ISSN (Print) : 0974-6846

Indian Journal of Science and Technology, Vol 9(48), DOI: 10.17485/ijst/2016/v9i47/97774, December 2016 ISSN (Online) : 0974-5645

A Fourth Algebraic Order Explicit Trigonometrically-

Fitted Modified Runge-Kutta Method for the
Numerical Solution of Periodic IVPs
F. A. Fawzi1*, N. Senu2, F. Ismail2 and Z. A. Majid2
Department of Mathematics, Faculty of Computer Science and Mathematics,
1

Tikrit University, Iraq; firasadil01@gmail.com

2
Department of Mathematics and Institute for Mathematical Research, Universiti Putra Malaysia, UPM, Serdang -
43400, Selangor, Malaysia

Abstract
This study has constructed an explicit Trigonometrically-Fitted Modified Runge-Kutta (TFMRK) method for solving first-
order differential equations with periodic solutions. The newly developed method was made according to the method of
Runge-Kutta Dorm and to fourth algebraic order. Numerical results for the new method were compared with the existing
method, showing the potential of the new method over other existing methods.

Keywords: Runge-Kutta Methods, Trigonometrically-Fitting and Periodic Solutions

the first order ordinary differential equations or systems

1. Introduction of first order ordinary differential equations. With respect
to the oscillatory nature of the problem (1), researchers
Since the last few years, many studies were conducted
have developed integrates with frequency dependent
on numerical solution for solving first-order initial value
coefficients using some techniques including trigono-
problems using the formula given:
metrical/exponential-fitting1-4. The early construction of
y ′( x) = f ( x, y ), y ( x0 ) = y 0 , these techniques is influenced by Gautschi5 and Lyche6.
Meanwhile, Vigo-Aguiar and Simos7 presented a simple
y ′ ( x0 ) = y0 , x ∈[a, b]  (1) procedure for adapting Cowell methods in any algebraic,
trigonometric and exponential order. Furthermore,
Where Bettis8 constructed a three-stage method and a four-
stage method, which solve the equation
y ( x) = [ y1 ( x), y 2 ( x),..., y s ( x)] T y ′ = iwy without truncation error. Paternoster9 devel-
oped a class of implicit methods for Runge-Kutta (RK)
f ( x, y ) = [ f1 ( x, y ), f 2 ( x, y ),..., f s ( x, y )] T and Runge-Kutta-Nyström (RKN) types using trigono-
metric fitting technique. On top of that, the study done
y 0 represents the initial condition given with an by Vanden Berghe et al.10,11 has suggested an exponen-
oscillatory solution. This problem always exists in sev- tially fitted explicit Runge-Kutta method, which exactly
eral applied science areas including quantum mechanics, integrates first-order systems with the solution that
astronomy, mechanics and electronics that are related to can be expressed as a linear combination of functions

*Author for correspondence

 A Fourth Algebraic Order Explicit Trigonometrically-Fitted Modified Runge-Kutta Method for the Numerical Solution of Periodic IVPs

from the set cos(w x ), sin(wx )

differential equation (1). By computing the value of y n +1 ,
with w > 0 . Recently, Yanwei Zhang et al. presented a
12
y n′ and substitute it in (3), the result became:
new trigonometrically-fitted two derivative Runge-Kutta
method for the numerical solution of the Schrödinger 7 7 i −1
equation. On the other hand, a trigonometrically-fitted ev = 1 + iv ∑ bi gi ∑ bi ∑ aij (Y j e− iwx n ) (4)
two derivative Runge-Kutta method was proposed by i =1 1 j =1

Yonglei Fang et al.13 to solve oscillatory differential equa-
tions. This paper is divided into four sections. Section where v = w
h . Using the formula:
2 describes the new trigonometrically fitted modified e v = cos(v) + i sin(v)
Runge-Kutta methods. While numerical results are (5)
reported in Section 3 to demonstrate the efficiency of comparing the real and imaginary part has resulted in:
the new methods over the existing methods from
i −1
literature, the last section deals with the conclusion and
remarks.
cos(v) = 1 − v 2 ∑aj =1
ij (Yi e −iwxn ) ,

1.1 Definition 7

Runge-Kutta method exactly or equivalently integrates

sin(v ) = v ∑b g
1
i i (6)

the function with the principal frequency of problem.

This method is said to be trigonometrically-fitted when Table 1. Butcher Tableau for 7-stage fourth order RK
applied to the test equation. method

2. Construction of New Method

A generalised explicit modified Runge-Kutta method for-
mula was given by:

Where
i −1
(2)
Yi = g i y n + h ∑a
j =1
ij f ( x n + ci hi Yi )

i −1
Yi = gi yn + h∑ aij f ( xn + ci hY
i i) (3) Solving (6) using the coefficients of Dorm and seven-
j =1
 stage fourth-order as given by Butcher14 while letting
and g 2 as free parameters, the solution was presented as:
For i = 1,2,..., k
49 4 127 4 383 6 1 6 1 23 4 1 6
Or in Butcher table as: cos(v) = 1 +
18000
v g2 −
3600
v a21 −
72000
v a21 −
4800
v g2 − v2 +
2 500
v −
1440
v ,

727 3 383 5 127 3 49 5 1 7 59 5

sin(v) = v −
3600
v +
72000
v g2 +
3600
v g2 +
18000
v g2 −
4800
v a21 +
18000
v
(7)

For optimised value of maximum global error, a 21 and
g 2 were chosen as the free parameters to construct the
following formula:
1625 v8 + 991440 v 6 − 1445680 v 4 − 108000 v3 sin(v) − 77976000 v 2 − 27576000 cos(v) v 2
a21 =
v 4 (6451600 + 1984056 v 2 + 140809 v 4 + 225v 6 )

14112000 cos(v) v − 108000 cos(v) v 3 − 19699200 v + 17491600 v 3 + 3780228 v 5 − 39728 v 7

g2 =
With k=7, a seven-stage MRK formula was produced.
v 4 (6451600 + 1984056 v 2 + 140809 v 4 + 225v 6 ) (8)
y n = e iwxn is the solution that integrates the ordinary As , and obtained the following Taylor expansions:

2 Vol 9 (48) | December 2016 | www.indjst.org Indian Journal of Science and Technology
 F. A. Fawzi, N. Senu, F. Ismail and Z. A. Majid

1
a21 = v −
886
v2 −
2625198759
v4 +
1266942444 682753 6
v 291 2 136934307 5496555899 76007
5 2016125 4552531217 5000 104896822 672250000 g2 = 1 − v − v4 + v6 .
12700 1433868100 00 2081417272 641000000
8420782953 5780616012 4243 8 8160959696 1076845429 7841380469
− v − v10
351742029 67695633 9750000000 172088498 44091854071 5952217500 0000000

+
3913969913 6770534266 31227520631092499
4163423083 0362714811389974086 2500000000 00
v12 (9) 1 1 1
R (hˆ) = 1 + hˆ − hˆ 2 + hˆ3 + hˆ 4 +
199 ˆ5
h +
233 ˆ 6
h −
96936193
hˆ 7
2 6 24 20000 180000 2032254000 000
5428429 4736401855 55507 20084941 592691
 +
5080635000 00
hˆ8 −
2950040228 9400000000
hˆ9 −
295004022 9400000000
hˆ10
291 2 136934307 5496555899 76007
g2 = 1 − v − v4 + v6 1535351040 8455371352 3 ˆ11 22125192663340586983
hˆ12
12700 1433868100 00
6555402750 04905504251
2081417272 641000000
8204014942 0857505847 0004321
+
951623977 514652000 0000000
h +
31720799235048840000000000 (16)
− v8 + v10
1230943236 9823119300 000000 7743014553 0222065563 9830000000 000 1266942444682753
+ hˆ13 + ...
1656183513 2723267091 5886779550 650347 503504749762680000000000
7867886148 7187956813 3335384100 0000000000
v12 (10) 
Secondly, the value of and were then taken up to
The results of different parameter were compared from their series solution.
and analysed to identify which problem that has contrib-
1 866 2625198753 9 4 1266942444 682753 6
uted the smallest error. This method is known as TFMRK. a21 = v 2 −
5 2016125
v2 −
4552531217 5000
v +
1048962286 72250000
v

8420782953 5780616012 4243

+ v8
2.1 Stability Identification 351742029 6795633997 500000000
291 2 136934307 5496555899 76007
A modified method based on k -stage Runge-Kutta (2) g2 = 1 −
12700
v −
1433868100 00
v4 +
2081417272 641000000
v6
and (3) was utilised in the test equation 6555402750 04905504251
− v8 .
1230943236 9823119300 000000
y ′ = iwy (11)
1 1 1 199 ˆ5 233 ˆ 6 96936193
R(hˆ) = 1 + hˆ − hˆ 2 + hˆ3 + hˆ 4 + h + h − hˆ 7
and produced 5428429
2 6 24
4736401855
20000
55507
180000 2032254000 000
20084941 592691
+ hˆ8 − hˆ9 − hˆ10
5080635000 00 2950040228 9400000000 295004022 9400000000
y n+1 = y n + hˆB
Y (12) +
1535351040 8455371352 3 ˆ11
h +
22125192663340586983
hˆ12
951623977 514652000 0000000 31720799235048840000000000 (17)
Y = y G + hˆA
n Y .(13) +
1266942444682753
503504749762680000000000
hˆ13 −
701228164318875043007087113
5065085231534817129564000000000000
hˆ14

8420782078295357806160124243
Where −
1688361743844939043188000000000000 
hˆ15 + ...

Y = [Y1 , Y2 ,..., Ys ] , G = [ g1 , g 2 ,..., g s ] , Lastly, the value of and were taken up to of their
and series solution.
B = [b1 , b2 ,..., bs ]T , hˆ = hv . 1
a21 = v 2 −
866
v2 −
2625198753 9 4
v +
1266942444 682753 6
v
5 2016125 4552531217 5000 1048962286 72250000
Equation (14) demonstrates that: +
8420782953 5780616012 4243
v8 −
8160959696 1076845429 7841380469
v10 .
351742029 6795633997 500000000 1720884984 4091854071 5952217500 00000 000

Y = ( I − hˆA) −1 y n G (14)
291 2 136934307 5496555899 76007
g2 = 1 − v − v4 + v6
Substituting Equation (14) into Equation (13), we 12700 1433868100 00 2081417272 641000000
6555402750 04905504251 8204014942 0857505847 0004321
obtained: −
1230943236 9823119300 000000
v8 +
7743014553 0222065563 9830000000 000
v10 .

y n+1 = R(hˆ) y n
Where R(hˆ) represents the stability function of this 1 1 1
R(hˆ) = 1 + hˆ − hˆ 2 + hˆ3 + hˆ 4 +
199 ˆ5
h +
233 ˆ 6
h −
96936193
hˆ 7
2 6 24 20000 180000 2032254000 000
new method such that: +
5428429
hˆ8 −
4736401855 55507
hˆ9 −
20084941 592691
hˆ10
5080635000 00 2950040228 9400000000 295004022 9400000000
1535351040 8455371352 3 ˆ11 22125192663340586983
+ h + hˆ12

R(hˆ) = 1 + hˆB(1 − hˆA) −1 G

951623977 514652000 0000000
22125192663340586983
31720799235048840000000000
1266942444682753
(18)
(15) +
31720799235048840000000000
hˆ12 +

701228164318875043007087113
503504749762680000000000
hˆ13

8420782078295357806160124243
− ˆ14
h − hˆ15
5065085231534817129564000000000000 1688361743844939043188000000000000
Using this new method, stability polynomial was +
6798297823482053780156340395901920
247807437754922698630971193200000000000000
hˆ16 −
8160959696107684542978413830469
82602479251640899543657064400000000000000
hˆ17

obtained by three different stages of solution. Firstly, the +..... 

value of and g 2 were taken up to h 6 of their series
solution. Stability region of the new method was then obtained
by using the formula of Euler to equate the above three
1 866 2625198753 9 4 1266942444 682753 6 stability polynomials and solving h utilising the maple
a21 = v 2 − v2 − v + v.
5 2016125 4552531217 5000 1048962286 72250000 package.

Vol 9 (48) | December 2016 | www.indjst.org Indian Journal of Science and Technology 3
 A Fourth Algebraic Order Explicit Trigonometrically-Fitted Modified Runge-Kutta Method for the Numerical Solution of Periodic IVPs

Equation (21) displays the obtained order for the new

R(hˆ) = e Iθ = cos(θ ) + I sin(θ ) method, which is four, since all terms for h that are lower
than h 5 have disappeared.
The stability region for the new method is shown in
Figure 1.
3. Tested Problems and Numerical
Results
This section describes the application of the new method
in solving different problems. Below are the MRK meth-
ods chosen for numerical comparison.
• TFMRK: Seven-stage RK of the fourth order given in
Section 3.
• MRK4: The modified RK method proposed by Simos
and Vigo-Aguiar15.
• RK5B: Fifth-order six-stage RK method based on the
study of Sakas and Simos16.
• RK4M: Fourth-order five-stage RK method proposed
by Butcher14.
• RK4Z: Fourth-order five-stage RK method developed
by Hairer et al17.
Problem 1: (Homogeneous)
y1′ = y2 , y1 (0) = 1 ,
Figure 1. The stability region of TFMRK using various
orders. Theoretical solution:
1
y1 ( x) = − sin(8 x) + cos(8 x)
4
Definition: y2 ( x) = −2 cos(8 x) − 8 sin(8 x)
Runge-Kutta method is considered station- Source: Chawla and Rao18.
ary if this new method is absolutely stable. Thus,
∀hˆ ∈ (−3.8,0), | R (hˆ) |< 1 was obtained by employing Problem 2: (Homogeneous)
the maple package.
y1′ = y2 , y1 (0) = 1 ,
2.2 Error analysis y2′ = −v 2 y1 + (v 2 − 1) sin( x) , y2 (0) = v + 1
The new method was further tested by applying the Local
Estimated frequency:
Truncation Error analysis (LTE) adopting Taylor expan-
sion series to observe any changes. Theoretical solution:

LTE = y n+1 − y ( x n + h) (19)

Source: Vander Howuen and Sommeijer19.
71 131 11 71
LTE = h5 [ ( f xxxx + f 4 f yyyy ) − f y f xyy y ′ + f y2 f xy y ′ + ( f x f xyy y ′ + f yyy y ′ (2) )

191
6480000
173
54000
71
40000 54000
1
Problem 3: (Periodic orbit system)
(20)
(2) (2) (2)
− f y f xy y ′ + f x f y f xy + ( f x f yyy y ′ + f xxyy y ′ + f x f xxy ) − f y f yyy y ′ (3)
1080000 36000 108000 27000
71 59
+ ( f xx f yy y ′ + f xxxy y ′ + f xyyy y ′ (3) + f xx f xy + f yy2 y ′ (3) ) + f x f y f yy y ′
y1′ = y2 , y1 (0) = 1
1620000 108000
887 2 (2) 71 2 37 2 97
+ f y f yy y ′ + f xy y ′ − ( f xxx f y + f y f xx ) − ( f x f y4 y ′ + f y3 y ( x) w2 )
2160000 810000 360000 120000
71 2 117079 2 2 6

y2′ = − y1 + 0.001 cos( x) , y 2 (0) = 0

+ f x f yy − ( f y f xy y ( x) w + f y f yy y ( x) w y ′ )] + O(h )
2160000 137160000

4 Vol 9 (48) | December 2016 | www.indjst.org Indian Journal of Science and Technology
 F. A. Fawzi, N. Senu, F. Ismail and Z. A. Majid

y3′ = y4 , y3 (0) = 0

y4′ = − y3 + 0.001sin( x) , y4 (0) = 0.9995

Theoretical solution:

y1 ( x) = cos( x) + 0.0005 x sin( x)

y2 ( x) = sin( x) − 0.0005 x cos( x)

y3 ( x) = − sin( x) + 0.0005 x cos( x)

Figure 2. Efficiency curves for all methods using problem
y4 ( x) = cos( x) + 0.0005 x sin( x) 1 with h = 0.00625, 0.0125, 0.025 and 0.05 for b = 10000 .

Source: Stiefel and Bettis20.

Problem 4: (Nonlinear system)
y1′ = y3 , y1 (0) = 1

y2′ = y4′ , y2 (0) = 0

− y1
y3′ = , y3 (0) = 0
( y12 + y22 ) 3

− y2
y4′ = , y4 (0) = 0
( y12 + y22 ) 3

Theoretical solution:
Figure 3. Efficiency curves for all methods utilising
y1 ( x) = cos( x) problem 2 with h = 0.00625, 0.0125, 0.025 and 0.05 for
y2 ( x) = sin( x) b = 10000 .

y3 ( x) = − sin( x)

y4 ( x) = cos( x)

Source: Moo et al21.

Problem 5: (Inhomogeneous)

y1′ = y2 , y1 (0) = 1

y2′ = − y1 + x , y2 (0) = 2

Theoretical solution:

y1 = sin( x) + cos( x) + x

y2 = cos( x) − sin( x) + 1 Figure 4 Efficiency curves for all methods

employingproblem 3 with h = 0.00625, 0.0125, 0.025 and
Source: Allen and Wing22. 0.05 for b = 10000

Vol 9 (48) | December 2016 | www.indjst.org Indian Journal of Science and Technology 5
 A Fourth Algebraic Order Explicit Trigonometrically-Fitted Modified Runge-Kutta Method for the Numerical Solution of Periodic IVPs

5. References
1. Simos TE. A family of fifth algebraic order trigono-
metrically fitted Runge-Kutta methods for the numerical
solution of the Schrodinger equation. Comput Mater Sci.
2005; 34:342–54.
2. Simos TE, Aguiar JV. A modified phase-fitted Runge-Kutta
method for the numerical solution of the Schrodinger equa-
tion. Journal of Mathematical Chemistry. 2001; 30:121–31.
3. Franco JM. Runge-Kutta methods adapted to the numeri-
cal integration of oscillatory problems. Numer Math Appl.
2004; 50:427–43.
4. de Vyver HV. An embedded exponentially fitted Runge-
Figure 5. Efficiency curves for all methods with Kutta-Nystom method for the numerical solution of orbital
h = 0.00625, 0.0125, 0.025 and 0.05 for b = 10000 in problems. New Astron. 2006; 11:577–87.
problem 4. 5. Gautschi W. Numerical integration of ordinary differential
equation based on trigonometrica polynomials. Numer
Math. 1961; 3:381–97.
6. Lyche T. Chebyshevian multistep methods for ordinary dif-
ferential equation. Numer Math. 1972; 19:65–75.
7. Aguiar JV, Simos TE. Review of multistep methods for the
numerical solution of the radial Schrodinger equation. Int J
Quant Chem. 2005; 103:278–90.
8. Bettis DG. Runge-Kutta algorithms for oscillatory prob-
lems. Zeitschrift fur Angewandte Mathematik und Physik.
1979; 30:699–704.
9. Paternoster B. Runge-Kutta (Nystrom) methods for ODEs
with periodic solutions based on trigonometric polynomi-
als. Applied Numerical Mathematics. 1998; 28:401–12.
10. Berghe GV, De Meyer H, Daele MV, Van Hecke T.
Exponentially fitted Runge-Kutta methods. Comput Phys
Figure 6. Efficiency curves for all methods in problem
Comm. 2000; 125:107–15.
5using h = 0.00625, 0.0125, 0.025 and 0.05 for b = 10000
11. Berghe GV, De Meyer H, Van Daele M, Van Hecke T.
Exponentially fitted explicit Runge-Kutta methods.
Comput Phys Comm. 1999; 123:7–15.
4. Discussion and Conclusion 12. Zhang Y, Haitao, Fang Y, You X. A new trigonometrically
fitted two derivative Runge-Kutta method for the numerical
This study has presented a newly developed solution of the Schrodinger equation and related problems.
Trigonometrically Fitted Modified Runge-Kutta Journal of Applied Mathematics. 2013; 937858:9.
(TFMRK) method in solving first-order ordinary differ- 13. Fang Y, You X, Ming Q. Trigonometrically-fitted two
ential equations with periodic solutions. Figures 2, 3, 4, derivative Runge-Kutta method for solving oscillatory dif-
5 and 6 demonstrate the numerical findings of this study. ferential equations. Numer Algor. 2014; 65:651–67.
Efficiency curves shown in these figures display that the 14. Butcher JC. Numerical Methods for Ordinary Differential
common logarithm of maximum global error throughout Equations. Wiley & Sons LTD., England. 2008.
the integration versus computational cost is measured by 15. Simos TE, Aguiar JV. A modi_ed Runge-Kutta method
with phase-lag of order infinity for the numerical solution
a number of function evaluations. This method was devel-
of the Schrodinger equation and related problems. Comput
oped based on Dormand’s seven stage fourth algebraic
Chem. 2001; 25:275–81.
order. From Figure 2-6, numerical results demonstrate 16. Sakas DP, Simos TE. A Fifth Order Trigonometrically-
that the new method is more accurate and efficient when Fitted Modified Runge-Kutta Zonneveld Method for the
solving first-order differential equations with oscillatory Numerical Solution of Orbital problems. Math Comp.
solutions compared to the existing methods. 2005; 42:903–20.

6 Vol 9 (48) | December 2016 | www.indjst.org Indian Journal of Science and Technology
 F. A. Fawzi, N. Senu, F. Ismail and Z. A. Majid

17. Hairer E, Norsett SP, Wanner G. Solving Ordinary 20. Stiefel E, Bettis DG. Stabilization of Cowell’s methods.
Differential Equations I, Nonstiff Problems. Springer, Numer Math. 1969; 13:154–75.
Berlin. 1993. 21. Moo KW, Senu N, Ismail F, Suleiman M. New phase-fit-
18. Chawla MM, Rao PS. High-accuracy P-stable methods for ted and amplification-fitted fourth-order and fifth-order
IMA Journal of Numerical Analysis. 1985; 5:21–220. Runge-Kutta-Nystrom methods for oscillatory problems.
19. van der Houwen PJ, Sommeijer BP. Explicit Runge-Kutta Abstract and Applied Analysis. 2013; 939367:9.
(-Nystrom) methods with reduced phase errors for com- 22. Allen RC Jr, Wing GM. An invariant imbedding algorithm
puting oscillating solutions. SIAM Journal on Numerical for the solution of inhomogeneous linear two-point bound-
Analysis. 1987; 24(3):595–617. ary value problems. J Computer Physics. 1974; 14:40–58.

Vol 9 (48) | December 2016 | www.indjst.org Indian Journal of Science and Technology 7