International Journal of Pure and Applied Mathematics

Volume 88 No. 1 2013, 15-34

ISSN: 1311-8080 (printed version); ISSN: 1314-3395 (on-line version)
url: http://www.ijpam.eu
doi: http://dx.doi.org/10.12732/ijpam.v88i1.2

AP
ijpam.eu

THE DETERMINANT OF AN INTERVAL MATRIX

USING GAUSSIAN ELIMINATION METHOD
T. Nirmala1 , D. Datta2 , H.S. Kushwaha3 , K. Ganesan4
1,4 Department

of Mathematics
Faculty of Engineering and Technology
SRM University
Kattankulathur, Chennai, 603203, INDIA
2,3 Health Physics Division
Bhabha Atomic Research Centre
Trombay, Mumbai, 400085, INDIA
Abstract: We introduce the notion of determinant and related results for interval matrices. We propose a Gaussian elimination like algorithm for computing the enclosures of the determinant of interval matrices. Numerical examples
are also provided to show the efficiency of the proposed algorithm.
AMS Subject Classification: 15A09, 65F05, 65G30
Key Words: interval number, interval arithmetic, interval matrix, determinant, Gaussian algorithm
1. Introduction
To model real-life problems and perform computations we must deal with uncertainty and inexactness. These uncertainty and inexactness are due to measurement inaccuracy, simplification of physical models, variations of the parameters
of the system, computational errors etc. Interval analysis is an efficient and reliable tool that allows us to handle such problems effectively. Determinants of
interval matrices plays a key role in the study of Eigen values and eigenvectors.
Received:

May 28, 2013

Correspondence

author

c 2013 Academic Publications, Ltd.

url: www.acadpubl.eu

16

T. Nirmala, D. Datta, H.S. Kushwaha, K. Ganesan

Eigen values of general interval matrices are useful to study robust stability
margins of analog and discrete systems and convergence rates in numerical
analysis. Determinants of real symmetric and Hermitian interval matrices play
a major role in signal processing, control, stochastic processes and statistics.
Evaluation of determinants of matrices in a straightforward way from the definition, requires about n.n! arithmetic operations for an (n n) matrix, where
2n3
operations which is very
as Gaussian elimination method requires only
3
much less when compared to the number n.n!. Hence we propose a Gaussian
elimination like algorithm for computing determinants of interval matrices.
Let IR = {
a = [a1 , a2 ] : a1 a2 and a1 , a2 R} be the set of all proper
a = [a1 , a2 ] : a1 > a2 and a1 , a2 R} be the set of all
intervals and IR = {
improper intervals on the real line R. If a1 = a2 = a, then a
= [a, a] is a
real number (or a degenerate interval). We shall use the terms interval and
interval number interchangeably. The mid-point and width(or half-width)

a1 + a2
and
of an interval number a
= [a1 , a2 ] are defined as m(
a) =
2


a2 a1
w(
a) =
. We denote the set of generalized intervals (proper and
2
improper) by D = IR IR = {[a1 , a2 ] : a1 , a2 R}. The set of generalized
intervals D is a group with respect to addition and multiplication operations,
while maintaining the inclusion monotonicity.
The dual is an important monadic operator proposed by Kaucher [9] that
reverses the end-points of the intervals and expresses an element-to-element
symmetry between proper and improper intervals in D. For a
= [a1 , a2 ] D,
its dual is defined by dual(
a) = dual[a1 , a2 ] = [a2 , a1 ]. The opposite of an
interval a
= [a1 ,a2 ] is opp{[a
1 , a2 ]} = [a1 , a2 ] which is the additive inverse

1 1
of [a1 , a2 ] and
,
is the multiplicative inverse of [a1 , a2 ], provided
 a1 a2
a1 + a2
6= 0.
m([a1 , a2 ]) =
2
That is

and

a
+ (dual a
) = a
dual(
a) = [a1 , a2 ] dual([a1 , a2 ])

1
dual a

= [a1 , a2 ] [a2 , a1 ] = [a1 a1 , a2 a2 ] = [0, 0]



1
1
=[a1 , a2 ]
= [a1 , a2 ]
dual([a1 , a2 ])
[a2 , a1 ]
 


a1 a2
1 1
,
,
=
= [1, 1]
=[a1 , a2 ]
a1 a2
a1 a2

In this paper we use the generalized interval arithmetic proposed by Nirmala

THE DETERMINANT OF AN INTERVAL MATRIX...

17

et.al [16] by extending Kauchers [9] interval arithmetic.This arithmetic operation satisfying group properties with respect to addition and multiplication
operations and satisfying the distributive relations between intervals, while
maintaining the inclusion monotonicity.
The rest of this paper is organized as follows: In Section 2, we extend the
Sengupta and Pals [2] method of comparison of interval numbers to the set
of all generalized intervals D. We recall the generalized interval arithmetic D
on the set of generalized interval numbers D proposed by Nirmala et.al [16].
In Section 3, we introduce the notion of interval matrices and the arithmetic
operations on interval matrices. We define the determinant of square interval
matrices and discuss some of its properties. In Section 4, we propose a Gaussian
elimination like algorithm for computing the enclosures of the determinant of
interval matrices. Numerical examples are also provided to show the efficiency
of the proposed algorithm.

2. Preliminaries
The aim of this section is to present some notations, notions and results which
are of useful in our further considerations.
2.1. Comparing Interval Numbers
Sengupta and Pal [2] proposed a simple and efficient index for comparing any
two intervals on IR through decision makers satisfaction. We extend this
concept to the set of all generalized intervals on D.
Definition 1. Let  be an extended order relation between the interval
numbers a
= [a1 , a2 ] and b = [b1 , b2 ] in D, then for m(
a) < m(b), we construct

a premise (
a  b) which implies that a
is inferior to b (or b is superior to a
).
Here, the term inferior to (superior to) is analogous to less than (greater
than).
Definition 2. An acceptability function A : D D [0, ) is defined
(m(b) m(
a))
, where w(b) + w(
a) 6= 0. A may
as: A (
a, b) = A(
a  b) =

(w(b) + w(
a))
be interpreted as the grade of acceptability of the first interval number a
to be
inferior to the second interval number b.
For any two interval numbers a
and b in D, either A(
a  b) > 0 or A(b 

a
) > 0 or A(
a  b) = A(b  a
) = 0 and A(
a  b) + A(b  a
) = 0. Also the

18

T. Nirmala, D. Datta, H.S. Kushwaha, K. Ganesan

proposed A-index is transitive; for any three interval numbers a

, b and c in D,

if A(
a  b) 0 and A(b  c) 0, then A(
a  c) 0. But it does not mean
that A(
a  c) max{A(
a  b), A(b  c)}. If A(
a  b) = 0, then we say that

the interval numbers a

and b are equivalent (or non-inferior to each other) and
we denote it by a
b. In particular, whenever A(
a  b) = 0 and w(
a) = w(b),

then a
= b. Also if A(
a  b) 0, then we say that a
 b and if A(b  a
) 0,
then we say that b  a
.
Remark 3. For any two interval numbers a
, b D, we have A(
a 
b) + A(b  a
) = 0. If m(
a) = 0 then we say that a
is a zero interval number.
In particular, if m(
a) = 0 and w(
a) = 0, then a
= [0, 0]. Also, if m(
a) = 0 and

w(
a) 6= 0, then a
0. It is to be noted that if a
= [0, 0] = 0, then a
0, but
the converse need not be true. If a
6 0 (i.e. a
is not equivalent to 0), then a

is said to be a non-zero interval number. It is to be noted that if a

6 0, then
a
6= 0, but the converse need not be true. If A(
a  0) 0 and A(
a  0) 6= 0,
that is, if a

0 and a
6
0, then a
is said to be a positive interval number and

is denoted by a
0.
2.2. A New Interval Arithmetic
Ganesan and Veeramani [4] proposed a new interval arithmetic on IR. We
extend this arithmetic operations to the set of generalized interval numbers D
and incorporating the concept of dual. For a
= [a1 , a2 ], b = [b1 , b2 ] D and

b) k, m(
for {+,
b = [m(
a) m(o
a) m(b) + k], where
n , , }, we define a
k = min (m(
a) m(b)) , (m(
a) m(b)) , and are the end points
of the interval a
b under the existing interval arithmetic. In particular
(i) Addition:
h
i
a
+ b = [a1 , a2 ] + [b1 , b2 ] = (m(
a) + m(b)) k, (m(
a) + m(b)) + k ,


(b2 + a2 ) (b1 + a1 )
.
where k =
2
(ii) Subtraction:
h
i
a
b = [a1 , a2 ] [b1 , b2 ] = (m(
a) m(b)) k, (m(
a) m(b)) + k
where k =


(b2 + a2 ) (b1 + a1 )
.
2

THE DETERMINANT OF AN INTERVAL MATRIX...

19

Also if a
= b, i.e. if [a1 , a2 ] = [b1 , b2 ], then
a
b = a
dual(
a) = [a1 , a2 ] [a2 , a1 ]
= [a1 a1 , a2 a2 ] = [0, 0]
(iii) Multiplication:
h
i
a
.b = a
b = [a1 , a2 ] [b1 , b2 ] = m(
a)m(b) k, m(
a)m(b) + k ,
n
o
where k = min (m(
a)m(b)) , (m(
a)m(b)) ,

= min(a1 b1 , a1 b2 , a2 b1 , a2 b2 ) and = max(a1 b1 , a1 b2 , a2 b1 , a2 b2 ).



1
1
1
1
(iv) Division: 1 a
= =
=
k,
+ k , where
a

[a1 , a2 ]
m(
a)
m(
a)



 
1 a2 a1
1 a2 a1
,
and 0 6 [a1 , a2 ].
k = min
a2 a1 + a2
a1 a1 + a2
Also if a
= b i.e. [a1 , a2 ] = [b1 , b2 ], then
a

1
=
= [a1 , a2 ] .
a

dual(
a)
[a2 , a1 ]

 

1 1
a1 a2
= [a1 , a2 ] .
,
,
=
= [1, 1]
a1 a2
a1 a2
=

From (iii), it is clear that

a=

[a1 , a2 ], for 0
[a2 , a1 ], for < 0.

It is to be noted that we use to denote the existing interval arithmetic

and to denote the modified interval arithmetic. But wherever there is no
confusion we use the same notation for both the cases. It is also to be noted
that a
b a
b = {a b/a a
, b b}, where {, , , } is the
existing interval arithmetic. For example if a
= [1, 2] and b = [3, 5], then

a
b = [1, 2] [3, 5] = [min(3, 5, 6, 10), max(3, 5, 6, 10)] = [5, 10] and
a
b = a
b = [1, 2][3, 5] = [5, 9] so that a
b a
b. It is also important
to note that by using this modified interval arithmetic we are able to prove the
distributive law for interval numbers and hence many other important results.

20

T. Nirmala, D. Datta, H.S. Kushwaha, K. Ganesan

3. Main Results

An interval matrix A is a matrix whose

elements are interval
numbers. An intera
11 a
1n
val matrix A will be written as: A = = (
aij )1im, 1jn,
a
m1 a
mn
where each a
ij = [aij , aij ] (or) A = [A, A] for some A, A satisfying A A.
mn
We use D
to denote the set of all (m n) interval matrices. The midpoint (center) of an interval matrixA is the matrix of midpoints
of its in
m(
a11 ) m(
a1n )
=
. The width
terval elements defined as m(A)

m(
am1 ) m(
amn )

of an interval matrix
A is the matrix of
widths of its interval elements de
w(
a11 ) w(
a1n )
=
which is always nonnegative. We

fined as w(A)
w(
am1 ) w(
a )

mn
0 0
to denote the
use O to denote the null matrix and O
0 0

0
0
null interval matrix . Also we use I to denote the iden
0
0

1 0
tity matrix
1 and I to denote the identity interval matrix
0 1

1
0
= m(B),
then the interval matrices A and B

1 . If m(A)
0
1
In particular if m(A)
=
are said to be equivalent and is denoted by A B.

m(B) and w(A) = w(B), then A = B. If m(A) = O, then we say that

A is a zero interval matrix. In

particular if m(A) = O and w(A) = O,
[0, 0] . . . [0, 0]
= O and w(A)
6= O, then
then A = . . . . . .
. . . . Also, if m(A)
[0, 0] . . . [0, 0]

0
0
If A 6 O
(i.e. A is not equivalent to O),
then
A = O.
0
0

= I, then we say that

A is said to be a non-zero interval matrix. If m(A)
= I and w(A)
= O,
A is a identity interval matrix. In particular if m(A)

THE DETERMINANT OF AN INTERVAL MATRIX...

21

[1, 1] [0, 0]
= I and w(A)
6= O, then
then A = [1, 1] . Also, if m(A)
[0, 0] . . . [1, 1]

1
0

A =
1 I.
0
1
3.1. Arithmetic Operations on Interval Matrices
We introduce the following arithmetic operations on interval matrices. As with
interval numbers, we define the arithmetic operations on interval matrices as
B
Dmn , x
Dn and
follows. If A,
D, then
(i).
A (
a
ij )1im,

1jn



a
(ii). A + B
ij + bij

1im, 1jn

(iii). A B

( 

a
ij bij

1im, 1jn

O
= O,
A dual(A)



Pn a
(iv). AB

b
k=1 ik kj

, if A 6 B

if A B

1im, 1jn

x
(v). A

P
n

x
j=1 ij

1im

3.2. Determinant of Interval Matrices

We define the determinant of a square interval matrix as in the case of real

square matrix except that the determinant
of an interval matrix is an interval
P
number. That is det A =| A |=
a
ij Aij , where Aij with usual meaning. It is
easy to see that most of the properties of determinants of classical matrices are
hold good (up to equivalent) for the determinants of interval matrices under
the modified interval arithmetic.
3.3. Properties of Determinants
Let A be an interval matrix of order (n n). Then

1. det A det AT , for all (n n) interval matrices A.

22

T. Nirmala, D. Datta, H.S. Kushwaha, K. Ganesan

2. If one row of A consists entirely of zero fuzzy numbers, then A 0.
is an interval matrix obtained from A by multiplying one row of A
3. If B
det A.

by the scalar then det B

is an interval matrix obtained from A by interchanging two rows of
4. If B
det A.

A then det B
5. If two rows of A are identical, then det A 0.

Theorem 4. If A is an (n n) upper triangular or lower triangular

is equivalent to the product of the elements in its
interval matrix, then |A|
main diagonal.
Proof. Let A be an upper triangular
intervalmatrix

 of order (n n). The

a
11 a
12
[a11 , a11 ] [a12 , a12 ]
.
=
proof is by induction on n.Let A =
0 a
22
[0, 0]
[a22 , a22 ]
Then by definition
| A | a
11 a
22 [a11 , a11 ][a22 , a22 ]

the product of the entries in the leading diagonal of A.

(1)

Hence the determinant of a (2 2) upper triangular interval matrix, is the

product of the entries in its leading diagonal. Therefore the result is true for
n = 2. Assume that the result is true for n = k, ie.


[a11 , a11 ] [a12 , a12 ] . . . [a1k , a1k ] 


 [0, 0]

[a
a
]
.
.
.
[a
a
]
,
,
22
2k
22
2k




.
.
. . .
.



| A |= 
(2)

.
.
. . .
.




.
.
. . .
.


 [0, 0]
[0, 0]
. . . [akk , akk ] 
Now for n = k + 1, determinant of upper triangular interval matrix A of size
(k + 1) by (k + 1) by expanding along the bottom row (since all entries in the
bottom row are zero except [ak+1 , ak+1 ]), we have


 [a11 , a11 ] [a12 , a12 ] . . . [ak+1 , ak+1 ] 


 [0, 0]
[a22 , a22 ] . . . [ak+1 , ak+1 ] 



.
.
. . .
.

| A | = 

.
.
.
.
.
.




.
.
. . .
.


 [0, 0]
[0, 0]
. . . [ak+1 , ak+1 ] 

THE DETERMINANT OF AN INTERVAL MATRIX...


 [a11 , a11 ] [a12 , a12 ]

 [0, 0]
[a22 , a22 ]


.
.
(1)(k+1)+(k+1) [ak+1 , ak+1 ] 
.
.


.
.

 [0, 0]
[0, 0]

23
.
.
.
.
.
.

.
.
.
.
.
.

. [a1k , a1k ]
. [a2k , a2k ]
.
.
.
.
.
.
. [akk , akk ]

| A | (1)(2k+2) [ak+1 , ak+1 ][a11 , a11 ][a22 , a22 ]...[akk , akk ], by equation (2)
(1)2(k+1) [a11 , a11 ][a22 , a22 ]...[akk , akk ][ak+1 , ak+1 ]
[a11 , a11 ][a22 , a22 ]...[akk , akk ][ak+1 , ak+1 ], since (1)2(k+1) = 1.

the product of the entries in the leading diagonal of A.

Therefore the result is true for n = k + 1 also. Hence the theorem. If A is lower
triangular interval matrix, then the theorem can also be proved in the similar
way.
is an interval matrix obtained from A by adding a
Theorem 5. If B
then det A det B.

multiple of one row of A to another row of A,

be an interval matrix of order (nn) obtained from an interval
Proof. Let B

matrix A of order (n n) by adding k times of jth row of A to the ith row of

where i, j = 1, 2, 3, ..., n. We have
A,

a
11
a
12
a
13
... ... ...
a
1n

a
21
a
22
a
23
... ... ...
a
2n

...
...
...
... ... ...
...

...
...
...
a

j1
j2
j3
jn
=

...
...
...
... ... ...
...

a
aj1 a
i2 + k
aj2 a
i3 + k
aj3 ... ... ... a
in + k
ajn
i1 + k

...
...
...
... ... ...
...
a
n1

a
n2

a
n3

... ... ...



a
11
a
12
a
13


a

a
23
21
22


...
...
...


a

j1
j2
j3
=
Now det B

...
...
...

 a
aj1 a
i2 + k
aj2 a
i3 + k
aj3
 i1 + k

...
...
...


a
n1
a
n2
a
n3

...
...
...
...
...
...
...
...

...
...
...
...
...
...
...
...

a
nn

...
a
1n
...
a
2n
...
...
...
a
jn
...
...
... a
in + k
ajn
...
...
...
a
nn

24
















T. Nirmala, D. Datta, H.S. Kushwaha, K. Ganesan

a
11 a
12
a
21 a
22
...
...
a
j1 a
j2
...
...
a
i1 a
i2
...
...
a
n1 a
n2
det A + k.
Hence



 a
a
13 ... ... ... a
1n 
 11
 a
a
23 ... ... ... a
2n 
 21
 ...
... ... ... ... ... 

 a


a
j3 ... ... ... a
jn 
+ k  j1

... ... ... ... ... 
 ...
 a

a
i3 ... ... ... a
in 
 j1
 ...

... ... ... ... ... 

 a

n1
a
n3 ... ... ... a
nn
properties (3) and (5).

0 det A,

a
12 a
13
a
22 a
23
...
...
a
j2 a
j3
...
...
a
j2 a
j3
...
...
a
n2 a
n3

...
...
...
...
...
...
...
...

det A.

det B

...
...
...
...
...
...
...
...

...
...
...
...
...
...
...
...

a
1n
a
2n
...
a
jn
...
a
jn
...
a
nn

4. Interval Gaussian Elimination Algorithm

We propose interval Gaussian elimination algorithm for computing the enclosures of the determinant of interval matrices. For simplicity, we consider a
four-by-four square interval matrix

a
11 a
12 a
13 a
14
[a11 , a11 ] [a12 , a12 ] [a13 , a13 ] [a14 , a14 ]
a

21 a
22 a
23 a
24
= [a21 , a21 ] [a22 , a22 ] [a23 , a23 ] [a24 , a24 ]
A =
a
31 a
32 a
33 a
34 [a31 , a31 ] [a32 , a32 ] [a33 , a33 ] [a34 , a34 ]
[a41 , a41 ] [a42 , a42 ] [a43 , a43 ] [a44 , a44 ]
a
41 a
42 a
43 a
44

Steps for reducing the square interval matrix A in to an upper triangular square
interval matrix.
Step 1: The Pivot is a
11
a
21
add the result to second row to get by
Multiply first row by m
21 =
a
11
using dual subtraction and dual division, a
21 = a
21 + m
21 a
11 = 0; a
22 =
a
22 + m
21 a
12 ; a
23 = a
23 + m
21 a
13 ; a
24 = a
24 + m
21 a
14 Multiply first row
a
31
add the result to third row to get a
31 = 0; a
32 = a
32 + m
31
by m
31 =
a
11
a
41
a
12 ; a
33 = a
33 + m
31
a13 ; a
34 = a
34 + m
31
a14 Multiply first row by m
41 =
a
11
add the result to fourth row to get a
41 = 0; a
42 = a
42 + m
41 a
12 ; a
43 =
a
43 + m
41 a
13 ; a
44 = a
44 + m
41 a
14
Step 2: The Pivot is a
22
a
32
add the result to third row to get a
31 =
Multiply second row by m
32 =
a
22
0; a
32 = 0; a
33 = a
33 + m
32 a
23 ; a
34 = a
34 + m
32 a
24 Multiply second row

THE DETERMINANT OF AN INTERVAL MATRIX...

25

a
42
add the result to fourth row to get a
41 = 0; a
42 = 0; a
43 =
a
22
a
43 + m
42 a
23 ; a
44 = a
44 + m
42 a
24
Step 3: The Pivot is a
33
a
43
add the result to fourth row to get a
41 =
Multiply third row by m
43 =
a
33
0; a
42 = 0; a
43 =
0; a
44 = a
44 + m
43 a
34 .
The reduced upper triangular square interval matrix is


 [a11 , a11 ] [a12 , a12 ] [a13 , a13 ] [a14 , a14 ] 


 [0, 0]
[a22 , a22 ] [a23 , a23 ] [a24 , a24 ] 


.
B=
[0, 0]
[a33 , a33 ] [a34 , a34 ] 
 [0, 0]
 [0, 0]
[0, 0]
[0, 0]
[a44 , a44 ] 
by m
42 =

|
Now by theorems (4) and (5), we have | A || B

Interval Gauss Elimination Algorithm Using Int Lab

A=input(Enter the interval matrix: )
[n,nn]=size(A)
for i=1:n-1
for k=i+1:n
a=inf(A(i,i))
b=sup(A(i,i))
c=infsup(a,b)
m1=(a+b)/2
if(a<=0 && b>=0)
if(m1>0)
m=m1/2
f=m1-m
g=m1+m
h=infsup(f,g)
m2=(f+g)/2
k1=(1/g)*((g-f)/(g+f))
k2=(1/f)*((g-f)/(g+f))
k=min(k1,k2)
n1=(1/m2)-k
n2=(1/m2)+k

26
n3=infsup(n1,n2)
a=inf(n1)
b=sup(n2)
c=inf(A(k,i))
d=sup(A(k,i))
m1=(a+b)/2
m2=(c+d)/2
I1=infsup(a,b)
I2=infsup(c,d)
m=I1*I2
alpha=inf(m)
beta=sup(m)
k1=(m1*m2)-alpha
k2=beta-(m1*m2)
k3=min(k1,k2)
n1=(m1*m2)-k3
n2=(m1*m2)+k3
m(k,i)=-infsup(n1,n2)
a=inf(m(k,i))
b=sup(m(k,i))
c=inf(A(i,:))
d=sup(A(i,:))
m1=(a+b)/2
m2=(c+d)/2
I1=infsup(a,b)
I2=infsup(c,d)
m=I1*I2
alpha=inf(m)
beta=sup(m)
k1=(m1*m2)-alpha
k2=beta-(m1*m2)
k4=min(k1,k2)
n1=(m1*m2)-k4
n2=(m1*m2)+k4
if(A(k,i)==A(k,i))
s1=infsup(0,0)
if(A(i,i)/A(i,i))
s2=infsup(1,1)
A(k,i)=s1*s2

T. Nirmala, D. Datta, H.S. Kushwaha, K. Ganesan

THE DETERMINANT OF AN INTERVAL MATRIX...

A(k,:)=A(k,:)+infsup(n1,n2)
A(k,i)=infsup(0,0)
full(A)
end
end
else
m3=-m1/2
f1=m1-m3
g1=m1+m3
h1=infsup(f1,g1)
m4=(f1+g1)/2
k3=(1/g1)*((g1-f1)/(g1+f1))
k4=(1/f1)*((g1-f1)/(g1+f1))
k1=min(k3,k4)
n4=(1/m4)-k1
n5=(1/m4)+k1
n6=infsup(n4,n5)
a=inf(n6)
b=sup(n6)
c=inf(A(k,i))
d=sup(A(k,i))
m1=(a+b)/2
m2=(c+d)/2
I1=infsup(a,b)
I2=infsup(c,d)
m=I1*I2
alpha=inf(m)
beta=sup(m)
k1=(m1*m2)-alpha
k2=beta-(m1*m2)
k3=min(k1,k2)
n1=(m1*m2)-k3
n2=(m1*m2)+k3
h=infsup(n1,n2)
m(k,i)=-h
a=inf(m(k,i))
b=sup(m(k,i))
c=inf(A(i,:))
d=sup(A(i,:))

27

28

T. Nirmala, D. Datta, H.S. Kushwaha, K. Ganesan

m1=(a+b)/2
m2=(c+d)/2
I1=infsup(a,b)
I2=infsup(c,d)
m=I1*I2
alpha=inf(m)
beta=sup(m)
k1=(m1*m2)-alpha
k2=beta-(m1*m2)
k4=min(k1,k2)
n1=(m1*m2)-k4
n2=(m1*m2)+k4
f=infsup(n1,n2)
if(A(k,i)==A(k,i))
s1=infsup(0,0)
if(A(i,i)/A(i,i))
s2=infsup(1,1)
A(k,i)=s1*s2
A(k,:)=A(k,:)+f
A(k,i)=infsup(0,0)
full(A)
end
end
end
else
k5=(1/b)*((b-a)/(a+b))
k6=(1/a)*((b-a)/(a+b))
k7=min(k5,k6)
n7=(1/m1)-k7
n8= (1/m1)+k7
n9=infsup(n7,n8)
a=inf(n9)
b=sup(n9)
c=inf(A(k,i))
d=sup(A(k,i))
m1=(a+b)/2
m2=(c+d)/2
I1=infsup(a,b)
I2=infsup(c,d)

THE DETERMINANT OF AN INTERVAL MATRIX...

m=I1*I2
alpha=inf(m)
beta=sup(m)
k1=(m1*m2)-alpha
k2=beta-(m1*m2)
k3=min(k1,k2)
n1=(m1*m2)-k3
n2=(m1*m2)+k3
h=infsup(n1,n2)
m(k,i)=-h
a=inf(m(k,i))
b=sup(m(k,i))
c=inf(A(i,:))
d=sup(A(i,:))
m1=(a+b)/2
m2=(c+d)/2
I1=infsup(a,b)
I2=infsup(c,d)
m=I1*I2
alpha=inf(m)
beta=sup(m)
k1=(m1*m2)-alpha
k2=beta-(m1*m2)
k4=min(k1,k2)
n1=(m1*m2)-k4
n2=(m1*m2)+k4
f=infsup(n1,n2)
if(A(k,i)==A(k,i))
s1=infsup(0,0)
if(A(i,i)/A(i,i))
s2=infsup(1,1)
A(k,:)=A(k,:)+f
A(k,i)=infsup(0,0)
full(A)
end
end
end
end
end

29

30

T. Nirmala, D. Datta, H.S. Kushwaha, K. Ganesan

%Product of the diagonal matrix

diag(A)
pro_dia =infsup(1,1)
for m=1:n
a=inf(pro_dia)
b=sup(pro_dia)
c=inf(diag(A(m,m)))
d=sup(diag(A(m,m)))
m1=(a+b)/2
m2=(c+d)/2
I1=infsup(a,b)
I2=infsup(c,d)
m=I1*I2
alpha=inf(m)
beta=sup(m)
k1=(m1*m2)-alpha
k2=beta-(m1*m2)
k4=min(k1,k2)
n1=(m1*m2)-k4
n2=(m1*m2)+k4
n3=infsup(n1,n2)
pro_dia=n3
end

5. Numerical Examples
Example 6. Using Gauss elimination
algorithm, find the determinant of

[3.7, 4.3]
[1.5, 0.5]
[0, 0]
where A = [1.5, 0.5]
the interval matrix A,
[3.7, 4.3]
[1.5, 0.5]
[0, 0]
[1.5, 0.5]
[3.7, 4.3]

Step 1: The Pivot element is m

21 = [0.1163, 0.3837]
a
21 = a
21

a
21
a
11
a
11

[1.5, 0.5]
[3.7, 4.3]
[3.7, 4.3]
=
0 by using dual division and dual subtraction

= [1.5, 0.5]

THE DETERMINANT OF AN INTERVAL MATRIX...

31

a
22 = [3.2581, 4.2419] ; a
23 = [1.5, 0.5]
The reduced matrix is

[3.7, 4.3]
[1.5, 0.5]
[0, 0]
[0, 0]
[3.2581, 4.2419] [1.5, 0.5]
[0, 0]
[1.5, 0.5]
[3.7, 4.3]

Step 2: The Pivot element is m

32 = [0.1179, 0.4154] a
32 = 0; a
33 = [3.2256,
4.2411]. Hence the reduced matrix is

[3.7, 4.3]
[1.5, 0.5]
[0, 0]
= [0, 0]
B
[3.2581, 4.2419]
[1.5, 0.5] .
[0, 0]
[0, 0]
[3.2256, 4.2411]

is the product of the elements in its main diagNow the determinant of B

onal and hence if we evaluate the determinant of the interval matrix A in a
straightforward way from the definition, we have



 [3.7, 4.3]
[1.5, 0.5]
[0, 0]


[3.7, 4.3]
[1.5, 0.5]
| A | = [1.5, 0.5]

[0, 0]
[1.5, 0.5]
[3.7, 4.3] 
[44.178, 75.822] + [7.075, 0.925]

[37.103, 74.897]

|) = 56. Hence by theorem (4) and (5), we

Here we see that m(| A |) = m(| B

have | A | [38.8846, 73.1159] .

Example 7. Using Gauss elimination
algorithm,

[4, 6]
[1, 1]
[1, 1] [6, 4]
where A =
the interval matrix A,
[1, 1] [1, 1]
[1, 1] [1, 1]

find the determinant

of
[1, 1]
[1, 1]
[1, 1]
[1, 1]

[9, 11]
[1, 1]
[1, 1] [11, 9]

Step 1: The Pivot element is m

21 = [0.2333, 0.2333]; a
21 = 0; a
22 =
[6.2334, 3.7666]; a
23 = [1.2334, 1.2334]; a
24 = [1.2334, 1.2334].
The Pivot element is m
31 = [0.2333, 0.2333]; a
31 = 0; a
32 = [1.2334, 1.2334];
a
33 = [8.7666, 11.2334]; a
34 = [1.2334, 1.2334].
The Pivot element is m
41 = [0.2334, 0.2334]; a
41 = 0; a
42 = [1.2334, 1.2334];
a
43 = [1.2334, 1.2334]; a
44 = [11.2334, 8.7666]. The reduced matrix is

[4, 6]
[1, 1]
[1, 1]
[1, 1]
[0, 0] [6.2334, 3.7666] [1.2334, 1.2334]
[1.2334, 1.2334]

.
[0, 0] [1.2334, 1.2334]
[8.7666, 11.2334]
[1.2334, 1.2334]
[0, 0] [1.2334, 1.2334] [1.2334, 1.2334] [11.2334, 8.7666]

32

T. Nirmala, D. Datta, H.S. Kushwaha, K. Ganesan

Step 2: The Pivot element is m

32 = [0.2955, 0.2955]; a
31 = 0; a
32 = 0;
a
33 = [8.4020, 11.5980]; a
34 = [1.5980, 1.5980].
The Pivot element is m
42 = [0.2955, 0.2955]; a
41 = 0; a
42 = 0; a
43 =
[1.5980, 1.5980]; a
44 = [11.5980, 8.4020].
The reduced matrix is

[4, 6]
[1, 1]
[1, 1]
[1, 1]
[0, 0] [6.2334, 3.7666] [1.2334, 1.2334]
[1.2334, 1.2334]

.
[0, 0]
[0, 0]
[8.4020, 11.5980]
[1.5980, 1.5980]
[0, 0]
[0, 0]
[1.5980, 1.5980] [11.5980, 8.4020]

Step 3: The Pivot element is m

43 = [0.1819, 0.1819];
a
41 = 0; a
42 =
0; a
43 = 0; a
44 = [11.8888, 8.1112]. Hence the reduced upper triangular
interval matrix is

[4, 6]
[1, 1]
[1, 1]
[1, 1]

[1.2334, 1.2334]
= [0, 0] [6.2334, 3.7666] [1.2334, 1.2334]

B
[0, 0]
[0, 0]
[8.4020, 11.5980]
[1.5980, 1.5980]
[0, 0]
[0, 0]
[0, 0]
[11.8888, 8.1112]

is the product of the elements in its main diagoNow the determinant of B

nal and hence | B |= [4, 6] [6.2334, 3.7666] [8.4020, 11.5980] [11.8888,

8.1112] [1026.8, 3973.2]. If we evaluate the determinant of the interval
matrix A in a straightforward way from the definition, we have


 [4, 6]
[1, 1] [1, 1]
[1, 1] 

[1, 1] [6, 4] [1, 1]
[1, 1] 
| A | = 
[1, 1] 
[1, 1] [1, 1] [9, 11]
[1, 1] [1, 1] [1, 1] [11, 9]
= [868, 4132]

|) = 2500. Hence by theorem (4) and (5),

Here we see that m(| A |) = m(| B

we have | A | [1026.8, 3973.2].

6. Conclusion
We have defined the concept of determinant for square interval matrices and
proved some of its properties. We have proposed an interval Gaussian elimination algorithm for computing the enclosures of determinant of interval matrices.
We have solved two numerical examples to show the efficiency of the proposed

THE DETERMINANT OF AN INTERVAL MATRIX...

33

algorithm. It is to be noted that the results obtained by applying the proposed algorithm are sharper (less width)than the results obtained by applying
straightforward way from the definition of determinant.

Acknowledgements
The authors gratefully acknowledge (BRNS - DAE) Board of Research in Nuclear Sciences, Department of Atomic Energy, Government of India for its support through the funded research project grant No.2008/36/35/BRNS/1999 for
the investigation presented here.

References
[1] G. Alefeld and J. Herzberger, Introduction to Interval Computations, Academic Press, New York (1983).
[2] Atanu Sengupta, Tapan Kumar Pal, Theory and Methodology: On comparing interval numbers, European Journal of Operational Research, 27
(2000), 28-43, doi: 10.1016/S0377-2217(99)00319-7.
[3] R. Baker Kearfott, V. Kreinovich, Applications of Interval Computations,
Springer (1996).
[4] K. Ganesan, P. Veeramani, On arithmetic operations of interval numbers,
International Journal of Uncertainty, Fuzziness and Knowledge - Based
Systems, 13, No. 6 (2005), 619-631, doi: 10.1142/S0218488505003710.
[5] K. Ganesan, On some properties of interval matrices, International Journal
of Computational and Mathematical Sciences, 1, No. 2 (2007), 92-99, doi:
10.11113.4379.
[6] E.R. Hansen, R.R. Smith, Interval arithmetic in matrix computations,
Part 2, SIAM. journal of Numerical Analysis, 4 (1967), 1-9, doi:
10.1137/0704001.
[7] E.R. Hansen, On the solution of linear algebraic equations with interval
coefficients, Linear Algebra Appl, 2 (1969), 153-165, doi: 10.1016/00243795.

34

T. Nirmala, D. Datta, H.S. Kushwaha, K. Ganesan

[8] E.R. Hansen, Bounding the solution of interval linear Equations, SIAM
Journal of Numerical Analysis, 29, No. 5 (1992), 1493-1503, doi:
10.1137/0729086.
[9] E. Kaucher, Interval analysis in the extended interval space IR, Computing,
Suppl., 2 (1980), 33-49, doi: 10.1007/978-3-7091-8577-3.
[10] J.R. Kuttler, A Fourth-order finite-difference approximation for the
fixed membrane eigen-problem, Math. Comp., 25 (1971), 237-256, doi:
10.1090/S0025-5718-1971-0301955-6.
[11] Luc Jaulin, Michel Kieffer, Olivier Didrit, Eric Walter, Applied Interval
Analysis, Springer-Verlag, London (2001).
[12] R.E. Moore, Methods and Applications of Interval Analysis, SIAM,
Philadelphia (1979).
[13] R.E. Moore, R.B. Kearfott, M. Cloud, Introduction to Interval Analysis,
SIAM, Philadelphia, January (2009).
[14] A. Neumaier, Interval Methods for Systems of Equations, Cambridge University Press, Cambridge (1990).
[15] S. Ning, R.B. Kearfott, A comparison of some methods for solving linear
interval Equations, SIAM Journal of Numerical Analysis, 34 (1997), 12891305, doi: 10.1137/S0036142994270995.
[16] T. Nirmala, D. Datta, H.S. Kushwaha, K. Ganesan, Inverse interval matrix:
A new approach, Applied Mathematical Sciences, 5, No. 13 (2011), 607-624,
doi: AIML092011005.
[17] J. Rohn, Inverse interval matrix, SIAM Journal of Numerical Analysis, 3
(1993), 864-870, doi: 10.1137/0730044.
[18] Thomas S. Shores, Applied linear Algebra and Matrix Analysis, SpringerVerlag, New York (2007).