Вы находитесь на странице: 1из 14

1

Some notes on Solving Delay Differential Equations by Continuous Runge-Kutta Method


using Hermite Interpolation

1
Nazeeruddin Yaacob and
2
Lim Tian Hwee
Department of Mathematics, Faculty of Science, Universiti Teknology Malaysia, 81310
UTM Skudai, Johor Darul Tazim.
e-mail:
1
ny@mel.fs.utm.my,
2
stken2002@yahoo.com.sg

Abstract In this report, we solved delay differential equations using continuous Runge-
Kutta method. This type of continuous extension was developed by Zennaro [11], namely the
natural continuous extensions (NCEs). Instead of using continuous Runge-Kutta method itself,
the delay argument is approximated by using Hermite interpolation. The numerical results based
on the methods are given.


1. Introduction

The first order delay differential equation (DDE) with one delay term can be written as


) ( ) (
)))) ( , ( ( ), ( , ( ) (
t t y
t y t t y t y t f t y


=
= '

)
`

s
>
0
0
t t
t t
(1.1)
where (t) is the initial function, (t,y(t)) is the delay argument and y(t (t,y(t)) is the solution
of the delay term. In recent years, many papers have been written regarding to solve DDEs (see
for example, [3], [5], [6], [7], [8], [9], [10]). In this report, we solve DDEs using a particular
class of continuous Runge-Kutta method which developed by Zennaro [11] using Hermite
interpolation.


2. Continuous Runge-Kutta Methods

Recall that an s-stage explicit Runge-Kutta (ERK) method has the form

=
=
+
+ + =
+ =
1
0
0
1
) , (
i
j
j ij n i i i
s
i
i i n n
K a y h c t f K
K b h y y
(2.1)

id141733093 pdfMachine by Broadgun Software - a great PDF writer! - a great PDF creator! - http://www.pdfmachine.com http://www.broadgun.com
2

As we know, ERK methods (2.1) provide approximations only at discrete point. In order to
obtain continuous approximation of an ODE, one strategy is to supply an already existing
discrete Runge-Kutta method with an interpolant. Through out the year, different approaches
have been proposed to obtain continuous extension (or interpolant) of Runge-Kutta methods
which are able to produce solution values at all intermediate points h t t + =
0
*

with 1 0 s s .
According to [1], interpolants can be classify into first and second class. Here, we focus our
attention on the first class interpolants proposed by Zennaro [11].


2.1. Natural Continuous Extension

Zennaro [11] developed a particular type of first class interpolants of Runge-Kutta method,
namely the natural continuous extension (NCEs).

Defination 2.1.1: The s-stage Runge-Kutta method (2.1) of order p has an NCE of degree q if
there exists polynomials ) (
i
b , s i , , 1 = , of degree less than or equal to q, independent of the
function f, such that, by putting

=
=
+ + =
s s + = +
1
0
0
) , (
1 0 ) ( ) (
i
j
j ij n i n i
s
i
i i n n
K a y h c t f K
K b h y h t u
(2.2)

where ) ( h t u
i
+

is a continuous approximation to y(t), satisfies the additional asymptotic
orthogonal condition


| | ( )
1
0
0
) ( ) ( ) (
+
+
= ' '
}
p
h t
t
h O dt t u t y t G

(2.3)

for every sufficient smooth matrix-valued function G and with the following statements hold

) ( ) ( ) ( max
) (
) (
0 0
1 0
0 0
q
h t t t
h O t u t y
y h t u
y t u
= ' '
= +
=
+ s s

(2.4)

3

Like discrete method, equation (2.2) require

=
=
1
1
i
j
ij i
a c , 0 ) 0 ( =
i
b and
i i
b b = ) 1 ( for
s i , , 1 = . It follows that one obtains a discrete method by putting 1 = leading to
1 0
) ( y h t u = + .

The uniform order for the continuous extension (2.2) can be viewed as discrete method
( )

) (
,
b
A
with steplength
1 + n
h . Thus we can get, from [1], the uniform order conditions for
polynomials ) (
i
b

from the well-known order conditions of the Runge-Kutta methods. The
conditions up to order 4 are summarized in Table 2.1.

Table 2.1: Order conditions for Continuous RK methods
Order Conditions Order Conditions
1
=

=
) (
1
s
i
i
b
4

4
24
1
1 , ,
1 ,
4
12
1
2
1 ,
4
8
1
4
4
1
3
1 ,
) (
) (
) (
) (




=
=
=
=

=
=
=
=
s
k j i
k jk ij i
s
j i
j ij i
s
j i
j ij i i
i
s
j i
i
c a a b
c a b
c a c b
c b

2
2
2
1
1
) ( =

=
i
s
i
i
c b
3

=
=
=
=
s
i
j ij i
i
s
i
i
c a b
c b
1
3
6
1
3
3
1 2
1
) (
) (




It is important to note that the collocation polynomial for any one-step collocation method is
an NCE of degree s q = (see [8]). Besides, with regard to the existence of NCEs and their
attainable uniform order, the following theorems hold.


Theorem 2.1: Every Runge-Kutta process (2.1) has a NCE of minimal degree
2
1 +
=
p
q .
Proof : See Zennaro [11].

Theorem 2.2: If the interpolant u(t) in (2.2) of order (and degree) q is an NCE of Runge-Kutta
method (2.2) of order p, then
2
1 +
>
p
q .
Proof : See Bellen and Zennaro [1].




4

Theorem 2.3: Every Runge-Kutta method (2.1) of order p, formula

0 ) 0 ( =
i
b and
i i i
r
c b d b = '
}

1
0
) (

(2.5)

furnished an NCE u(t) of minimal order (and degree)
2
1 +
=
p
q .
Proof : See Bellen and Zennaro [1].

Note that the NCE of an Runge-Kutta method is not unique and that formula (2.5) simply
provides a rule for obtaining one of them. In general, to obtain the NCE of an Runge-Kutta
method, one can use the order conditions of Table 2.1 along with the orthogonality condition
(2.3).

By using information above, we would like to derive NCEs for some ERK method. From [2],
the coefficient tableau for three stage ERK of order 3 = p (with ) 0
3
2
2 3 2
= = = = c c c are
summarized as follows

3
2
0
c
c
( )
( )
( )
( )
2 2
2 3 3
2 2
3
2
2 2 3
2
3 2 3 2
3 3
0
c c
c c c
c c
c c c c
c




( ) ( )
2 3 3
2
2 3 3
3
3 2
2 3 2 3
6
3 2
6
2 3
6
3 2 6 3
c c c
c
c c c
c
c c
c c c c

+ +


From Theorem 2.1, this method has a NCE of minimal degree 2
2
1
= =
+ p
q . Thus we can
write
i i i
b + =
2
) ( , where
i i i
b = since
i i
b b = ) 1 ( , for 3 , , 1 = i , satisfies the order
conditions in Table 2.1, up to order 2. Then we get two equations

=

=
) (
3
1 i
i
b

(2.6)
2
2
1
3
1
) ( =

=
i
i
i
c b

(2.7)

From (2.7) we get


5

) (
3
b ) (
2
1
2
3
2 2
3
b
c
c
c
=
( )



2
2
3
3
2 3
2 2
2
3
2 3 1
3
2 2
2
3
2 2
2
2
3
2 2
3
1
2
1
1 1
2
1
) 1 (
2
1
) (
2
1
c
c
c
b
c c
c
c
c
c b b
c
c
c
c
b
c
c
c
+
|
|
.
|

\
|
|
|
.
|

\
|
|
|
.
|

\
|

|
|
.
|

\
|
=
+
|
|
.
|

\
|
=
+ =



|
|
.
|

\
|
+
|
|
.
|

\
|
=
2
3
3
2
2
3
2
1
2
1
c
c
b c
c
(2.8)

where 9 e =
3
2
c

. Substitute (2.8) to (2.7) we will get



) (
2
b
3 2
2
3 3
3
2 2
2
) (
2
1
c b c b
c
c
c
+ = = (2.9)

Similarly, we can get ) (
1
b by substituting (2.8) and (2.9) to (2.6)


) (
1
b ) ( ) (
3 2
b b =

|
|
.
|

\
|
+
|
|
.
|

\
|
= ) (
2
1
) (
2
1
2 3
3
1
2
2 3
3
c c
c
b c c
c



This is an example of non-uniqueness of NCEs of minimal order q, since we have a one-
parameter family of NCEs. In particular, the polynomial furnished by the conditions (2.5) is


i i i i i
b c b c b ) 3 2 ( 2 ) 1 2 ( 3 ) (
2
+ =

3 , , 1 = i

(2.10)


Which corresponds to the value
2
3
2
) 1 2 ( 3
b
c
c
=

For 3 = = p q , we write
i i i i
b + + =
2 3
) ( . In this case, ) (
i
b , for 3 , , 1 = i , satisfies
the order conditions in Table 2.1, up to order 3. Then we get four equations

6

=

=
) (
3
1 i
i
b

(2.11)

2
2
1
3
1
) ( =

=
i
i
i
c b

(2.12)
3
3
1
2
3
1
) ( =

=
i
i
i
c b

(2.13)

=
=
3
1
3
6
1
) (
i
j ij i
c a b

(2.14)

From the (2.14), we get
3
3 3
) ( b b = and at the end we will get
3
2 2
) ( b b = ,
3
1 1
) ( b b = ,
which do not satisfy the first and second equation. Thus there is no NCE for order 3 = q .

The same way to determine the NCE for ERK method of order 4 = q , which has the form

4
3
2
0
c
c
c

43 42 41
32 31
21
a a a
a a
a



4 3 2 1
b b b b

where

) 1 )( 1 ( 12
3 4 6 4
) )( 1 ( 12
1 2
) )( 1 ( 12
1 2
12
1 2 2 6
2 3
2 2 3 3
4
2 3 3 3
2
3
2 3 2 2
3
2
2 3
2 3 2 3
1

+ +
=


=


=
+
=
c c
c c c c
b
c c c c
c
b
c c c c
c
b
c c
c c c c
b
(2.15)


According to Theorem 2.1, we know that this method has NCE with minimal order
2
2
1
= =
+ p
q . Thus we can write ) ( ) (
2
i i i i
b b + = , for 4 , , 1 = i .

Similarly, the NCE is
furnished by the condition (2.5).

7


i i i i i
b c b c b ) 3 2 ( 2 ) 1 2 ( 3 ) (
2
+ =

4 , , 1 = i

(2.16)


For 3 = q , we can write
i i i i
b + + =
2 3
) ( , for 4 , , 1 = i . We know that the NCE
satisfies the order conditions in Table 2.1 up to order 3. Thus we get four equations

=

=
) (
4
1 i
i
b

(2.17)

2
2
1
4
1
) ( =

=
i
i
i
c b

(2.18)
3
3
1
2
4
1
) ( =

=
i
i
i
c b

(2.19)

=
=
4
1
3
6
1
) (
i
j ij i
c a b

(2.20)

By doing some calculation, we get

+ + =
2 3
1
) 1 3 ( 3 ) 4 1 ( 2 ) (
i i
b b b

2 3
) 4 3 ( 3 ) 2 3 ( 4 ) (
i i i i i
b c b c b + =

4 , , 2 = i

(2.21)


For 4 = = = p s q , like the previous case, we get
4
) (
i i
b b = , for 4 , , 1 = i , which do not
satisfy some order condition in Table 2.1.

Thus, there are no NCEs for order 4 = = = p s q .

By using equation (2.10), (2.16) and (2.21), one can easily derive the NCE for any third and
fourth order ERK method. The following are NCEs of some third and fourth order ERK method.

Table 2.2: Third order ERK method interpolated by its unique NCE of order 2
1
0
2
1

2 1
2
1





3
1
2
2
1
3
3
2
2
3
2
2
2
1
1
) (
) (
) (
=
=
+ =
b
b
b



6
1
3
2
6
1








8

Table 2.3: Classical fourth order ERK method interpolated by its unique NCE of order 2
1
0
2
1
2
1

2
1
2
1
2
1
0 0
0





3
1
2
2
1
4
3
1
3
3
1
2
3
2
2
2
1
1
) (
) (
) (
) (
=
=
=
+ =
b
b
b
b


6
1
3
1
3
1
6
1


Table 2.4: Classical fourth order ERK method interpolated by its unique NCE of order 3
1
0
2
1
2
1

2
1
2
1
2
1
0 0
0

2
2
1
3
3
2
4
2 3
3
2
3
2 3
3
2
2
2
2
3
3
3
2
1
) (
) (
) (
) (




=
+ =
+ =
+ =
b
b
b
b


6
1
3
1
3
1
6
1


3. Numerical Method For Solving DDEs

Most numerical methods for solving ordinary differential equation (ODE) can be adapted to
solve DDE. The idea is, when solving DDE (1.1), the ERK method in (2.1) can be generalized
into the following form

=
=
+ + + =
+ = +
1
0
0
)) ( , , (
) (
i
j
i n j ij n i i i
s
i
i i n n
h c t y K a y h c t f K
K b h y h t u

(3.1)

Similarly, when method (2.2) is adapted to DDE (1.1) we have

=
=
+ + + =
+ = +
1
0
0
)) ( , , (
) ( ) (
i
j
i n j ij n i i i
s
i
i i n n
h c t y K a y h c t f K
K b h y h t u


(3.2)

When solving DDEs, one difficulty is to approximate the delay term, ) ( + h c t y
i n
on
] , [
j i
t t . Virk [10], in his paper, shows that the mentioned difficulty can be solved by reducing the
DDE to an ODE so that it can be solved by using normal Runge-Kutta method. Another way,
preferable by most authors, is to approximate the delay term by using interpolation method. For
example, Ismail et al. [6], Karoui [7] and Karoui et al. [8] used 3-point Hermite interpolation;
9

Orbele et al. [9] used appropriate multipoint Hermite interpolation; while Enright et al. [3] and
Hayashi [5] considered using continuous Runge-Kutta method itself to approximate the delay
term.

In order to obtain high accuracy approximation, we decided to use 3-point Hermite
interpolation, instead of using NCEs, to approximate the delay term.


4. Numerical Examples

We have written experimental program in MAPLE to solve the following test problem using
method described in Table 2.2, 2.3 and 2.4.

Problem 1: (Karoui [5])
1 . 0 0 ), ln( ) (
1 . 0 ) (
) (
s < =
> = '

t t t y
t ye t t y
t y


|
.
|

\
|

=
t
e t y t
1
1
) , (
Exact solution: ) ln( ) ( t t y =
Results are given for ] 10 , 1 . 0 [ e t

Problem 2: (Ismail [3])
0 , 1 ) (
0 ), ( ) cos( ) (
s =
> = '
t t y
t t y t t y

2 ) ( ) , ( + = t y t y t
Exact solution: 1 ) sin( ) ( + = t t y

Results are given for ] 10 , 0 [ e t

Problem 3: (Karoui [4], page 9)
), ( ) (
1
) ( = ' t y t y
t
t y

1 > t
, 1 ) ( = t y 1 s t

2 ) (
) (
) , (

=
t y
e
t y
t
t y t
Exact solution:


=
) 3 log( 2
) (
t
t
t y

6 . 2 2 ,
2 1 ,
s s
s s
t
t

Results are given for ] 6 . 2 , 1 [ e t
10

Problem 4: (Bellen & Zennaro [1], page 86)
), (
2
1
) ( = ' t y
t
t y

1 > t
, 1 ) ( = t y 1 s t

1 2 ) ( ) , ( + = t y y t
Exact solution:

( )

+ +
=
t
t
t y
t
2
2
2
1
4
1
) (

3 2 ,
2 1 ,
s s
s s
t
t

Results are given for ] 3 , 1 [ e t

The numerical results are obtained when the problems are solved using NCEs of third and
fourth order Runge-Kutta method developed by Zennaro [8] using 3-point Hermite interpolation
to approximate the delay term. Different stepsize was used in each problem.

The numerical results are presented in Tables 4.1 and 4.2 and the notations used are as
follows:

Method:
M1: Third order ERK method with NCE of order 2
M2: classical Runge-Kutta method with NCE of order 2
M3: classical Runge-Kutta method with NCE of order 3

The notation 8.103989 (-3) means 8.103989 10
-3


Table 4.1: Maximum error for Runge-Kutta solutions with its unique NCE using stepsize h= 0.1

Problem

Maximum absolute error
) ( ) ( max
1 0
h t y h t u
i i

+ +
s s

M1 M2 M3
1 8.103989 (-3) 8.103989(-3) 4.257116(-3)
2 9.6376 (-6) 8.7586(-6) 1.0148(-6)
3 5.36078120 (-3) 5.3605628(-3) 5.36056232(-3)
4 1.96722570 (-2) 1.96710421(-2) 1.96710428(-2)





11

Table 4.2: Maximum error for Runge-Kutta solutions with its unique NCE using stepsize h= 0.05

Problem

Maximum absolute error
) ( ) ( max
1 0
h t y h t u
i i

+ +
s s

M1 M2 M3
1 1.338854 (-3) 1.338855 (-3) 4.67560(-4)
2 1.10356(-6) 1.05381(-6) 2.235(-7)
3 1.293306375(-3) 1.293301425(-3) 1.29330245(-3)
4 1.96711546(-2) 1.96710314(-2)

1.96710311(-2)



5. Conclusions

In this report, we discussed the NCEs for Runge-Kutta method which developed by Zennaro
[11]. We have also derived NCEs for the third and fourth order ERK method.

In general, we obtained good continuous approximation to the solution by using NCE of
Runge-Kutta method, since the continuous Runge-Kutta solutions, either third or fourth order,
cannot be distinguished graphically from the exact solution (see Appendix), except question 4.
However, from the numerical results, we observed that the fourth order method gives smaller
errors if compare with third order method. The numerical results also shows, for fourth order
ERK method, the greater the order of NCE gives the better continuous approximation.


Acknowledgement

The authors acknowledge the Research Management Center, Universiti Teknologi Malaysia
(UTM), for funding this work under Project Vote: 78337.












12

References

[1] Bellen, A and Zennaro, M. (2003). Numerical methods for delay differential equations.
Clarendon press, Oxford.
[2] Butcher, J.C. (2008). Numerical method for ordinary differential equations. Wiley,
London.
[3] Enrigh W. H. amd Hu M. (1994). Interpolating Runge-Kutta methods for vanishing delay
differential equations. Technical Report 292/94. Department of Computer Science.
University of Toronto.
[4] Enrigh W. H., Jackson K. R., Nrsett S. P. and Thomsen P. G. (1986). Interpolants for
Runge-Kutta formulas. ACM Trans. Math. Soft., 12: 193-218.
[5] Hayashi H. (1996). Numerical solution of retarded and neutral delay differential
equations using continuous Runge-Kutta methods. PhD thesis, University of Toronto.
[6] Ismail, F., Al-Khasawneh, R.A., Aung San Lwin and Suleiman, M. (2002). Numerical
treatment of delay differential equations by Runge-Kutta method using Hermite
interpolation, Matematika UTM, 18(2): 79-90.
[7] Karoui, A. (1992). On the numerical solution of delay differential equations, Master
Thesis, University of Ottawa, Ontario.
[8] Karoui, A and Vaillancourt, R. (1995). A Numerical Method for Vanishing-lag delay
differential equations. Appl. Numer. Math., 17: 383-395.
[9] Orbele, H. J. and Pesch, H. J. (1981). Numerical treatment of delay differential equations
by Hermite interpolation, Numer. Math., 31: 235-255.
[10] Virk, G.S. (1985). Runge-Kutta method for delay-differential systems. IEE processings,
132(3): 119-123.
[11] Zennaro, M. (1986). Natural continuous extensions of Runge-Kutta methods. Math.
Comput., 46: 119-133.
[12] Zennaro, M. (1988). Natural Runge-Kutta and projection methods. Numer. Math., 53:
423-438.





13

Appendix

Figure 1: Continuous Runge-Kutta solution for problem 1


Figure 2: Continuous Runge-Kutta solution for problem 2

Figure 3: Continuous Runge-Kutta solution for problem 3
14


Figure 4: Continuous Runge-Kutta solution for problem 4, 1 . 0 = h

Figure 4: Continuous Runge-Kutta solution for problem 4, 05 . 0 = h

Вам также может понравиться