Sie sind auf Seite 1von 11

Proceedings of the 12th International Conference

on Computational and Mathematical Methods


in Science and Engineering, CMMSE 2012
July, 2-5, 2012.

A rational Falkner method for solving special second order


IVPs
Higinio Ramos1 and Ces
areo Lorenzo2
1
2

Scientic Computing Group, University of Salamanca. Spain

Escuela Politecnica Superior de Zamora, University of Salamanca. Spain


emails: higra@usal.es, cesareo@usal.es

Abstract
In this paper we present the construction of a rational non-standard explicit algorithm for solving initial-value problems of the special second order. The main formula
is a rational one to follow the solution. The appearance of the rst derivative in this
main formula requires the use of a second formula to follow the derivative, as is usually
done in Falkner methods. Local truncation errors and linear stability analysis are addressed. Some numerical experiments are considered in order to check the behavior of
the proposed method.
Key words: rational method, Falkner method, special second order ODEs
MSC 2000: 65L05

Introduction

Second-order dierential equations deserve special consideration because they appear frequently in applied sciences. Examples of that are the mass movement under the action of
a force, problems of orbital dynamics, or in general, any problem involving Newtons law.
There is a vast literature addressing the numerical solution of the so-called special
second-order initial value problem (I.V.P.)
y (x) = f (x, y(x)),

y(x0 ) = y0 ,

y (x0 ) = y 0 ,

x [x0 , xN ] ,

(1)

(see for example the classical books by Henrici [6], Collatz [1], Lambert [7], Shampine and
Gordon [10] or Hairer et al. [5]).

c
CMMSE

ISBN:978-84-615-5392-1

Rational Falkner method

Although it is possible to integrate a second-order I.V.P. by reducing it to a rst-order


system and applying one of the methods available for such systems, it seems more natural to
provide numerical methods in order to integrate the problem directly. The Stormer-Cowell
methods is a well-known class of schemes of this type.
The advantage of this procedure lies in the fact that they are able to exploit special
information about ODES, and this results in an increase in eciency. For instance, it is wellknown that Runge-Kutta-Nystrom methods for (1) involve a real improvement as compared
to standard Runge-Kutta methods for a given number of stages ([5], p. 285), although the
computational cost remains high because of the number of function evaluations. On the
other hand, a linear k-step method for rst-order ODEs becomes a 2k-step method for (1),
([5], p. 461), increasing the computational work.
One of the methods for numerically solving the problem in (1) is the explicit method
due to Falkner [2], which can be written in the form
yn+1 = yn + h yn + h2

k1

j j fn ,

j=0

yn+1
= yn + h

k1

(2)

j j fn ,

j=0

where h is the stepsize, yn and yn are approximations to the values of the solution and
its derivative at xn = x0 + n h , fn = f (xn , yn , yn ) and j fn is the standard notation for
the backward dierences. The coecients j and j can be obtained using appropriate
generating functions [3].
The implicit Falkner formulas may also be used for solving the above problem, (see [1])
and may be written as
yn+1 = yn +

h yn

+h

j j fn+1 ,

j=0

yn+1
= yn + h

(3)

j j fn+1 ,

j=0

where the coecients can be obtained using appropriate generating functions [3].
For the problem in (1) the above formulas may be used more eciently taking the
explicit formula in (2) to approximate the solution and the corresponding implicit formula
in (3) to approximate the derivative (for details see [3]). In this way, the two-step procedure
provided by the Falkner formulas is given by

c
CMMSE

ISBN:978-84-615-5392-1

H. Ramos, C. Lorenzo

yn+1 = yn + h yn +

yn+1

yn

)
h2 (

4yn yn1
6

(4)

+
(5yn+1
+ 8yn yn1
),
12

which result in an explicit method due to the absence of the derivative in the function
f (x, y). The local truncation errors for these formulas are given respectively by
1 (4)
y (xn )h4 + O(h5 )
8

and

1 (5)
y (xn )h4 + O(h5 ) .
24

A similar strategy was used by Beeman [9] to develop the well-known semi-implicit
method
yn+1 = yn + h yn +

yn+1

)
h2 (

4yn yn1
6

(5)

= yn + (2yn+1
+ 5yn yn1
).
6

which has been commonly used in molecular dynamics, when the acceleration at time only
depends on position and not on velocity. In this case the local truncation errors are given
by
1 (4)
1 (4)
y (xn )h4 + O(h5 )
and
y (xn )h3 + O(h4 ) .
8
12

Rational Falkner method

In this section we derive a rational method to approximate the solution y(xn+1 ) for the
problem in (1) based on the ideas in [4]. We suggest an approximation to the theoretical
solution y(xn+1 ) by
y(xn + h) = y(xn ) + h y (xn ) +

h2 y (xn )
,
y(xn ) + a(h)y (xn )

(6)

where a(h) is a suciently dierentiable unknown function of the step size that has to be
determined and it is assumed that y(xn ) + a(h)y (xn ) = 0.
From (6) it results that
(
)(
)
Fn (y, a, h) = y(xn + h) y(xn ) hy (xn ) y(xn ) + a(h)y (xn ) h2 y (xn ) = 0 .

c
CMMSE

ISBN:978-84-615-5392-1

Rational Falkner method

Now, if we consider a(h) expanded in Taylor series about h = 0, after expanding Fn (y, a, h)
in Taylor series about xn the following expression is obtained:
Fn (y, a, h) =

]
1 [
2 + y(xn ) + a(0)y (xn ) y (xn ) h2
2
(
)]
1 [
+
y(xn ) y (3) (xn ) + y (xn ) 3 a (0) y (xn ) + a(0) y (3) (xn )
h3
6
+ O(h4 ) .

(7)

Imposing that the coecients of h2 and h3 in (7) vanish, we obtain a system of equations
from which it is readily deduced that
a(0) =

2 y(xn )
,
y (xn )

a (0) =

2y (3) (xn )
,
3 y (xn ) y (xn )

provided that y (xn ), y (xn ) = 0. Introducing the above values in the Taylor series of a(h)
we have
a(h) =

2 y(xn )
2hy (3) (xn )

+ O(h2 ) .

y (xn )
3y (xn )y (xn )

(8)

From (6) and (8) it is readily deduced the numerical scheme, which may be written in the
form
yn+1 = yn + hyn +

3 h2 fn2
,
6 fn 2h fn

(9)

where yn = y(xn ) , yn+1 y(xn+1 ) , fn = f (xn , yn ) and


fn =

f
f
(xn , yn ) +
(xn , yn ) yn .
x
y

It can be easily checked that the method in (9) is exact when the solution of the dierential
equation is of the form
c3
y(x) = c1 + c2 x +
c4 + x
where the ci are arbitrary constants.
The presence of the derivative in the formula in (9) forces to consider a second formula
to follow the derivative on each step. Due to the form of the function f (x, y) we include the

c
CMMSE

ISBN:978-84-615-5392-1

H. Ramos, C. Lorenzo

value y (xn + h) in the formula. Following a similar procedure as for the formula in (9) we
have obtained the formula

yn+1
= yn +

4 h fn2 (2fn + fn+1 )


.
12 fn2 2h fn fn + h2 (fn )2

(10)

The rational Falkner method consists in both formulas (9) and (10)
yn+1 = yn + hyn +

yn+1

3 h2 fn2
,
6 fn 2h fn

(11)

4 h fn2 (2fn + fn+1 )


= yn +
.
12 fn2 2h fn fn + h2 (fn )2

Local truncation errors

To obtain the expressions for the local truncation errors we follow the procedure as in [7].
We consider the functional given by
L(z(x), h) = z(x + h) z(x) h z (x)

3 h2 (z (x))2
,
6 z (x) 2h z (3) (x)

(12)

where z(x) is an arbitrary function dened on [x0 , xN ] and dierentiable as often as we


need, after expanding in Taylor series about x and collecting terms in h we obtain
(
( (3) )2 )
4
z (x)
1
3z (4) (x)
h4 + O(h5 ) ,
(13)
L(z(x), h) =
72
z (x)
which means that the formula in (9) has at least second-order of accuracy. The local
truncation error of the method may be written as

(
)
(3) 2
4
y
n
1 (4)
4
5
Tn+1 =
3yn
h + O(h ) ,

72
yn
where yn , yn , yn denote respectively the numerical approximations to the fourth, third
and second derivatives of y(x) at the point xn .
For the formula in (10) we consider the functional given by
(4)

(3)

Lp (z(x), h) = z (x + h) z (x)

c
CMMSE

4 h (z (x))2 (2z (x) + z (x + h))


,
12 (z (x))2 2h z (x) z (3) (x) + h2 (z (3) (x))2

(14)

ISBN:978-84-615-5392-1

Rational Falkner method

After expanding in Taylor series about x and collecting terms in h results


(
)
(3) (x)3 z (5) (x)z (x)2 2z (4) (x)z (3) (x)z (x)
3z
Lp (z(x), h) =
h4 + O(h5 ) ,
72z (x)2

(15)

and the local truncation error for the formula in (10) may be written as
(
)
(3)
(5)
(4) (3)
3(yn )3 yn (yn )2 2yn yn yn
p
Tn+1 =
h4 + O(h5 ) ,
72(yn )2
with the obvious notations for the derivatives of y(x) at the point xn .

Stability analysis

The linear stability analysis of the above formulas follow a similar procedure as for linear
multistep methods. Considering the Dahlquists test given by
y (x) = 2 y(x),

y(xn ) = yn ,

y (xn ) = yn .

(16)

The solution of this problem is y(x) = e(xxn ) yn , from which we obtain that y(xn + h) =
eh y(xn ). This means that for < 0 the exact solution decreases. It should be expected
that the application of the numerical method to this problem has a similar behavior. If we
apply the method in (9) to the problem in (16) we obtain the dierence equation
yn+1 =

h2 2 + 4h + 6
yn .
6 2h

Setting H = h the stability function is obtained as


R(H) =

H 2 + 4H + 6
.
6 2H

If H were allowed to be in the complex plane, than the stability region of the method
consists of the region in the H-complex plane for which it is |R(H)| < 1, which is plotted
in Figure 1.
In particular, for < 0 the stability interval results from the intersection with the real
axis, and is given by 6 < H < 0.

Numerical experiments

In order to evaluate the performance of the rational Falkner method in this article, we have
considered the methods in (4) and (5) for comparison purposes. As they are all two-step

c
CMMSE

ISBN:978-84-615-5392-1

H. Ramos, C. Lorenzo

-2

-4

-8

-6

-4

-2

Figure 1: Stability region for the method in (9).


methods, apart from the known values y0 , y0 , we need the value y1 in order to initialize the
methods. This additional value may be obtained by any one-step method, but as in the
examples we know the exact solution, we have used the true values. The errors have been
dened as the maximum of the absolute errors on the nodal points of the solution over the
integration interval
Emax (y) = max {|y(xj ) yj |} ,
xj [x0 ,xN ]

and the maximum of the absolute errors on the nodal points of the derivative over the
integration interval
{
}
Emax (y ) = max
|y (xj ) yj | ,
xj [x0 ,xN ]

where N I refers to the number of steps used in the integration.

5.1

Example 1

Consider the nonlinear problem


y (x) = 6 y(x)2 ,

c
CMMSE

y(0) = 1 ,

y (0) = 2 ,

ISBN:978-84-615-5392-1

Rational Falkner method

1
. We have integrated this problem over the interval
(x + 1)2
[0, 2]. In Table 1 appear the errors for the solution and the derivative for dierent number of
steps. We have included a nal column named CPU with the computational time needed,
in seconds. We observe that the computational times are similar for the tree methods
considered, but the errors are smaller with the rational method.
whose exact solution is y(x) =

5.2

Example 2

Consider the problem


y (x) = 6 y(x)2 6 ,

y(0) = 2 ,

y (0) = 0 ,

( )
whose exact solution is y(x) = 2 + 3 tan2
3x . We have integrated this problem over
the interval [0, 0.8]. In Table 2 appear for dierent number of steps the maximum of the
relative errors on the grid points over the integration interval for the solution. We see that
the rational Falkner method shows a better performance.

Conclusions

A two-step rational Falkner method for solving initial-value problems of the special form
y = f (x, y) has been presented. The method consists in a couple of rational formulas,
one to follow the solution, and the other to follow the derivative. The analysis of the local
truncation errors and the stability region are presented. The numerical examples show the
good performance of the method.

References
[1] L. Collatz, The Numerical treatment of Dierential Equations, Springer, Berlin,
1966.
[2] V. M. Falkner, A method of numerical solution of dierential equations, Phil. Mag.
S. 7, 21 (1936) 621-640.
[3] H. Ramos, C. Lorenzo, Review of explicit Falkner methods and its modications for
solving special second-order I.V.P.s, Computer Physics Communications 181 (2010)
1833-1841.
[4] H. Ramos, A non-standard explicit integration scheme for initial-value problems, Applied Mathematics and Computation 189 (2007) 710718.

c
CMMSE

ISBN:978-84-615-5392-1

H. Ramos, C. Lorenzo

[5] E. Hairer, S. P. Norsett and G. Wanner, Solving Ordinary Dierential Equations


I , Springer, Berlin, 1987.
[6] P. Henrici, Discrete variable Methods in Ordinary Dierential Equations, John Wiley,
New York, 1962.
[7] J. D. Lambert, Computational Methods in Ordinary Dierential Equations, John
Wiley & sons, London, 1973.
[8] M. K. Jain, Numerical Solution of Dierential Equations, Wiley Eastern Limited,
New Delhi, 1984.
[9] D. Beeman, Some multistep methods for use in molecular dynamics calculations, J.
Comput. Phys. 20(1976) 130.
[10] L. F. Shampine and M. K. Gordon, Computer solution of Ordinary Dierential
Equations. The initial Value Problem, (1975), Freeman, San Francisco, CA.

c
CMMSE

ISBN:978-84-615-5392-1

Rational Falkner method

Beeman (5)

NI
100
200
400
800
1600
3200

Falkner (4)

NI
100
200
400
800
1600
3200

Method (11)

NI
100
200
400
800
1600
3200

Emax (y)
5.70 103
1.43 103
3.60 104
9.04 105
2.25 105
5.63 106
Emax (y)
6.78 104
9.04 105
1.16 105
1.48 106
1.86 107
2.34 108
Emax (y)
5.26 105
7.31 106
9.63 107
1.24 107
1.56 108
1.96 109

Emax (y )

CP U

7.59 103
1.92 103
4.82 104
1.21 104
3.02 105
7.54 106

0.000
0.016
0.016
0.031
0.125
0.203

Emax (y )

CP U

9.02 104
1.20 104
1.55 105
1.97 106
2.48 107
3.12 108

0.015
0.016
0.031
0.063
0.140
0.313

Emax (y )

CP U

7.06 105
9.79 106
1.29 106
1.65 107
2.09 108
2.63 109

0.000
0.015
0.016
0.047
0.109
0.219

Table 1: Data for the problem y (x) = 6 y(x)2 . Maximum absolute errors for dierent
number of steps.

c
CMMSE

ISBN:978-84-615-5392-1

H. Ramos, C. Lorenzo

NI

Beeman (5)

100
200
400
800
1600
3200

5.26 103
1.42 103
3.72 104
9.49 105
2.42 105
6.05 106

Falkner (4)
1.17 103
1.60 104
2.09 105
2.67 106
3.37 107
4.24 108

Method (11)
1.48 104
2.18 105
2.97 106
3.89 107
5.04 108
6.33 109

Table 2: Data for the problem y (x) = 6 y(x)2 6. Maximum relative errors for dierent
number of steps.

c
CMMSE

ISBN:978-84-615-5392-1

Das könnte Ihnen auch gefallen