This action might not be possible to undo. Are you sure you want to continue?

# c Copyright, Darbose

**International Journal of Applied Mathematics and Computation
**

Volume 2(4),pp 1–8, 2010

http://ijamc.psit.in

Numerical solution of fuzzy diﬀerential equations by third

order Runge-Kutta method

K. Kanagarajan, M. Sambath

Department of Mathematics, Sri Ramakrishna Mission Vidyalaya,

College of Arts and Science,Coimbatore 641 020, India.

Email: kanagarajank@gmail.com

Abstract:

In this paper we give a numerical algorithm for solving fuzzy diﬀerential equations based

on Seikkala’s derivative of fuzzy process. We discuss in detail a numerical method based on

Runge-Kutta method of order 3. The algorithm is illustrated by solving some fuzzy diﬀerential

equations.

Keywords: Numerical solution, Fuzzy diﬀerential equation, Runge-Kutta method.

1 Introduction

Fuzzy set theory is a tool that makes possible to describe vague and uncertain notions. The

concept of a fuzzy derivative was ﬁrst introduced by Chang and Zadeh [4], later Dubois and Prade

[5] deﬁned the fuzzy derivative by using Zadeh’s extension principle and was followed by Puri and

Ralescu [10]. Fuzzy diﬀerential equations have been suggested as a way of modelling uncertain

and incompletly speciﬁed systems and were studied by many researchers [6, 7, 8]. The existence of

solutions of fuzzy diﬀerential equations has been studied by several authors [2, 3]. It is diﬃcult to

obtain exact solution for fuzzy diﬀerential equations and hence several numerical methods where

proposed [9]. Abbasbandy and Allahviranloo [1] developed numerical algorithms for solving fuzzy

diﬀerential equations based on Seikkala’s derivative of fuzzy process introduced in [12]. In this

paper we apply Runge-Kutta method of order 3 to solve fuzzy diﬀerential equations and establish

that this method is better than Euler method. The structure of the paper is organized as follows:

In Section 2, we give some basic deﬁnitions and results. In Section 3, we deﬁne the initial value

problem and discuss the Runge-Kutta method of order 3. In Section 4, we apply the third order

Runge-Kutta method to solve the initial value problem and give the convergence result. Finally

in section 5, we provide some examples to illustrate our results.

2 Preliminaries

Consider the initial value problem

_

y

(t) = f(t, y(t)); a ≤ t ≤ b,

y(a) = α.

(2.1)

Corresponding author: K. Kanagarajan

2

The basis of all Runge-Kutta method is to express the diﬀerence between the value of y at t

n+1

and t

n

as

y

n+1

−y

n

=

m

i=1

w

i

k

i

, (2.2)

where w

i

’s are constants and for i = 1, 2, · · · m,

k

i

= hf

_

_

t

n

+c

i

h, y

n

+h

i−1

j=1

a

ij

k

j

_

_

. (2.3)

Equations (2.2) is to be exact for powers of h through h

m

, because it is to be coincident with

Taylor series of order m. Therefore, the truncation error T

m

, can be writtern as

T

m

= γ

m

h

m+1

+O(h

m+2

).

The true magnitude of γ

m

will generally be much less than the bound of Theorem 2.1. Thus, if

the O(h

m+2

) term is small compared with γ

m

h

m+1

, as we expect, to be so if h is small, then the

bound on γ

m

h

m+1

, will usually be a bound on the error as a whole. The famous nonzero constants

c

i

, a

ij

in the Runge-Kutta method of order three are

c

1

= 0, c

2

= 1/2, c

3

= 1, a

21

= 1/2, a

31

= −1, a

32

= 2,

where m = 3. Hence we have, see[11]

k

1

= hf

_

t

i

, y

i

_

,

k

2

= hf

_

t

i

+

h

2

, y

i

+

1

2

k

1

_

,

k

3

= hf (t

i

+h, y

i

−k

1

+ 2k

2

) ,

y

i+1

= y

i

+

1

6

(k

1

+ 4k

2

+k

3

),

(2.4)

where

a = t

0

≤ t

1

≤ · · · ≤ t

N

= b and h =

(b −a)

N

= t

i+1

−t

i

. (2.5)

Theorem 2.1. Let f(t, y) belongs to C

3

[a, b] and it’s partial derivatives are bounded and assume

that there exists positive constants L, M, such that

|f(t, y)| < M,

¸

¸

¸

¸

∂

i+j

f

∂t

i

∂y

j

¸

¸

¸

¸

<

L

i+j

M

j−1

, i +j ≤ m.

Then in the Runge-Kutta method of order 3, we have, see [11]

y(t

i+1

) −y

i+1

≈ γ

3

h

4

+ O(h

5

),

y(t

i+1

) −y

i+1

≈

13

36

h

4

ML

3

+ O(h

5

).

A triangular fuzzy number v, is deﬁned by three numbers a

1

< a

2

< a

3

where the graph of v(x),

the membership function of the fuzzy number v, is a triangle with base on the interval [a

1

, a

3

] and

vertex at x = a

2

. We specify v as (a

1

/a

2

/a

3

). We will write: (1) v > 0 if a

1

> 0; (2) v ≥ 0 if

a

1

≥ 0; (3) v < 0 if a

3

< 0; and (4) v ≤ 0 if a

3

≤ 0.

3

Let E be the set of all upper semicontinuous normal convex fuzzy numbers with bounded

r−level intervals. It means that is v ∈ E then r−level set

[v]

r

= {s | v(s) ≥ r}, 0 < r ≤ 1,

is a closed bounded interval which is denoted by

[v]

r

= [v

1

(r), v

2

(r)].

Let I be a real interval. A mapping x : I → E is called a fuzzy process and its r−level set is

denoted by

[x(t)]

r

= [x

1

(t; r), x

2

(t; r)], t ∈ I, r ∈ (0, 1].

The derivative x

**(t) of a fuzzy process x(t) is deﬁned by
**

[x

(t)]

r

= [x

1

(t; r), x

2

(t; r)], t ∈ I, r ∈ (0, 1],

provided that this equation deﬁnes a fuzzy number, as in Seikkala[13].

Lemma 2.1. Let v, w ∈ E and s scalar, then for r ∈ (0, 1]

[v +w]

r

= [v

1

(r) +w

1

(r), v

2

(r) +w

2

(r)],

[v −w]

r

= [v

1

(r) −w

1

(r), v

2

(r) −w

2

(r)],

[v · w]

r

=

_

min{v

1

(r) · w

1

(r), v

1

(r) · w

2

(r), v

2

(r) · w

1

(r), v

2

(r) · w

2

(r)},

max{v

1

(r) · w

1

(r), v

1

(r) · w

2

(r), v

2

(r) · w

1

(r), v

2

(r) · w

2

(r)}

_

,

[sv]

r

= s[v]

r

.

3 Fuzzy Cauchy Problem

Consder the fuzzy initial value problem

_

y

(t) = f(t, y(t)), t ∈ I = [0, T],

y(0) = y

0

,

(3.1)

where f is a continuous mapping from R

+

×R into R and y

0

∈ E with r-level sets

[y

0

]

r

= [y

1

(0; r), y

2

(0; r)], r ∈ (0, 1].

The extension principle of Zadeh leads to the following deﬁnition of f(t, y) when y = y(t) is a fuzzy

number

f(t, y)(s) = sup{y(τ)|s = f(t, r)}, s ∈ R.

It follows that

[f(t, y)]

r

= [f

1

(t, y; r), f

2

(t, y; r)], r ∈ (0, 1],

where

f

1

(t, y; r) = min{f(t, u)| u ∈ [y

1

(r), y

2

(r)]},

f

2

(t, y; r) = max{f(t, u)| u ∈ [y

1

(r), y

2

(r)]}.

(3.2)

Theorem 3.1.[12] Let f satisfy

|f(t, v) −f(t, v)| ≤ g(t, |v −v|), t ≥ 0, v, v ∈ R,

where g : R

+

×R

+

is a continuous mapping such that r →g(t, r) is nondecreasing and the initial

value problem

u

(t) = g(t, u(t)), u(0) = u

0

, (3.3)

has a solution on R

+

for u

0

> 0 and that u(t) = 0 is the only solution of (3.3) for u

0

= 0. Then

the fuzzy initial value problem (3.1) has a unique solution.

4

4 Runge-Kutta method of order 3

Let the exact solution [Y (t)]

r

= [Y

1

(t; r), Y

2

(t; r)] is approximated by some [y(t)]

r

= [y

1

(t; r), y

2

(t, r)].

From (2.2),(2.3) we deﬁne

y

1

(t

n+1

; r) −y

1

(t

n

; r) =

3

i=1

w

i

k

i,1

(t

n

, y(t

n

; r)),

y

2

(t

n+1

; r) −y

2

(t

n

; r) =

3

i=1

w

i

k

i,2

(t

n

, y(t

n

; r)),

(4.1)

where the w

i

’s are constants and

[k

i

(t, y(t; r))]

r

= [k

i,1

(t, y(t; r), k

i,2

(t, y(t; r))], i = 1, 2, 3

k

i,1

(t

n

, y(t

n

; r)) = hf

_

_

t

n

+c

i

h , y

1

(t

n

) +

i−1

j=1

a

ij

k

j,1

(t

n

, y(t

n

; r))

_

_

,

k

i,2

(t

n

, y(t

n

; r)) = hf

_

_

t

n

+c

i

h , y

2

(t

n

) +

i−1

j=1

a

ij

k

j,2

(t

n

, y(t

n

; r))

_

_

,

(4.2)

and

k

1,1

(t, y(t; r)) = min

_

hf (t, u) |u ∈ [y

1

(t; r), y

2

(t; r)]

_

,

k

1,2

(t, y(t; r)) = max

_

hf (t, u) |u ∈ [y

1

(t; r), y

2

(t; r)]

_

,

k

2,1

(t, y(t; r)) = min

_

hf

_

t +

1

2

h, u

_

¸

¸

¸u ∈ [z

1,1

(t, y(t; r)), z

1,2

(t, y(t; r)]

_

,

k

2,2

(t, y(t; r)) = max

_

hf

_

t +

1

2

h, u

_

¸

¸

¸u ∈ [z

1,1

(t, y(t; r)), z

1,2

(t, y(t; r)]

_

,

k

3,1

(t, y(t; r)) = min

_

hf (t +h, u)

¸

¸

¸u ∈ [z

2,1

(t, y(t; r)), z

2,2

(t, y(t; r)]

_

,

k

3,2

(t, y(t; r)) = max

_

hf (t +h, u)

¸

¸

¸u ∈ [z

2,1

(t, y(t; r)), z

2,2

(t, y(t; r)]

_

,

(4.3)

where in the Runge-Kutta method of order 3

z

1,1

(t, y(t; r)) = y

1

(t; r) +

1

2

k

1,1

(t, y(t; r)),

z

1,2

(t, y(t; r)) = y

2

(t; r) +

1

2

k

1,2

(t, y(t; r)),

z

2,1

(t, y(t; r)) = y

1

(t; r) −k

1,1

(t, y(t; r)) + 2k

2,1

(t, y(t; r)),

z

2,2

(t, y(t; r)) = y

2

(t; r) −k

1,2

(t, y(t; r)) + 2k

2,2

(t, y(t; r)).

(4.4)

Deﬁne

F[t, y(t; r)] = k

1,1

(t, y(t; r) + 4k

2,1

(t, y(t; r)) +k

3,1

(t, y(t; r)),

G[t, y(t; r)] = k

1,2

(t, y(t; r) + 4k

2,2

(t, y(t; r)) +k

3,2

(t, y(t; r)).

(4.5)

The exact and approximate solutions at t

n

, 0 ≤ n ≤ N are denoted by [Y (t

n

)]

r

= [Y

1

(t

n

; r), Y

2

(t

n

; r)]

and [y(t

n

)]

r

= [y

1

(t

n

; r), y

2

(t

n

; r)] respectively. The solution is calculated by grid points at (2.5).

5

By (4.1) and (4.5) we have

Y

1

(t

n+1

; r) ≈ Y

1

(t

n

; r) +

1

6

F [t

n

, Y (t

n

; r)] ,

Y

2

(t

n+1

; r) ≈ Y

2

(t

n

; r) +

1

6

G[t

n

, Y (t

n

; r))] .

(4.6)

We deﬁne

y

1

(t

n+1

; r) = y

1

(t

n

; r) +

1

6

F [t

n

, y(t

n

; r)] ,

y

2

(t

n+1

; r) = y

2

(t

n

; r) +

1

6

G[t

n

, y(t

n

; r)] .

(4.7)

The following lemmas will be applied to show convergence of these approximates. That is

lim

h→0

y

1

(t; r) = Y

1

(t; r),

lim

h→0

y

2

(t; r) = Y

2

(t; r).

Lemma 4.1.[9] Let the sequence of numbers {W

n

}

N

n=0

satisfy

|W

n+1

| ≤ A|W

n

| +B, 0 ≤ n ≤ N −1,

for some give positive constants A and B. Then

|W

n

| ≤ A

n

|W

0

| +B

A

n

−1

A−1

, 0 ≤ n ≤ N.

Lemma 4.2.[9] Let the sequence of numbers{W

n

}

N

n=0

, {V

n

}

N

n=0

satisfy

|W

n+1

| ≤ |W

n

| +Amax{|W

n

|, |V

n

|} +B,

|V

n+1

| ≤ |V

n

| +Amax{|W

n

|, |V

n

|} +B,

for some given positive constants A and B, and denote

U

n

= |W

n

| +|V

n

|, 0 ≤ n ≤ N.

Then

U

n

≤ A

n

U

0

+B

A

n

−1

A−1

, 0 ≤ n ≤ N,

where A = 1 + 2A and B = 2B.

Let F(t, u, v) and G(t, u, v) are obtained by substituting [y(t)]

r

= [u, v] in (4.5),

F[t, y(t; r)] = k

1,1

(t, y(t; r)) + 4k

2,1

(t, y(t; r)) +k

3,1

(t, y(t; r)),

G[t, y(t; r)] = k

1,2

(t, y(t; r)) + 4k

2,2

(t, y(t; r)) +k

3,1

(t, y(t; r)).

The domain where F and G are deﬁnd is therefore

K = {(t, u, v)|0 ≤ t ≤ T, −∞< v < ∞ , −∞< u ≤ v}.

Theorem 4.1. Let F(t, u, v) and G(t, u, v) belong to C

3

(k) and let the partial derivatives of F

and G be bounded over K. Then, for arbitrary ﬁxed r, 0 ≤ r ≤ 1, the approximately solutions(4.6)

converge to the exact solutions Y

1

(t; r) and Y

2

(t; r) uniformly in t.

Proof: It is suﬃcient to show

lim

h→0

y

1

(t

N

; r) = Y

1

(t

N

; r),

lim

h→0

y

2

(t

N

; r) = Y

2

(t

N

; r),

6

where t

N

= T. For n = 0, 1, · · · , N −1, by using Taylor theorem we get

Y

1

(t

n+1

; r) = Y

1

(t

n

; r) +

1

6

F [t

n

, Y (t

n

; r)] +

13

36

h

4

ML

3

+O(h

5

),

Y

2

(t

n+1

; r) = Y

2

(t

n

; r) +

1

6

G[t

n

, Y (t

n

; r))] +

13

36

h

4

ML

3

+O(h

5

),

(4.8)

W

n

= Y

1

(t

n

; r) −y

1

(t

n

; r),

V

n

= Y

2

(t

n

; r) −y

2

(t

n

; r).

Hence from (4.7) and (4.8)

W

n+1

= W

n

+

1

6

{F [t

n

, Y

1

(t

n

; r), Y

2

(t

n

; r)] −F [t

n

, y

1

(t

n

; r), y

2

(t

n

; r)]}

+

13

36

h

4

ML

3

+O(h

5

),

V

n+1

= V

n

+

1

6

{G[t

n

, Y

1

(t

n

; r), Y

2

(t

n

; r)] −G[t

n

, y

1

(t

n

; r), y

2

(t

n

; r)]}

+

13

36

h

4

ML

3

+O(h

5

).

Then

|W

n+1

| ≤ |W

n

| +

1

3

Ph · max{|W

n

|, |V

n

|} +

13

36

h

4

ML

3

+O(h

5

),

|V

n+1

| ≤ |V

n

| +

1

3

Ph · max{|W

n

|, |V

n

|} +

13

36

h

4

ML

3

+O(h

5

),

for t ∈ [0, T] and P > 0 is a bound for the partial derivatives of F and G. Thus by Lemma 4,

|W

n

| ≤ (1 +

2

3

Ph)

n

|U

0

| +

_

13

18

h

4

ML

3

+O(h

5

)

_

(1 +

2

3

Ph)

n

−1

2

3

Ph

,

|V

n

| ≤ (1 +

2

3

Ph)

n

|U

0

| +

_

13

18

h

4

ML

3

+O(h

5

)

_

(1 +

2

3

Ph)

n

−1

2

3

Ph

,

where |U

0

| = |W

0

| +|V

0

|. In particular

|W

N

| ≤ (1 +Ph)

N

|U

0

| +

_

13

12

h

3

ML

3

+O(h

4

)

_

(1 +

2

3

Ph)

T

h

−1

P

,

|V

N

| ≤ (1 +Ph)

N

|U

0

| +

_

13

12

h

3

ML

3

+O(h

4

)

_

(1 +

2

3

Ph)

T

h

−1

P

.

Since W

0

= V

0

= 0, we obtain

|W

N

| ≤

13

12

ML

3

_

e

2

3

Ph

−1

P

_

h

3

+O(h

4

),

|V

N

| ≤

13

12

ML

3

_

e

2

3

Ph

−1

P

_

h

3

+O(h

4

),

and if h →0 we get W

N

→0 and V

N

→0 which completes the proof.

7

5 Numerical Examples

Example 5.1. Consider the fuzzy diﬀerential equation

_

y

(t) = y(t), t ≥ 0,

y(0) = [0.75 + 0.25r, 1.5 −0.5r].

(5.1)

The exact solution is given by

Y (t; r) =

_

(0.75 + 0.25r)e

t

, (1.5 −0.5r)e

t

¸

.

At t = 0.1 we get

Y (0.1; r) =

_

(0.75 + 0.25r)e

0.1

, (1.5 −0.5r)e

0.1

¸

.

By using Runge-kutta method of order 3, the approximate solution is

y

1

(t

n+1

; r) = y

1

(t

n

; r)

_

1 +h +

h

2

2!

+

h

3

3!

_

,

y

2

(t

n+1

; r) = y

2

(t

n

; r)

_

1 +h +

h

2

2!

+

h

3

3!

_

.

The exact, approximate solutions by Euler method and the Runge-Kutta method of order 3 are

compared and plotted in ﬁgure 1.

0.8 1 1.2 1.4 1.6 1.8

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

t

r

Exact

Euler

RK 3rd order

Figure 1: h = 0.5

0.8 1 1.2 1.4 1.6 1.8

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

t

r

Exact

Euler

RK 3rd order

Figure 2: h = 0.5

Example 5.2. Consider the second order fuzzy diﬀerential equation

_

_

_

y

(t) = ty(t), t ∈ [0, 1],

y(0) =

_

√

e −0.5(1 −r),

√

e + 0.5(1 −r)

_

.

(5.2)

The exact solution is given by

Y (t; r) =

_

(

√

e −0.5(1 −r))e

t

2

2

, (

√

e + 0.5(1 −r))e

t

2

2

_

.

At t = 0.1 we get

Y (0.1; r) =

_

(

√

e −0.5(1 −r))e

0.005

, (

√

e + 0.5(1 −r))e

0.005

¸

.

The exact and approximate solutions by using Runge-Kutta method of order 3 equations (4.3-

??eq15]), are compared and plotted in ﬁgure 2.

8

6 Conclusions

In this work we applied Runge-Kutta method of order 3 for ﬁnding the numerical solution

of fuzzy diﬀerential equations. Taking into account the convergence order of the Euler method is

O(h) (as given in [9]), a higher order of convergence O(h

3

) is obtained by the proposed method.

Comparison of solutions of examples (5.1) and (5.2) shows that our proposed method gives better

solution than Euler method.

Acknowledgements

This work has been supported by University Grants Commission, MRP-3346/10, SERO, Hy-

derabad, India.

References

[1] S. Abbasbandy, T. Allahviranloo, Numerical solution of fuzzy diﬀerential equations by Taylor method, Journal

of Computational Methods and Applied Mathematics, 2(2002)113-124.

[2] K. Balachandran, P. Prakash, Existence of solutions of fuzzy delay diﬀerential equations with nonlocal condition,

Journal of the Korea Society for Industrial and Applied Mathematics, 6(2002)81-89.

[3] K. Balachandran, K. Kanagarajan, Existence of solutions of fuzzy delay integrodiﬀerential equations with

nonlocal condition, Journal of the Korea Society for Industrial and Applied Mathematics, 9(2005)65-74.

[4] S.L. Chang, L.A. Zadeh, On fuzzy mapping and control, IEEE Trans, Systems Man Cybernet. 2(1972)30-34.

[5] D. Dubois, H. Prade, Towards fuzzy diﬀerential calculus part 3 : Diﬀerentiation, Fuzzy Sets and Systems,

8(1982)225-233.

[6] O. Kaleva, Fuzzy diﬀerential equations, Fuzzy Sets and Systems, 24(1987)301-317.

[7] O. Kaleva, The Cauchy problem for fuzzy diﬀerential equations, Fuzzy Sets and Systems, 35(1990)389-396.

[8] V. Lakshmikantham and R. Mohapatra, Theory of Fuzzy Diﬀerential Equations and Inclutions, Taylor and

Francis, London,(2005).

[9] M. Ma, M. Friedman, A. Kandel, Numerical solutions of fuzzy diﬀerential equations, Fuzzy Sets and Systems,

105(1999)133-138.

[10] M.L. Puri, D.A. Ralescu, Diﬀerentials of fuzzy functions, Journal of Mathematical Analysis and Applications,

91(1983)552-558.

[11] A. Ralston, P. Rabinowitz, First Course in Numerical Analysis, McGraw Hill International Edition, (1978).

[12] S. Seikkala, On the fuzzy initial value problem, Fuzzy Sets and Systems, 24(1987)319-330.

We specify v as (a1 /a2 /a3 ). aij in the Runge-Kutta method of order three are c1 = 0. where m = 3. ∂ti ∂y j M Then in the Runge-Kutta method of order 3. M. · · · m. c3 = 1. Li+j ∂ i+j f < j−1 .1. The famous nonzero constants ci . i=1 (2. 13 4 h M L3 + O(h5 ). where a = t0 ≤ t1 ≤ · · · ≤ tN = b and h = (b − a) = ti+1 − ti . 1 yi+1 = yi + 6 (k1 + 4k2 + k3 ). the truncation error Tm .4) k3 = hf (ti + h. if the O(hm+2 ) term is small compared with γm hm+1 . Let f (t. 36 A triangular fuzzy number v. will usually be a bound on the error as a whole. Hence we have. then the bound on γm hm+1 . a3 ] and vertex at x = a2 . 2. c2 = 1/2. see [11] y(ti+1 ) − yi+1 y(ti+1 ) − yi+1 ≈ ≈ γ3 h4 + O(h5 ). We will write: (1) v > 0 if a1 > 0. a32 = 2. see[11] k1 = hf ti . we have.2) is to be exact for powers of h through hm . The true magnitude of γm will generally be much less than the bound of Theorem 2. the membership function of the fuzzy number v. a31 = −1. (2.2) i−1 Equations (2. i + j ≤ m. yi − k1 + 2k2 ) . a21 = 1/2.2 The basis of all Runge-Kutta method is to express the diﬀerence between the value of y at tn+1 and tn as m yn+1 − yn = where wi ’s are constants and for i = 1. Therefore. b] and it’s partial derivatives are bounded and assume that there exists positive constants L. . such that |f (t. (3) v < 0 if a3 < 0. (2) v ≥ 0 if a1 ≥ 0. is a triangle with base on the interval [a1 . can be writtern as Tm = γm hm+1 + O(hm+2 ). 1 k2 = hf ti + h . and (4) v ≤ 0 if a3 ≤ 0. to be so if h is small. yi . is deﬁned by three numbers a1 < a2 < a3 where the graph of v(x).5) Theorem 2. w i ki . 2 ki = hf tn + ci h. y)| < M. because it is to be coincident with Taylor series of order m. as we expect. Thus. y) belongs to C 3 [a.3) (2.1. N (2. yi + 2 k1 . yn + h j=1 aij kj .

r) = max{f (t. It means that is v ∈ E then r−level set [v]r = {s | v(s) ≥ r}. v2 (r) · w1 (r). 1]. is a closed bounded interval which is denoted by [v]r = [v1 (r).2) . v)| ≤ g(t. = s[v]r . 3 Fuzzy Cauchy Problem y (t) = f (t. The extension principle of Zadeh leads to the following deﬁnition of f (t.3) for u0 = 0. r) is nondecreasing and the initial value problem u (t) = g(t. r). 1].3) has a solution on R+ for u0 > 0 and that u(t) = 0 is the only solution of (3. r) = min{f (t. t ∈ I.1) has a unique solution. y2 (0.1. t ≥ 0. A mapping x : I → E is called a fuzzy process and its r−level set is denoted by [x(t)]r = [x1 (t. Let I be a real interval. (3.1. Theorem 3. as in Seikkala[13]. It follows that where [f (t. w ∈ E and s scalar. The derivative x (t) of a fuzzy process x(t) is deﬁned by [x (t)]r = [x1 (t. provided that this equation deﬁnes a fuzzy number. r)]. t ∈ I = [0. v1 (r) · w2 (r). r)]. f2 (t. t ∈ I. r ∈ (0. Then the fuzzy initial value problem (3. y2 (r)]}. r)]. x2 (t. r). y(t)). v2 (r)]. y)(s) = sup{y(τ )|s = f (t. r). where g : R+ × R+ is a continuous mapping such that r → g(t. r ∈ (0. max{v1 (r) · w1 (r). s ∈ R. Lemma 2.1) where f is a continuous mapping from R+ × R into R and y0 ∈ E with r-level sets [y0 ]r = [y1 (0. u(t)). y)]r = [f1 (t. u(0) = u0 . y2 (r)]}. 1] [v + w]r [v − w]r [v · w]r [sv]r = [v1 (r) + w1 (r). v2 (r) · w1 (r). v2 (r) · w2 (r)} . v2 (r) + w2 (r)]. (3. v2 (r) − w2 (r)]. y. u)| u ∈ [y1 (r). r ∈ (0. v. y) when y = y(t) is a fuzzy number f (t. f1 (t. |v − v|). r). T ]. v) − f (t. Let v. y. y. y. 0 < r ≤ 1. Consder the fuzzy initial value problem (3. r ∈ (0. r)}. r)]. y(0) = y0 . v2 (r) · w2 (r)}. then for r ∈ (0.[12] Let f satisfy |f (t.3 Let E be the set of all upper semicontinuous normal convex fuzzy numbers with bounded r−level intervals. = [v1 (r) − w1 (r). v ∈ R. x2 (t. u)| u ∈ [y1 (r). f2 (t. = min{v1 (r) · w1 (r). 1]. v1 (r) · w2 (r). 1].

r)) = max hf t + 1 h. r) − k1. r)] .1 (t. r)) . y(t. .2 (t.2 (t. y(t.2 (t. y2 (tn ) + j=1 i−1 aij kj. r)).1 (t. r)) = max hf (t. i=1 3 (4. r)). 2 z2. r)] . r) + 1 k1. y(t.1 (t. G[t. y(tn . r)) + 2k2. r)). y(t.1 (tn . y1 (tn ) + ki.5) The exact and approximate solutions at tn . i = 1. 1 k2. r)) = y2 (t. u) |u ∈ [y1 (t. r)). y(tn .1 (t. r)) = min hf (t + h. aij kj. u u ∈ [z1.2 (t. y(t. r) + 4k2. z1. r)). r). r).4 4 Runge-Kutta method of order 3 Let the exact solution [Y (t)]r = [Y1 (t. r) − y1 (tn . r)] is approximated by some [y(t)]r = [y1 (t. 2.2 (t. y(t. y(tn .(2.1 (t. u) u ∈ [z2. where in the Runge-Kutta method of order 3 1 z1. y(t. r) + 2 k1. From (2.2 (t. y(t. r)). y(t. y(t. The solution is calculated by grid points at (2. y(t.2 (t. y(t. r)). y(t. y(t.1 (tn .1 (t. r)). z1. r)) = hf tn + ci h . r)] . r))]r = [ki. r)] = k1.5). y(tn . y(t. r).1 (t.1 (t. y(t. r)) = min hf (t. Y2 (tn . y(t. y2 (t. y(t.2 (tn . (4.1) wi ki. r)] .1 (t. r)) . k1. r).2 (t. r) − k1. Y2 (t.2 (t.2) j=1 and k1. y(t. u) u ∈ [z2. r)). r)] respectively. (4. r)] and [y(tn )]r = [y1 (tn .3) k2.1 (t. r)) = min hf t + 2 h. y2 (t. r)] . r)) = hf tn + ci h . r)). r)) + 2k2. r)) = y2 (t. z1. r) − y2 (tn . r)]. z2. y(t. y(t. y(t. y(t. r). r)).2 (t. y(t. y(t.1 (t. k3. z2. y(t. y(tn . r) + 4k2. r)) + k3. y(t.1 (t.4) (4. (4. 0 ≤ n ≤ N are denoted by [Y (tn )]r = [Y1 (tn .2 (t. r)] . y(t. r) = where the wi ’s are constants and wi ki.2 (t. y(tn . r)] = k1. r).2).1 (t. r)) + k3. y(t. y(t.2 (t.1 (t. r) = y2 (tn+1 . y2 (t. r)) = y1 (t. y(t.2 (tn . u) |u ∈ [y1 (t.2 (t.2 (t.2 (t. r)). z2. u u ∈ [z1. 3 i−1 ki.2 (tn . y(t. 2 k3.1 (t. r)) = y1 (t. y(t. y(t. Deﬁne F [t. r). r)) = max hf (t + h. y(t. ki. i=1 [ki (t.1 (t.3) we deﬁne 3 y1 (tn+1 .1 (tn . y2 (tn . r))].

Let F (t. for some given positive constants A and B. (4. r). v)|0 ≤ t ≤ T. u. y(t. v) belong to C 3 (k) and let the partial derivatives of F and G be bounded over K. Y (tn . y(tn . Then. for some give positive constants A and B.[9] Let the sequence of numbers{Wn }N . r).7) The following lemmas will be applied to show convergence of these approximates. r) + 1 F [tn . v) are obtained by substituting [y(t)]r = [u. F [t. −∞ < u ≤ v}. r)) + 4k2. v) and G(t.1. 0 ≤ n ≤ N − 1. −∞ < v < ∞ . the approximately solutions(4. r) = Y1 (tN . r)] . 6 1 y2 (tn+1 . r)). r)).5) we have Y1 (tn+1 . r)) + k3. Theorem 4. Let F (t.6) We deﬁne y1 (tn+1 . r) = y1 (tn . 0 ≤ n ≤ N. y(t. The domain where F and G are deﬁnd is therefore K = {(t.2. r) and Y2 (t. G[t. r) = y2 (tn . r) = Y2 (t. r) = Y2 (tN . limh→0 y2 (tN . r)] = k1. A−1 . u. y(t. 0 ≤ r ≤ 1.[9] Let the sequence of numbers {Wn }N satisfy n=0 |Wn+1 | ≤ A|Wn | + B.1 (t. u. Y (tn .1) and (4. y(t. h→0 lim y2 (t. Proof: It is suﬃcient to show limh→0 y1 (tN . and denote Un = |Wn | + |Vn |. for arbitrary ﬁxed r. r) + 6 G [tn .2 (t. |Vn |} + B. Then |Wn | ≤ An |W0 | + B An − 1 . {Vn }N satisfy n=0 n=0 |Wn+1 | ≤ |Wn | + A max{|Wn |. r) ≈ Y1 (tn .1 (t. |Vn |} + B. 0 ≤ n ≤ N.6) converge to the exact solutions Y1 (t. y(t. u. r)] . Then Un ≤ A U0 + B n n 0 ≤ n ≤ N. |Vn+1 | ≤ |Vn | + A max{|Wn |. r).5 By (4. Lemma 4. r)] = k1. r).5). r)) + 4k2. A −1 . y(t.1 (t. y(t. That is h→0 lim y1 (t.2 (t.1. r))] . A−1 Lemma 4. 6 1 Y2 (tn+1 . y(t. r)) + k3. v] in (4. (4. y(tn .1 (t. r) uniformly in t. u. where A = 1 + 2A and B = 2B. r) = Y1 (t. v) and G(t. r) ≈ Y2 (tn . r) + 1 F [tn . r) + 6 G [tn . r)] .

y2 (tn . 3 36 13 4 1 |Vn+1 | ≤ |Vn | + P h · max{|Wn |. Y (tn .6 where tN = T . r). Y (tn . P T T N h3 + O(h4 ). we obtain |WN | ≤ |VN | ≤ 13 M L3 12 13 M L3 12 e3Ph − 1 P e3Ph − 1 P 2 2 13 4 h M L3 + O(h5 ) 18 13 4 h M L3 + O(h5 ) 18 2 (1 + 3 P h)n − 1 . and if h → 0 we get WN → 0 and VN → 0 which completes the proof. 2 3Ph 13 3 h M L3 + O(h4 ) 12 13 3 h M L3 + O(h4 ) 12 2 (1 + 3 P h) h − 1 . r). r)]} 6 13 + h4 M L3 + O(h5 ). r) = Y2 (tn .8) Wn+1 = Wn + 1 {F [tn . Thus by Lemma 4. r). Hence from (4. r) − y1 (tn . Y2 (tn . r) = Y1 (tn . Vn = Y2 (tn .8) Wn = Y1 (tn . |Vn |} + h M L3 + O(h5 ). r)]} 6 13 + h4 M L3 + O(h5 ). r). 1 Y2 (tn+1 . 36 1 = Vn + {G [tn . 1. y1 (tn . r). P 2 (1 + 3 P h) h − 1 . · · · . Y2 (tn . + O(h5 ). In particular |WN | ≤ (1 + P h)N |U0 | + |VN | ≤ (1 + P h) |U0 | + Since W0 = V0 = 0. N − 1. |Vn |} + h4 M L3 + O(h5 ). by using Taylor theorem we get 1 Y1 (tn+1 .7) and (4. 2 3Ph 2 (1 + 3 P h)n − 1 . y2 (tn . . Y1 (tn . 2 |Wn | ≤ (1 + P h)n |U0 | + 3 2 |Vn | ≤ (1 + P h)n |U0 | + 3 where |U0 | = |W0 | + |V0 |. Y1 (tn . h3 + O(h4 ). r)] + 13 4 3 36 h M L + O(h5 ). r) + 6 F [tn . For n = 0. T ] and P > 0 is a bound for the partial derivatives of F and G. 36 Vn+1 Then 1 13 |Wn+1 | ≤ |Wn | + P h · max{|Wn |. r))] + 13 4 3 36 h M L (4. r)] − G [tn . r) + 6 G [tn . r) − y2 (tn . r)] − F [tn . r). 3 36 for t ∈ [0. y1 (tn .

+ y2 (tn+1 .5 − 0.5(1 − r) .8 1 1.3 0. ( e + 0.5 0.3??eq15]). Example 5.5(1 − r))e 2 . t ≥ 0.2 t 1.75 + 0. r) 1 + h + 2! 3! Y (0.5 0.8 0. At t = 0.005 . .5(1 − r))e0. 1].4 r Exact Euler RK 3rd order 1.6 0. Consider the second order fuzzy diﬀerential equation y (t) = ty(t). The exact. (1.4 0.7 0. √ √ t2 t2 Y (t.2.9 0.6 1.5 Figure 2: h = 0. √ √ y(0) = e − 0. 1 0. The exact solution is given by Y (t.6 0. the approximate solution is y1 (tn+1 . + 2! 3! h3 h2 . r) = y2 (tn .5 − 0.1.75 + 0.1) y(0) = [0.9 0.1 we get By using Runge-kutta method of order 3.5(1 − r).5 − 0. (5.2 0.8 0.1 0 0.1 .1 . r) = (0. approximate solutions by Euler method and the Runge-Kutta method of order 3 are compared and plotted in ﬁgure 1.2 t 1.3 0.7 0. r) = ( e − 0. The exact and approximate solutions by using Runge-Kutta method of order 3 equations (4.25r.6 1.1 0 0.2 0.4 1. r) = ( e − 0.7 5 Numerical Examples y (t) = y(t). ( e + 0. r) = (0.4 0.5(1 − r))e0.5r)et .8 Figure 1: h = 0.5(1 − r))e 2 .5r)e0. (1. r) = y1 (tn . r) 1 + h + h2 h3 .25r)et .1.25r)e0.1. t ∈ [0.8 1 1.75 + 0.005 . Consider the fuzzy diﬀerential equation (5. 1.8 r 1 Exact Euler RK 3rd order 0.2) At t = 0.5r].1 we get √ √ Y (0.5 The exact solution is given by Example 5. e + 0. are compared and plotted in ﬁgure 2.

Puri. IEEE Trans. Seikkala. [7] O. 8(1982)225-233.(2005). Balachandran. [3] K.1) and (5. Towards fuzzy diﬀerential calculus part 3 : Diﬀerentiation. Balachandran. Journal of the Korea Society for Industrial and Applied Mathematics. Prade. [2] K. A. 2(1972)30-34. Zadeh. 105(1999)133-138.L. India. [6] O. T. [8] V. Abbasbandy. 91(1983)552-558. 35(1990)389-396. Kanagarajan. Kaleva. Theory of Fuzzy Diﬀerential Equations and Inclutions. Rabinowitz. 24(1987)301-317. Journal of the Korea Society for Industrial and Applied Mathematics. P. L. Chang. Allahviranloo. Fuzzy Sets and Systems. Taking into account the convergence order of the Euler method is O(h) (as given in [9]). . Journal of Mathematical Analysis and Applications. [12] S. 6(2002)81-89. 9(2005)65-74. [10] M. Fuzzy Sets and Systems. The Cauchy problem for fuzzy diﬀerential equations. [11] A. Kandel.A. Taylor and Francis.2) shows that our proposed method gives better solution than Euler method. Fuzzy diﬀerential equations. Fuzzy Sets and Systems.L. MRP-3346/10. Existence of solutions of fuzzy delay diﬀerential equations with nonlocal condition. First Course in Numerical Analysis. [5] D. Lakshmikantham and R. Hyderabad. Mohapatra. Fuzzy Sets and Systems. (1978). On fuzzy mapping and control. Journal of Computational Methods and Applied Mathematics. Numerical solution of fuzzy diﬀerential equations by Taylor method. Dubois. Diﬀerentials of fuzzy functions. London. a higher order of convergence O(h3 ) is obtained by the proposed method. McGraw Hill International Edition. Acknowledgements This work has been supported by University Grants Commission. [4] S. P. Comparison of solutions of examples (5. Kaleva. Systems Man Cybernet. References [1] S. Ralescu. Prakash. Ralston. Numerical solutions of fuzzy diﬀerential equations.A. Fuzzy Sets and Systems. On the fuzzy initial value problem. D. SERO. M. 2(2002)113-124. Existence of solutions of fuzzy delay integrodiﬀerential equations with nonlocal condition. Friedman. [9] M. Ma. 24(1987)319-330.8 6 Conclusions In this work we applied Runge-Kutta method of order 3 for ﬁnding the numerical solution of fuzzy diﬀerential equations. K. H.