See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/257143364

Third-Order Improved Runge-Kutta Method for Solving Ordinary Differential

Equation

Article in International Journal of Applied Physics and Mathematics · January 2011

DOI: 10.7763/IJAPM.2011.V1.37

CITATIONS READS

18 11,140

1 author:

Faranak Rabiei
Texas A&M University - Kingsville
56 PUBLICATIONS 1,092 CITATIONS

SEE PROFILE

All content following this page was uploaded by Faranak Rabiei on 20 May 2014.

The user has requested enhancement of the downloaded file.

 International Journal of Applied Physics and Mathematics, Vol. 1, No. 3, November 2011

Third-Order Improved Runge-Kutta Method

for Solving Ordinary Differential Equation
Faranak rabiei and Fudziah Ismail

order three with two stages. Rabiei et al [6] constructed the

Abstract—In this paper we constructed the sets of explicit Improved Runge-Kutta method for solving ordinary
third-order Improved Runge-Kutta (IRK) methods. The differential equations. The order conditions of the methods
method used in two and three stage which indicated as the up to order five were derived also the convergence and
required number of function evaluations per step. The
third-order IRK method in two-stage has a lower number of
stability region of the methods were discussed. Rabiei and
function evaluations than the classical third-order RK method Ismail [7] developed the fifth-order Improved Runge-Kutta
while maintaining the same order of local accuracy. In method for solving ordinary differential equations. The
three-stages, the new method is more accurate compared to the method used only five stages. Rabiei et al [8] proposed the
classical third-order RK method. The stability region of numerical solution of ordinary differential equations using
methods are given and numerical examples are presented to fifth-order Improved Runge- Kutta method .
illustrate the efficiency and accuracy of the new methods.
In section 2 we presented the derivation of IRK3 with tow
Index Terms—Improved Runge-Kutta Methods, two-step stages using the Taylor’s series expansion. In section 3 we
runge-kutta methods, order conditions, stability region, proposed the method with three-stages which will be denoted
ordinary differential equations. as IRK3-3. In section 4 the stability region of the methods are
discussed and in the last section, numerical examples are
carried out and performance of the new method is compared
I. INTRODUCTION with the classical RK2 and RK3 methods.
Consider the numerical solution of the initial value
problem for the system of ordinary differential equation.
II. THIRD ORDER METHOD WITH TWO-STAGES (IRK3)
y '( x ) = f ( x, y ( x )), x ∈ ⎡⎣ x0 X ] , In two-stages (s=2), the general form of IRK3 can be
written as:
y ( x0 ) = y0 . (1)
yn +1 = yn + h(b1k1 − b−1k−1 + b2 (k2 − k−2 )),
One of the most common method for solving numerically k1 = f ( xn , yn ),
(1) is Runge-Kutta (RK) method. Most efforts to increase the
order of RK method, have been accomplished by increasing k−1 = f ( xn −1 , yn −1 ),
the numberof Taylor’s series terms used and thus the number k2 = f ( xn + c2 h, yn + ha21k1 ),
of function evaluations. In third-order classical Runge-Kutta
k −2 = f ( xn −1 + c2 h, yn −1 + ha21k−1 ).
method (RK3) three number of function evaluations is (2)
required per step. Many authors have attempted to increase
the efficiency of RK methods with a lower number of where 0 ≤c2 ≤ 1. In the derivation of the method we will use
function evaluations required. ∑ 1 ; which is called the row sum condition of RK
As a result, Goeken et.al [1] proposed a class of method, so here we have c2 = a21.
Runge-Kutta method with higher derivatives approximations To determine the coefficients of method given by (2), the
for the third and fourth-order method. Xinyuan [2] presented IRK method expression (2) is expanded using the Taylor’s
a class of Runge-Kutta formulae of order three and four with series expansion. After some algebraic simplifications this
reduced evaluations of function. Phohomsiri and Udwadia [3] expansion is equated to the true solution that is given by
constructed the Accelerated Runge-Kutta integration Taylor’s series expansion. This result will be in the system of
schemes for the third-order method using two functions nonlinear algebraic equations which is denoted as order
evaluation per step. Udwadia and Farahani [4] developed the conditions. Finally we tried to solve as many order conditions
Accelerated Runge-Kutta methods for higher orders. as possible because the highest power of h for which all of the
However most of the presented methods are obtained for the order equations are satisfied is the order of resulting IRK
autonomous system while the Improved Runge-kutta method. A great deal of algebraic and numeric calculations is
methods (IRK) can be used for autonomous as well as required for the above processes which were mainly
non-autonomous systems. Rabiei and Ismail [5] constructed performed using Maple.
the New Improved Runge-Kutta method with reduced Consider equations:
number of function evaluations. The method proposed of y ' = f ( x, y ) = f ,
y '' = f x + ff y ,
Manuscript received October 10, 2011; revised November 20, 2011.
F. Rabiei is with the Department of Mathematics, Faculty of Science, y ''' = f xx + 2 f xy + f yy f 2 + f y ( f x + ff y ). (3)
University Putra Malaysia (email: rabiei@math.upm.edu.my)

191
 International Journal of Applied Physics and Mathematics, Vol. 1, No. 3, November 2011

The values of y’(x), y’’(x), . . . are obtained by substituting, 5 (11)

b2 c2 = .
x = xn. The Taylor’s series expansion of y(xn + h) is 12

yn +1 = y ( xn + h) By satisfying the above order equations, it is possible to

2 achieve the third-order method with two functions evaluation
h
= y ( xn ) + hy '( xn ) + y ''( xn ) per step. Here, we solved the order conditions (9) - (11) in
2! terms of b-1, b 1, b2 and choose c2 as a free parameter. In table
h3 (I) we choose c2 = 4 and 1 which leads to two different sets
+ y '''( xn ) + O ( h 4 ). (4)
3! 5 2
of coefficients for IRK3.
Substituting (3) into (4) we have
TABEL I: IRK3 PARAMETERS SETS
Coefficient Set1 Set2
2
h 4 1
yn +1 = yn + hf + ( f x + ff y ) c2
2! 5 2
h3 4 1
+
6
( f xx + 2 f xy + f yy f 2 + f y ( f x + ff y ) )
a21
5 2
−1 −1
b−1
+O ( h 4 ). (5) 48 3
47 2
b1
Define F = fx + ffy and G = fxx + 2fxy +fyyf2; thus, giving 48 3
25 5
b2
y '' = F , y ''' = G + f y F . 48 6

After simplifying, can be written (5) as follows

III. THIRD ORDER METHOD WITH THREE-STAGES (IRK3-3)

h2 h3 In three-stages (s=3), the general form of IRK3 is as

yn +1 − yn = hf + F + ( G + f y F ) + O(h 4 ). (6) follows:
2 6

Now, by using the Taylor’s series expansion of k1, k2 , k-1 yn +1 = yn + h(b1k1 − b−1k−1 + b2 (k2 − k−2 )),
and k-2 which are used in (2) we have +b3 ( k3 − k−3 ),
i −1
k1 = f , ki = f ( xn + ci h, yn + h∑ aij k j ),
j =1
h2
k−1 = f − hF + ( G + f y F ) + O (h3 ), i −1

2 k− i = f ( xn −1 + ci h, yn −1 + h∑ aij k− j ). (12)
j =1
h2
k2 = f + hc2 F + (c22G ) + O(h3 ),
2 For i = 1, 2, 3 and c1 = 0, c2, c3 ∈ [0 1]. Also we considered
h2
k−2 = f − h(1 − c2 ) F + ( (1 − c2 ) 2 G c2 = a21 ,
2
c3 = a31 + a32 .
+(1 − 2c2 ) f y F ) + O (h3 ).
(7) Using the Taylor’s series expansion in section 2, we
obtained the following order conditions for IRK3-3:
Substituting the above formulas in (2) we obtain
b1 − b−1 = 1, (13)
yn +1 − yn = h ( b1 − b−1 ) f + h 2 ( b−1 + b2 ) F
h3 1 (14)
( −b−1 − (1 − 2c2 ) b2 ) ( G + f y F ) +O ( h ). b−1 + b2 + b3 = ,
4
+ (8) 2
2
5 (15)
Comparing (8) with (6) we obtained the following order b2 c2 + b3c3 = .
conditions 12

(9) By satisfying order equations (13)-(15), we obtained the

b1 − b−1 = 1,
third-order method with three function evaluations per step
which we will be noted as IRK3-3. In Table II we presented
1 (10)
b−1 + b2 = , the two sets of coefficients of IRK3-3.
2

192
 International Journal of Applied Physics and Mathematics, Vol. 1, No. 3, November 2011

TABEL II：IRK3-3 PARAMETERS SETS ⎛5 2 3 ⎞ 5 h

Coefficient Set1 Set2 ρ (ζ , h ) = ζ 2 − ⎜ h + h + 1⎟ ζ + h 2 + . (17)
1 1 ⎝ 12 2 ⎠ 12 2
c2
3 2
2 The three-stages method, using the same process the
c3 1 stability polynomial for fourth order method (IRK3-3) is
3
1 1
a21 ⎛1 5 2 3 ⎞
3 2 ρ (ζ , h ) = ζ 2 − ⎜ h 3 +h + h + 1⎟ ζ
2 −1 ⎝ 6 12 2 ⎠
a31
21 3 1 3 5 2 h
+ h + h + . (18)
4 4 6 12 2
a32
7 3
1 −1 Stability region of the methods is the set of the values of
b−1
8 12 such that all the roots of the stability polynomial are inside a
9 11 unit circle. Here, the stability region of IRK3 which is plotted
b1
8 12 in Figure. 1 is done using Matlab.
−1 1
b2
2 3
7 1
b3
8 4

In the last section, numerical results of IRK3 and IRK3-3

are compared with Butcher’s classical second and third-order
Runge-Kutta method [9] which are denoted as RK2 and RK3
in the Table III.

TABEL III: BUTCHER’S RK2 AND RK3 PARAMETERS

0
1 1 Fig. 1. Stability region of IRK3 (thin line) and RK3 (thick line) for
0
2 2
h = λ h.
1 4 3 3
0
2 9 4 4
2 1 4 V. NUMERICAL RESULTS
0 1
9 3 9
(RK2) (RK3) In this section, we tested a standard set of initial value
problems to show the efficiency and accuracy of the proposed
methods. The exact solution y(x) is used to estimate the
IV. STABILITY ANALYSIS global error as well as to approximate starting values of y1 at
the first step [x0 x1]. The following problems are solved for x
Generally, to define the stability region of the method we
applied the test problem y0 = λy, where λ is a complex number.
∈ [0 10].
Problem 1 (see [10] )
Here, by applying the test problem to (2) we have

y ' = λ y, y ' = y cos( x ), y (0) = 1,

k1 = λ yn ,
Exact solution: y ( x) = esin( x ) . (19)
k −1 = λ yn −1 ,
Problem 2 (see [10] )
k2 = λ (1 + λ ha21 ) yn ,
k −2 = λ (1 + λ ha21 ) yn −1. y '1 = −2 y1 + y2 + 2sin( x),
y '2 = y1 − 2 y2 + 2 ( cos( x ) − sin( x) ) ,
Substituting all the above values in (2) we have y1 (0) = 2, y2 (0) = 3,

( (
yn+1 = 1 + h b1 + b2 (1 + ha21 ) yn )) Exact solution: y1 ( x) = 2e − x + sin( x),

(
− b−1 + b2 (1 + ha21 ) yn −1. ) (16)
y2 ( x) = 2e − x + cos( x). (20)

The maximum global error of the tested problems 1 and 2

against the step size are shown in Table IV and V. In Table VI,
where . After substituting the order conditions given we compared the computational effort required for the
by (9)-(11) in (16) we obtained the following stability classical Runge-kutta method to that required for the method
polynomial of IRK3 presented.

193
 International Journal of Applied Physics and Mathematics, Vol. 1, No. 3, November 2011

TABEL IV: MAXIMUM GLOBAL ERROR FOR PROBLEM1 VI. CONCLUSION

Step IRK3 IRK3 IRK3-3 IRK3-3 From tables IV and V we obtain that IRK3 has a lower
RK2 RK3
size Set1 Set2 set1 Set2
number of function evaluations than RK3 method while
9.88 7.46 4.93 2.02 maintaining the same order of local accuracy and IRK3-3 is
0.5 4.22 E-3 1.18 E-3
E-2 E-3 E-2 E-2 more accurate compared with the classical RK3 method. In
conclusion, third order improved Runge-Kutta methods, have
8.28 6.15 1.62 1.90 been developed for numerical integration of first order
0.1 8.79 E-6 2.26 E-5
E-4 E-4 E-3 E-4 ordinary differential equations. They are almost two-step in
nature and they are computationally more efficient and
0.05
1.02 8.02
5.80 E-7 1.43 E-6
3.91 2.44 produced smaller errors compared to the existing
E-4 E-5 E-4 E-5 Runge-Kutta methods of the same order.

0.01
8.08 6.53 9.76
2.32 E-9
1.52 1.75 REFERENCES
E-7 E-7 E-10 E-5 E-7
[1] D. Goeken and O. Johnson, “Runge-Kutta with higher order erivative
approximations,” Appl. Numer. Math, vol. 34, pp. 207-218, 2000.
1.00 7.90 3.13 1.45 3.79 2.43 [2] W. Xinyuan, “A class of Runge-Kutta formulae of order three and four
0.005
E-7 E-8 E-11 E-10 E-6 E-8 with reduced evaluations of function,” Appl. Math. Comput, vol. 146,
pp. 417-432, 2003.
[3] P. Phohomsiri and F. E. Udwadia, “Acceleration of Runge-Kutta
integeration schemes,” Discrit. Dynamic. Nature. Soci. vol. 2, pp.
TABEL V: MAXIMUM GLOBAL ERROR FOR PROBLEM2 307-314, 2004.
Step IRK3 IRK3 IRK3-3 IRK3-3 [4] F. E. Udwadia and A. Farahani, “Accelerated Runge- Kutta methods,”
RK2 RK3 Discrit. Dynamic. Nature. Soci. 2008, doi:10.1155/2008/790619, 2008
size Set1 Set2 set1 Set2
[5] F. Rabiei and F. Ismail, “New Improved Runge- Kutta method with
reducing number of function evaluation,” ASME Press (2011),
3.73 3.04 7.44 1.88 ISSN:9780791859797, DOI: 10.1115/1.859797. paper14
0.5 4.79 E-2 8.21 E-2
E-2 E-2 E-2 E-2 [6] F. Rabiei, F. Ismail, M. Suleiman, and N. Arifin, “Improved
Runge-Kutta method for solving ordinary differential equation,”
2.42 2.22 1.60 1.07 Journal of Applied Mathematics, submitted 2011.
0.1 9.28 E-6 1.20 E-5 [7] F. Rabiei and F. Ismail, “Fifth-order Improved Runge-Kutta method
E-4 E-4 E-3 E-4
for solving ordinary differential equation,” Proceeding of WSEAS
Conference, Penang, Malaysia, ISBN: 978-1- 61804-039-8, 2011, pp.
3.05 2.79 3.77 1.25 129-133.
0.05 5.80 E-7 7.56 E-7 [8] F. Rabiei and F. Ismail, “Numerical solution ofordinary differential
E-5 E-5 E-4 E-5
equation using fifth-order Improved Runge-Kutta method,”
International Journal of Applied Applied Mathematics and Informatics,
2.45 2.25 9.30 1.44 4.53 Submitted 2011.
0.01 1.21 E-9
E-7 E-7 E-10 E-5 E-7 [9] J. C. Butcher, “On Runge-Kutta processes of highh order,” J.
Australian. Math. Soci. vol.4, pp.179- 194, 1964.
[10] L. F. Shampine, “Numerical solution of ordinary differential equations,”
3.07 2.81 5.82 7.58 3.59 1.18 Chapman and Hall, New York, NY, USA, 1994.
0.005
E-8 E-8 E-11 E-11 E-6 E-8

TABEL VI: SUMMARY OF ORDER AND NUMBER OF FUNCTION Faranak Rabiei was born on July 10 1982. She
EVALUATIONS FOR IRK AND RK METHOD PER STEP studied in Applied Mathematics for bachelor and
IRK3 IRK3-3 RK2 RK3 Master degrees from Iran, she is PhD student from
Department of Mathematics, Universiti Putra
Order of method 3 3 2 3 Malaysia. She is research assistant with the grant
No function evaluations 2 3 2 3 scholarship from Universiti Putra Malaysia.

194

View publication stats