7.5.

Fig. 7.5 Defuzzification

using max-membership
method
124 7 Defuzzification

7.1.3 Centroid Method

Here, we find x ∗ ∈ X using the relation


µc̃ (x) · x d x
x∗ =  . (7.3)
µc̃ (x) d x

It may be noted that the integration is the usual algebraic integration. Accordingly,
based on Eq. (7.3), MATLAB code for performing defuzzification using centroid
method has been given in Program 7.3.

Program 7.3: Write a MATLAB function “fcentroid” to perform defuzzification

using centroid method.

%Centroid defuzzification
function fcentroid(n)
%n is the number of fuzzy sets
%mf is the membership function of aggregated fuzzy set
z = 0;
[x,mf] = aggregate(n);
nx = length(x);
z = (sum(mf(1:nx).*x(1:nx))/(sum(mf(1:nx))));
disp('The defuzzified crisp value is:');
disp(z);
end

Using Program 7.3, the defuzzification of fuzzy set is obtained at x ∗ = −2.9882

using centroid method.

Input:
>> fcentroid(3)
Enter the fuzzy set 1
Select 1 for triangular or 2 for trapezoidal or 3 for Gaussian fuzzy sets
1
Enter a TFN (a,b,c)
[4, 5, 6]
Enter membership scaling factor in interval [0,1] for triangular type fuzzy set
0.2
Enter the fuzzy set 2
Select 1 for triangular or 2 for trapezoidal or 3 for Gaussian fuzzy sets
2
7.1 Defuzzification Methods 125

Enter a TrFN (a,b,c,d)

[-9, -6, -2, 2]
Enter membership scaling factor in interval [0,1] for trapezoial type fuzzy set
0.9
Enter the fuzzy set 3
Select 1 for triangular or 2 for trapezoidal or 3 for Gaussian fuzzy sets
3
Enter a GFN (a,b,c)
[-1,2,3]
Enter membership scaling factor in interval [0,1] for Gaussian type fuzzy set
0.4

Output:
The defuzzified crisp value is:
-2.9882

Accordingly, the defuzzification using centroid method with respect to Program

7.3 has been depicted in Fig. 7.6.
There exist already various predefined MATLAB functions to perform defuzzifi-
cation. The defuzzification centroid technique in Program 7.3 may also be obtained
by MATLAB command “z = defuzz(x,mf,‘centroid’)” to obtain the defuzzification
point, where “mf” is the corresponding membership values with respect to “x”.
It is worth mentioning that defuzzification using above command “z =
defuzz(x,mf,‘centroid’)” yields equivalent crisp value as obtained using Program
7.3. There exist various other methods, viz., weighted-average and mean–max meth-
ods discussed in Sects. 7.1.4 and 7.1.5.

Fig. 7.6 Defuzzification

using centroid method
126 7 Defuzzification

7.1.4 Weighted-Average Method

Here, we have x ∗ ∈ X as
n
i=1 µC ( x̄ i ) · x̄ i
x∗ =  n , (7.5)
i=1 µC ( x̄ i )

where x̄i is the centroid of maximum for each ith membership function of the aggre-
gated fuzzy set.
n
Further, i=1 represents the algebraic sum with respect to membership functions
of n fuzzy sets. The method seems appealing but has a limitation that it may be
applicable only in case of fuzzy sets having symmetric membership functions.

7.1.5 Mean–Max Method

Mean–max defuzzification method is also sometimes termed as middle of maxima

as given in Fig. 7.7.
The corresponding defuzzified crisp value is obtained using

a+b
x∗ = . (7.5)
2
MATLAB command to obtain the defuzzification point using mean–max method
is “z = defuzz(x,mf,‘mom’)”, where “mf” is the aggregated membership function
corresponding to “x”.
There exist various other defuzzification methods. Readers may refer defuzzi-
fication commands in MATLAB for the other defuzzification techniques based on
bisector, middle, smallest, and largest of maximum methods.
Exercise
1. Define MATLAB function “wedavg” to compute defuzzification of fuzzy set
using weighted-average method.

Fig. 7.7 Defuzzification

using mean–max method
7.1 Defuzzification Methods 127

2. Write a MATLAB function “meanmax” to implement the weighted-average

defuzzification technique for a given fuzzy set.

References

H. Hellendoorn, C. Thomas, Defuzzification in fuzzy controllers. J. Intell. Fuzzy Syst. 1(2), 109–123
(1993)
G.J. Klir, B. Yuan, Fuzzy Sets and Fuzzy Logic, Theory and Applications (Prentice Hall, London,
2008)
K.H. Lee, First Course on Fuzzy Theory and Applications (Springer, New York, 2006)
S.N. Sivanandam, S.N. Deepa, Principles of Soft Computing (Wiley India Pvt. Ltd., New Delhi,
2012)
H.J. Zimmermann, Fuzzy control, in Fuzzy Set Theory—And Its Applications (Springer, Dordrecht,
1996), pp. 203–240
H.J. Zimmermann, Fuzzy Set Theory and Its Applications (Springer, New York, 2001)
Chapter 8
Interval System of Linear Equations

Differential equations serve as the backbone of various science and engineering

problems viz. structure analysis, control theory, image processing, circuit analysis,
robotics, etc. and there exist variety of methods for solving such differential equa-
tions. Generally, under static conditions, the governing differential equations get
converted to system of equations. But in actual practice, as a result of errors in mea-
surements, observations, calculations, or due to maintenance induced errors, etc., the
differential equations and the corresponding system of equations may have uncer-
tain parameters that may be handled using probabilistic approach, interval analysis,
and fuzzy set theory. Unfortunately, the probabilistic approach may not deliver reli-
able results at the required precision without sufficient experimental data. As such,
interval analysis and fuzzy set theory serve as powerful tools for handling uncer-
tain parameters or values having infimum and supremum. In Chap. 1, the notion of
interval matrix has been introduced and as such, this chapter deals with application
of interval matrices for solving interval system of linear equations (ISLEs). Further,
the chapter includes certain solved numerical examples along with MATLAB codes
and also consists of few unsolved problems for self-evaluation.

8.1 System of Linear Equations

System of equations is of great importance in solving practical problems of various

engineering fields to obtain the values of unknown parameter vector x. For instance, in
structural analysis, system of equations helps in computation of nodal displacements
for defined structures. Also, in circuit analysis problem V = IR, system of equations
helps in computation of current I based on known voltage V and resistance R. When
the dependency between the parameters in any physical system is linear, then such
system of equations is referred as system of linear equations. Generally, a system of
linear equation is formulated as

© Springer Nature Singapore Pte Ltd. 2019 129

S. Chakraverty et al., Concepts of Soft Computing,
https://doi.org/10.1007/978-981-13-7430-2_8
130 8 Interval System of Linear Equations

Ax = b, (8.1)

where A is the linear relation coefficient matrix having dimension m × n, b is the right-
hand side vector having m elements, and x is the solution vector having n variables.
But due to insufficiency and incompleteness of data, the parameters in such system
of equations may be uncertain which leads to uncertain system of linear equations.
According to Chaps. 1 and 3, the uncertainty may be handled using interval analysis
and fuzzy set approaches.
Note: Any fuzzy system of linear equations may be converted into appropriate ISLEs
since a fuzzy number is approximately represented by a set of closed intervals using
α-cut.

8.2 Interval System of Equations

System of equations having interval uncertainties is referred as interval system of

equations. Now, the ISLE is defined as

[A][x] = [b], (8.2)

where [A] is the linear interval coefficient matrix having dimension m × n, [b] is the
interval vector of size m, and [x] is the corresponding solution vector consisting of n
variables. Further, Eq. (8.2) may be written as


n
[ai j ] = [bi ].
j=1

Rohn (1989) discussed the computation of solution ISLE, in terms of x i =

min{xi |xi ∈ [x]} and x̄i = max{xi |xi ∈ [x]} (Kreinovich et al. 2013) that is generally
non-convex in nature. In case of narrow intervals, Kahl (1996) discussed that solving
such an ISLE is NP-hard. There also exist few algebraic or formal (Shary 2002)
and numerical Neumaier (1990) solution approaches (Hansen 1969; Chakraverty
et al. 2017). As such, three cases based on the solution procedure discussed in (Das
and Chakraverty 2012) with respect to vertex approach have been considered in
Sect. 8.2.1.

8.2.1 Methods for Solving ISLEs

Let us consider n × n ISLE as given in Eq. (8.2).

8.2 Interval System of Equations 131
⎛ ⎞⎧ ⎫ ⎧ ⎫
[a11 ] [a12 ] . . . [a1n ] ⎪⎪ [x1 ] ⎪
⎪ ⎪
⎪ [b1 ] ⎪
⎪
⎜ [a21 ] [a22 ] . . . [a2n ] ⎟⎪
⎨ ⎬ ⎪
⎪ ⎨ [b2 ] ⎪
⎬
⎜ ⎟ [x2 ]
⎜ . . . . ⎟ . = . . (8.3)
⎝ .. .. . . .. ⎠⎪ ⎪ .. ⎪⎪ ⎪
⎪ .. ⎪⎪
⎪
⎩ ⎭ ⎪
⎪ ⎩ ⎪
⎭
[an1 ] [an2 ] . . . [ann ] [xn ] [bn ]

Then, Eq. (8.3) is further written as

[a11 ][x1 ] + [a12 ][x2 ] + . . . [a1n ][xn ] = [b1 ]

[a21 ][x1 ] + [a22 ][x2 ] + . . . [a2n ][xn ] = [b2 ]
.. . (8.4)
.
[an1 ][x1 ] + [an2 ][x2 ] + . . . [ann ][xn ] = [bn ]

In terms of lower and upper bounds, Eq. (8.4) may be written as

[a 11 , ā 11 ][x 1 , x̄ 1 ] + [a 12 , ā 12 ][x 2 , x̄ 2 ] + . . . + [a 1n , ā 1n ][x n , x̄ n ] = [b1 , b̄1 ]

[a 21 , ā 21 ][x 1 , x̄ 1 ] + [a 22 , ā 22 ][x 2 , x̄ 2 ] + . . . + [a 2n , ā 2n ][x n , x̄ n ] = [b2 , b̄2 ]
.. .
.
[a n1 , ā n1 ][x 1 , x̄ 1 ] + [a n2 , ā n2 ][x 2 , x̄ 2 ] + . . . + [a nn , ā nn ][x n , x̄ n ] = [bn , b̄n ]
(8.5)

8.2.1.1 Method I

If the elements of coefficient interval matrix [A] in Eq. (8.5) are positive, that is,
∀a i j , āi j ∈ + for i, j = 1, 2, . . . , n, and then Eq. (8.5) may be further written as

a 11 x 1 + a 12 x 2 + . . . + a 1n x n = b1
a 21 x 1 + a 22 x 2 + . . . + a 2n x n = b2
.. , (8.6a)
.
a n1 x 1 + a n2 x 11 + . . . + a nn x 11 = bn
ā11 x̄ 1 + ā12 x̄ 2 + . . . + ā1n x̄ n = b̄1
ā21 x̄ 1 + ā22 x̄ 2 + . . . + ā2n x̄ n = b̄2
.. . (8.6b)
.
ān1 x̄ 1 + ān2 x̄ 2 + . . . + ānn x̄ n = b̄n

In terms of interval matrix notation, Eqs. (8.6a, 8.6b) is given as

132 8 Interval System of Linear Equations
⎛ ⎞⎧ ⎫ ⎧ ⎫
a 11 a 12 . . . a 1n 0 0 ... 0 ⎪ ⎪ x1 ⎪ ⎪ ⎪
⎪ b1 ⎪⎪
⎜a ⎪ ⎪ ⎪ ⎪
⎜ 21 a 22 . . . a 2n 0 0 ... 0 ⎟ ⎟⎪
⎪
⎪ x ⎪
⎪
⎪
⎪
⎪
⎪ b ⎪
⎪
⎪
⎟⎪ ⎪ ⎪ ⎪
2 2
⎜ . .. ⎪
⎪ . ⎪
⎪ ⎪
⎪ . ⎪
⎪
⎜ .. . ⎟⎪ .
⎪ . ⎪ ⎪
⎪ ⎪ ⎪ .
. ⎪
⎪
⎜ ⎟⎪ ⎪ ⎪
⎜ ⎪
⎟⎨ ⎬ ⎨ ⎪ ⎪ ⎪ ⎬
⎜ a n1 a n2 . . . a nn 0 0 . . . 0 ⎟ xn bn
⎜ ⎟ = . (8.7)
⎜ 0 0 . . . 0 ā11 ā12 . . . ā1n ⎟⎪ x̄1 ⎪ ⎪ b̄1 ⎪
⎜ ⎟⎪⎪
⎪ x̄ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ b̄ ⎪
⎪
⎪
⎪
⎜ 0 0 . . . 0 ā21 ā22 . . . ā2n ⎟ ⎪
⎜ ⎟⎪⎪
⎪ . ⎪
2⎪
⎪ ⎪
⎪ 2⎪
⎪ . ⎪ ⎪
⎜ .. .. ⎟⎪ ⎪ ⎪
⎪ . ⎪
⎝ . . ⎠⎪⎪
⎪ .. ⎪⎪ ⎪
⎪ ⎪ . ⎪
⎪
⎪
⎪
⎩ ⎭ ⎩ ⎪⎪ ⎪ ⎭
0 0 . . . 0 ān1 ān2 . . . ānn x̄n b̄n

Accordingly, it may be noted that the ISLE in Eq. (8.3) gets converted to crisp
system of linear equations having double dimension. For instance, in case of a 2 ×
2 ISLE, Eq. (8.7) gets reduced to 4 × 4 crisp system of linear equation as
⎛ ⎞⎧ ⎫ ⎧ ⎫
a 11 a 12 0 0 ⎪ ⎪ x ⎪ ⎪ b ⎪
⎜a ⎨ 1⎪ ⎬ ⎪⎨ 1⎪ ⎬
⎜ 21 a 22 0 0 ⎟⎟ x b
⎝ 0
2 = 2 . (8.8)
0 ā11 ā12 ⎠⎪
⎪ x̄ ⎪ ⎪ b̄ ⎪
⎩ 1⎪ ⎭ ⎪⎩ 1⎪ ⎭
0 0 ā21 ā22 x̄2 b̄2

Now, there exists a unique solution to Eq. (8.8) as given below:

a 22 b1 − a 12 b2
x1 = , (8.9a)
a 11 a 22 − a 12 a 21
ā22 b̄1 − ā12 b̄2
x̄1 = , (8.9b)
ā11 ā22 − ā12 ā21
a b − a 21 b1
x 2 = 11 2 , (8.9c)
a 11 a 22 − a 12 a 21
ā11 b̄2 − a 21 b̄1
x2 = . (8.9d)
ā11 ā22 − ā12 ā21

Example 8.1 Let us consider a 2 ×2 ISLE [A][x] = [b] as given in (Chakraverty

et al. 2017),
    
[4, 6] [5, 8] [x 1 , x̄1 ] [40, 67]
= , (8.10)
[6, 7] [4, 5] [x 2 , x̄2 ] [43, 55]
   
[4, 6] [5, 8] [40, 67]
where [A] = and [b] = .
[6, 7] [4, 5] [43, 55]

Solution: Using Eqs. (8.8), (8.10) gets converted to crisp system of linear equations.
8.2 Interval System of Equations 133

⎛ ⎞⎧ ⎫ ⎧ ⎫
4 5 0 0 ⎪ ⎪ x ⎪ ⎪ 40 ⎪
⎜6 ⎨ 1⎪ ⎬ ⎪⎨ ⎪ ⎬
⎜ 4 0 0⎟⎟ x 43
⎝0
2 = .
0 6 8 ⎠⎪⎪ x̄ ⎪ ⎪ 67 ⎪
⎩ 1⎪ ⎭ ⎪⎩ ⎪ ⎭
0 0 7 5 x̄2 55

Accordingly, using Eqs. (8.9a, 8.9b, 8.9c, and 8.9d), the bounds of solution vector
are obtained as [x1 ] = [3.9286, 4.0385] and [x2 ] = [4.8571, 5.3462].

Program 8.1: MATLAB program for solving interval system of linear equations
using Eq. (8.8).

%Interval system of linear equations

%Enter the dimension of system of linear equations
n = input('Enter the dimension of coefficient matrix A');
%Enter the lower and upper bounds of coefficient matrix A
Al = input('Enter the lower bound of each element of interval aij: \n ');
Au = input('Enter the upper bound of each element of interval aij: \n ');
%Enter the lower and upper bounds of right-hand side vector
bl = input('Enter the lower bounds of interval vector b:\n ');
bu = input('Enter the upper bounds of interval vector b:\n ');
%Crisp system of linear equations
disp('The crisp coefficient matrix is');
O=zeros(n);
A=[Al O;O Au]
disp('The crisp right-hand side vector is');
b=[bl;bu]
x=inv(A)*b
Using Program 8.1, the interval system of linear equations given in Example 8.1
has been solved.

Input:
Enter the dimension of coefficient matrix A
2
Enter the lower bound of each element of interval aij:
[4 5; 6 4]
Enter the upper bound of each element of interval aij:
[6 8;7 5]
Enter the lower bounds of interval vector b:
[40;43]
Enter the upper bounds of interval vector b:
[67;55]
134 8 Interval System of Linear Equations

Output:
The crisp coefficient matrix is
A= 4 5 0 0
6 4 0 0
0 0 6 8
0 0 7 5
The crisp right-hand side vector is
b = 40
43
67
55

x= 3.9286
4.8571
4.0385
5.3462

8.2.1.2 Method II

Equation (8.8) may further be written in terms of lower and upper bounds, systems
of linear equations as given below:
    
a 11 a 12 x1 b1
,= (8.11a)
a 21 a 22 x2 b2
    
ā11 ā12 x̄1 b̄1
= . (8.11b)
ā21 ā22 x̄2 b̄2

On solving, Eqs. (8.11a, 8.11b) yields

a 22 b1 − a 12 b2
x1 = ,
a 11 a 22 − a 12 a 21
ā22 b̄1 − ā12 b̄2
x̄1 = ,
ā11 ā22 − ā12 ā21
a b − a 21 b1
x 2 = 11 2 ,
a 11 a 22 − a 12 a 21
ā11 b̄2 − ā21 b̄1
x̄2 = ,
ā11 ā22 − ā12 ā21

which is equivalent to Eqs. (8.9a, 8.9b, 8.9c, and 8.9d).

8.2 Interval System of Equations 135

Program 8.2: A MATLAB program for solving interval system of linear equations
using Eqs. (8.11a, 8.11b).

%Interval system of linear equations

%Enter the dimension of system of linear equations
n = input('Enter the dimension of coefficient matrix A\n');
%Enter the lower and upper bounds of coefficient matrix A
Al = input('Enter the lower bound of each element of interval aij: \n ');
Au = input('Enter the upper bound of each element of interval aij: \n ');
%Enter the lower and upper bounds of right-hand side vector
bl = input('Enter the lower bounds of interval vector b:\n ');
bu = input('Enter the upper bounds of interval vector b:\n ');
%Solution bounds
disp('The lower and upper bounds of solution vector x');
xl=inv(Al)*bl
xu=inv(Au)*bu
Accordingly, the output with respect to Example 8.1 has been obtained using
Program 8.2.

Input:
Enter the dimension of coefficient matrix A
2
Enter the lower bound of each element of interval aij:
[4 5; 6 4]
Enter the lower bound of each element of interval aij:
[6 8;7 5]
Enter the upper bounds of interval vector b:
[40;43]
Enter the upper bounds of interval vector b:
[67;55]

Output:
The lower and upper bounds of solution vector x
xl = 3.9286
4.8571
xu = 4.0385
5.3462

8.2.1.3 Method III

If the elements of coefficient interval matrix [A] are either positive or negative, that
is, ∀a i j , āi j ∈ + or a i j , āi j ∈ − for i, j = 1, 2, . . . , n.
136 8 Interval System of Linear Equations

Substituting [ci j , c̄i j ] = −[a i j , āi j ] if a i j , āi j ∈ − for i, j = 1, 2, . . . , n,

Eq. (8.5) is rewritten as

[a 11 , ā 11 ][x 1 , x̄ 1 ] − [c12 , c̄12 ][x 2 , x̄ 2 ] + . . . + [a 1n , ā 1n ][x n , x̄ n ] = [b1 , b̄1 ]

−[c21 , c̄21 ][x 1 , x̄ 1 ] + [c22 , c̄22 ][x 2 , x̄ 2 ] + . . . − [c2n , c̄2n ][x n , x̄ n ] = [b2 , b̄2 ]
.. .
.
[a n1 , ā n1 ][x 1 , x̄ 1 ] − [cn2 , c̄n2 ][x 2 , x̄ 2 ] + . . . + [a nn , ā nn ][x n , x̄ n ] = [bn , b̄n ]
(8.12)

Now, using matrix multiplication of intervals, [a i j , āi j ][x i , x̄i ] is rewritten as

[āi j x i , a i j x̄i ]. So, in terms of interval matrix notation, Eq. (8.12) is given by
⎛ ⎞⎧ ⎫ ⎧ ⎫
a 11 ā12 . . . a 1n 0 0 . . . 0 ⎪ ⎪ x ⎪ ⎪ b ⎪
⎜ ā a . . . ā ⎪ 1⎪
⎟⎪ ⎪ ⎪
⎪ ⎪ 1⎪
⎪ ⎪
⎪
⎜ 21 22 0 0 . . . 0 ⎪
⎟⎪ x ⎪
2⎪
⎪
⎪ b ⎪
2⎪
⎜ .
2n
⎟⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎜ .. .
.. ⎪
⎟⎪ .
.. ⎪⎪ ⎪
⎪ .
.. ⎪
⎪
⎜ ⎟⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎜ ⎟⎪
⎨ ⎪ ⎬ ⎪ ⎨ ⎪ ⎬
⎜ a n1 ān2 . . . a nn 0 0 . . . 0 ⎟ x n bn
⎜ ⎟ = . (8.13)
⎜ 0 0 . . . 0 ā11 a 12 . . . ā1n ⎟⎪ x̄1 ⎪ ⎪ b̄1 ⎪
⎜ ⎟⎪
⎪
⎪ x̄ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ b̄ ⎪
⎪
⎪
⎪
⎜ 0 0 . . . 0 a 21 ā22 . . . a 2n ⎟⎪
⎜ ⎟⎪
⎪
⎪ . ⎪
2⎪
⎪ ⎪
⎪ 2⎪
⎪ . ⎪ ⎪
⎜ .. .. ⎟⎪ ⎪ ⎪
⎪ . ⎪
⎝ . . ⎠⎪
⎪
⎪ .. ⎪
⎪ ⎪
⎪ ⎪
⎪
. ⎪
⎪
⎪
⎩ ⎭ ⎩ ⎪⎪ ⎪ ⎭
0 0 . . . 0 ān1 a n2 . . . ānn x̄n b̄n

Example 8.2 Let us now consider an application problem of a three-stepped fixed-

free bar as given in Fig. 8.1 having force [P3 ] applied on the free end (Behera
and Chakraverty 2013). The interval values of the material parameters of the three-
stepped bar element viz. Young’s modulus [E i ], area [Ai ], and length [L i ] for i =
1, 2, 3 are considered in Table 8.1.

Fig. 8.1 A three-stepped bar with force applied at the free end
8.2 Interval System of Equations 137

Table 8.1 Interval values of Parameters Interval values

parameters of the
three-stepped bar [A1 ](in.2 ) [2.99, 3.01]

[A2 ](in.2 ) [1.99, 2.01]

[A3 ](in.2 ) [0.99, 1.01]

[L 1 ](in.) [11.95, 12.05]

[L 2 ](in.) [9.95, 10.05]
[L 3 ](in.) [5.95, 6.05]
[E 1 ], [E 2 ], [E 3 ]( psi) [2.8, 3.1]
[P3 ](lb) [7500, 12500]

Table 8.2 Interval i Translational displacements

translational displacements of
[u i ] ui ū i
2 1.0794 × 104 1.6007 × 104
3 2.4323 × 104 3.5969 × 104
4 4.0691 ×104 5.9724 ×104

Solution: Using finite element method for the three-stepped bar under static con-
dition yields the interval system of equation as given in (Behera and Chakraverty
2013)
⎛ ⎞
[1.2492, 1.4071] −[0.5544, 0.6262] 0
⎝ −[0.5544, 0.6262] [1.0126, 1.1524] −[0.4582, 0.5262] ⎠
0 −[0.4582, 0.5262] [0.4582, 0.5262]
⎧ ⎫ ⎧ ⎫ .
⎨ [u 2 , ū 2 ] ⎬ ⎨ 0 ⎬
× 107 [u 3 , ū 3 ] = 0 .
⎩ ⎭ ⎩ ⎭
[u 4 , ū 4 ] [7500, 12500]

Applying Method III, the interval translational displacements [u i ] at nodes 2, 3,

and 4 are computed and the results are given in Table 8.2.

Program 8.3: A MATLAB program for solving interval system of linear equations
using Method III.
138 8 Interval System of Linear Equations

%Interval system of linear equations

%Enter the dimension of system of linear equations
n = input('Enter the dimension of coefficient matrix A\n');
%Enter the lower and upper bounds of coefficient matrix A
Al = input('Enter the lower bound of each element of interval aij: \n ');
Au = input('Enter the upper bound of each element of interval aij: \n ');
%Enter the lower and upper bounds of right-hand side vector
bl = input('Enter the lower bounds of interval vector b:\n ');
bu = input('Enter the upper bounds of interval vector b:\n ');
for i=1:n
for j=1:n
if Au(i,j)<0
t=Al(i,j);
Al(i,j)=Au(i,j);
Au(i,j)=t;
end
end
end
%Crisp system of linear equations
disp('The crisp coefficient matrix is');
O=zeros(n);
A=[Al O;O Au];
disp('The crisp right-hand side vector is');
b=[bl;bu];
%Solution bounds
disp('The bounds of solution vector x');
x=inv(A)*b
Further, the output corresponding to Example 8.2 has been obtained and incorpo-
rated as given below:

Input:
Enter the dimension of coefficient matrix A
3
Enter the lower bound of each element of interval aij:
[1.2492 -0.6262 0;-0.6262 1.0126,-0.5262;0 -0.5262 0.4582]
Enter the upper bound of each element of interval aij:
[1.4071 -0.5544 0;-0.5544 1.1524 -0.4582; 0 -0.4582 0.5262]
Enter the lower bounds of interval vector b:
[0;0;7500]
Enter the upper bounds of interval vector b:
[0;0;12500]

Output:
The crisp coefficient matrix is
A = 1.2492 -0.5544 0 0 0 0
8.2 Interval System of Equations 139

-0.5544 1.0126 -0.4582 0 0 0

0 -0.4582 0.4582 0 0 0
0 0 0 1.4071 -0.6262 0
0 0 0 -0.6262 1.1524 -0.5262
0 0 0 0 -0.5262 0.5262
The crisp right-hand side vector is
b= 0
0
7500
0
0
12500
The bounds of solution vector x
x = 1.0e+004 *
1.0794
2.4323
4.0691
1.6007
3.5969
5.9724

The methods considered may not yield tighter interval bounds. Readers interested
in having more detail usage of various methods for solving interval system of linear
equations are encouraged to see the references Rohn (1989), Neumaier (1990) and
Chakraverty et al. (2017)
Similar procedures involving interval computations may be applied to solve fuzzy
system of linear equations using α-cut approach. As such, literature viz. (Abbasbandy
et al. 2006), (Behera and Chakraverty 2015), and the references mentioned therein
may be referred for solving fuzzy system of equations.
Exercise

1. Solve the ISLE [A][x] = [b], where the coefficient matrix and vector is given
by
   
[9, 11] −[3.9, 4.1] [26, 30]
[A] = and [b] = .
−[3.9, 4.1] [15, 16] [35, 39]

2. Generate MATLAB code for solving ISLE if 0 belongs to elements of coefficient

matrix [A].
140 8 Interval System of Linear Equations

References

S. Abbasbandy, E. Reza, J. Ahmad, LU decomposition method for solving fuzzy system of linear
equations. Appl. Math. Comput. 172, 633–643 (2006)
D. Behera, S. Chakraverty, Fuzzy finite element analysis of imprecisely defined structures with
fuzzy nodal force. Eng. Appl. Artif. Intell. 26(10), 2458–2466 (2013)
D. Behera, S. Chakraverty, New approach to solve fully fuzzy system of linear equations using
single and double parametric form of fuzzy numbers. Sadhana 40, 35–49 (2015)
S. Chakraverty, M. Hladík, N.R. Mahato, A sign function approach to solve algebraically interval
system of linear equations for nonnegative solutions. Fundam. Inform. 152, 13–31 (2017)
S. Das, S. Chakraverty, Numerical solution of interval and fuzzy system of linear equations. Appl.
Appl. Math. 7, 334–356 (2012)
E. Hansen, On the solution of linear algebraic equations with interval coefficients. Linear Algebra
Appl. 2, 153–165 (1969)
P.T. Kahl, Solving narrow-interval linear equation systems is NP-hard. MS thesis. University of
Texas at El Paso, 1996
V. Kreinovich, A.V. Lakeyev, J. Rohn, P.T. Kahl, Computational Complexity and Feasibility of Data
Processing and Interval Computations, vol. 10 (Springer, 2013)
A. Neumaier, Interval Methods for Systems of Equations, vol. 37 (Cambridge University Press,
1990)
J. Rohn, Systems of linear interval equations. Linear Algebr. Appl. 126, 39–78 (1989)
S.P. Shary, A new technique in systems analysis under interval uncertainty and ambiguity. Reliable
Comput. 8, 321–418 (2002)