numerical solution of boundary-value and initial-boundary-value problems using...
TRANSCRIPT
Ghulam Ishaq Khan Institute of
Engineering Sciences and Technology
NUMERICAL SOLUTION OF
BOUNDARY-VALUE
AND
INITIAL-BOUNDARY-VALUE PROBLEMS
USING SPLINE FUNCTIONS
By
Fazal-i-Haq
Supervised by
Professor Dr. Syed Ikram Abbas Tirmizi
Co-Supervisor
Dr. Siraj-ul-Islam
This thesis is submitted in partial fulfillment of
The requirement of Degree of Doctor of Philosophy (PhD)
in Engineering Sciences
May 2009
Faculty of Engineering Sciences
GIK Institute of Engineering Sciences and Technology,
Topi, Pakistan.
ii
Dedicated to
My Late Mother,
Wife and Children
iii
iv
Acknowledgements
I would like to bow my head before Allah Almighty, the Most Gracious and the Most
Merciful, whose benediction bestowed upon me talented teachers, provided me
sufficient opportunity and enabled me to undertake and execute this research work.
Throughout the completion of this work, I have been supported and guided by
several people. I would like to take this opportunity to express my gratitude to all those
people. My deepest and sincere gratitude and appreciation goes to my supervisor
Professor Dr. S. I. A. Tirmizi for his guidance at each stage of this work. His patience,
encouragement and support have been very valuable in the completion of this thesis.
His advices were always stimulating and helpful when I was facing difficulties in my
research. His mission of producing high-quality work will always help me grow and
expand my thinking. Secondly, I wish to express a sincere acknowledgement to my co-
supervisor Dr. Siraj-ul-Islam for all the guidance, kind comments and constructive
criticism which have made the long hours of labor worthwhile. From the very
beginning, his dedication to the work, his concern for my well being and his confidence
in me and my work gave me the energy and inspiration to work through the difficulties.
I am also thankful to all my teachers in the Faculty of Engineering Sciences for their
encouragement during this research work.
I wish to record my deepest obligations to friends, parents, brothers and sisters for
their unfailing support and many sacrifices at every stage during these years. I would
like to thank my late mother, wife and children for their love, care, understanding,
encouragement and enduring support. I am also grateful to entire staff and research
students of faculty of Engineering Sciences, GIK Institute of Engineering Sciences and
Technology, Topi, for their help and support in one way or another during the course of
my research.
I would also like to gratefully acknowledge NWFP Agricultural University
Peshawar for study leave and to Higher Education Commission (HEC) Islamabad for
financial assistance during my Ph.D.
Fazal-i-Haq
May, 2009
v
Published and submitted papers
Certain aspects of this dissertation are based on the following published/submitted
papers:
Published Journal papers
[1] S. Islam, S. I. A. Tirmizi, F. Haq (2007): Quartic non-polynomial splines
approach to the solution of a system of second-order boundary-value problems.
Int J. High Performance Computing Applications 21 (1), 42-49.
[2] S. Islam, S. I. A. Tirmizi, F. Haq, M. A. Khan (2008): Non-Polynomial splines
approach to the solution of sixth-order boundary-value problems. Appl. Math.
Comput. 195, 270-284.
[3] S. Islam, S. I. A. Tirmizi, F. Haq, S. K. Taseer (2008): Family of numerical
methods based on non-polynomial splines for solution of contact problems.
Comm. Nonlinear Science and Numer. Simul. 13, 1448-1460.
[4] S. I. A. Tirmizi, F. Haq, S. Islam (2008): Non-polynomial spline solution of
singularly perturbed boundary-value problems. Appl. Math. Comput. 196, 6-16.
Submitted Journal papers
[5] F. Haq, S. I. A. Tirmizi, S. Islam (2009): A numerical solution of Kuramo-
Sivashinsky equation using quartic and quintic B-splines. J. Comput. Math.
[6] F. Haq, S. Islam, S. I. A. Tirmizi (2009): Numerical solution of MRLW
equation using quintic B-spline functions. Appl.Math. Comput.
[7] S. Islam, S. I. A. Tirmizi, F. Haq (2009): Collocation method using quartic B-
Splines for numerical solution of the modified equal width equation. J. Comput.
Appl. Math.
[8] S. I. A. Tirmizi, F. Haq, S. Islam (2009): A numerical technique for solution of
the MRLW equation using quartic B-splines. Appl. Math. Modell.
Fazal-i-Haq
vi
Abstract
The following two types of problems in differential equations are investigated:
(i) Second and sixth-order linear and nonlinear boundary-value problems in
ordinary differential equations using non-polynomial spline functions.
(ii) One dimensional nonlinear Initial-boundary-value problems in partial
differential equations using B-spline collocation method.
Polynomial splines, non-polynomial splines and B-splines are introduced. Some
well known results and preliminary discussion about convergence analysis of
boundary-value problems and stability theory are described.
Quartic non-polynomial spline functions are used to develop numerical methods for
computing approximations to the solution of linear, nonlinear and system of second-
order boundary-value problems and singularly perturbed boundary-value problems.
Convergence analysis of the method is discussed.
Numerical methods for computing approximations to the solution of linear and
nonlinear sixth-order boundary-value problems with two-point boundary conditions are
developed using septic non-polynomial splines. Second-, Fourth- and Sixth-order
convergence is obtained.
Numerical method based on collocation method using quartic B-spline functions for
the numerical solution of one-dimensional modified equal width (MEW) wave equation
is developed. The scheme is shown to be unconditionally stable using Von-Neumann
approach. Propagation of a single wave, interaction of two waves and Maxwellian
initial condition are discussed.
Algorithms based on quartic and Quintic B-spline collocation methods are designed
for the numerical solution of the modified regularized long wave (MRLW) equation.
Stability analysis is performed. Propagation of a solitary wave, interaction of multiple
solitary waves, and generation of train of solitary waves are also investigated.
Quartic and quintic B-spline functions have been used to develop collocation
methods for the numerical solution of Kuramoto-Sivashinsky (KS) equation. Also,
using splitting technique, the equation is reduced to a problem of second order in space.
Using error norms L2 and L∞ and conservative properties of mass, momentum and
energy, accuracy and efficiency of the suggested methods is established through
comparison with the existing numerical techniques. Performance of the algorithms is
tested through application of the methods on benchmark problems.
vii
CONTENTS
Acknowledgement iv
Published and submitted papers v
Abstract vi
Contents vii
CHAPTER 1 Introduction
1.1 Historical note and literature survey 1
1.2 Existence and uniqueness of two-point boundary-value problems 3
1.3 Polynomial and non-polynomial spline functions 5
1.4 B-Splines 7
1.5 Von Neumann stability analysis 8
1.6 Convergence of two-point boundary-value problems 9
1.7 Newton’s method 13
1.8 Summary 14
CHAPTER 2 Non-Polynomial Splines Methods for the Solution of
Second-Order Boundary-Value Problems
2.1 Introduction 17
2.2 Numerical Methods 20
2.2.1 Second-order convergence 23
2.2.2 Fourth-order convergence 23
2.2.3 Sixth-order convergence 24
2.3 Convergence Analysis 24
2.4 Numerical results and discussion 27
2.4.1 Linear problems 28
viii
2.4.2 Non-linear problem 29
2.4.3 System of differential equations 29
2.4.4 Singular perturbation problems 30
2.5 Conclusion 51
CHAPTER 3 Non-Polynomial Splines Approach to the Solution of
Sixth-Order Boundary-Value Problems
3.1 Introduction 52
3.2 Numerical Methods 54
3.3 Numerical methods of different orders 60
3.3.1 Second-order convergence 60
3.3.2 Fourth-order convergence 60
3.3.3 Sixth-order convergence 61
3.3.4 Eight-order convergence 61
3.4 Properties of the coefficient matrix A0 62
3.5 Convergence 66
3.6 Numerical results and discussion 68
3.6.1 Non-linear problems 68
3.6.2 Linear problems 68
3.7 Conclusion 76
CHAPTER 4 Collocation Method using Quartic B-Spline for
Numerical Solution of the Modified Equal Width
Equation
4.1 Introduction 77
4.2 Quartic B-spline solution 78
4.3 Stability analysis 81
4.4 Test problems and discussion 82
4.4.1 A single solitary wave 83
3.3.2 Interaction of two solitary waves 84
3.3.3 The Maxwellian initial condition 85
4.5 Conclusion 95
ix
CHAPTER 5 Solitary Wave Solutions of the Modified Regularized
Long Wave Equation
5.1 Introduction 96
5.2 The B-spline collocation methods 97
5.2.1 Quartic B-spline collocation method 97
5.2.2 Quintic B-spline collocation method 99
5.3 Stability of the proposed scheme 101
5.3.1 Stability of scheme based on quartic B-spline
collocation method
101
5.3.2 Stability of scheme based on quintic B-spline
collocation method
102
5.4 Numerical Tests and Results 103
5.4.1 Single solitary wave 103
5.4.2 Two solitary waves 105
5.4.3 Three solitary waves 106
5.4.4 The Maxwellian initial condition 106
5.5 Conclusion 122
CHAPTER 6 Collocation Method using B-Splines for Numerical
Solution of Kuramoto-Sivashinsky Equation
6.1 Introduction 123
6.2 The B-spline collocation methods 124
6.2.1 Quintic B-Spline Collocation Method 1 124
6.2.2 Quintic B-Spline Collocation Method 1I 126
6.2.3 Quartic B-Spline Collocation Method 129
6.3 Numerical validation 131
6.4 Conclusion 137
Future Work 138
References 140
Chapter 1
Introduction
1.1 Historical note and literature survey
The rapid development of spline functions is primarily due to their great usefulness in
applications. Classes of spline functions possess many nice structural properties as well
as excellent approximation powers. Splines have many applications in the numerical
solution of a variety of problems in applied mathematics and engineering. Some of
them are, data fitting, function approximation, Integro-differential equations, optimal
control problems, Computer-Aided Geometric Design (CAGD), Wavelets and so on.
Programs based on spline functions have found their way in most of computer
applications.
It is commonly accepted that the first mathematical reference to splines was made in
the year 1946 in an interesting paper by Schoenberg (Schoenberg [121]) , which is
probably the first place that the word "spline" is used in connection with smooth,
piecewise polynomial approximation. However, the ideas have their roots in the aircraft
and shipbuilding industries. Splines are types of curves, originally developed for ship-
building in the days before computer modeling. Naval architects needed a way to draw
a smooth curve through a set of points. The solution was to place metal weights
(called knots) at the control points, and bend a thin metal or wooden beam (called
a spline) through the weights. The physics of the bending spline meant that the
influence of each weight was greatest at the point of contact and decreased smoothly
further along the spline. To get more control over a certain region of the spline, the
draftsman simply added more weights. This scheme had obvious problems with data
exchange. There was a need for mathematical way to describe the shape of the
curve. Cubic Polynomials splines are the mathematical equivalent of the draftsman's
wooden beam. Through the advent of computers, splines have gained more importance.
They were first used as a replacement for polynomials in interpolation and then as a
tool to construct smooth and flexible shapes in computer graphics.
Chapter 1 Introduction
2
As late as 1960, there were no more than a handful of papers mentioning spline
functions by name. Some of the papers which have made great contributions in the
development of splines include (Ahlberg and Nilson [6], Brikhoff and Garabedian [25],
Loscalzo and Talbot [85], Maclaren [86], Rubin and Khosla [115], Sastry [120],
Schoenberg [122]). Univariate splines were studied intensely in the 60s, and by the
mid-70s they were sufficiently well understood to permit a fairly comprehensive
treatment in books form. Some of the books which discuss splines thoroughly include
(Ahlberg et al. [7], deBoor [35], Prenter [102], Schumaker [123], Shikin and Plis [124],
Spath [136]). Some of the earliest papers using spline functions for smooth approximate
solution of ordinary and partial differential equations (PDEs) include (Albasiny and
Hoskins [11], Bickely [21], Crank and Gupta [32], Jain and Aziz [64], Jain and Aziz
[65], Rubin and Khosla [115], Sastry [120], Usmani [152], Usmani and Sakai [153],
Usmani and Warsi [154]) . These papers demonstrate the approximate methods of
solving second-, third- and fourth-order linear boundary-value problems (BVPs) and
solution of elliptic and parabolic equations by spline functions of various degrees.
Today, there are hundreds of research papers on this subject, and it remains an active
research area. Recently, non-polynomial spline method has turned out to be an effective
tool for solving ordinary and partial differential equations. In many papers various
techniques using quadratic, cubic, quartic, quintic, sextic, septic and higher degree non-
polynomial splines have been discussed for the numerical solution of linear and non-
linear BVPs. Non-polynomial splines were used for numerical solution of system of
second-order BVPs in (Islam et al. [55], Islam and Tirmizi [56], Islam et al. [59]) and
numerical solution of second-order singularly perturbed BVPs has been obtained in
(Rashidinia et al. [108], Rashidinia and Mahmoodi [112], Tirmizi et al. [143]) . Third-
order BVPs have been treated in (Islam et al. [54], Islam and Tirmizi [57]) and fourth-
order equations are discussed in (Islam et al. [58], Islam et al. [61], Ramadan et al.
[105], Rashidinia and Jalilian [109], Siddiqi and Akram [127], Siddiqi and Akram
[129], Van-Daele et al. [155]) . Fifth-order BVPs have been considered in (Islam et al.
[62], Siddiqi and Akram [126], siddiqi et al. [131]) and sixth-order problems are
discussed in (Akram and Siddiqi [8], Islam et al. [60], Ramadan et al. [106]). A survey
on recent spline techniques for solving boundary value problems in ordinary differential
equations using cubic, quintic and sextic polynomial and non-polynomial splines is
given in (Kumar and Srivastava [82]) . Techniques for the solution of eight-, tenth- and
twelfth- order BVPs using non-polynomial spline are applied in (Rashidinia et al.
Chapter 1 Introduction
3
[110], Siddiqi and Akram [128], Siddiqi and Akram [130]). The use of non-polynomial
splines for the numerical solution of parabolic equations was proposed in (Ramadan et
al. [104], Rashidinia and Mahmoodi [111]).
1.2 Existence and uniqueness of two-point BVPs
In chapters 2 and 3 numerical methods for the solution of boundary value problems of
second and sixth orders are considered. Since existence and uniqueness theory plays an
important role in analyzing numerical methods for solving BVPs, therefore, some of its
aspects are studied here. Consider the second-order linear differential equation
0 1 2 , , ,p x y x p x y x p x y x r x x a b (1.1)
where the functions 0 1 2, ,p x p x p x and r x are continuous in ,a b . Consider
the general boundary conditions of the form:
1 0 1 0 1
2 0 1 0 1
,
,
l y a y a a y a b y b b y b A
l y c y a c y a d y b d y b B
(1.2)
where , , , , 1, 2i i i ia b c d i and ,A B are given constants. The particular types of the
boundary conditions (1.2) are:
i. Dirchlet boundary conditions,
, ;y a A y b B (1.3)
ii. Neuman’s boundary conditions,
, ,
or , ;
y a A y b B
y a A y b B
(1.4)
iii. Mixed boundary conditions,
2 20 1 0 1
2 20 1 0 1
, 0
, 0;
a y a a y a A a a
d y b d y b B d d
(1.5)
iv. Periodic boundary conditions,
, .y a y b y a y b (1.6)
The homogenous BVP associated with (1.1)-(1.2) takes the form
0 1 2 0, , ,p x y x p x y x p x y x x a b (1.7)
1 20, 0.l y l y (1.8)
Chapter 1 Introduction
4
Theorem 1.1 (Agarwal and O'Regan [5]) Let 1y x and 2y x be any two linearly
independent solutions of the DE (1.7). Then, the homogeneous boundary value problem
(1.7)-(1.8) has only the trivial solution if and only if
1 1 1 2
2 1 2 2
0.l y l y
l y l y
(1.9)
Theorem 1.2 (Agarwal and O'Regan [5]) The non-homogeneous boundary value
problem (1.1)-(1.2) has a unique solution if and only if the homogeneous boundary
value problem (1.7), (1.8) has only the trivial solution.
Consider the second-order nonlinear DE
, , , , ,y f x y y x a b (1.10)
subject to the boundary conditions
1 20 ,y a A y b A (1.11)
where constants 1A and 2A are real finite and arbitrary.
The following theorem given in (Isaacson and Keller [52]) guarantees the existence of
a unique solution of the BVP (1.10)-(1.11).
Theorem 1.3 (Isaacson and Keller [52]) Suppose the function f in BVP (1.10)-(1.11)
satisfies the following conditions
(i). f and its partial derivatives y
ff
y
and y
ff
y
are all continuous on
, , , , ;D x y y a x b y y
(ii). , , 0yf x y y on some D for some ;
(iii). constant K exists, with
, ,
for all , ,f x y y
K x y y Dy
.
Then solution of BVP (1.10)-(1.11) exists and is unique.
For existence and uniqueness of BVPs of 6-th order we refer to (Agarwal [4]); which
provides a detailed description of the existence and uniqueness theorems for the
solution of n-the order linear and nonlinear boundary value problems in ordinary
differential equations (ODEs).
Chapter 1 Introduction
5
1.3 Polynomial and non-polynomial spline functions
Ordinary and partial differential equations are useful in describing mathematical
models for various physical processes. While there are many theoretical results on
existence, uniqueness, and properties of solutions of such equations, usually only the
simplest specific problems can be solved explicitly, especially when the nonlinear
terms are involved and we usually construct approximate solutions. Since only limited
classes of the equations are solved by analytical means, numerical solution of these
differential equations is of practical importance. Polynomials have long been the
functions most widely used to approximate other functions mainly because of their
simple mathematical properties. However, it is well-known that polynomials of high
degree tend to oscillate strongly and in many cases they are liable to produce very poor
approximations. With a spline function we combine low degree and hence weakly
oscillating polynomials in such a way as to obtain a function which is as smooth as
possible in the sense that it has maximal continuity without being globally a
polynomial. Spline functions can be integrated and differentiated due to being
piecewise polynomials and can be easily stored and implemented on digital computers.
Thus, spline functions are adapted to numerical methods to get the solution of the
differential equations. Numerical methods with spline functions in getting the
approximate solution of the differential equations lead to band matrices which are
solvable easily with algorithms having low cost of computation.
Definition 1.1 Let be the partition 0 1 ... 1nx x x n m of the interval ,a b .
Then a spline s of degree m with knots at is a function possessing the following
two properties
(i). In each subinterval 1[ , )i ix x of ,a b , s is a polynomial of degree m or less.
(ii). s and its derivatives of order 1,2,..., 1m are continuous on ,a b , i.e.
1 ,ms C a b .
The space all such functions is denoted by mS .
Thus a spline function is a series of polynomial arcs of degree m or less, joined
together in such a way that the function and its derivatives of orders 1m or less are
Chapter 1 Introduction
6
continuous everywhere. The spline is, in general, a different polynomial in each of the
subintervals 1[ , )i ix x and the continuity constraint 1 ,ms C a b imposes maximal
continuity on this piecewise defined function.
Piecewise non-polynomial splines are a blend of trigonometric as well as
polynomial basis functions which form a complete extended Chebyshev space. This
approach ensures enhanced accuracy and general form to the existing spline based
algorithms. A parameter is introduced in the trigonometric part of the spline function.
The C -differentiability of the trigonometric part of non-polynomial splines
compensates for the loss of smoothness inherited by polynomial splines. It is well
known that the Bézier basis is a basis for the space of degree-n algebraic polynomials
21, , ,..., .nnP span x x x (1.12)
A new basis, called the C-Bézier basis, is constructed in (Chen and Wang [30]) for the
space
2 21, , ,..., , cos ,sin ,nn span x x x x x (1.13)
in which 1nx and nx in (1.12) are replaced by cos x and sin x .
A part of this thesis presents numerical methods for the numerical solution of linear
and nonlinear second and sixth-order ordinary differential equations as well as system
of BVPs based on the use of non-polynomial spline functions. Consider the numerical
solution of Nth-order linear and non-linear ordinary differential equations of the form
, ,... 0,N
N
d yF x y
dx
(1.14)
In order to develop the numerical method for approximating solution of boundary value
problems like the one given in Eq. (1.14), the interval [ , ]a b is partitioned into 1n
uniformly spaced points ix such that 0 1 1... and n n
b aa x x x x b h
n
. For
each i-th segment, the non-polynomial spline function xPi , equivalent to m-th degree
polynomial, has the form,
2 2
0 1 2 ( 2)
( 1) ( ) ,
0,1,..., 1,
...
sin cos
mi i i ii i i m i
i im i m i
i n
P x a a x x a x x a x x
b k x x c k x x
(1.15)
where the coefficients are arbitrary real and finite constants and k is free parameter.
iP x defined in Eq. (1.15) reduces to usual m-th degree spline in ,a b when 0k .
Chapter 1 Introduction
7
The continuity conditions for non-polynomial spline iP x and its derivatives upto
order 1m , are exploited at the point ,i ix y to get the main scheme for the numerical
solution of the BVPs. The main advantage of non-polynomial based schemes is to
obtain methods of different order with higher accuracy. Computational cost of non-
polynomial spline based methods is the same as that of polynomial splines. In order to
get a unique solution, we require more equations at each end of the range of integration.
These equations are developed by Taylor series and method of undetermined
coefficients. The schemes generally lead to systems of algebraic equations involving
banded matrices and can easily be solved. Newton method is used in the case of
nonlinear system.
1.4 B-Splines
The term B-spline was coined by Isaac Jacob Schoenberg and is short for basis spline.
A B-spline is a spline function that has minimal support with respect to a given degree,
smoothness, and domain partition. A detailed description of B-splines can be found in
(deBoor [35], Prenter [102]). We specifically consider the type of basis for which the
knots are equidistant. In such a case the B-spline is said to be uniform. A fundamental
theorem states that every spline function of a given degree, smoothness, and domain
partition can be represented as a linear combination of B-splines of that same degree
and smoothness, and over that same partition. This feature immediately implies the
band structure of matrices that appear in interpolation and collocation problems;
moreover some of the properties of B-splines imply total positivist. In consequence,
these matrices can be efficiently stored in a computer and inverted by Gaussian
elimination without pivoting. The problem of stable calculation of polynomial B-
splines and their derivatives has been completely solved since 1975. The derivation of
B-splines may be found, for example, in (Prenter [102]).
The B-spline functions are used as a basis functions in finite element method,
Galerkin method and collocation method to construct numerical methods for the
approximate solutions of BVPs occurring in various engineering applications.
Application of the Galerkin method accompanied by higher-degree polynomials results
in a higher-degree matrix system. That brings a burden for numerical analysis, and the
computational cost of the matrix system increases in the evaluation of both linear and
Chapter 1 Introduction
8
nonlinear systems. On the other hand, the collocation method together with B-spline
approximations represents an economical alternative since it only requires the
evaluation of the unknown parameters at the grid points. Papers on solving ODEs and
PDEs using B-splines are quite numerous of which we mention some with quartic and
quintic B-splines. In Refs. (Caglar et al. [26], Saka and Dag [117]) quartic B-splines
are used and solution of various PDEs is obtained by employing quintic B-spline
collocation method in (Ramadan and El-Danaf [103], Raslan [113], Saka [116], Saka
and Dag [118], Zaki [161]). In this thesis methods are developed for the numerical
solution of initial-BVPs in PDEs. The time derivative of given equation is discretized
by a first order accurate forward difference formula and the -weighted, (0 1)
scheme is applied to the space derivative at two adjacent time levels. The B-spline
collocation method over finite intervals is also applied to the space-split equations.
These methods provide satisfactory results in terms of accuracy and ease of
implementation.
1.5 Von Neumann stability analyses
The stability of numerical schemes is closely associated with numerical error. A
solution is said to be unstable if errors introduced at some stage in the calculations (for
example, from erroneous initial conditions or local truncation or round-off errors) are
propagated without bound throughout subsequent calculations. Thus a method is stable
if the difference between the theoretical and numerical solutions remains bounded at a
given time t , as time and space steps tend to zero or time step remains fixed at every
level and t . There are three common methods of investigating stability: the
energy method, the matrix method, and the Von Neumann method. A detailed
description of these methods may be found, for example, in (Mitchell and Griffiths
[89], Twizell [147]).
In chapters 4 and 5 the stability of numerical schemes is investigated by Von
Neumann method developed in the early 1940s. Here we mention this method briefly.
The method is linearized before applying Von Neumann method. The descritized
solution in the nonlinear term is replaced by a constant value. Following (Twizell
[147]) , suppose the errors are given by
,n n nm m mZ u U (1.16)
where nmu is the theoretical and n
mU is the approximate solution of the scheme so that
Chapter 1 Introduction
9
nmU contains approximate errors. The approach is to find the conditions under which
Eq. (1.16) remains bounded as n increases for fixed step size h and time step t . Von
Neumann method expresses the initial line of errors 0mZ in terms of finite Fourier series
and finds the criterion governing the growth of a function which reduces to this series
for 0t . The finite Fourier series for the errors at the 2N points is given by
10
0
exp( ),N
m r rr
Z A i mh
(1.17)
where 0,1,..., 1, 1m N i and rA denote the Fourier coefficients. Suppose the
scheme is linear (If it is nonlinear, the approximate solution in the nonlinear term is
replaced by a constant value), then by the superposition principal it is only necessary to
consider the error propagation due to a single term. Consider the term associated with
the typical frequency s and the coefficient sA being constant can be neglect. To
find the propagation of this single, typical error as t increases, it is necessary to find the
solution of the scheme which will reduce to exp i mh at 0t . Such a solution is
given by
exp exp exp exp ,t i x n t i mh (1.18)
where is a function of and is complex. It is clear that error term exp i mh will
not grow with time when
exp 1 .t (1.19)
Introducing exp t , we have
1. (1.20)
Condition (1.20) is called von Neumann criterion for stability. Here is called the
amplification factor.
1.6 Convergence of two-point BVPs
Consider the solution of nth-order ordinary differential equations of the form
2
2, , , ,... 0
n
n
dy d y d yF x y
dx dx dx
(1.21)
with the prescribed boundary conditions. Let the error between the approximate and
exact solutions be denoted by
Chapter 1 Introduction
10
, 0,1,..., .i i ie y y x i n (1.22)
A numerical procedure is said to be convergent if
0lim 0,ih
e
(1.23)
where h is the step size.
The following definitions and some basic results are needed to establish
convergence of the numerical methods.
Definition 1.2 A matrix norm . : m n is a nonnegative real valued function
which satisfies the following axioms:
0 when 0 and 0 iff 0;
for all ;
;
( ) .
i A A A A
ii A A
iii A B A B
iv AB A B
Definition 1.3 Suppose that is a linear space over the field of real numbers. The
nonnegative real-valued function . is called a norm on the space if it satisfies the
following axioms:
0 when 0 and 0 iff 0 in ;
for all and in ;
for all and in (Triangle inequality).
i v v v v
ii v v v
iii u v u v u v
A linear space , equipped with a norm, is called a normed linear space.
Definition 1.4 The linear space 2 ,L a b is the collection of all measurable functions
f x defined on ,a b for which
2
,b
af x dx
where integration is Lebesgue integration.
Definition 1.5 Let ,f x C a b (The linear space of all continuous functions on
,a b ) or 2 ,f x L a b . The real valued function 2
. defined by
122
2
b
af f x dx
Chapter 1 Introduction
11
is called 2L norm of f .
In this thesis we have used the following approximation for 2L error norm:
12 22
21
,Nb
i iai
L u U dx h u U
where u and U are exact and approximate solutions respectively.
Definition 1.6 A vector norm . : n is a nonnegative function satisfying the
following axioms:
0 when 0 and 0 iff 0 in ,
,
.
n
n
i
ii
iii
v v v v
v v
u v u v u,v
The following vector norms are in common use in numerical linear algebra: the 1-norm
1. , the 2-norm ( Euclidean norm)
2. , norm .
. If T
1 2, ,..., nnv v v v ,
these norms are defined as follows:
Definition 1.7 The 1-norm 1
. of the vector v is defined by
11
.n
ii
v
v
Definition 1.8 The 2-norm ( or Euclidean norm) 2
. of the vector v is defined by
12
12
2T
21
.n
ii
v
v v v
Definition 1.9 The norm .
of the vector v is defined by
1
max .n
ii
v
v
Definition 1.10 A matrix A is convergent to zero if the sequence of matrices
2 3, , ,...A A A converges to the null matrix 0.
Theorem 1.4 (Twizell [147]) For a given matrix A , the necessary and sufficient
condition for the sequence 2, , ,...I A A to converge to the zero matrix is that every eigen
value of A should be less than unity in modulus.
Chapter 1 Introduction
12
Theorem 1.5 A necessary and sufficient condition for the series 2 3 ...I A A A
converges to 11 A
is that lim 0r
rA
.
Proof The necessity of the hypothesis is obvious. To establish sufficiency, Theorem
1.4 is used; this shows that if lim 0r
rA
, then all eigen values of A are less than unity
in modulus. This shows that the determinant of the matrix I A is not zero and thus the
matrix 1I A
exists.
Consider next the formula
2 3 1... .r rI A I A A A A I A
Since 1I A
exists, it follows that
12 3 1
1 1 1
1
...
,
r r
r
I A A A A I A I A
I A I A A
I A
since 1 0rA as .r
The series 2 3 ...I A A A thus converges to 1I A
if 0rA as .r
Lemma 1.1 For any matrix norm with 1,I then the following inequality holds,
111 .I A A
Proof The proof follows from Theorem 1.5 by replacing A by .A
Lemma 1.2 A sufficient condition for convergence of the series 2 3 ...I A A A is
that 1A for any norm of a matrix A .
Lemma 1.3 Let A be some matrix; then, if 1A for some norm, then the following
inequality holds,
1 1... 1 .
rrI A I A A A A
Proof It is easy to see the from Theorem 1.5 that
1 1 2... ...r r rI A I A A A A
Chapter 1 Introduction
13
2
1 1 2
1
1 2
1
... ...
...
1 ...
if 1.
r
r r r
r
r
r
I A I A A A A
A A
A A A
A I A A
The previous results can be summarized in the following theorem.
Theorem 1.6 The following statements are equivalent:
(i) A is convergent matrix,
(ii) lim 0n
nA
for some norm . ,
(iii) 1A where A is spectral radius of A .
Definition 1.11 (Agarwal and O'Regan [5]) A problem consisting of a PDE in a domain
with a set of initial and / or boundary conditions is said to be well-posed if the
following three properties hold:
i. Existence: There exists at least one solution of the problem.
ii. Uniqueness: There is at most one solution of the problem.
iii. Stability: The unique solution depends continuously on data (initial and boundary
conditions): i.e., a slight change in the data leads to only a small change in the solution.
Definition 1.12 A class of nonlinear partial differential equations possesses a special
type of elementary solution. These solutions known as solitons, which have the form of
localized waves that conserve their properties even after interaction among them, and
then act somewhat like particles. These equations have a number of local conserved
quantities, like energy, mass and momentum. Solitons were first appeared in the
solution of the Korteweg-de Vriese equation. When a soliton encounters another
soliton of arbitrary size or velocity it can change beyond recognition for a short or long
period, but ultimately it will revert to its original shape.
1.7 Newton’s method
For solution of nonlinear systems, the well known Newton’s method is used to solve
the nonlinear system of equations. For the sake of completeness the following results
Chapter 1 Introduction
14
are quoted from (Suli and Mayers [137])
Definition 1.13 The recursion defined by
1
, 0,1, 2,...,k+1 k k kfJ k
x x x f x
(1.24)
where n0x , is called Newton’s method (or Newton’s iteration) for the system of
equations 0f x . It is implicitly assumed that the Jacobian matrix kfJ x exists
and is non-singular for each 0,1, 2,...k
Definition 1.14 Suppose that kx is a convergent sequence in n and lim k
kx .
We say that kx converges to with at least order 1q , if there exist a sequence
k of positive real numbers converging to 0, and 0 , such that
, 0,1, 2,...,kk k
x and 1lim .k
qkk
(1.25)
If (1.25) holds with , 0,1, 2,...,kk k k
x then the sequence kx is said
to converge to with order q. In particular, if 2q , then we say that the sequence
kx converges to quadratically.
The next theorem is concerned with the convergence of Newton’s method.
Theorem 1.7 (Suli and Mayers [137]) Suppose that 0f , that in some (open)
neighborhood N of , where f is defined and continuous, all the second-order
partial derivatives of f are defined and continuous, and that the Jacobian matrix
fJ of f at the point is non-singular. Then, the sequence kx defined by
Newton’s method (1.24) converges to the solution provided that 0x is sufficiently
close to ; the convergence of the sequence kx to is at least quadratic.
1.8 Summary
The research work presented in this thesis consists of six chapters. Following is the
Chapter 1 Introduction
15
chapter-wise brief summary:
In chapter 2 quartic non-polynomial spline functions are used to develop numerical
methods for computing approximations to the solution of general linear, nonlinear and
system of second-order BVPs associated with obstacle, unilateral, and contact
problems. Singularly perturbed BVPs are also considered. It is shown that the
suggested methods give approximations, which are better than those produced by other
collocation, finite-difference and spline methods. Convergence analysis of the method
for obstacle problem is discussed. Numerical examples from literature are given to
illustrate practical usefulness of new approach.
In chapter 3 Non-polynomial splines, which are equivalent to seventh-degree
polynomial splines, are used to develop a class of numerical methods for computing
approximations to the solution of linear and nonlinear sixth-order BVPs with two-point
boundary conditions. Second-, Fourth- and Sixth-order convergence is obtained by
using standard procedure. The first sections explain the development of numerical
methods of various orders while the rest of the sections discuss standard convergence
analysis and numerical validation. Numerical results are given to verify the order of
convergence as predicted by the analysis.
Chapter 4 covers construction of numerical scheme based on collocation method
using quartic B-spline functions for the numerical solution of the one-dimensional
modified equal width (MEW) wave equation. Using Von-Neumann approach the
scheme is shown to be unconditionally stable. Performance of the method is validated
through test problems including single wave, interaction of two waves and use of
Maxwellian initial condition. Using error norms L2 and L∞ and conservative properties
of mass, momentum and energy, accuracy and efficiency of the suggested method is
established through comparison with the existing numerical techniques.
In chapter 5 algorithms based on quartic and quintic B-spline collocation methods
are designed for the numerical solution of the modified regularized long wave (MRLW)
equation. Stability analysis is performed using Von-Neumann approach. Performance
of the method is validated through numerical examples. Using error norms L2 and L∞
and conservative properties, accuracy of the new method is established through
comparison with the existing techniques to prove its simple applicability and
superiority. Propagation of a solitary wave, interaction of multiple solitary waves, and
generation of train of solitary waves are also investigated.
In chapter 6 quartic and quintic B-spline functions have been used to develop
Chapter 1 Introduction
16
collocation methods for the numerical solution of a class of nonlinear partial
differential equations known as Kuramoto-Sivashinsky (KS) equation. Also, by the
virtue of splitting technique, the equation is reduced to a problem of second order in
space. Performance of the schemes is tested through application of the methods on
benchmark problems. Accuracy of the methods is established through error norms L2
and L∞.
***************************************************
Chapter 2
Non-Polynomial Splines Methods for
the Solution of Second-Order
Boundary-Value Problems
2.1 Introduction
In this chapter we investigate numerical solutions of the following three types (A, B, C)
of BVPs.
Type A:
Second-order linear and nonlinear differential equations of the type
subject to the boundary conditions
Type B:
System of second-order BVPs of the type
with the boundary conditions (2.2) and the continuity conditions of yy and at c and
d. Here, f and g are continuous functions on ,a b and ,c d respectively. The
parameters 1 2, and r are real finite constants.
Type C:
The self-adjoint singularly perturbed two-point BVP of the form;
, , ,y g x y x f x y f x y a x b (2.1)
1 2,( ) and ( )y a y b (2.2)
,
f x a x c
y g x y x f x r c x d
f x d x b
(2.3)
Chapter 2 Second-order BVPs
18
where 0 1, are given constants, is a small positive parameter such that 0 1
and f and g are sufficiently smooth functions.
Type A belongs to general second second-order BVPs. Problems of this type occur
frequently in mechanical problems without dissipation (Henrici [47]). These special
BVPs also occur in other engineering context, for example Troesh’s problem relating to
the confinement of a plasma column by radiation pressure (Jones [67], Kubicek and
Hlavacek [79]). Methods of order two, four and six are described by various authors
(See (Al-Said [10], Chawala [28], Jain [63], Tirmizi [142], Tirmizi and Twizell [144],
Tirmizi and Twizell [145], Twizell and Tirmizi [150], Usmani [151], Usmani and
Warsi [154]) ).
Problems of type B arise in the study of obstacle problems. This model represents
an elastic obstacle lying over elastic string which is characterized by a sequence of
boundary value problems (2.3) with the values of y and/or y are unknown at points c
and d. The locations of free boundary become intrinsic part of the problem. The
analytical solution of the problems (2.1)-(2.3) cannot be obtained for arbitrary choices
of ,f x g x and ,f x y . In such cases, we usually resort to some numerical methods
to obtain an approximate solution. Numerical solution of problems of type B is
discussed in a series of papers (Al-Said [9], Al-Said [10], Aziz et al. [16], Islam et al.
[55], Islam et al. [59], Noor and Khalifa [95], Noor and Tirmizi [96]) using splines,
finite difference and collocation methods.
Type C belongs to singular perturbation problems (SPPs) which have applications
in various disciplines of science and engineering, for instance fluid mechanics, fluid
dynamics, elasticity, quantum mechanics, chemical reactor theory, convection-diffusion
processes, optimal control etc. SPPs are thoroughly discussed in the books (Doolan et
al. [36], Miller et al. [88], Morton [92]). The solution of SPPs exhibits a multi-scale
character; that is, there are thin layer(s) where the solution varies rapidly, while away
from the layer(s) the solution behaves regularly and varies slowly. So the numerical
treatment of SPPs give major computational difficulties and the standard methods do
not yield accurate results for all x when is very small. A variety of numerical
methods are available in the literature to solve SPPs for second-order ordinary
differential equations. For details, one may refer to survey articles by (Kadalbajoo and
0 1,
'' , 0, 0 1,
(0) and (1)
y g x y x f x g x x
y y
(2.4)
Chapter 2 Second-order BVPs
19
Patidar [70], Kumar and Singh [81]). These methods include finite difference method
(Natesan and Ramanujam [93], Niijima [94], Shishkin [125]), finite element method
(Boglaev [22], O'Riordan and Stynes [97], Vukoslavcevic and Surla [156]), boundary
value approach (BVM) (Kumar et al. [80]) and various forms of splines based methods
(Aziz and Khan [14], Aziz and Khan [15], Kadalbajoo and Bawa [69], Khan et al. [77],
Mohanty and Jha [90], Rashidinia et al. [108], Surla et al. [138], Surla and Stojanovic
[139], Surla and Vukoslavcevic [140]).
In the present chapter, we have developed a class of methods based on quartic non-
polynomial spline functions for the numerical solution of the aforementioned types of
problems. The spline function proposed here has the form:
24 1, , , cos , sin ,T Span x x kx kx
where k is the frequency of the trigonometric part of the splines function which can be
real or pure imaginary and which will be used to raise the accuracy of the method. Thus
in each subinterval 1i ix x x , we have:
21, , , cos , sin ,Span x x kx kx
21, , , cosh , sinh ,Span x x kx kx or
2 3 41, , , , , when 0 .Span x x x x k
The above correlation can be better explained by the following equation:
24
2
23 4
1, , , sin( ), cos( )
6 241, , , -sin ( ) , cos( ) 1 .
2
T Span x x kx kx
kxSpan x x kx kx kx
k k
(2.5)
Spline solution has its own advantages. For example, once the solution has been
computed, the information required for spline interpolation between mesh points is
available. This is particularly important when the solution of the BVP is required at
different locations in the interval [a, b]. This approach has the added advantage that it
not only provides continuous approximations to xy , but also to y and higher
derivatives at every point of the range of integration. Also, the C -differentiability of
the trigonometric part of non-polynomial splines compensates for the loss of
smoothness inherited by polynomial splines.
The organization of this chapter is as follows. In Section 2.2 the numerical methods
are described and methods of different orders are categorized. Convergence analysis of
Chapter 2 Second-order BVPs
20
problems of first two types is given in section 2.3 and numerical results are reported in
section 2.4. Some conclusions are drawn in section 2.5.
2.2 Numerical Methods
Let the partition of the space interval ,a b into uniformly spaced points mx be such
that 0 1 1... N Na x x x x b and b a
hN
. For simplicity we take
3
4
a bc
and 3
4
a bd
. For each i-th segment, the quartic non-polynomial spline function
xPi has the form,
2,
0,1,..., 1,
cos sini i i i i i i i i i
i n
P x a k x x b k x x c x x d x x e
(2.6)
where , , , ,i i i i ia b c d e are arbitrary real and finite constants and k is free parameter.
iP x defined in Eq. (2.6) reduces to usual quartic spline in ,a b when 0k as
mentioned in Eq. (2.5).
To derive expression for the coefficients of Eq. (2.6) in terms of
, , , ,i j i j i j i j i jy D M T S , the following relations are defined for the relevant values of j:
, , ,
, .
i i j i j i i j i j i i j i j
ivi i j i j i i j i j
P x y P x D P x M
P x T P x S
(2.7)
We obtain the following expressions from Eq. (2.6):
1 2 1/ 2 1/ 2
4 3 3 2
221/ 2
1/ 2 1/ 22 4 2 2
tan 2, , ,
cos 2 2 2
1, , .
8 8 2 2
i i i i ii i i
i i i ii i i i i
S T T M Sa b c
k k k k
T h M hT hDhd D e y S kh
k k k k
(2.8)
The continuity conditions for non-polynomial spline iP x and its first, second and
third derivatives are exploited at the point ,i ix y to get the following consistency
relations:
2
1 1/ 2 1/ 2 1/ 2 1/ 2 13 2
2 2
1 2 1/ 24 2 4 4 2 4
tan / 2( ) 3
2 8 2
cos1 1 3 1
cos 2 8 cos( / 2) 8,
i i i i i i i i
i i
h h hD D y y M M T T
k k
h hS S
k k k k k k
(2.9)
Chapter 2 Second-order BVPs
21
1 1/ 2 1 23 2
2sin / 2,i i i i
hD D hM S
k k
(2.10)
1 1/2 1/2 1/2 1/22 2 2 2
tan /2 1 1 cos( ) 1
cos( /2) cos( /2)i i i i i iT T M M S Sk k k k k
(2.11)
1 1/ 2
2sin / 2.i i iT T S
k
(2.12)
Eqs. (2.9)-(2.11) yield the following equation,
2
1/ 2 1/ 2 1/ 2 1/ 22
2
1/ 2 1/ 23 3 2
1
2 tan / 2 8
.2 sin / 2 2 tan / 2 8
i i i i i
i i
h hhD y y M M
k k
h h hS S
k k k
(2.13)
Likewise from Eqs. (2.11) and (2.12) it follows that,
1/ 2 1/ 2 1/ 2 1/ 2
1 1.
2 tan / 2 2 tan / 2 2 sin / 2i i i i i
kT M M S S
k k
(2.14)
Elimination of T’s from Eqs. (2.12) and (2.14), D’s from Eqs. (2.10) and (2.13) yield
1/ 2 1/ 2 3 / 2 3 / 2 1/ 2
1/ 2
1 12
2 tan / 2 2 tan / 2 2 sin / 2
2sin / 21 1
sin / 2 tan / 2,
i i i i i
i
kM M M S S
k k
Sk k k
(2.15)
2 2 2
1/ 2 1/ 2 1/ 2 3/ 2 1/ 22 2
2
1/ 2 1/ 22
1 cos / 2 12 2
2 8(1 cos / 2 2
1
8(1 cos / 2
i i i i i
i i
h h hS y y y M
k k
hM M
k
(2.16)
When 0k , formula derived in (Usmani [152]) becomes special case of Eq. (2.15);
that is,
4
21/ 2 1/ 2 3/ 2 1/ 2 1/ 2 3/ 22 6 .
8i i i i i i
hh M M M S S S
(2.17)
Eliminating 'iS s from Eqs. (2.16) and (2.17), we get
Chapter 2 Second-order BVPs
22
3/ 2 1/ 2 2
1/ 2 3/ 2 5/ 2 3/ 22 2
5/ 2
2 2 2 2
3/ 2 1/ 22 2
2 2 2 2
1/ 22 2
41
10 416sin / 4
4 3 2 4 cos / 2 4cos
4 sin / 4
8 19 24( 1 )cos / 2 16cos.
8 sin / 4
i i
i i i i
i
i i
i
y yh
y y M Mk
y
h k h kM M
k
h k h kM
k
(2.18)
Eq. (2.18) in simpler form renders the following equation,
5/ 2 3/ 2
25/ 2 3/ 2 3/ 2 1/ 2 1/ 2 3/ 2 1/ 2
1/ 2
4 10
i i
i i i i i i i
i
M M
y y y y y h M M
M
(2.19)
where
2 2
2 2
2 2
2 2
2 2
1 1,
16sin / 4
4 3 2 4 cos / 2 4cos,
4 sin / 4
8 19 24( 1 )cos / 2 16cos,
8 sin / 4
3, 4,..., 2.i n
(2.20)
According to Eq. (2.5), 2
0lim , , 1,76,230
48k
h
which is the formula based on
quartic polynomial spline functions derived in (Usmani [152]).
The local truncation errors , 2, 4,..., 2,it i n associated with scheme (2.19) are,
2 (2) 3 (3) 4 (4) 5 (5) 6 (6) 7 (7) 8 (8) 9
2 3 4 5 6 7 8 ,i i i i i i i it C h y C h y C h y C h y C h y C h y C h y O h (2.21)
2
3
4
5
6
7
8
8 2 2 ,
18 2 2 ,
21
64 3 34 10 ,241
48 98 26 ,48
12408 15 706 82 ,
57601
504 2882 242 ,3840
113232 7 1635 730 .
322560
C
C
C
C
C
C
C
Chapter 2 Second-order BVPs
23
Thus for different choices of , , in scheme (2.19), methods of different order are
obtained. The Eq. (2.19) gives 4n equations in n unknowns 1/ 2 , 0,1,..., 1.iy i n We
require four more equations, two at each end of the range of integration, for the direct
computation of 1/ 2iy . These equations are developed by Taylor series and method of
undetermined coefficients. The general form of initial boundary equations for the main
scheme is as follows:
250 0 0 1 2 3 41/2 3/2 5/2 1/2 3/2 5/2 7/2 9/2 010 14 3 ,y y y y h M M M M M Mh h h h h h (2.22)
250 0 0 1 2 3 41/2 3/2 5/2 7/2 1/2 3/2 5/2 7/2 9/2 0.2 3 10 4y y y y y h M M M M M Mk k k k k k (2.23)
The remaining two equations at the other end can be obtained from above equations by
writing them in reverse order. The constants ,i ia b are parameters to be determined.
2.2.1 Second-order convergence
For second-order convergence we put
2 3 271, ,
55 2 55 in the main Eq. (2.19), which gives
2 3 4
70,
330C C C .
The values of the coefficients of the first boundary equation are given by
0 1 2 3 4 5
5, 5, 1, 0.
4h h h h h h
(2.24)
The values of the coefficients of the second boundary equation are given by
0 1 2 3 4 5
11, 0, 1, 1, 0.
4k k k k k k
(2.25)
2.2.2 Fourth-order convergence
For fourth-order convergence we put
1 13, , 5
18 9r in the main Eq. (2.19), which gives
2 4 6
10,
180C C C .
The values of the coefficients of the first boundary equation are given by
0 1 2 3 4 5
2 154 75 1, , , , 0.
48 48 48 48h h h h h h
(2.26)
Chapter 2 Second-order BVPs
24
The values of the coefficients of the second boundary equation are given by
0 1 2 3 4 5
10 75 230 76 1, , , , , 0.
48 48 48 48 48k k k k k k
(2.27)
2.2.3 Sixth-order convergence
For sixth-order convergence we put
1 22 149
, ,20 15 30
r in the main Eq. (2.19), which gives
2 4 6 8
10,
15120C C C C .
The values of the coefficients of the first boundary equation are given by
0 1 2 3 4 5 1
1 1 1 1 1 1
608 40439 24059 2009 869 126, , , = , , 13440.,h h h h h h cd
cd cd cd cd cd cd
(2.28)
The values of the coefficients of the second boundary equation are given by
0 1 2 3 4 5 2
2 2 2 2 2 2
10784 157059 612759 170163 8649 406, , , = , , 120960.,k k k k k k cd
cd cd cd cd cd cd
(2.29)
2.3 Convergence analysis
Now we investigate the error analysis of the quartic non-polynomial spline method
described in section 2.2 for problems of first two types. To do so we let, ,i ic t C T
and 1/2ieE be n-dimensional column vectors. Here ~
1/ 2 1/ 2 1/ 2i i ie y y is the
discretization error for 0,1,..., 1i n . Thus, we can write our method in the matrix
form:
~
AY = C + T
A Y = C
AE = T,
(2.30)
where
20 0 , , , , ij ijh p m A A BD A PM P M (2.31)
and
Chapter 2 Second-order BVPs
25
3, 1, ,
2, 2,3,..., 1,
1, 1,
0, otherwise,
ij
i j n
i j np
i j
(2.32)
5, 1, ,
6, 2, 3,..., 1,
1, 1,
0, otherwise.
ij
i j n
i j nm
i j
(2.33)
The five-diagonal matrix 0A is given by
14 3 1
3 10 4 1
1 4 10 4 1
1 4 10 4 1
1 4 10 3
4 3 1
0A
(2.34)
and the matrix B is given by
40439 24059 2009 869 126
157059 612759 170163 8649 406
406 8649 170163 612759 157059
126 869 2009 24059 40439
B
(2.35)
and 1/2 1/2diag , 1,2,..., with 0i ig i n g D for /4 3 /4n i n . The vector C is defined by
Chapter 2 Second-order BVPs
26
21 1
21 2
2
2
2
2
110 , 1,13440
12 , 2,120960
3, 2 and 1 14 4
3, 1 and 2,4 4
3, and 1,4 4
, 1 an4
i
i
i
i
c
h F i
h F i
n nh F i i n
n nh F r i
n nh F r i
nh F r i
2
2
22 1
22
3d ,432 , 2 1 ,
4 432 2 , 3 2 ,
4 412 , 1,
120960110 , ,
13440
i
i
n
n
n
n nh F r i and
n nh F r i i
h F i n
h F i n
(2.36)
where
5 3 1 1 32 22 2 2
0 1/ 2 3/ 2 5/ 2 7/ 2 9/ 2
9/ 2 7/ 2 5/ 2
0 1/2 3/2 5/2 7/2 9/2
608 40439 24059 2009 869 126 , 1,
, 2,
, 3 2,
406 8649 170163 6
10784 157059 612759 170163 8649 406
i i ii i i
n n n
f f f f f f i
i
F f f f f f i n
f f f
f f f f f f
3/ 2 1/ 2
9/2 7/2 5/2 3/2 1/2
12759 157059 10784 1,
, .
,
126 869 2009 24059 40439 608n n n
nn n n n n
f f f i n
i nf f f f f f
(2.37)
The analysis of the convergence for our method depends on the properties of the
matrix 0A . It has been shown in (Al-Said [10]) and (Usmani [152]) that the matrix 0A
is nonsingular and that its inverse satisfies the inequality
2 21
2
3.
64
b a h
h
0A (2.38)
Thus from Eqs (2.30) and (2.31) , we have
.
2 2
2
h h
h
-1 -1-1 -1 -10 0 0
-1-1 -10 0
E = A T = A + BD T = I + A BD A T,
E I + A BD A T
Using the result 1I and 11 1
I + M M with M 1, we get the
following expression
2h
-1
0
-10
A TE .
1- A B D
(3.39)
Chapter 2 Second-order BVPs
27
We consider the following two cases for T :
(i) For 1 13
, , 518 9
, LTE is 66
1
180h MT
(ii) For 1 22 149
, ,20 15 30
, LTE is 88
1
15120h MT
where 6
6 max xxM y and
88 max xx
M y .
Using Eq. (2.38) and the fact that
B is some finite number, it follows from standard
procedure, that:
Case i 2h
E , hence the method (2.19) with second-order boundaries is second-
order convergent,
Case ii 4h
E , hence the method (2.19) with fourth-order boundaries is fourth-
order convergent,
Case ii 6h
E , hence the method (2.19) with sixth-order boundaries is sixth-
order convergent.
It is remarked that in the case of problems of type (2.3), established numerical
methods (Al-Said [9], Al-Said [10], Aziz et al. [16], Islam et al. [55], Islam and
Tirmizi [56], Noor and Khalifa [95], Noor and Tirmizi [96]) lag behind theoretical
convergence irrespective of the order of the method. The reason for this failure is that
in most of the cases, maximum error lies around the breakup points c and d of the
interval ,a b where solution and its first derivative satisfy extra continuity conditions.
Since there is no break up point in the domain set of problems of type (2.1), all the
numerical methods of various orders perform well and exhibit accuracy predicted
theoretically.
2.4 Numerical results and discussion
In order to test the efficiency of the new scheme developed in section 2.2, we apply it
on a number of problems from the literature. We use double precision arithmetic in
order to reduce the round-off errors to a minimum.
Chapter 2 Second-order BVPs
28
2.4.1 Linear problems
Problem 2.1 Consider the linear BVP
2 2 , 2 3, 2 0 and 3 0,x y y x x y y (2.40)
for which the analytic solution is
21 365 19 .
38y x x x
x
(2.41)
The above problem is the type where both f x and g x in Eq. (2.1) are variables.
Table 2.1 contains results for second-order method 2.2.1. The results using the fourth-
order method 2.2.2 are summarized in Table 2.2 and compared with quartic and quintic
spline methods (Al-Said [10], Usmani and Warsi [154]) of the same order. It is seen
that our method performs better than these methods. Table 2.3 contains results for
sixth-order method 2.2.3 which is compared with sixth-order method given in (Usmani
[151]) that uses a quindiagonal scheme. It is clear from the table that method 2.2.3 is far
superior to this method. Fig. 2.1 shows the exact and approximate solution for 1
64h .
Problem 2.2 Consider the linear BVP
with analytical solution
The above example is the type where only f x in Eq. (2.1) is variable. The results
using the fourth-order method 2.2.2 are summarized in Table 2.4 and compared with
quintic spline method (Usmani and Warsi [154]) of the same order. It is seen that
present method produces better results. Exact and approximate solution is shown in Fig.
2.2 for 1
128h .
Problem 2.3 Consider the linear BVP
with analytical solution
4 , 0 1, 0 1 0,xy y xe x y y (2.42)
1 .xy x x x e (2.43)
2 2, 0 1, 0 1 1,y y x x y y (2.44)
22sinh.
sinh1
xy x x (2.45)
Chapter 2 Second-order BVPs
29
Table 2.5 contains the results for fourth-order method. We have compared our results
with fourth-order method based on (1,3) padeapproximant (Tirmizi [142]) . It is seen
that our method performs better than pade approximant. Fig. 2.3 shows the exact and
approximate solution for 1
64h .
2.4.2 Nonlinear problem
Problem 2.4 Consider the following nonlinear problem:
311 , 0 1
2y x y x , 0 1 0y y
(2.46)
for which the exact solution is
21.
2y x x
x
(2.47)
For the sake of comparison with fourth-order method, we have applied the quartic
splines based scheme given in (Al-Said [10]) to the nonlinear problem and the results
are summarized in Table 2.6. It is seen from the table that method 2.2.2 produces better
results. We have also compared the sixth-order method with higher order finite-
difference method (Tirmizi and Twizell [145]) and the results are given in Table 2.7. It
is observed that four iterations were enough for present methods to converge. The plot
of exact and approximate solution is shown in Figs. 2.4-2.5 for 1 1
,32 64
h .
2.4.3 System of differential equations
Problem2.5 Consider the system of differential equations (2.3) when 0f and
1r .
30, for 0 and ,
4 43
1, for ,4 4
x xu
u x
(2.48)
with the boundary conditions 0 0,u u and the condition of continuity of
3 and at and .
4 4u u x
The analytical solution for this problem is given by
Chapter 2 Second-order BVPs
30
1
2
1
1 2
4, 0
4
4 31 cosh ,
2 4 4
4 3, ,
4
where = +4coth and = sinh +4cosh , 4 4 4
x x
x xu x
x x
(2.49)
We have compared our method with different methods like cubic spline (Al-Said [9]),
quartic spline (Al-Said [10]), quintic spline (Aziz et al. [16]), non-polynomial quadratic
spline (Islam et al. [55]), non-polynomial cubic spline (Islam and Tirmizi [56]),
collocation based on cubic spline (Noor and Khalifa [95]) and finite difference method
(Noor and Tirmizi [96]) . The results are displayed in Table 2.8. It is clear from the
table that method 2.2.3 gives better results than the existing spline and finite difference
methods except in the case of non-polynomial quadratic spline (Islam et al. [55]).
Graph of exact and approximate solution is provided in Fig. 2.6 for 80
h
.
2.4.4 Singular perturbation problems
Problem 2.6 Consider the following problem with constant coefficients:
2 2'' cos 2 cos 2 , 0 1, 0 1 0.y y x x x y y (2.50)
The theoretical solution for this problem is
2exp (1 ) / exp /
cos .1 exp 1/
x xy x x
(2.51)
The results of maximum absolute errors for this problem are tabulated in Table 2.9 for
different values of the parameters and n . We have compared our method with
various methods from literature including cubic, quintic, sextic splines, splines in
tension, family of exponential spline difference scheme (Aziz and Khan [14], Aziz and
Khan [15], Khan et al. [77], Rashidinia et al. [108], Surla et al. [138], Surla and
Stojanovic [139], Surla and Vukoslavcevic [140]) and the results are summarized in
Tables 2.10-2.16. The Tables confirm better performance against the existing methods.
Figs. 2.7-2.26 show the exact and approximate solutions for various values of and n .
Chapter 2 Second-order BVPs
31
Problem 2.7 Consider the following equation with variable coefficients:
2
2
'' 1 1 1 1 2 1 exp 1 /
2 1 exp / ,
y x x y x x x x x
x x x
(2.52)
subject to the boundary conditions:
0 1 0.y y
The exact solution is given by
1 1 exp / exp (1 ) / .y x x x x x (2.53)
The above problem contains variable coefficients. We have solved it by fourth- and
sixth-order methods. The results of maximum absolute errors are tabulated in Table
2.17 and 2.18 for different values of the parameters and n .
Chapter 2 Second-order BVPs
32
Table 2.1
Observed maximum absolute errors for linear problem 2.1
h Second-order method
1/8 3.8203E-3
1/16 1.0646E-3
1/32 2.8473E-4
1/64 7.3859E-5
1/128 1.8824E-5
1/256 4.7523E-6
1/512 1.1940E-6
1/1024 2.9924E-7
Table 2.2
Observed maximum absolute errors for linear problem 2.1
h Method 2.2.2 (Al-Said [10]) (Usmani and Warsi [154])
1/16 8.74E-10 9.56E-9 2.92E-9
1/32 9.45E-11 5.99E-10 2.16E-10
1/64 6.81E-12 3.75E-11 1.40E-11
1/128 4.40E-13 2.34E-12 8.91E-12
Table 2.3
Observed maximum absolute errors for linear problem 2.1
h Method 2.2.3 (Usmani [151])
1/8 1.1779E-9 4.97E-8
1/16 7.6235E-12 1.03E-9
1/32 3.7773E-14 1.86 E-11
1/64 4.3021E-16 3.13E-13
Chapter 2 Second-order BVPs
33
Table 2.4
Observed maximum absolute errors for the linear problem 2.2
h Method 2.2.2 (Usmani and Warsi [154])
1/8 1.36E-6 1.10E-6
1/16 3.81E-8 1.09E-7
1/32 3.49E-9 7.51E-9
1/64 2.36E-10 4.81E-10
1/128 1.51E-11 3.03E-11
1/256 1.01E-12 1.85E-12
1/512 6.00E-13 6.48E-13
1/1024 2.72E-13 7.64E-13
Table 2.5
Observed maximum absolute errors for the linear problem 2.3
h Method 2.2.2 (1,3) pade approximant (Tirmizi [142])
1/8 2.71E-8 6.25E-8
1/16 6.98E-10 3.92E-9
1/32 6.31E-11 2.46E-10
1/64 4.29E-12 1.53E-11
1/128 2.69E-13 7.75E-12
Table 2.6
Observed maximum absolute error for the nonlinear problem 2.4
h Method 2.2.2 (Al-Said [10])
1/8 8.69E-6 1.43E-5
1/16 2.20E-7 9.2E-7
1/32 7.08E-9 5.77E-8
1/64 6.14E-10 3.62E-9
1/128 4.18E-11 2.26E-10
Chapter 2 Second-order BVPs
34
Table 2.7
Observed maximum absolute errors for the nonlinear problem 2.4
h Method 2.2.3 (Tirmizi and Twizell [145])
1/8 4.96E-7 6.32E-7
1/16 4.94E-9 6.33E-9
1/32 3.20E-11 1.57E-10
1/64 1.58E-13 2.72E-12
Table 2.8
Observed maximum absolute error for problem 2.5
Method / 20h / 40h / 80h
Method 2.2.3 3.35E-4 1.01E-4 2.14E-5
(Al-Said [9]) 1.26E-3 3.29E-4 8.43E-5
(Al-Said [10]) 8.74E-4 1.19E-4 2.98E-5
(Aziz et al. [16]) 2.57E-3 7.31E-4 1.94E-4
(Islam et al. [55]) 2.39E-4 6.23E-5 1.62E-5
(Islam and Tirmizi [56]) 6.43E-4 1.83E-4 4.87E-5
(Noor and Khalifa [95]) 1.42E-2 7.71E-3 4.04E-3
(Noor and Tirmizi [96]) 2.50E-2 1.29E-3 6.58E-3
Table 2.9
Maximum absolute errors for problem 2.6 method 2.2.3
n=16 n=32 n=64 n=128 n=256
1/16 1.19E-07 6.74E-10 4.26E-12 6.18E-14 2.09E-14
1/32 4.08E-08 2.19E-10 3.00E-12 4.61E-14 9.60E-15
1/64 5.84E-07 4.92E-09 2.81E-11 1.27E-13 1.62E-14
1/128 6.48E-06 6.4E-08 4.24E-10 2.1E-12 2.84E-14
Chapter 2 Second-order BVPs
35
Table 2.10
Maximum absolute errors for problem 2.6 reported in (Aziz and Khan [14])
n=16 n=32 n=64 n=128 n=256
1/16 1.57E-05 8.79E-07 5.32E-08 3.30E-09 2.05E-10
1/32 8.27E-06 4.41E-07 2.62E-08 1.62E-09 1.00E-10
1/64 1.84E-05 8.67E-07 6.65E-08 4.39E-09 2.78E-10
1/128 1.03E-04 2.61E-06 2.23E-07 1.54E-08 9.44E-10
Table 2.11
Maximum absolute errors for problem 2.6 reported in (Aziz and Khan [15])
n=16 n=32 n=64 n=128 n=256
1/16 4.074E-05 2.532E-06 1.581E-07 9.878E-09 6.172E-10
1/32 2.005E-05 1.242E-06 7.746E-08 4.839E-09 3.023E-10
1/64 5.456E-05 3.429E-06 2.146E-07 1.343E-09 8.397E-10
1/128 1.834E-04 1.226E-05 7.689E-07 4.814E-08 3.011E-09
Table 2.12
Maximum absolute errors for problem 2.6 reported in (Khan et al. [77])
n=16 n=32 n=64 n=128 n=256
1/16 1.224E-06 6.458E-09 3.407E-11 1.037E-12 8.881E-15
1/32 2.004E-06 1.683E-08 1.367E-10 1.093E-12 1.901E-14
1/64 8.898E-06 1.168E-07 1.203E-09 1.083E-11 9.076E-14
1/128 5.746E-05 9.984E-07 1.182E-08 1.141E-10 9.919E-13
Table 2.13
Maximum absolute errors for problem 2.6 reported in (Rashidinia et al. [108])
n=16 N=32 n=64 n=128 n=256
1/16 4.07E-05 2.53E-06 1.58E-07 9.87E-09 6.17E-10
1/32 2.00E-5 1.24E-06 7.74E-08 4.83E-09 3.02E-10
1/64 5.45E-05 3.42E-06 2.14E-07 1.34E-08 8.39E-10
1/128 1.83E-04 1.22E-05 7.68E-07 4.81E-08 3.01E-9
Chapter 2 Second-order BVPs
36
Table 2.14
Maximum absolute errors for problem 2.6 reported in (Surla et al. [138])
n=16 n=32 n=64 n=128 n=256
1/16 4.14E-03 1.02E-03 2.54E-04 6.35E-05 1.58E-05
1/32 3.68E-03 9.03E-04 5.61E-05 1.40E-05 3.50E-06
1/64 3.45E-03 8.40E-04 2.08E-04 5.20E-05 1.30E-05
1/128 3.43E-03 8.21E-04 2.03E-04 5.06E-05 1.26E-05
Table 2.15
Maximum absolute errors for problem 2.6 reported in (Surla and Stojanovic [139])
n=16 n=32 n=64 n=128 n=256
1/16 8.06E-03 2.02E-03 5.08E-04 1.27E-04 3.17E-05
1/32 7.11E-03 1.79E-03 4.48E-04 1.12E-04 2.80E-05
1/64 6.58E-03 1.66E-03 4.15E-04 1.04E-04 2.60E-05
1/128 6.36E-03 1.61E-03 4.03E-04 1.01E-04 2.52E-05
Table 2.16
Maximum absolute errors for problem 2.6 reported in (Surla and Vukoslavcevic [140])
n=16 n=32 n=64 n=128 n=256
1/16 1.20E-04 7.47E-06 4.67E-07 2.90E-08 4.39E-09
1/32 1.28E-04 8.00E-06 5.00E-07 3.14E-08 1.99E-09
1/64 1.60E-04 1.00E-05 6.26E-07 3.92E-08 2.31E-09
1/128 2.34E-04 1.47E-05 9.23E-07 5.77E-08 3.72E-09
Table 2.17
Maximum absolute errors for problem 2.7, 1 13
, , 518 9
n=16 n=32 n=64 n=128 n=256
1/16 1.64E-06 6.81E-08 4.84E-09 3.19E-10 2.02E-11
1/32 8.96E-06 1.79E-07 1.24E-08 8.73E-10 5.63E-11
1/64 4.41E-05 1.62E-06 3.46E-08 2.75E-09 1.84E-10
1/128 1.92E-04 6.78E-06 1.34E-07 9.05E-09 6.42E-10
Chapter 2 Second-order BVPs
37
Table 2.18
Maximum absolute errors for problem 2.7, 1 22 149
, ,20 15 30
n=16 n=32 n=64 n=128 n=256
1/16 1.63E-08 9.22E-10 4.14E-13 3.00E-15 1.08E-14
1/32 1.47E-07 1.00E-09 5.04E-12 2.38E-14 1.99E-14
1/64 1.25E-06 1.06E-08 6.07E-11 2.75E-13 1.22E-14
1/128 1.03E-05 1.05E-07 7.14E-10 3.58E-12 1.87E-14
Chapter 2 Second-order BVPs
38
0 10 20 30 40 50 60 700
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
0.05
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.1: Plot of solution for problem 2.1 , 1
64h
0 20 40 60 80 100 120 1400
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.2: Plot of solution for problem 2.2 , 1
128h
Chapter 2 Second-order BVPs
39
0 10 20 30 40 50 60 700
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.3: Plot of solution for problem 2.3 , 1
64h
0 5 10 15 20 25 30 35-0.18
-0.16
-0.14
-0.12
-0.1
-0.08
-0.06
-0.04
-0.02
0
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.4: Plot of solution for problem 2.4 , 1
32h
Chapter 2 Second-order BVPs
40
0 10 20 30 40 50 60 70-0.18
-0.16
-0.14
-0.12
-0.1
-0.08
-0.06
-0.04
-0.02
0
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.5: Plot of solution for problem 2.4 , 1
64h
0 10 20 30 40 50 60 70 800
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.6: Plot of solution for problem 2.5 , 80
h
Chapter 2 Second-order BVPs
41
0 2 4 6 8 10 12 14 16-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.7: Plot of solution for problem 2.6 , 1
, 1616
n
0 5 10 15 20 25 30 35-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.8: Plot of solution for problem 2.6, 1
, 3216
n
Chapter 2 Second-order BVPs
42
0 10 20 30 40 50 60 70-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.9: Plot of solution for problem 2.6 , 1
, 6416
n
0 20 40 60 80 100 120 140-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.10: Plot of solution for problem 2.6 , 1
, 12816
n
Chapter 2 Second-order BVPs
43
0 50 100 150 200 250 300-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.11: Plot of solution for problem 2.6 , 1
, 25616
n
0 2 4 6 8 10 12 14 16-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.12: Plot of solution for problem 2.6 , 1
, 1632
n
Chapter 2 Second-order BVPs
44
0 5 10 15 20 25 30 35-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.13: Plot of solution for problem 2.6 , 1
, 3232
n
0 10 20 30 40 50 60 70-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.14: Plot of solution for problem 2.6 , 1
, 6432
n
Chapter 2 Second-order BVPs
45
0 20 40 60 80 100 120 140-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.15: Plot of solution for problem 2.6 , 1
, 12832
n
0 50 100 150 200 250 300-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.16: Plot of solution for problem 2.6 , 1
, 25632
n
Chapter 2 Second-order BVPs
46
0 2 4 6 8 10 12 14 16-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.17: Plot of solution for problem 2.6 , 1
, 1664
n
0 5 10 15 20 25 30 35-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.18: Plot of solution for problem 2.6 , 1
, 3264
n
Chapter 2 Second-order BVPs
47
0 10 20 30 40 50 60 70-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.19: Plot of solution for problem 2.6 , 1
, 6464
n
0 20 40 60 80 100 120 140-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.20: Plot of solution for problem 2.6 , 1
, 12864
n
Chapter 2 Second-order BVPs
48
0 50 100 150 200 250 300-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.21: Plot of solution for problem 2.6 , 1
, 25664
n
0 2 4 6 8 10 12 14 16-0.7
-0.6
-0.5
-0.4
-0.3
-0.2
-0.1
0
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.22: Plot of solution for problem 2.6 , 1
, 16128
n
Chapter 2 Second-order BVPs
49
0 5 10 15 20 25 30 35-0.7
-0.6
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.23: Plot of solution for problem 2.6 , 1
, 32128
n
0 10 20 30 40 50 60 70-0.7
-0.6
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.24: Plot of solution for problem 2.6 , 1
, 64128
n
Chapter 2 Second-order BVPs
50
0 20 40 60 80 100 120 140-0.7
-0.6
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.25: Plot of solution for problem 2.6 , 1
, 128128
n
0 50 100 150 200 250 300-0.7
-0.6
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 2.26: Plot of solution for problem 2.6 , 1
, 256128
n
Chapter 2 Second-order BVPs
51
2.5 Conclusion
Quartic non-polynomial spline functions are used to develop a class of numerical
methods for solving linear, nonlinear, system of two point BVPs and SPPs. Second-,
fourth- and sixth-order accurate methods have been obtained. Numerical results
confirm the orders of the methods except for the system of second-order BVPs due to
extra continuity conditions associated with the system. Numerical results show better
accuracy of the new methods over some existing methods of the same order reported in
the literature. The methods are computationally efficient and the algorithm can easily
be implemented on a computer.
***************************************************
Chapter 3
Non-Polynomial Splines Approach to the
Solution of Sixth-Order
Boundary-Value Problems
3.1 Introduction
In this chapter, non-polynomial spline functions are applied to develop numerical
methods for obtaining smooth approximations for the following BVP:
6 , , , / ,D y x f x y a x b D d dx (3.1)
subject to the boundary conditions:
2 40 2 4
2 40 2 4
, , ,
, , .
y a A D y a A D y a A
y b B D y b B D y b B
(3.2)
where y x and ( , )f x y are continuous functions defined in the interval ,x a b . It
is assumed that 6, [ , ]f x y C a b is real and that iA , , 0, 2, 4,iB i are finite real
numbers. The literature on the numerical solution of sixth-order BVPs is sparse. Such
problems are known to arise in astrophysics; the narrow convicting layers bounded by
stable layers, which are believed to surround A-type stars, may be modeled by sixth-
order BVPs (Toomre et al. [146]). Also in (Glatzmaier [44]) it is given that dynamo
action in some stars may be calculated by such equations. (Chandrasekhar [27])
determined that when an infinite horizontal layer of fluid is heated from below and is
under the action of rotation, instability sets in. When this instability is an ordinary
convection, the ordinary differential equation is sixth order.
Theorems, which list the conditions for the existence and uniqueness of solutions of
sixth-order BVPs, are thoroughly discussed in the book by Agarwal (Agarwal [4]).
Non-numerical techniques for solving such problems are contained in papers (Baldwin
[18], Baldwin [19]). Numerical methods of solution are contained implicitly in
Chapter 3 Sixth-order BVPs
53
(Chawala and Katti [29]), although those authors concentrated on numerical methods
for fourth-order problems. E. H. Twizell (Twizell [148]) developed a second-order
method for solving special and general sixth-order problems and in his later work
(Twizell and Boutayeb [149]) developed finite-difference methods of order two, four,
six and eight for solving such problems. The authors in (Siddiqi and Twizell [132])
used sixth-degree splines, where spline values at the mid knots of the interpolation
interval and the corresponding values of the even order derivatives are related through
consistency relations. Sinc-Galerkin method for the solutions of sixth order BVPs was
used in (Gamel et al. [41]) whereas decomposition and modified domain
decomposition methods were used in (Wazwaz [157]) to investigate solution of the
sixth-order BVPs.
The spline function proposed in this chapter has the form:
2 3 4 57 =Span 1, , , , , , cos , sinT x x x x x kx kx
where k is the frequency of the trigonometric part of the splines function. Thus in each
subinterval 1i ix x x , we have:
2 3 4 5
2 3 4 5
2 3 4 5 6 7
Span 1, , , , , , cos , sin ,
Span 1, , , , , , cosh , sinh or
Span 1, , , , , , , , when 0 .
x x x x x kx kx
x x x x x kx kx
x x x x x x x k
The above correlation can be better explained by the following equation:
2 3 4 5
7
3 5
2 3 4 5
7
2 4
6
1, , , , , , sin( ), cos( )
7!1, , , , , - sin ( )
6 120Span
6!, 1 cos( )
2 24
Span
,
T x x x x x kx kx
kx kxx x x x x kx kx
k
kx kxkx
k
(3.3)
In the next section the special sixth-order, BVP to be solved is transformed using
non-polynomial spline techniques into linear or non-linear algebraic system. In section
3.3 methods of different orders are categorized. In section 3.4, properties of the inverse
of coefficient matrix are discussed. In section 3.5, convergence of the method is
established and numerical results are reported in section 3.6. In the last section some
conclusions are drawn.
Chapter 3 Sixth-order BVPs
54
3.2 Numerical Methods
To develop spline approximation to the problem (3.1)-(3.2), the interval ,a b is divided
into n equal subintervals, using the grid points , 0,1,..., ,ix a ih i n where b a
hn
. For
each segment 1,i ix x , the polynomial 1/ 2 ( )iP x has the form:
51/ 2 1/ 2 1/ 2 1/ 2 1/ 2 1/ 2 1/ 2
4 3 21/ 2 1/ 2 1/ 2 1/ 2 1/ 2 1/ 2 1/ 2 1/ 2 1/ 2
( ) sin ( ) cos ( ) ( )
( ) ( ) ( ) ( )
i i i i i i i
i i i i i i i i i
P x a k x x b k x x c x x
d x x e x x f x x g x x r
(3.4)
where 1/ 2 1/ 2 1/ 2 1/ 2 1/ 2 1/ 2 1/ 2 1/ 2, , , , , , andi i i i i i i ia b c d e f g r are constants and k is free
parameter. Let 1/ 2iy be an approximation to 1/ 2( )iy x , obtained by the segment
1/ 2 ( )iP x of the mixed splines function passing through the points
1/ 2 1/ 2 1/ 2 1/ 2( , ) and ( , ).i i i ix y x y To determine the coefficients in (3.4) at the common
nodes 1/ 2 1/ 2,i ix y , we first define:
( )1/ 2 1/ 2 1/2 1/ 2( ) , ( ) , ( ) , ( ) , 1,2,..., , 1/ 2.vi
i i j i j i i j i j i i j i j i i j i jP x y P x Z P x M P x S i n j
From algebraic manipulation we get the following expressions for coefficients:
5 6
1/ 2 1/ 2 1/ 26
1/ 21/ 2 6
2 2 2 2 2 6
1/ 2 1/ 2 1/ 2
2 6
1/ 2 1/ 2 1/ 2 1/ 2
1/ 2
1/ 2 1
csc( )
2
1( cos( ) )csc( ),
,
sin( ) sin( ) sin( )
sin( ) 6 6 cos( ) 6 cos( )
6 6
i i i
ii
i i i
i i i i
i
i ih k
a S Sk
Sb
k
h k S h S k h M k
h k M hS k hk S hkSc
hkS hZ
4 6
6 6
/ 2 1/ 2 1/ 2
6 6
1/ 2 1/ 2 1/ 2
2 2 2 2 2 6
1/ 2 1/ 2 1/ 2
2 6
1/ 2
1/ 2
csc( )
2
,sin( ) 6 sin( ) 12 sin( )
12 sin( ) 12 sin( ) 12 sin( )
2 sin( ) 3 sin( ) 3 sin( )
2 sin( ) 14
i i
i i i
i i i
i
ih k
k hk Z S
S y k y k
h k S h S k h M k
h k M hSd
1/ 2 1/ 2 1/ 2
6 6
1/ 2 1/ 2 1/ 2 1/ 2
6 6
1/ 2 1/ 2 1/ 2
6
1/ 2
1/ 2 3 6
14 cos( ) 16 cos( ),
16 16 sin( ) 14 sin( ) 30 sin( )
30 sin( ) 30 sin( ) 30 sin( )
20 sin(
csc( )
2
i i i
i i i i
i i i
i
i
k hk S hkS
hkS hZ k hk Z S
S y k y k
y k
eh k
6 2 6
1/ 2 1/ 2
2 2 6
1/ 2 1/ 2 1/ 2 1/ 2
2 6 2 2
1/ 2 1/ 2 1/ 2 1/ 2
6
1/ 2 1/ 2 1/
) 20 sin( ) sin( )
sin( ) 12 8 cos( ) 8 sin( )
8 3 sin( ) 3 sin( ) 12 cos( )
12 sin( ) 20 sin( ) 20
i i
i i i i
i i i i
i i i
y k h k M
h k S hkS hk S hk Z
hS k h M k h S k hkS
hZ k S S
2
,
sin( )
(3.5)
Chapter 3 Sixth-order BVPs
55
4
5
41/ 2 -1/ 2 1/ 2
51/ 2 1/ 2 1/ 2 1/ 2
61/ 2 1/ 2
1/ 2 6
1
2csc( )
,
cos( ) sin( ) ,
whereby .,
i i i
i i i i
i ii
k
k
f S M k
g S S Z k
S y kr kh
k
Using the continuity condition of the third, fourth and fifth derivatives at 1/2 1/2( , )i ix y , i.e.,
( ) ( )1 1 2 1 2( ) ( )n n
i i i iP x P x where 3,4,5n , we obtain,
1/2 1/2 1/23 2 5 2 5
3/2 1/2 1/22 5 3 3
1/2 1/2 1/2 1/2 1/24 4 2 5
1/2 12 5 2
1 72 24sin( ) cot( ) cot( )
24 1 2cot( ) cos( )cot( ) cot( )
18 3 3 18 36csc( )
48 24csc( )
i i i
i i i
i i i i i
i i
S S Sk h k h k
S S Sh k k k
S S M M Shk hk h h h k
S Zh k h
/2 1/2 1/2 1/2 1/23 6 3 6 3 3
3/2 3/2 3/2 3/2 3/2 3/22 5 3 4 2 3 6
3/23
120 60 60 120
36 60 3 3 24 60csc( )
1csc( )
0,
i i i i
i i i i i
i
S S y yh k h k h h
S y M S Z Sh k h h hk h h k
Sk
(3.6)
3/ 2 1/ 2 3/ 2 3/ 23 5 3 5 2 4 2
3/ 2 3/ 2 3/ 2 3/ 2 1/ 2 1/ 23 5 3 4 6 4 2 4 2
1/ 2 1/ 2 1/ 23 5 3 3 4
168 168 24 24cot( ) 168cot( )
192 168 360 360 24 24csc( )
192 168 384 360csc( ))
i i i i
i i i i i i
i i i
S S S Mh k h k h k h
S Z S y S Mh k h h k h h k h
S Z Zh k h h h
1/ 2 1/ 26 4
360
0,
i iS yk h
(3.7)
1/2 1/2 1/2 1/23 4
1/2 -1/2 1/2 1/2 1/23 4 3 3 4 5 4 5
5 61/2 1/2 1/2 1/24 5 6 5
1 1 2 120csc( ) cos( )cot( ) cot( )
60 60 120 360 720csc( ) csc( )
360 720 720 1441440/( )
i i i i
i i i i i
i i i i
S S S Sk k k h k
S M M S Sh k h h h k h k
Z h k S S yh h k k
1/25
3/2 3/2 3/2 3/2 3/23 4 3 4 5 4 5 6
3/2 1/2 1/2 3/25 4 5 4 5 4 5
3/2
0
0.60 60 360 360 720
csc( )
720 720 360 360cot( ) cot( ) cot( )
1csc( )
i
i i i i i
i i i i
i
yh
S M S Z Sh k h h k h h k
y S S Sh h k h k h k
Sk
(3.8)
Replacing i by 1, 1, 2, 2i i i i , in Eqs. (3.6)-(3.8) we get the following equations:
Chapter 3 Sixth-order BVPs
56
1 / 2 3 / 2 1 / 23 2 5 2 5
5 / 2 1 / 2 3 / 22 5 3 3
3 / 2 1 / 2 1 / 2 3 / 2 1 / 24 4 2 5
3 / 2 12 5 2
1 72 24sin( ) cot( ) cot( )
24 1 2cot( ) cos( ) cot( ) cot( )
18 3 3 18 36csc( )
48 24csc( )
i i i
i i i
i i i i i
i i
S S Sk h k h k
S S Sh k k k
S S M M Shk hk h h h k
S Zh k h
/ 2 3 / 2 1 / 2 1 / 2 3 / 23 6 3 6 3 3
5 / 2 5 / 2 5 / 2 5 / 2 5 / 2 5 / 22 5 3 4 2 3 6
5 / 23
0,120 60 60 120
36 60 3 3 24 60csc( )
1csc( )
i i i i
i i i i i i
i
S S y yh k h k h h
S y M S Z Sh k h h hk h h k
Sk
(3.9)
3 / 2 1 / 2 3 / 23 2 5 2 5
1 / 2 3 / 2 1 / 22 5 3 3
1 / 2 3 / 2 3 / 2 1 / 2 3 / 2 1 / 24 4 2 5 2 5
32
1 72 24sin( ) cot( ) cot( )
24 1 2cot( ) cos( ) cot( ) cot( )
18 3 3 18 36 48csc( ) csc( )
24
i i i
i i i
i i i i i i
i
S S Sk h k h k
S S Sh k k k
S S M M S Shk hk h h h k h k
Zh
/ 2 1 / 2 3 / 2 3 / 2 1 / 2 1 / 23 6 3 6 3 3 2 5
1 / 2 1 / 2 1 / 2 1 / 2 1 / 2 1 / 23 4 2 3 6 3
0,
120 60 60 120 36csc( )
60 3 3 24 60 1csc( )
i i i i i
i i i i i i
S S y y Sh k h k h h h k
y M S Z S Sh h hk h h k k
(3.10)
3 / 2 5 / 2 3 / 23 2 5 2 5
7 / 2 3 / 2 5 / 22 5 3 3
5 / 2 3 / 2 3 / 2 5 / 2 3 / 2 5 / 24 4 2 5 2 5
32
1 72 24sin( ) cot( ) cot( )
24 1 2cot( ) cos( ) cot( ) cot( )
18 3 3 18 36 48csc( ) csc( )
24
i i i
i i i
i i i i i i
i
S S Sk h k h k
S S Sh k k k
S S M M S Shk hk h h h k h k
Zh
/ 2 5 / 2 3 / 2 3 / 2 5 / 2 7 / 23 6 3 6 3 3 2 5
7 / 2 7 / 2 7 / 2 7 / 2 7 / 2 7 / 23 4 2 3 6 3
0,
120 60 60 120 36csc( )
60 3 3 24 60 1csc( )
i i i i i
i i i i i i
S S y y Sh k h k h h h k
y M S Z S Sh h hk h h k k
(3.11)
5 / 2 3 / 2 5 / 23 2 5 2 5
1/ 2 5 / 2 3 / 22 5 3 3
3 / 2 -5 / 2 5 / 2 3 / 2 5 / 2 3 / 24 4 2 5 2 5
52
1 72 24sin( ) cot( ) cot( )
24 1 2cot( ) cos( ) cot( ) cot( )
18 3 3 18 36 48csc( ) csc( )
24
i i i
i i i
i i i i i i
i
S S Sk h k h k
S S Sh k k k
S S M M S Shk hk h h h k h k
Zh
/ 2 3 / 2 5 / 2 5 / 2 3 / 2 1/ 23 6 3 6 3 3 2 5
1/ 2 1/ 2 1/ 2 1/ 2 1/ 2 1/ 23 4 2 3 6 3
0,
120 60 60 120 36csc( )
60 3 3 24 60 1( )csc
i i i i i
i i i i i i
S S y y Sh k h k h h h k
y M S Z S Sh h hk h h k k
(3.12)
Chapter 3 Sixth-order BVPs
57
5/ 2 1/ 2 5/ 2 5/ 23 5 3 5 2 4 2
5/ 2 5/ 2 5/ 2 5/ 2 1/ 2 1/ 23 5 3 4 6 4 2 4 2
1/ 2 1/ 2 3/ 23 5 3 3 4 6
168 168 24 24cot( ) cot( )
192 168 360 360 24 24csc( )
192 168 384 360csc( )
i i i i
i i i i i i
i i i i
S S S Mh k h k h k h
S Z S y S Mh k h h k h h k h
S Z Z Sh k h h h k
1/ 2 1/ 24
0,
360iy
h
(3.13)
1/ 2 3/ 2 1/ 2 1/ 23 5 3 5 2 4 2
1/ 2 1/ 2 1/ 2 1/ 2 3/ 23 5 3 4 6 4 2 4
3/ 2 3/ 2 3/ 2 1/ 22 3 5 3 3 4 6
168 168 24 24cot( ) cot( )
192 168 360 360 24csc( )
24 192 168 384 360sin( )
i i i i
i i i i i
i i i i i
S S S Mh k h k h k h
S Z S y Sh k h h k h h k
M S Z Z Sh h k h h h k
-3/ 2 3/ 24
0,
360iy
h
(3.14)
7 / 2 3/ 2 7 / 2 7 / 23 5 3 5 2 4 2
7 / 2 7 / 2 7 / 2 7 / 2 3/ 23 5 3 4 6 4 2 4
3/ 2 3/ 2 3/ 2 5/ 22 3 5 3 3 4 6
168 168 24 24cot( ) cot( )
192 168 360 360 24csc( )
24 192 168 384 360csc( )
i i i i
i i i i i
i i i i i
S S S Mh k h k h k h
S Z S y Sh k h h k h h k
M S Z Z Sh h k h h h k
3/ 2 3/ 24
0,
360iy
h
(3.15)
-1/ 2 5/ 2 1/ 2 1/ 23 5 3 5 2 4 2
1/ 2 1/ 2 1/ 2 1/ 2 5/ 23 5 3 4 6 4 2 4
5/ 2 5/ 2 5/ 2 3/ 22 3 5 3 3 4 6
168 168 24 24cot( ) cot( )
192 168 360 360 24csc( )
24 192 168 384 360csc( )
i i i i
i i i i i
i i i i i
S S S Mh k h k h k h
S Z S y Sh k h h k h h k
M S Z Z Sh h k h h h k
-5 / 2 5/ 24
0,
360iy
h
(3.16)
1/ 2 1/ 2 3/ 2 3/ 23 4
1/ 2 1/ 2 3/ 2 1/ 2 3/ 23 4 3 3 4 5 4 5
1/ 2 3/ 2 1/ 2 1/ 24 5 6 5 6 5 5
1 1 2 120sin( ) cos( )cot( ) cot( )
60 60 120 360 720sin( ) csc( )
360 1440 720 720 1440
i i i i
i i i i i
i i i i
S S S Sk k k h k
S M M S Sh k h h h k h k
Z S S yh h k h k h h
3/ 2 5/ 23 4
5/ 2 5/ 2 5/ 2 5/ 2 5/ 23 4 5 4 5 6 5
3/ 2 1/ 2 5/ 2 5/ 24 5 4 5 4 5
600,
60 360 360 720 720sin( ))
720 360 360 1cot( ) cot( ) cot( ) csc( )
i i
i i i i i
i i i i
y Sh k
M S Z S yh h k h h k h
S S S Sh k h k h k k
(3.17)
Chapter 3 Sixth-order BVPs
58
3/ 2 3 / 2 1/ 2 1/ 23 4
3/ 2 3 / 2 1/ 2 3/ 2 1/ 23 4 3 3 4 5 4 5
3/ 2 1/ 2 3 / 2 3/ 24 5 6 5 6 5 5
1 1 2 120sin( ) cos( ) cot( ) cot( )
60 60 120 360 720csc( ) csc( )
360 1440 720 720 1440
i i i i
i i i i i
i i i i
S S S Sk k k h k
S M M S Sh k h h h k h k
Z S S yh h k h k h h
1/ 2 1/ 23 4
1/ 2 1/ 2 1/ 2 1/ 2 1/ 23 4 5 4 5 6 5
1/ 2 3/ 2 1/ 2 1/ 24 5 4 5 4 5
600,
60 360 360 720 720csc( )
720 360 360 1cot( ) cot( ) cot( ) csc( )
i i
i i i i i
i i i i
y Sh k
M S Z S yh h k h h k h
S S S Sh k h k h k k
(3.18)
3/ 2 3/ 2 5 / 2 5 / 2 3/ 23 4 3 4
3/ 2 5 / 2 3/ 2 5 / 2 3 / 23 3 4 5 4 5 4
5 6
5 / 2 3/ 25 6 5 6 5
1 1 2 120 60sin( ) cos( ) cot( ) cot( )
60 120 360 720 360csc( ) csc( )
1440 720 7201440 /( )
i i i i i
i i i i i
i i i
S S S S Sk k k h k h k
M M S S Zh h h k h k h
h k S S yh k h k h
3/ 2 5 / 2 7 / 25 3 4
7 / 2 7 / 2 7 / 2 7 / 2 7 / 23 4 5 4 5 6 5
5 / 2 3 / 2 7 / 2 7 / 24 5 4 5 4 5
1440 600,
60 360 360 720 720csc( )
720 360 360 1cot( ) cot( ) cot( ) csc( )
i i
i i i i i
i i i i
y Sh h k
M S Z S yh h k h h k h
S S S Sh k h k h k k
(3.19)
5 / 2 5 / 2 3 / 2 5 / 23 4 3 4
5 / 2 3 / 2 5 / 2 3 / 2 5 / 23 3 4 5 4 5 4
3/ 2 5 / 2 5 / 2 3 / 25 6 5 6 5 5
1 1 2 120 60sin( ) cos( ) cot( ) cot( )
60 120 360 720 360sin( ) csc( )
1440 720 720 1440
i i i i
i i i i i
i i i i
S S S Sk k k h k h k
M M S S Zh h h k h k h
S S y yh k h k h h
1/ 2 1/ 23 4 3
1/ 2 1/ 2 1/ 2 1/ 2 3/ 24 5 4 5 6 5 5 6
5 / 2 1/ 2 1/ 24 5 4 5
60 600.
360 360 720 720 720csc( ) cot( )
360 360 1cot( ) cot( ) csc( )
i i
i i i i i
i i i
S Mh k h
S Z S y Sh k h h k h h k
S S Sh k h k k
(3.20)
We eliminate Z’s and M’s by solving Eqs. (3.6)-(3.19) simultaneously. The values of
' and 'Z s M s obtained from these equations are used in Eq. (3.20), which results, after
lengthy calculations the following recurrence relation,
7 / 2 5/ 2 3/ 2 1/ 2 1/ 2 3/ 2 5/ 2
67 / 2 5/ 2 5/ 2 3/ 2 3/ 2 1/ 2 1/ 2
6 15 20 15 6
,
i i i i i i i
i i i i i i i
y y y y y y y
h S S S S S S S
(3.21)
where
6 3 5
1 120 20 1 120,
120 sin sin sin
(3.22)
Chapter 3 Sixth-order BVPs
59
3 6 5 5 3
3 5 5 6 3
6 5 3
1 40 2cos 720 480 240 cos 26 40 cos,
120 sin sin sin sin sin sin
1 100 67 960cos 840 1800 52cos 80cos,
120 sin sin sin sin sin sin
1 2400 960 80
120 sin sin
3 5
132cos 240 cos 52 1440 cos,
sin sin sin sin
for 4,5, 6,..., 3.i n
Here 0
1lim , , , 1, 120, 1191, 2416
5040
, which is the polynomial case as mentioned
in Eq. (3.3).
The local truncation error ,it associated with the scheme developed in (3.21) is,
6 (6) 7 (7) 8 (8) 9 (9) 10 (10) 11 (11)
6 7 8 9 10 11
12 (12) 13 (13) 14 (14) 1512 13 14 ,
i i i i i i i
i i i
t C h y C h y C h y C h y C h y C h y
C h y C h y C h y O h
(3.23)
where
6
7
8
9
10
11
12
13
1 2 2 2 ,
1 2 2 2,
23 74 34 10
,8
7 218 98 26,
48121 5(3026 706 82
,1920
77 13682 2882 242,
38406227 21(133274 16354 730
,967680
3353 3(7454
C
C
C
C
C
C
C
C
14
18 75938 2186,
19353604681 6155426 397186 6562
.10321920
C
(3.24)
Thus for different choices of , , , in scheme (3.21), methods of different order
are obtained.
The relation (3.21) gives 6n algebraic equations in n unknowns 1/ 2my ,
0,1,..., 1.m n We require six more equations, three at each end of the range of
integration, for the direct computation of 1/ 2my . These equations are developed by
Taylor series and method of undetermined coefficients. General form of boundary
equations for the main scheme is as follows.
Chapter 3 Sixth-order BVPs
60
0 1/ 2 3/ 2 5 / 2 7 / 2 9 / 2
(2) (4) 6
1 0 1 0 1 1 1 1 1 1
2 4 (6) (6) (6) (6) (6) (6)1/ 2 3/ 2 5/ 2 7 / 2 9 / 2 11/ 2
20 35 21 7
,
y y y y y y
y y hh h a y b y c y d y e y f y
(3.25)
(2) (4) 6
2 0 2 0 2 2 2 2 2 2
0 1/ 2 3/ 2 5/ 2 7 / 2 9/ 2
2 4 (6) (6) (6) (6) (6) (6)1/ 2 3/ 2 5/ 2 7 / 2 9 / 2 11/ 2
10 21 21 15 6
,y y h
y y y y y y
h h a y b y c y d y e y f y
(3.26)
(2) (4) 6
3 0 3 0 3 3 3 3 3 3
0 1/ 2 3/ 2 5/ 2 7 / 2 9/ 2 11/ 2
2 4 (6) (6) (6) (6) (6) (6)1/ 2 3/ 2 5/ 2 7 / 2 9 / 2 11/ 2 .
2 7 15 20 15 6
y y h
y y y y y y y
h h a y b y c y d y e y f y
(3.27)
The remaining three equations at the other end can be obtained from (3.25)-(3.27) by
writing them in reverse order. The constants , , , , , , ,i i i i i i i ia b c d e f and are
parameters which must be chosen so that the local truncation errors of (3.25)-(3.27) are
identical with (3.21).
3.3 Numerical methods of different orders
3.3.1 Second-order convergence
1 722 10543 23548, , ,
46080 46080 46080 46080 in the main Eq. (3.21) gives
6 7 8
10, .
24 C C C
The values of the coefficients for these methods are:
1 1 1 1 1 1 1 1
2 2 2 2 2 2
2 2 3 3 3 3 3
3 3
7 77 13005 9821 721 1, , , , , , 0, 0,
2 963 25 9821 23547 10543 722
, , , , , ,4 64
1 1 1 721 10543 23548, 0, , , , , ,
4 19210543 722
, ,
a b c d e fcd cd cd cd
a b c dcd cd cd cd
e f a b ccd cd cd cd
d e fcd cd 3
1, 0, 46080. cd
cd
(3.28)
3.3.2 Fourth-order convergence
7 3 2 17 8 9,
20 5 5 20 5 10
in the main Eq. (3.21) gives
6 7 8 9 10
151 30, 2
120 2C C C C C
for arbitrary constants , .
Chapter 3 Sixth-order BVPs
61
The values of the coefficients of boundary equations are given by
1 1 1 1 1 1
1 1 2 2 2 2
2 2 2 2 3 3
3 3
7 77 2685830 2704842 166502 50582, , , , ,
2 96
3 25 12161085 353329650, 0, , , , ,
4 64
64014771 41166795 10321920 1 1, , , 0, , ,
4 192
51608615,
,
a b c dcd cd cd cd
e f a bcd cd
c d e fcd cd cd
a bcd
3 3
3 3
193535007 273530281 141926257, , ,
10321920 10321920, 10321920.0,,
c dcd cd cd
e f cdcd cd
(3.29)
3.3.3 Sixth-order convergence
11 47 3 151 3 , and
400 20 300 10 240 4
in the main Eq. (3.21) gives
6 7 8 9 10 11 12
41530,
24 151200
C C C C C C C for arbitrary constant .
The values of the coefficients for these methods are:
1 1 1 1 1
1 1 1 2 2
2 2 2 2
2 2
7 77 119271901 128562861 18967984, , , , ,
2 96
11537036 3523149 483175 3 25, , , , ,
4 64
80655078 277312338 77419113 19393863, , , ,
5790447 920775, ,
a b ccd cd cd
d e fcd cd cd
a b c dcd cd cd cd
e fcd cd
3 3 3
3 3 3 3 3
1 1 3815581, , ,
4 192
100731801 255355121 100719251 3839346 5140, , , ,
464486400.
,
1,
acd
b c d e fcd cd cd cd cd
cd
(3.30)
3.3.4 Eight-order convergence
1 41 2189 4153, , ,
30240 5040 10080 7560 in the main Eq. (3.21) gives
6 7 8 9 10 11 12 13 140, 0, 0, 0. C C C C C C C C C
Boundary equations for eight order convergence are not constructed; however these
equations can be obtained following the procedure adapted in the previous cases.
Chapter 3 Sixth-order BVPs
62
3.4 Properties of the Coefficient matrix A0
Consider the following seven-band matrix A0 of order n :
0
35 21 0 0
21 21 0 0
,
0 0 21 21
0 0 21 35
a
b
A a b M b a
b
a
T T TR TR
R
R
(3.31)
Where [ 7,1,0,...,0], [15, 6,1,0,...,0].a b Here T denotes the operation of
transposition and for a row vector 1 2 1 1, ,..., , , ,..., . v vRn n nv v v v v v Further, a, b
are 4n dimensional row vectors. The matrix ,M i jm is a seven-band matrix of
order 4n given by,
,
20, 1,2,..., 4,
15, 1,
6, 2,
1, 3,
0, 3.
i j
i j n
i j
m i j
i j
i j
(3.32)
In this section we find matrix 1A 0 and 1A
0 . The matrix 1A 0 is likewise of the
form,
10 ,
c
d
A c d N d c
d
c
T T TR TR
R
R
(3.33)
where ,i i= c dc d are 4n dimensional row vectors. T and R have the same
meaning as defined above and , , , , , , , are scalars. The matrix
,N i jn is of order 4n . Using 0 0A A I-1n , we get the following equations:
35 21 1
35 21 0
21 21 0
21 21 0
T
TR
T T TR T TR
T
TR
i
ii
iii
iv
v
0
ac
ac
M c a a b b
bc
bc
(3.34)
Chapter 3 Sixth-order BVPs
63
4
21 21 1
21 21 0
35 21 0
35 21 0
.
T
TR
T
TR
T T TR T TR
T TR R T TR Rn
vi
vii
viii
ix
x
xi
0
bd
bd
ad
ad
Md a a b b
MN a c a c b d b d I
Here 0 denotes a 4n dimensional null column vector.
In order to find the unknowns in above equations, we need to use 1, .M i jm
From (Hoskins and Ponzo [49]) it is given by,
,
3 2 1 1 2
240 1 1 2 3
2 2 1 3 3 1,
1 2 2 1 1 1 3 4 3 2
3 2 1 1 2
240 1 1 2 3
2 2 1 3 3 1
1 2 2 1 1 1 3 4 3 2
i j
n i n i n i j j j
n n n n n
i i j j n ni j
i i j j n n i i j j n nm
n j n j n j i i i
n n n n n
j j i i n n
j j i i n n j j i i n n
, .i j
(3.35)
Algebraic manipulation of Eqs. (3.34(i-v)) yields the following expressions. 4 2
4 2
4 2
4 2
16 40 11,
2880
16 40 9,
960
14 20 11,
2880
14 20 9,
960
n n
n
n n
n
n n
n
n n
n
(3.36)
and
3 2 2
3 2 4 3 2
16 24 16 48 362 2 3,
2880 24 108 112 6 6 36 56 6 9
1, 2, ..., 4 .
i
i n i i nn i
n i i i n i i i ic
i n
(3.37)
Eqs. (3.34(vi-x)) yield the following results:
4 2
4 2
16 40 9,
960
16 120 160 51,
320
n n
n
n n n
n
(3.38)
Chapter 3 Sixth-order BVPs
64
4 2
4 2
14 20 9,
960
14 60 51,
320
n n
n
n n
n
and
3 2 2
3 2 4 3 2
16 24 16 48 362 2 3,
960 24 108 192 126 6 36 96 126 51
1,2,..., 4.
i
i n i i nn id
n i i i n i i i i
i n
(3.39)
From Eq. (3.34(xi)), we have
= + ,-1N M U (3.40)
where
1 .U M a c + b d + b d a c T T TR R TR R (3.41)
Hence ,U i ju is given by:
2
5 41 2
4 3 23
2 34 5 6
,1
2880 3 2 1
12 30 3
3 5 5 5 6 ,
60 3 20 4
i jn n n n
i z i n z
n n n n z
i n z i z iz
u
(3.42)
where
5 4 3 2 2 21
4 3 2 4 3 2
5 4 3 22
2 2 4 3 2
4 3 2
5 4 23
12 30 3 20 8 12 30 2 9 9
2 5 15 75 60 3 5 5 5 6 ,
12 88 32 12 2 14 19
8 9 35 31 2 2 7 20 79 54
3 2 7 4 7 6 ,
153 12 90 60 480 360 48
z j j n j n n j n n
j n n n n n n n n
z j j n j n n
j n n j n n n n
n n n n
z j j n n n
4
2 2 3 2
4 2
5 4 24
2 3 2 3 3 2
2 4 3 2
5 4 3 2
180 2 5 3 20 4 18 15
8 4 75 135 60 ,
6 2 3 4 9 35 31
9 1 2 5 6 6 5 38 81 51
6 69 420 679 338
12 64 41 238 413 186 ,
n
j n n j n n
j n n n
z j n j n n
n n n n j n n n
j n n n n
j n n n n n
Chapter 3 Sixth-order BVPs
65
5 2 4 3 25
2 4 3 2
3 4 3 2
6 5 4 3 2
6 5 4 3 2
5 4 3 26
4 5
12 8 12 18 2 20 61 57
18 5 53 195 294 153
4 8 103 490 959 645
3 4 38 125 145 39 183 90
4 2 19 49 67 489 711 315 ,
6 5 15 75 60
15 2 13
z j n n j n n n
j n n n n
j n n n n
n n n n n n
j n n n n n n
z j n n n n
j n n
3 6 5 4 3 2
2 6 5 4 3 2
8 7 6 5 4 3
2
8 7 6 5 4 3
4 3 2
20 2 19 49 67 489 711 315
15 12 100 233 115 1127 1425 558
4 20 55 530 1134 6006
825 1110 360
8 40 200 1825 4230 16502
64
139 291 162
j n n n n n n
j n n n n n n
n n n n n n
n n
n n n n n nj
n n n
2.
80 8595 3150n n
We summarize the results of this section in the following lemma.
Lemma 3.1 The matrix A0 is nonsingular and if 1,A 0 i ja , then , 0i ja and ,i ja
are given by ((3.36)-(3.40)).
In order to find 1A
0 , we first find the sum of elements in the i-th row of 1A 0 .
The sum of elements in the i-th row of 1M is given by,
11 2 1 2 3 ,
720 i i i n i n i n i
(3.43)
and the sum of the elements of i-th row of U is:
4 3 3 2 2
4 3 2
4 3 2
4 6 16 24 8 34 52 291
4 4 12 104 192 116 .960
18 108 258 288 120
i n i n i i i n
n i i i i n
i i i i
(3.44)
Let iS designate the sum of the elements in the i-th row of 1A 0 in modulus, then
, ,1 1
n n
i i j i jj j
S a a
is given by
5 3 2 3
5 4 3 2
6 5 4 3 2
12 6 20 30 20 151
12 30 40 90 16 242880
4 12 20 60 16 48 .
.
i
i n i i i n
i i i i i n
i i i i i i
S
(3.45)
Chapter 3 Sixth-order BVPs
66
Consider iS as a function of the real variable i. It turns out that iS is maximum for
11
2i n . The infinity norm of 1A
0 must be restricted to an integral value of i and
therefore
1( 1) / 2maxA
0 i i nS S .
Here equality will hold only if n is odd. We substitute 11
2i n into (3.45) to get
( 1) / 2nS . Hence,
6 4 21
( 1) / 2
6 4 22 4 66
6
61 175 259 225
46080
61 175 259 225.
46080
A
0 n
n n nS
b a h b a h b a hh
h
(3.46)
3.5 Convergence
Clearly, the family of numerical methods is described by the Eqs. (3.21), boundary
equations and the solution vector 1/ 2 3/ 2 1/ 2, ,...,Y T
ny y y . T denoting transpose, is
obtained by solving a nonlinear algebraic system of order n which has the form,
6 , .h f x 00A Y B Y C = (3.47)
Now we investigate the error analysis of the seven-degree non-polynomial spline
method described in section 3.2. To do so we let,
1/ 2 1/ 2( ) , , ,y Y C T i i i iy x y c t and 1/ 2E ie ,be n-dimensional column
vectors. Here 1/ 21/ 2 1/ 2ii ie y x y is the discretization error for 4,..., 4i n . Thus,
we can write our method in the matrix form:
The matrix B is given by
Chapter 3 Sixth-order BVPs
67
1 1 1 1 1 1
2 2 2 2 2 2
3 3 3 3 3 3
3 3 3 3 3 3
2 2 2 2 2 2
1 1 1 1 1 1
.B
a b c d e f
a b c d e f
a b c d e f
f e d c b a
f e d c b a
f e d c b a
(3.48)
The column vector C is given by
0 1 1
0 2 2
0 3 3
0 3 3
0 2 2
0 1 1
2 42 4
2 42 4
2 42 4
2 42 4
2 42 4
2 42 4
20
10
2
0
.
0
2
10
20
C
=
A A A
A A A
A A A
B B B
B B B
B B B
h h
h hh h
h h
h hh h
(3.49)
The vector 1/ 2 3/ 2 1/ 2, ,...,y T
ny x y x y x satisfies
6 , ,h f x 0 - 0A y B y C T =
where 1 2, ,...,T T
nt t t is the vector of local truncation errors, and a conventional
convergence analysis shows that the norm of the vector
E y Y
satisfies:
6
6 4 22 4 6
6 7 8 9 10 11 12
6 6 7 7 8 8 9 9 10 10 11 11 12 12
46080 61 175 259 225
... ,
Eb a
b a h b a h b a h B F
C h V C h V C h V C h V C h V C h V C h V
(3.50)
where max for 1, 2,..., , and max .i
i ia x b a x b
d y x fV i B F
dx y x
B
Chapter 3 Sixth-order BVPs
68
The order of convergence of the numerical method is, thus p, if 6pC is the first non-
vanishing constant on the right hand side of (3.23) provided
*
6 4 22 4 6
46080
61 175 259 225F
b a h b a h b a h
(3.51)
3.6 Numerical Results and Discussion
The numerical methods outlined in the previous sections were tested on the following
nonlinear and linear problems.
3.6.1 Non-linear problems Problem 3.1 Consider the nonlinear BVP
6 2 , 0 1 xD y x e y x x (3.52)
subject to the boundary conditions:
2 4 2 40 0 0 1, 1 1 1 .y D y D y y D y D y e (3.53)
The theoretical solution for this problem is
.xy x e (3.54)
The results of maximum absolute errors for this problem are listed in Table 3.1.
Problem 3.2 We consider the nonlinear BVP
66 20exp 36 40 1 , 0 1 D y x y x x x (3.55)
with boundary conditions:
2 4 2 41 1 1 10 0, 0 , 0 1, 1 ln 2, 1 , 1 .
6 6 24 16 y D y D y y D y D y
(3.56)
for which the theoretical solution is
1ln 1 .
6 y x x
(3.57)
The results of maximum absolute errors for this problem are listed in Tables 3.2-3.3.
3.6.2 Linear problems
Problem 3.3 Consider the linear BVP
Chapter 3 Sixth-order BVPs
69
6 324 11 exp , 0 1 D y x xy x x x x (3.58)
with boundary conditions:
2 2 4 40 0 1 , 0 0, 1 4 , 0 8, 1 16 . y y D y D y e D y D y e (3.59)
The analytical solution of the above differential equation is
1 exp . y x x x x (3.60)
The maximum errors (in absolute value) for different orders are shown in Table 3.4.
Problem 3.4 Finally we consider the linear BVP
6 6 2 cos 5sin , 1 1 D y x y x x x x x (3.61)
with boundary conditions,
2 2
4 4
1 0 1 , 1 4cos 1 2sin 1 , 1 4cos 1 2sin 1 ,
1 8cos 1 12sin 1 , 1 8cos 1 12sin 1 .
y y D y D y
D y D y
(3.62)
The analytical solution of the above differential equation is
2 1 sin . y x x x (3.63)
The maximum errors (in absolute value) are listed in Table 3.5.
The interval 0 1x was divided into n equal subintervals each of width 2 mh
with 3,4,5m so that 7,15,31n respectively. The value of y Y , where Y is
numerical solution, was computed for each value of n . The results for all second-,
fourth-, and sixth-order methods are given in Tables 3.1-3.5. In Table 3.1 the new
method (3.21) is applied to problem 3.1 and the results are compared with (Wazwaz
[157]), where as in Table 3.2 the new method is applied to problem 3.2 and the results
are compared with (Boutayeb and Twizell [24]). The new methods of different orders
are applied to problems 3.2-3.4 and the results are shown in Tables 3.3-3.5 respectively.
Performance of the new method is better than (Wazwaz [157]) and it is comparable in
case of (Boutayeb and Twizell [24]). In (Akram and Siddiqi [8]) problem 3.4 is solved
with the same solution but different boundary conditions. These results are reported in
Table 3.6. This table shows improved performance of our method. Figs. 3.1-3.4 show
graphs of exact and approximate solutions for various values of n. Error graphs are
shown in Figs. 3.5-3.8.
Chapter 3 Sixth-order BVPs
70
Table 3.1
Maximum absolute errors corresponding to problem 3.1
n Sixth-order method (3.3.3) (Wazwaz [157])
10 3.577×10-13 1.299×10-4
Table 3.2
Maximum absolute errors corresponding to problem 3.2
n Sixth-order method (3.3.3) (Boutayeb and Twizell [24])
7 4.55×10-7 2.41×10-7
15 5.96×10-9 7.56×10-10
31 2.68×10-11 2.25×10-11
Table 3.3
Maximum absolute errors corresponding to problem 3.2
n Second-order method (3.3.1) Fourth-order method (3.3.2) Sixth-order method (3.3.3)
7 2.3×10-3 3.0×10-3 4.55×10-7
15 3.4×10-4 1.38×10-4 5.96×10-9
31 1.54×10-4 4.30×10-6 2.68×10-11
Table 3.4
Maximum absolute errors corresponding to problem 3.3
n Second-order method (3.3.1) Fourth-order method (3.3.2) Sixth-order method (3.3.3)
7 2.99×10-2 2.39×10-4 1.15×10-9
15 7.00×10-3 3.43×10-6 3.95×10-12
31 1.80×10-3 7.34×10-8 4.41×10-11
Chapter 3 Sixth-order BVPs
71
Table 3.5
Maximum absolute errors corresponding to problem 3.4
n Second-order method (3.3.1) Fourth-order method (3.3.2) Sixth-order method (3.3.3)
7 1.23×10-2 6.97×10-4 1.78×10-8
15 2.80×10-3 3.60×10-5 1.37×10-10
31 1.6×10-3 7.44×10-7 9.45×10-11
Table 3.6
Maximum absolute errors corresponding to problem 3.4 and results of (Akram and Siddiqi [8])
n (Akram and Siddiqi [8]) Sixth-order method (3.3.3)
8 1.04×10-4 9.2×10-9
16 6.58×10-6 8.45×10-11
32 6.86×10-7 2.75×10-11
64 1.51×10-7 1.97×10-9
Chapter 3 Sixth-order BVPs
72
1 2 3 4 5 6 7 8 9 101
1.2
1.4
1.6
1.8
2
2.2
2.4
2.6
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 3.1: Plot of solution for problem 3.1 , 10n
0 5 10 15 20 25 30 350
0.02
0.04
0.06
0.08
0.1
0.12
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 3.2: Plot of solution for problem 3.2 , 31n
Chapter 3 Sixth-order BVPs
73
0 5 10 15 20 25 30 350
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 3.3: Plot of solution for problem 3.3 , 31n
0 2 4 6 8 10 12 14 16-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
Grid Points
So
luti
on
Exact solutionCalculated values
Fig. 3.4: Plot of solution for problem 3.4 , 15n
Chapter 3 Sixth-order BVPs
74
1 2 3 4 5 6 7 8 9 100.5
1
1.5
2
2.5
3
3.5
4
4.5x 10
-13
Grid points
Err
or
Fig. 3.5: Error graph for problem 3.1 , 10n
0 5 10 15 20 25 30 35-5
-4.5
-4
-3.5
-3
-2.5
-2
-1.5
-1
-0.5
0x 10
-11
Grid points
Err
or
Fig. 3.6: Error graph for problem 3.2 , 31n
Chapter 3 Sixth-order BVPs
75
0 5 10 15 20 25 30 35-5
-4.5
-4
-3.5
-3
-2.5
-2
-1.5
-1
-0.5
0x 10
-11
Grid points
Err
or
Fig. 3.7: Error graph for problem 3.3 , 15n
0 5 10 15 20 25 30 35-5
-4.5
-4
-3.5
-3
-2.5
-2
-1.5
-1
-0.5
0x 10
-11
Grid points
Err
or
Fig. 3.8: Error graph for problem 3.4 , 64n
Chapter 3 Sixth-order BVPs
76
3.7 Conclusion
Non-polynomial spline functions are used to develop a class of numerical methods for
approximate solution of sixth-order linear and non-linear BVPs, with two point
boundary conditions. Second-, Fourth- and Sixth- order convergence is obtained. It has
been shown that the relative errors in absolute value confirm the theoretical
convergence.
***************************************************
Chapter 4
Collocation Method using
Quartic B-Spline for Numerical Solution
of the Modified Equal Width Equation
4.1. Introduction
We consider a model for nonlinear waves, the modified equal width (MEW) equation
2 0 t x xxtu u u u (4.1)
subject to the following physical boundary conditions
1 2, , , , u a t u b t (4.2a)
along with collocation boundary conditions necessary for unique quartic B-spline
solution
, , 0, , , 0, x x xx xxu a t u b t u a t u b t (4.2b)
and initial condition
,0 , .u x f x a x b (4.3)
The parameter is a positive constant and is an arbitrary constant, ( )f x is a
localized disturbance inside the interval [ , ]a b and 0 asU x . Here the
subscripts t and x denote differentiation with respect to t and x respectively. The
MEW equation was introduced by (Morrison et al. [91]) as a model for nonlinear
dispersive waves and is related to the modified regularized long wave (MRLW)
equation (Abdulloev et al. [3]) and the modified Korteweg-de Vries (MKdV) equation
(Gardner et al. [42]). Many authors have investigated numerical solution of the problem
(4.1)-(4.3). These include finite difference method given by (Esen and Kutluay [39]),
He's variational iteration method (Junfeng [68]), tanh and sine-cosine methods
(Wazwaz [159]), mesh free methods (Haq et al. [46]) and various forms of finite
element methods including collocation and Galerkin methods, (see (Esen [38], Evans
Chapter 4 MEW equation
78
and Raslan [40], Raslan [114], Saka [116], Zaki [162]) and the references therein).
In Refs. (Evans and Raslan [40], Raslan [114], Saka [116], Zaki [162]) the MEW
equation is solved numerically by collocation methods based on quadratic, cubic and
quintic B-splines. The present method solves Eqs. (4.1)-(4.3) by using quartic B-spline
collocation method. In Section 4.2, a new numerical method is developed. The stability
analysis of the method is established in section 4.3 and test problems are reported in
section 4.4 to validate performance of the method. Some conclusions are drawn in
section 4.5.
4.2 Quartic B-spline solution
In order to develop the numerical method for approximating solution of boundary value
problems like the one given in Eqs. (4.1)-(4.3), the interval [ , ]a b is partitioned into
1N uniformly spaced points mx such that 0 1 1... N Na x x x x b and
b a
hN
. The quartic B-splines , 2, 3,... 1mB x m N at the knots mx are defined
as (See (Prenter [102])):
4
1 2 2 1
4
2 1 1 1
4
3 2 14 4 4
3 2 1 2
4
3 2 3
, , ,
5 , , ,
10 , , ,1
5 , , ,
, , ,
0 otherwise,
m m m
m m m
m m mm
m m m m
m m m
d x x x x
d d x x x x
d d x x x xB x
h x x x x x x
x x x x
(4.4)
and the set 2 1 1, ,..., NB B B of quartic B-splines forms a basis over the interval ,a b .
The numerical solution ,U x t to ,u x t is given as:
1
2
, ,
N
m mm
U x t t B x (4.5)
where m t are time dependent parameters to be determined at each time level. The
nodal values , and m m mU U U at the knots mx are derived from Eqs. (4.4)-(4.5) in the
following form
Chapter 4 MEW equation
79
2 1 1
2 1 1
2 1 12
11 11 ,
43 3 ,
12.
m m m m m
m m m m m
m m m m m
U
Uh
Uh
(4.6)
where dashes represent differentiation with respect to space variable. Eq. (4.1) can be
rewritten as
2 0. xx xtu u u u (4.7)
We discretize the time derivative of Eq. (4.7) by a first order accurate forward
difference formula and apply the -weighted, (0 1) scheme to the space
derivative at two adjacent time levels to obtain the equation
1 1
12 21 0,
n n n nn nxx xx
x x
U U U UU U U U
t
(4.8)
where t is time step and the superscripts n and n+1 are successive time levels. In this
work we take 1/ 2 . Hence Eq. (4.8) takes the form
11 1 2 2
0,2
n nn n n nxx xx x xU U U U U U U U
t
(4.9)
The nonlinear term in Eq. (4.9) is approximated using the Taylor series:
2 1 1 2 1 1 2( ) ( ) 2 2( ) . n n n n n n n n nx x x xU U U U U U U U U (4.10)
At the nth time step, we denote , and m m mU U U at the knots mx by the following
expressions:
1 2 1 1
2 2 1 1
3 2 1 12
11 11 ,
43 3 ,
12.
n n n nm m m m m
n n n nm m m m m
n n n nm m m m m
L
Lh
Lh
(4.11)
Using the knots , 0,1,.., ,mx m N as collocation points, the following recurrence
relation at point mx is obtained using Eq. (4.9)-(4.11):
21 1 1 1 21 2 2 1 3 4 1 1 1 2 32 2n n n n
m m m m m m m m m m m mha a a a L tL L L (4.12)
where
Chapter 4 MEW equation
80
2 21 0 1 2 0 1
2 23 0 1 4 0 1
20 1 2
4 24 , 11 12 24 ,
11 12 24 , 4 24 ,
L 2 1 and 0,1,..., .
m m m m m m
m m m m m m
m m m
a L thL a L thL
a L thL a L thL
h L L m N
The Eq. (4.12) relates parameters at adjacent time levels and gives 1N equations in
4N unknowns , 2, 1,..., 1.i i N In order to get a unique solution; we eliminate
the parameters 2 1 1, , N . Using Eq. (4.6) and the boundary conditions (4.2), the
values of the parameters take the form
12 0 1
11 0 1
1 2 1 2
333 7,
4 4 87 1
,4 4 8
11 11 .N N N N
(4.13)
Elimination of the above parameters from Eq. (4.12) yields a 4-banded linear system of
(N+1) equations in (N+1) unknown parameters , 0,1,...,i i N . The linear system can
be solved by a four-diagonal solver successively for , 1, 2,..., ;ni n once we calculate
the initial parameters 0m . Finally the approximate solution ,U x t will be obtained
from Eq. (4.6).
Using the initial and boundary conditions, the values of the initial parameters 0m at
the initial time are determined with the help of the following expressions:
0 0 0 0
2 1 1
0 0 0 00 2 1 1 2
0 0 0 00 2 1 0 12
0 0 0 02 1 1 2
0 0 0 02 1 12
,0 ,0 11 11 , 0,1,..., ,
4,0 3 3 0,
12,0 0,
4,0 3 3 0,
12,0 0.
m m m m m m m
N N N N N
N N N N N
U x u x f x m N
U xh
U xh
U xh
U xh
(4.14)
Eq. (4.14) consists of a linear 4 4N N system which can also be solved by a
four-diagonal solver.
Chapter 4 MEW equation
81
4.3 Stability Analysis
In this section we apply the Von-Neumann stability method (Mitchell and Griffiths
[89]) for the stability of scheme developed in the previous section. Since this method is
applicable to linear schemes, the nonlinear term 2xU U is linearized by taking U as a
constant value k . The linearized form of proposed scheme takes the form
1 1 1 11 2 2 1 3 4 1 4 2 3 1 2 1 1,
n n n n n n n nm m m m m m m mpp p p p p p p (4.15)
where
2 21 2
2 22 24 4 , 22 24 12 ,p h h t k p h h t k
2 23 4
2 222 24 12 , 2 24 4 , 0,1,..., .p h h t k p h h t k m N
Substitution of exp , 1,n nm i mh i into Eq. (4.15) leads to
1 2 3 4
4 3 2 1
exp 2 exp exp
exp 2 exp exp .
p i h p i h p p i h
p i h p i h p p i h
(4.16)
Simplifying Eq. (4.16), we get
2 2 2
2 2
2 2
2 2
2 2 2
2 2
2 2
2
2
22
22
where
A= 12 24 24 8 cos
2 4 24 cos 2 ,
20 16 48 sin
2 4 24 sin 2 ,
= 12 24 24 8 cos
2 4 24 cos 2 ,
20 16
h
C h
A iB
C iD
dth k h dth k h
h dth k h
B h dth k h
h dth k h
dth k h dth k h
h dth k h
D h dth k
2 2
48 sin
2 4 24 sin 2 .
h
h dth k h
(4.17)
After simplification, we obtain same expressions for 2 2A B and 2 2C D in the
following form:
Chapter 4 MEW equation
82
2 2 2 2
4 2 2 2 4 2 2
4 2 2 2 4 2
4 2 2 2 2 4
4 2 2 2 4 2 2
122 40 240 2888 cos
143 12 2 144
8 22 288 24 10 cos 2
8 4 24 144 cos 3 ,
A B C D
h dt h k hh
h h dt k
h h dt k h
h dt h k h h
(4.18)
so that 2
1 and the linearized numerical scheme for the MEW equation is
unconditionally stable.
4.4 Test problems and discussion
In this section the numerical method outlined in the previous section is tested for a
single solitary wave and interactions of two solitary waves. Moreover, the Maxwellian
initial condition is also considered. The accuracy of the scheme is measured in terms of
the following discrete forms of 2L and L error norms:
2
21
, ,N
i i i ii
i
L Max u U L h u U
(4.19)
where u and U are exact and approximate solutions respectively. The exact solitary
wave solution of MEW equation is given in (Esen and Kutluay [39]):
0
2
where
( , ) sec ,
1 , ,
6
u x t A h k x x ct
Ac k
(4.20)
and A, c represent the amplitude and velocity of a single solitary wave initially centered
at 0x . The initial condition for the above problem is given by
0( ,0) sec ,u x A h k x x (4.21)
and the boundary conditions are taken from Eq. (4.2a) with 1 2 0 . We examine
the conservation properties of the MEW equation related to mass, momentum and
energy by calculating the following three invariants (See (Zaki [162])):
2 2 41 2 3, , .
b b b
x
a a a
C udx C u u dx C u dx (4.22)
Chapter 4 MEW equation
83
Integrals in Eq. (4.22) can be approximated by the trapezoidal rule. 4.4.1 A single solitary wave Problem 4.1 To compare our results with (Esen [38], Esen and Kutluay [39], Evans
and Raslan [40], Raslan [114]), we choose the following parameters:
00, 80, 3, 0.25, 0.1, 0.2,0.05, 1, 30.a b A h t x
In order to find error norms and the invariants 1 2 3, ,C C C at different times, the
computations are carried out for times upto 20.t We have compared present method
with earlier published papers (Esen [38], Esen and Kutluay [39], Evans and Raslan
[40], Raslan [114]) at t=20 and the results are reported in Table 4.1. At t=20 the error
norms of the present method are -4 -40.010451 10 , 0.009269 10 ,L -4
2 0.015789 10 ,L
-40.007867 10 for the time steps 0.2 and 0.05 respectively. It is clear from Table 4.1
that the errors in 1 2 3, ,C C C approach zero during the simulation, showing excellent
conservation properties of the new method. Hence the performance of the new method
is better than the above mentioned methods. Fig. 4.1 shows the graphs of the single
solitary wave solutions at 0 and 10t . Initially the centre of solitary wave of
amplitude 0.25 is located at 30x . At time 20t its magnitude is 0.249922 centered
at 30.6x . The absolute difference in amplitudes over the time interval 0, 20 is
observed to be 57.8 10 while it is 52 10 in the case of velocities. It can be
concluded that the solitary wave moves to the right with almost constant magnitude and
velocity. The error graph at time 20t is reported in Fig. 4.2. It can be observed from
the graph that the maximum errors occur around the central position of the solitary
wave.
The same problem is also considered for different values of the amplitude at time
step of 0.01. In Table 4.2 the error norms and invariants are summarized for
0.25,0.5,0.75,1.0A . It is observed that the errors are smaller and the invariants
remain constant during the simulation. The new method is compared with (Esen and
Kutluay [39]) and the comparison of error norms declares superiority of present
scheme. Fig. 3 shows the graphs of the solutions for 0.25,0.5,0.75,1.0A at 20t .
Chapter 4 MEW equation
84
Problem 4.2 In order to compare our method with earlier work (Esen [38], Saka [116],
Zaki [162]) we choose the parameters 0, 70, 3, 0.25,0.5,1.0, 0.1, 0.05,a b A h t
01, 30x . The simulation is performed upto time t=20. Error norms and the
invariants 1 2 3, ,C C C are recorded for different values of t and tabulated in Table 4.3. It
is observed that the accuracy of the scheme in terms of error norms increases for
decreasing values of A. For 1,0.25A and t=20 the error norms of the present method
are found as -3 -71.095929 10 ,9.27 10L and -3 -72 1.747622 10 ,7.878 10L . The
invariant quantities 1 2 3, ,C C C are almost constant during the simulation. In Table 4.3
we have also compared our results with lumped Galerkin method using quadratic B-
spline functions (Esen [38]), collocation and petro-Galerkin methods using quintic B-
splines (Saka [116], Zaki [162]) at t=20. In this problem it is observed that the accuracy
of different schemes also depends on amplitude A. Performance of the present method
is better than the methods given in (Esen [38], Zaki [162]) when A=0.25 and
comparable with (Zaki [162]) for 0.5A . It is observed that error norms are less in the
methods given in (Saka [116], Zaki [162]) for 1.A
4.4.2 Interaction of two solitary waves
Problem 4.3 To study the interaction of two solitary waves we use the following initial
condition:
2
1
( ,0) sec ,i iu x A h k x x (4.23)
where
6 1, .i
i
cA k
Parameters 1 2 1 23, 1, 0.5, 15, 30, 1, 0.1, 0.2, 0 80A A x x h t x are chosen
for the sake of comparison with the results of (Esen and Kutluay [39], Evans and
Raslan [40], Saka [116]). These parameters give two solitary waves having amplitudes
of ratio 2:1 and their peak positions are located at 15 and 30.x The analytical values
of the invariants 1 2 3, ,C C C for the above parameters are given in (Evans and Raslan
[40], Saka [116]) as:
Chapter 4 MEW equation
85
2 21 1 2 2 1 2
4 43 1 2
84.712389, 3.333333,
34
1.416667.3
C A A C A A
C A A
(4.24)
The calculations are performed from 0 to 80.t t Values of the invariant quantities
1 2 3, ,C C C are tabulated in Table 4.4 for the present method and are compared with
Refs. (Esen and Kutluay [39], Evans and Raslan [40]). It can be seen from the table that
the invariants remain satisfactorily constant throughout the simulation. The upper
bounds for absolute error in the invariants 1 2 3, ,C C C from t=0 to t=55 are less than
7 3 31.0 10 ,2.6 10 and 2.6 10 respectively. In Refs. (Esen and Kutluay [39], Evans
and Raslan [40]), the same errors are less than 6 3 32.0 10 ,3.1 10 ,2.6 10 and
3 3 21.8 10 ,8.7 10 ,1.8 10 . Fig. 4.4 shows the state of interaction and then separation
of solitary waves at times 30,35,40t and 55,80t in sequel. Initially the larger
wave of amplitude 1 is centered at 15x and the smaller one of amplitude 0.5 at
30x . Since the velocity of larger wave is 0.5 and that of the smaller 0.125, the larger
wave moves faster than the smaller and hence collides with later. At 80t the
amplitude of the larger wave is 0.9993 centered at 56.8x and that of smaller 0.4988
with peak position located at 37.7.x Hence during this interaction the amplitude is
almost unchanged. The absolute difference in amplitude for larger wave is 47.0 10
and that of smaller 31.2 10 , consequently the velocities of the waves are almost
maintained after the interaction. Thus the waves interact and then emerge from the
collision by preserving their shapes and velocities. We have solved the same problem
with 0.025t and the invariants are reported in Table 4.5 along with the invariants
reported in Ref. (Saka [116]). It is evident from the comparison of Tables 4.4-4.5 that
the conservation properties of the present method are excellent when time step is
reduced.
4.4.3 The Maxwellian initial condition
Problem 4.4 The birth of solitary waves is considered using the Maxwellian initial
condition
2,0 exp .u x x (4.25)
Chapter 4 MEW equation
86
The parameters 3, 1,0.5,0.1,0.05,0.02,0.005, 0.1, 0.01, 20 20h t x used in
Refs. (Saka [116], Zaki [162]) are chosen. In the case of Maxwellian condition the
behavior of the solution depends on the values of . The Maxwellian does not break up
into solutions for c where c is some critical number, and exhibits rapidly
oscillating wave packets. When ,c a mixed type of solution is obtained consisting
of a leading soliton with an oscillating tail (See (Zaki [162])). The Maxwellian breaks
up into a number of solitons according to the value of when .c Simulations
are performed upto time 12.t For 1,0.5 the Maxwellian shows an oscillatory
behavior and no clean waves are obtained as shown in Fig. 4.5.
For 0.1,0.05,0.02,0.005 the number of observed solitary waves is 1, 2, 3 and 7
respectively as shown in Fig. 4.5. The graphs are in good agreement with earlier work
(Saka [116], Zaki [162]). It is also clear from Fig. 4.5 that the peaks of solitary waves
lie on the straight line. For various values of the conservative quantities are tabulated
in Table 4.6 which remain almost constant during the simulation.
Chapter 4 MEW equation
87
Table 4.1
Invariants and error norms for single solitary wave, problem 4.1
t Time L 410 2L 410 1C 2C 3C
0.2 0 0.0 0.0 0.785398 0.166667 0.005208
5 0.002706 0.004075 0.785398 0.166667 0.005208
10 0.005377 0.008094 0.785398 0.166667 0.005208
15 0.007944 0.012009 0.785398 0.166667 0.005208
20 0.010451 0.015789 0.785398 0.166667 0.005208
20 (Esen and Kutluay [39]) 2.576377 2.701647 0.785398 0.166474 0.005208
20 (Evans and Raslan [40]) 1.569539 2.021476 0.785286 0.166582 0.005206
20 (Raslan [114]) 1.744330 1.958879 0.784668 0.166434 0.005194
0.05 0 0.0 0.0 0.785398 0.166667 0.005208
5 0.002357 0.002128 0.785398 0.166667 0.005208
10 0.004788 0.004186 0.785398 0.166667 0.005208
15 0.007141 0.006114 0.785398 0.166667 0.005208
20 0.009269 0.007867 0.785398 0.166667 0.005208
20 (Esen [38]) 0.465523 0.796940 0.785390 0.166761 0.005208
20 (Esen and Kutluay [39]) 2.569972 2.692812 0.785398 0.166474 0.005208
20 (Evans and Raslan [40]) 2.498925 2.905166 0.784955 0.166477 0.005200
3, 0.25, 0.1, 1, 30,0 80oA h x x
Chapter 4 MEW equation
88
Table 4.2
Invariants and error norms for single solitary wave for various values of A, problem 4.1
A Time L 410 2L 410 1C 2C 3C
0.25 0 0.0 0.0 0.785398 0.166667 0.005208
5 0.002431 0.002174 0.785398 0.166667 0.005208
10 0.004952 0.004282 0.785398 0.166667 0.005208
15 0.007409 0.006267 0.785398 0.166667 0.005208
20 0.009651 0.008087 0.785398 0.166667 0.005208
20 (Esen and Kutluay [39]) 2.569562 2.692249 -- -- --
0.50 0 0.0 0.0 1.570796 0.666667 0.083333
5 0.018825 0.015873 1.570796 0.666667 0.083333
10 0.028470 0.026252 1.570796 0.666667 0.083333
15 0.028486 0.032155 1.570796 0.666667 0.083333
20 0.028502 0.036027 1.570796 0.666667 0.083333
20 (Esen and Kutluay [39]) 14.57568 18.26059 -- -- --
0.75 0 0.0 0.0 2.356194 1.500000 0.421875
5 0.035184 0.039624 2.356194 1.500000 0.421875
10 0.033608 0.047029 2.356194 1.500000 0.421875
15 0.035750 0.049007 2.356194 1.500000 0.421875
20 0.036156 0.051698 2.356194 1.500000 0.421875
20 (Esen and Kutluay [39]) 30.91793 43.95711 -- -- --
1.0 0 0.0 0.0 3.141593 2.666667 1.333333
5 0.095600 0.133043 3.141593 2.666667 1.333333
10 0.185455 0.261374 3.141593 2.666667 1.333333
15 0.276137 0.403847 3.141593 2.666667 1.333333
20 0.366993 0.549764 3.141593 2.666667 1.333333
20 (Esen and Kutluay [39]) 56.82131 82.85314 -- -- --
3, 0.1, 0.01, 1, 30,0 80oh t x x
Chapter 4 MEW equation
89
Table 4.3
Invariants and error norms for single solitary wave, problem 4.2
A Time L 310 2L 310 1C 2C 3C
0.25 0 0.0 0.0 0.785398 0.166667 0.005208
5 0.000236 0.000213 0.785398 0.166667 0.005208
10 0.000479 0.000419 0.785398 0.166667 0.005208
15 0.000714 0.000611 0.785398 0.166667 0.005208
20 0.000927 0.000787 0.785398 0.166667 0.005208
20 (Esen [38]) 0.046009 0.080145 0.785397 0.166764 0.005208
20 (Saka [116]) 0.00032 0.00027 0.785398 0.166667 0.005208
20 (Zaki [162]) 0.00203 0.00345 0.78539 0.16667 0.00521
0.5 0 0.0 0.0 1.570796 0.666667 0.083333
5 0.002090 0.003158 1.570796 0.666667 0.083333
10 0.004018 0.005902 1.570796 0.666667 0.083333
15 0.005900 0.008422 1.570796 0.666667 0.083333
20 0.007787 0.010999 1.570796 0.666667 0.083333
20 (Saka [116]) 0.00640 0.00920 1.570796 0.666667 0.083333
20 (Zaki [162]) 0.00852 0.01172 1.57078 0.66666 0.08333
1.0 0 0.0 0.0 3.141593 2.666668 1.333333
5 0.266834 0.410369 3.141593 2.666663 1.333329
10 0.541117 0.845886 3.141593 2.666659 1.333324
15 0.817482 1.294345 3.141593 2.666654 1.333320
20 1.095929 1.747622 3.141593 2.666650 1.333316
20 (Saka [116]) 0.65318 1.04778 3.141593 2.666667 1.333334
20 (Zaki [162]) 0.08360 0.14465 3.14165 2.66676 1.33343
3, 0.1, 0.05, 1, 30,0 70oh t x x
Chapter 4 MEW equation
90
Table 4.4
Invariant quantities for interaction of two waves, problem 4.3
Present method (Esen and Kutluay [39]) (Evans and Raslan [40])
Time 1C 2C 3C 1C 2C 3C 1C 2C 3C
0 4.712389 3.333337 1.416669 4.712388 3.329462 1.416669 4.712388 3.332357 1.416670
10 4.712389 3.332777 1.416108 4.712389 3.328927 1.416103 4.712022 3.324678 1.400768
20 4.712389 3.332191 1.415520 4.712387 3.328361 1.415523 4.711697 3.324210 1.401182
30 4.712389 3.330775 1.413861 4.712388 3.327818 1.413882 4.711242 3.346583 1.424847
40 4.712389 3.330942 1.414043 4.712385 3.327112 1.414050 4.711017 3.321250 1.398239
50 4.712389 3.330976 1.414314 4.712388 3.326632 1.414330 4.710804 3.320956 1.398729
55 4.712389 3.330701 1.414043 4.712386 3.326393 1.414062 4.710630 3.323628 1.399068
60 4.712389 3.330417 1.413763 4.712388 3.326228 1.413785 -- -- --
70 4.712389 3.329849 1.413199 4.712388 3.325891 1.413228 -- -- --
80 4.712389 3.329283 1.412635 4.712389 3.325434 1.412671
1 2 1 23, 1, 0.5, 15, 30, 0.1, 0.2, 0 80A A x x h t x
Table 4.5
Invariants for interaction of two solitary waves, problem 4.3
Present method (Saka [116])
Time 1C 2C 3C 1C 2C 3C
0 4.712389 3.333336 1.416669 4.7123884 3.3333358 1.4166697
5 4.712389 3.333336 1.416669 4.7123895 3.3333358 1.4166697
10 4.712389 3.333336 1.416669 4.7123896 3.3333358 1.4166697
15 4.712389 3.333335 1.416668 4.7123896 3.3333358 1.4166697
20 4.712389 3.333334 1.416667 4.7123896 3.3333358 1.4166696
25 4.712389 3.333332 1.416664 4.7123896 3.3333358 1.4166690
30 4.712389 3.333318 1.416647 4.7123896 3.3333359 1.4166648
35 4.712389 3.333295 1.416615 4.7123897 3.3333359 1.4166568
40 4.712389 3.333325 1.416655 4.7123896 3.3333358 1.4166669
45 4.712389 3.333332 1.416664 4.7123896 3.3333357 1.4166695
50 4.712389 3.333332 1.416665 4.7123896 3.3333357 1.4166698
55 4.712389 3.333332 1.416665 4.7123896 3.3333357 1.4166698
1 2 1 23, 1, 0.5, 15, 30, 0.1, 0.025, 0 80A A x x h t x
Chapter 4 MEW equation
91
Table 4.6
Invariants for Maxwellian initial condition for various values of , problem 4.4
Time 1C 2C 3C
1 0 1.772454 2.506634 0.886227
3 1.772453 2.506637 0.886223
6 1.772456 2.506629 0.886231
9 1.772456 2.506627 0.886232
12 1.772456 2.506627 0.886232
0.5 0 1.772454 1.879974 0.886227
3 1.772452 1.879979 0.886219
6 1.772454 1.879972 0.886227
9 1.772454 1.879972 0.886227
12 1.772454 1.879972 0.886226
0.1 0 1.772454 1.378646 0.886227
3 1.772437 1.378721 0.886144
6 1.772435 1.378718 0.886138
9 1.772430 1.378715 0.886130
12 1.772427 1.378712 0.886123
0.05 0 1.772454 1.315980 0.886227
3 1.772355 1.316331 0.885265
6 1.772273 1.315714 0.884471
9 1.772224 1.315072 0.883746
12 1.772167 1.314636 0.883001
3, 0.1, 0.01, 20 20h t x
Chapter 4 MEW equation
92
0 10 20 30 40 50 60 70 800
0.05
0.1
0.15
0.2
0.25
x
U
t=0
0 10 20 30 40 50 60 70 800
0.05
0.1
0.15
0.2
0.25
x
U
t=20
Fig. 4.1: Single solitary wave solution at t = 0, 20.
Chapter 4 MEW equation
93
0 10 20 30 40 50 60 70 80-8
-6
-4
-2
0
2
4
6
8
10x 10
-7
x
Err
or
Fig. 4.2: Error graph at 20.t
0 20 40 60 800
0.05
0.1
0.15
0.2
0.25
x
U
A=0.25
0 20 40 60 80-0.2
0
0.2
0.4
0.6
x
U
A=0.50
0 20 40 60 800
0.2
0.4
0.6
0.8
x
U
A=0.75
0 20 40 60 800
0.5
1
1.5
x
U
A=1.0
Fig. 4.3: Solitary wave solution for various values of A at 20t .
Chapter 4 MEW equation
94
0 20 40 60 800
0.2
0.4
0.6
0.8
1
t=0
0 20 40 60 800
0.2
0.4
0.6
0.8
1
t=30
0 20 40 60 800
0.2
0.4
0.6
0.8
1
t=35
0 20 40 60 800
0.2
0.4
0.6
0.8
1
t=40
0 20 40 60 800
0.2
0.4
0.6
0.8
1
t=55
0 20 40 60 800
0.2
0.4
0.6
0.8
1
t=80
Fig. 4.4: Interaction of solitary waves at selected times.
-20 -10 0 10 20-0.5
0
0.5
1=1
-20 -10 0 10 20-0.5
0
0.5
1=0.5
-20 -10 0 10 20
0
0.5
1
1.5=0.1
-20 -10 0 10 20
0
0.5
1
1.5=0.05
-20 -10 0 10 20
0
0.5
1
1.5=0.02
-20 -10 0 10 20
0
0.5
1
1.5=0.005
Fig. 4.5: Maxwellian initial condition for different values of at 12t .
Chapter 4 MEW equation
95
4.5 Conclusion
Quartic B-spline collocation method is employed to simulate the motion and interaction
of solitary waves of MEW equation. Four test problems are chosen from literature to
validate performance of the proposed method. The Maxwellian initial condition is also
studied. The accuracy of the method is checked through 2 ,L L error norms and the
invariants 1 2 3, ,C C C . It has been observed that the errors are sufficiently small and the
invariants are almost kept constant during simulation. The results obtained from
numerical experiments are in agreement with some earlier results available in the
literature. Linear stability analysis proved that the new method is unconditionally stable
theoretically and this has been supported by the test problems as well.
***************************************************
Chapter 5
Solitary Wave Solutions of the Modified
Regularized Long Wave Equation
5.1 Introduction
The generalized regularized long wave (GRLW) equation has the form
0,pt x x xxtU U U U U (5.1)
where parameters ,p and are positive constants. This equation has a major role in
the propagation of nonlinear dispersive waves and many authors have investigated its
numerical solution. These include finite difference method (Zhang [163]), Adomian
decomposition method (ADM) (Kaya [71], Kaya and El-Sayed [72]), mesh free
methods (Ali [12]) and Quasi linearization method (Ramos [107]). Exact solution was
obtained by using He’s variational iteration method (Soliman [135]) and Cosine-
function algorithm (Ali et al. [13]). A special case of Eq. (5.1) for 1p is the
regularized long wave (RLW) equation which is used to model a large number of
problems in various areas of science. The equation was originally introduced to explain
behavior of the undular bore development (Pergrine [101]). The analytical solution of
the RLW equation for some initial and boundary conditions are given in (Benjamin et
al. [20], Bona and Bryant [23]). It has been solved numerically by finite difference
methods (Eilbeck and McGuire [37], Pergrine [101]), Meshfree methods (Islam et al.
[53]), Fourier pseudospectral methods (Gou and Cao [45]) and various forms of finite
element methods including collocation, Galerkin and least square methods, (see (Dag
et al. [33], Dag et al. [34], Saka et al. [119]) for the details). We consider another
special case of Eq. (5.1) for 2p , the modified regularized long wave (MRLW)
equation given by
2 0t x x xxtu u u u u (5.2)
subject to the boundary conditions:
Chapter 5 MRLW equation
97
0 as .u x
We will consider the following periodic boundary conditions
, , 0u a t u b t (5.3)
and the collocation boundary conditions in order to get a unique B-spline solution,
given by
, , 0, , , 0.x x xx xxu a t u b t u a t u b t (5.4)
The initial condition is taken as
,0 , ,u x f x a x b (5.5)
where ( )f x is a localized disturbance inside the interval [ , ]a b . In Eq. (5.2) the
subscripts t and x denote differentiation with respect to t and x respectively. Various
methods have been used for the numerical solution of MRLW equation; for instance
cubic B-spline finite element method due to (Gardner et al. [43]), finite difference
method by (Khalifa et al. [74]), Adomian decomposition method in (Khalifa et al. [76])
and collocation method by (Khalifa et al. [75]).
We solve Eqs. (5.2)-(5.5) by quartic and quintic B-Spline collocation methods. The
outline of this chapter is as follows. In Section 5.2, the numerical methods are
presented. In section 5.3, stability is established and numerical results are reported in
section 5.4 to validate performance of the new methods. Conclusion is given in section
5.5.
5.2 The B-spline collocation methods
Let the partition of the space interval ,a b into uniformly spaced points mx be such
that 0 1 1... and N N
b aa x x x x b h
N
.
5.2.1 Quartic B-spline collocation method
We use the quartic B-splines given in Eq. (4.4). The approximate solution ,U x t of
MRLW equation to the exact solution ,u x t is given as:
1
2
, ,N
m mm
U x t t B x
(5.6)
Chapter 5 MRLW equation
98
where m t are time dependent parameters which will be determined for each time
level. Eq. (5.2) can be rewritten as
2 0.xx x xtu u u u u (5.7)
Discretizing the time derivative of Eq. (5.7) by a forward difference formula and
applying the -weighted scheme to the space derivative at two adjacent time levels, we
obtain the equation
1 111 2
21 0.
n n n nnnxx xx
x x
nn
x x
U U U UU U U
t
U U U
(5.8)
Substituting 1/ 2 in Eq. (5.8), we get
11 1 2 21
0.2 2
n nn n n n n n
xx xx x xx xU U U U U U U UU U
t
(5.9)
The nonlinear term in Eq. (5.9) is approximated by the Eq. (4.10). Using the knots
, 0,1,.., ,mx m N as collocation points, the following recurrence relation at point mx is
obtained using above equation and Eqs. (4.10)-(4.11):
2 2
1 2 1 3
1 1 1 1
1 2 2 1 3 4 1 2 1 2 ,m m m m
n n n n
m m m m m m m m h L tL L La a a a (5.10)
where
2 21 0 1 2 0 1
2 23 0 1 4 0 1
20 1 2
4 1 24 , 11 12 1 24 ,
11 12 1 24 , 4 1 24 ,
L 2 1 and 0,1,..., .
m m m m m m
m m m m m m
m m m
a L h t L a L h t L
a L h t L a L h t L
h L L m N
The Eq. (5.10) relates parameters at adjacent time levels and gives 1N equations in
4N unknowns , 2, 1,..., 1.i i N To assure a unique solution; we eliminate the
parameters 2 1 1, , N from Eq. (5.10) by making use of Eq. (4.13) with 1 2 0 .
After the elimination, Eq. (5.10) will give a 4-banded linear system of (N+1) equations
in (N+1) unknown parameters , 0,1,...,i i N . The linear system can be solved by a
four-diagonal solver successively for , 1, 2,...,ni n . Finally the approximate solution
,U x t will be obtained from Eq. (4.6). Before the commencement of the solution
process, we can find initial parameters 0m with the help of Eq. (4.14).
Chapter 5 MRLW equation
99
5.2.2 Quintic B-spline collocation method
Quintic B-spline functions , 2, 3,... 2mB x m N are defined at the knots mx by
(See (Prenter [102])):
5
1 3 3 2
5
2 1 2 2 1
5
3 2 1 1
5
4 3 15
5
5 4 1 1 2
5
6 5 2 2 3
, ,
6 , ,
15 , ,1
20 , ,
15 , ,
6 , ,
0 otherwise.
m m m
m m m
m m m
m m m m
m m m
m m m
e x x x x
e e x x x x
e e x x x x
B x e e x x x xh
e e x x x x
e e x x x x
(5.11)
The set of quintic B-splines 2 1 1 2, ,..., ,N NB B B B forms a basis over the interval
,a b . We take the approximate solution as
2
2
, ,N
m mm
U x t t B x
(5.12)
where m t are time dependent unknown real coefficients and mB x are five-degree
B-spline functions . Using Eqs. (5.11)-(5.12), the nodal values , and m m mU U U at the
knots mx can be written as:
2 1 1 2
2 1 1 2
2 1 1 22
26 66 26 ,
510 10 ,
202 6 2 .
m m m m m m
m m m m m
m m m m m m
U
Uh
Uh
(5.13)
At the nth time step, mU and its derivatives at the knots mx are denoted by the
following expressions
1 2 1 1 2
2 2 1 1 2
3 2 1 1 22
26 66 26 ,
510 10 ,
202 6 2 .
n n n n nm m m m m m
n n n nm m m m m
n n n n nm m m m m m
H
Hh
Hh
(5.14)
Using the knots , 0,1,.., ,mx m N as collocation points, we obtain the following
Chapter 5 MRLW equation
100
recurrence relation at point mx form Eq. (5.9) by making use of Eqs. (4.10) and (5.14):
2
1 2 3
1 1 1 1 1 2
1 2 2 1 3 4 1 5 2 12 1 2 ,m m m
n n n n n
m m m m m m m m m m mh H tH H Ha a a a a (5.15)
where
1 1 2 2 1 2
3 1 4 1 2
25 1 2 1 1 2
22 1
5 40 , 26 50 80 ,
66 240 , 26 50 80 ,
5 40 , 2 2 ,
1 and 0,1,..., .
m m m m m m
m m m m m
m m m m m m
m m
a K K a K K
a K a K K
a K K K h H H
K h t H m N
The Eq. (5.15) relates parameters at adjacent time levels and gives 1N equations in
5N unknowns , 2, 1,..., 2.i i N In order to have a closed form system; we
need to eliminate the parameters 2 1 1 2, , ,N N from the Eq. (5.15). Using Eq.
(5.13) and the boundary conditions, we arrive at the following values of the parameters:
2 0 1 2 1 0 1 2
1 2 1 2 2 1
165 65 9 33 9 1, ,
4 2 4 8 4 81 9 33 9 65 165
, .8 4 8 4 2 4N N N N N N N N
(5.16)
Elimination of the above parameters from Eq. (5.15) yields five-diagonal linear system
of (N+1) equations in (N+1) unknown parameters , 0,1,..., .i i N The linear system
can be solved by the penta-diagonal solver successively for , 1, 2,..., ;ni n once we
calculate the initial parameters 0m .Consequently the approximate solution ,U x t will
be obtained from Eq. (5.13).
The values of the initial parameters 0m at the initial time are determined with the
help of initial conditions and derivatives at boundaries in the following manner:
0 0 0 0 0
2 1 1 2
0 0 0 00 2 1 1 2
0 0 0 0 00 2 1 0 1 22
0 0 0 02 1 1 2
0 02 12
, 0 , 0 26 66 26 , 0,1,..., ,
5,0 10 10 0,
20,0 2 6 2 0,
5,0 10 10 0,
20,0 2
m m m m m m m m
N N N N N
N N N
U x u x f x m N
U xh
U xh
U xh
U xh
0 0 01 26 2 0.N N N
(5.17)
Eq. (5.17) consists of a five-diagonal linear system of 5N equations in
5N unknowns 0 , 2, 1,..., 2,m m N which can also be solved by penta-diagonal
Chapter 5 MRLW equation
101
algorithm.
5.3 Stability of the proposed scheme
Von-Neumann stability method (Mitchell and Griffiths [89]) is used for the stability of
scheme developed in the previous section. Being applicable to only linear schemes, the
nonlinear term 2xU U is linearized by taking U as a locally constant value k .
5.3.1 Stability of scheme based on quartic B-spline collocation method
The linearized form of proposed scheme using quartic B-spline collocation method
is given as
1 1 1 11 2 2 1 3 4 1 4 2 3 1 2 1 1,
n n n n n n n nm m m m m m m mpp p p p p p p (5.18)
where
2 2
2 21 2
2 23 4
2 2
1 1
1 12 24 4 , 22 24 12 ,
22 24 12 , 2 24 4 ,
0,1,..., .
k k
k kp h h t p h h t
p h h t p h h t
m N
Substitution of exp , 1,n nm i mh i into Eq. (5.18) leads to
1 2 3 4
4 3 2 1
exp 2 exp exp
exp 2 exp exp .
p i h p i h p p i h
p i h p i h p p i h
(5.19)
Simplifying Eq. (5.19), we get
,A iB
C iD
(5.20)
where
2 2 2
2 2
2 2
2 2
2 2 2 2
2 2
2
22 2A = 1 2 1 2 4 2 4 8 1 c o s
2 4 1 2 4 c o s 2 ,
2 0 1 6 1 4 8 s i n
2 4 1 2 4 s i n 2 ,
= 2 2 1 2 1 2 4 2 4 8 1 c o s
2 4 1 2 4 c o s 2 ,
2 0 1 6 1
h d t h k h d t h k h
h d t h k h
B h d t h k h
h d t h k h
C h d t h k h d t h k h
h d t h k h
D h d t h
2
2 2
4 8 s i n
2 4 1 2 4 s i n 2 .
k h
h d t h k h
After algebraic manipulation, we obtain similar expressions for 2 2A B and 2 2C D in
Chapter 5 MRLW equation
102
the following form:
2 2 2 2
4 2 2 2 2 2 2 2 4 2 2 2
4 2 2 2 2 2 2 2 4 2 2 2
4 2 2 2 2 2 2 2 4 2 2 2
4 2 2 2 2 2 2
16 61 20 40 20 120 144
8 143 12 24 12 24 144 cos
16 11 12 24 12 120 144 cos 2
8 4 8 4
A B C D
h dt h dt h k dt h k h
h dt h dt h k dt h k h h
h dt h dt h k dt h k h h
h dt h dt h k dt
2 4 2 2 224 144 cos 3 ,h k h h
so that 2
1, which proves unconditional stability of the linearized numerical
scheme for the MRLW equation.
5.3.2 Stability of scheme based on quintic B-spline collocation method
The linearized form of proposed scheme using quintic B-spline collocation method
is given by
1 1 1 1 11 2 2 1 3 4 1 5 2
5 2 4 1 3 2 1 1 2 ,
n n n n nm m m m m
n n n n nm m m m mp
p p p p p
p p p p
(5.21)
where
2 2 21 2 3
2 24 5
2 2
2 2
2 40 5 (1 ), 52 80 50 (1 ), 132 240 ,
52 80 50 (1 ), 2 40 5 (1 ), 0,1,..., .
p h h t k p h h t k p h
p h h t k p h h t k m N
Substitution of exp , 1,n nm i mh i into Eq. (5.21) leads to
1 2 3 4 5
5 4 3 2 1
exp 2 exp exp exp 2
exp 2 exp exp exp 2
m m m m m
m m m m m
p i h p i h p p i h p i h
p i h p i h p p i h p i h
(5.22)
Simplification of Eq. (5.22) yields
A iB
A iB
(5.23)
where
2 2 2
2 2
A= 4 80 cos 2 104 160 cos 132 240 ,
10 (1 )sin 2 100 (1 )sin ,
h h h h h
B h t k h h t k h
so that 1 and the linearized numerical scheme for the MRLW equation is
unconditionally stable.
Chapter 5 MRLW equation
103
5.4 Numerical Tests and Results
The numerical methods outlined in the previous section are tested for single solitary
wave and interactions of multiple solitary waves. The Maxwellian initial condition
which generates a train of solitary waves is also considered. The accuracy is checked
using the error norms L and 2L . The analytical solution of MRLW equation is given as
(See (Khalifa et al. [74])):
0
6( , ) sec 1 ,
1
c cu x t h x c t x
c
(5.24)
where 0x is an arbitrary constant. The initial condition is given by
0
6( ,0) sec ,
1
c cu x h x x
c
(5.25)
and the boundary conditions are extracted from the exact equation. The conservation
properties of the MRLW equation related to mass, momentum and energy are
determined by finding the following three invariants (Gardner et al. [43], Olver [98]):
2 2 4 21 2 3
6, , ,
b b b
x x
a a a
C udx C u u dx C u u dx
(5.26)
5.4.1 Single solitary wave
Problem 5.1 We take 01, 0.03, 0.2, 0.025, 1, 0, 40 60c h t x x to
compare our results with (Khalifa et al. [74]). In order to find error norms and the
invariants 1 2 3, ,C C C at different times, the computations are carried out for times upto
10t . The results are given in Table 5.1. At 10t the error norms are
-4 0.033611 10L and -42 0.048674 10L . The error in invariant 3C approaches
zero during the simulation and maximum absolute errors in 1 2,C C remain less than
3 63.55 10 and 2.0 10 throughout the simulation. In Table 5.1 the performance of
the new method is compared with finite difference method due to (Khalifa et al. [74]) at
10t . It is observed that errors of the method in (Khalifa et al. [74]) are considerably
larger than those obtained with method 5.2.1. Fig. 5.1(a) shows the solutions at
0,2,4,6,8,10t and illustrates the motion of solitary wave to the right along the
Chapter 5 MRLW equation
104
interval 40 60x when 0 10t . Initially the centre of the solitary wave of
amplitude 0.424264 is located at 0x . At time 10t its amplitude is 0.424202
centered at 10.2x . The absolute difference in amplitudes over the time interval
0,10 is found to be 56.2 10 . It can be inferred that the solitary wave moves to the
right with constant amplitude and velocity. Error graph is shown in Fig. 5.1(b).
Problem 5.2 For the sake of comparison with earlier work of authors (Gardner et al.
[43], Khalifa et al. [75]), we choose the following parameters for the solution of
MRLW equation using method 5.2.2:
00, 100, 6, 1, 0.2, 0.025, 1, 40.a b c h t x In order to find error norms
and the invariants 1 2 3, ,C C C at different times, the computations are carried out for
times upto 10.t The amplitude of the solitary wave is 0.9999644 located at 60.x
The absolute difference between numerical and exact peak values is 53.56 10 . The
results are summarized in Table 5.2. At 10t the error norms are -32.0248 10L
and -32 3.9314 10L . It is clear from Table 5.2 that 1C is constant and 2 3,C C remain
almost constant with the maximum error of 55.0 10 . In Table 5.2 the new method is
compared with (Gardner et al. [43], Khalifa et al. [75]) which shows superiority of our
scheme. Fig. 5.2(a) shows the graphs of the solutions at 0, 2,5,10t and depicts the
motion of solitary wave from left to the right along the interval 0 100.x The error
graph is reported in Fig. 5.2(b).
Problem 5.3 The parameters 06, 0.3, 0.1, 0.01, 1, 40,0 100c h t x x are
chosen to enable comparison with collocation method based on cubic splines (Khalifa
et al. [75]). The results are given in Table 5.3. At 20t the error norms are
-5 2.222848 10L and -52 5.089274 10L . The absolute error in invariants
1 2 3, ,C C C approach zero during the simulation. In Table 5.3 the new method is
compared with collocation method based on cubic splines given in (Khalifa et al. [75]).
Again errors are less in the present method, however both methods 5.2.1 and (Khalifa
et al. [75]) produce excellent conservation properties. Fig. 5.3(a) shows the solutions at
0,5,10,15,20t and illustrates the motion of solitary wave to the right along the
interval 0 100x when 0 20t . Initially the amplitude of solitary wave is
0.547723 and its peak position is located at 40x . At 20t its amplitude is recorded
Chapter 5 MRLW equation
105
as 0.547721 with center 66x . Hence the absolute difference in amplitudes over the
time interval 0, 20 is observed as 62.0 10 Error graph is displayed in Fig. 5.3(b).
The same problem is also solved by method 5.2.2 and the results are reported in
Table 5.4. Fig. 5.4(a) shows the solution at 0,6,12, 20t and the error graph is shown
in Fig. 5.4(b).
5.4.2 Two solitary waves
Problem 5.4 We consider interaction of two solitary waves using the following initial
condition:
2
1
6( ,0) sec .
1i i
ii i
c cu x h x x
c
(5.27)
In our numerical experiment, we choose 1 2 1 26, 4, 1, 25, 55, 1, 0.2,c c x x h
0.025,t 0 250.x The parameters give solitary waves of different amplitudes of
ratio 2:1 having centers at 25 and 55x x to make the interaction possible. For the
case of interaction of two solitary waves the analytical values of invariants are given in
(Khalifa et al. [75]) as 1 2 311.467698, 14.629243, 22.880466C C C . Computations
for both methods 5.2.1 and 5.2.2 are done up to time 20t and same values for
invariant quantities are obtained. The values of invariants 1 2 3, ,C C C are tabulated in
Table 5.5 and compared with (Khalifa et al. [75]) at 20t . The invariants are almost
constant , however it is observed that collocation method using cubic splines by
(Khalifa et al. [75]) has better conservative properties in this case. In Fig. 5.5 the
interaction of solitary waves is shown at different times 0,4,8,10,14,20t
respectively. It is clear from the figure that the two solitary waves interact at times
4,8,10t and then separate at 14,20t preserving their original shapes.
We have also taken 1 2 1 26, 0.03, 0.01, 18, 58, 40 180c c x x x to
compare our results with ADM method in (Khalifa et al. [76]). The results are tabulated
in Table 5.6. It is clear from the table that the conservation properties of present method
are better than ADM when smaller amplitudes and times are considered. Moreover, no
interaction of solitary waves is observed over the time interval 0, 2 .
Chapter 5 MRLW equation
106
5.4.3 Three solitary waves
Problem 5.5 Interaction of three solitary waves is considered using the initial
condition
3
1
6( ,0) sec ,
1i i
ii i
c cu x h x x
c
(5.28)
and the following parameters for the purpose of comparison.
1 2 3 1 2 36, 4, 1, 0.25, 15, 45, 60, 1, 0.2, 0.025, 1,0 250.c c c x x x h t x These parameters give solitary waves of different amplitudes 4, 2 and 1 having their
peaks located at 15,45 and 60x moving along the same direction. The analytical
values of invariants for the interaction of three waves are given in (Khalifa et al. [75])
as 1 2 314.9801, 15.8218, 22.9923C C C . Computations for both methods 5.2.1 and
5.2.2 are done from 0 to 45t t and same values for invariant quantities are
obtained. The values of the invariant quantities during the simulation are given in Table
5.7. It is evident from this table that the values produced by our method for 1 2,C C are
better when compared with the exact values whereas the performance of the method in
(Khalifa et al. [75]) is better in case of 3C . Fig. 5.6 shows the interaction of solitary
waves at times 0,5,8,15,20,40t respectively. The three solitary waves interact at
times 5,8,15t and then separate at 20,40t emerging unchanged.
The parameters 1 2 2 1 2 36, 0.03, 0.02, 0.01, 18, 48, 88, 40 180c c c x x x x are
chosen to compare our results with ADM of (Khalifa et al. [76]). The results are
tabulated in Table 5.8 along with those given in (Khalifa et al. [76]). It is again
observed that the conservation properties of present method are very good as compared
to ADM when smaller amplitudes and times are considered.
5.4.4 The Maxwellian initial condition
Problem 5.6 As the last problem, the Maxwellian initial condition
2,0 exp 40u x x , (5.29)
is used with boundary conditions given by Eq. (5.3) when 0a and 100b . The
numerical solutions of the MRLW equation are carried out for generation of a train of
solitary waves for various values of . We consider for the set of values
Chapter 5 MRLW equation
107
0.1,0.04,0.015,0.01, which is used in the earlier work(See (Gardner et al. [43], Khalifa
et al. [74])) and the simulations are performed up to time 15t . For 0.1 we
observe generation of a single solitary wave along with an oscillating tail. However
when is reduced, Maxwellian initial condition breaks up into a number of solitary
waves. For 0.04,0.015,0.01 the number of observed solitons is 2,3,4 respectively.
An oscillating wave behind the train of solitary waves is observed in each case. The
various cases for are shown in Fig. 5.7 and the graphs are in agreement with earlier
work (Gardner et al. [43], Khalifa et al. [74]). It is also clear from the Fig. 5.7 that the
peaks of solitary waves lie on the straight line. The conservative quantities are given in
Table 5.9 which remain almost constant during the simulation
Chapter 5 MRLW equation
108
Table 5.1 Accuracy test for problem 5.1 method 5.2.1 Time L
410 2L 410 1C 2C 3C
0 0.0 0.0 7.804400 2.129885 0.130251
1 0.001589 0.004743 7.805221 2.129885 0.130251
2 0.003456 0.009481 7.805897 2.129886 0.130251
3 0.005569 0.014214 7.806451 2.129886 0.130251
4 0.007960 0.018951 7.806898 2.129887 0.130251
5 0.010731 0.023719 7.807254 2.129887 0.130251
6 0.013944 0.028526 7.807528 2.129887 0.130251
7 0.017671 0.033387 7.807729 2.129887 0.130251
8 0.022125 0.038349 7.807864 2.129887 0.130251
9 0.027374 0.043428 7.807936 2.129887 0.130251
10 0.033611 0.048674 7.807948 2.129887 0.130251
10 (Khalifa et al. [74]) 1.99524 6.98280 7.80932 2.12988 0.130315
Invariants and error norms for single solitary wave, 1, 0.03, 0.2, 0.025,c h t 40 60x
Chapter 5 MRLW equation
109
Table 5.2 Accuracy test for problem 5.2 method 5.2.2 Time L
310 2L 310 1C 2C 3C
0 0.0 0.0 4.44288 3.29983 1.41421
1 0.2209 0.4036 4.44288 3.29983 1.41421
2 0.4283 0.8041 4.44288 3.29982 1.41420
3 0.6289 1.2018 4.44288 3.29982 1.41420
4 0.8277 1.5953 4.44288 3.29981 1.41419
5 1.0261 1.9859 4.44288 3.29980 1.41419
6 1.2248 2.3749 4.44288 3.29980 1.41418
7 1.4239 2.7633 4.44288 3.29980 1.41418
8 1.6236 3.1520 4.44288 3.29979 1.41417
9 1.8239 3.5413 4.44288 3.29979 1.41417
10 2.0248 3.9314 4.44288 3.29978 1.41416
10 (Khalifa et al. [75]) 5.43718 9.30196 4.44288 3.29983 1.41420 10 (Gardner et al. [43]) 9.24 16.39 4.442 3.299 1.413
Invariants and error norms for single solitary wave, 6, 1, 0.2, 0.025, 1, 40,0 100oc h t xx
Chapter 5 MRLW equation
110
Table 5.3 Accuracy test for problem 5.3 method 5.2.1 Time L
510 2L 510 1C 2C 3C
0 0.0 0.0 3.581967 1.345076 0.153723
2 0.267928 0.556873 3.581967 1.345076 0.153723
4 0.520815 1.103371 3.581967 1.345076 0.153723
6 0.768640 1.637097 3.581967 1.345076 0.153723
8 0.994829 2.156048 3.581967 1.345076 0.153723
10 1.209189 2.662302 3.581967 1.345076 0.153723
12 1.417087 3.158828 3.581967 1.345076 0.153723
14 1.621213 3.648135 3.581967 1.345076 0.153723
16 1.823033 4.132113 3.581967 1.345076 0.153723
18 2.023404 4.612153 3.581967 1.345076 0.153723
20 2.222848 5.089274 3.581967 1.345076 0.153723
20 (Khalifa et al.
[75])
29.6650 60.6885 3.58197 1.34508 0.153723
Invariants and error norms for single solitary wave, 6, 0.3, 0.1, 0.01, 40oc h t x
Chapter 5 MRLW equation
111
Table 5.4
Accuracy test for problem 5.2 method 5.2.2
Time L 510 2L 510 1C 2C 3C
0 0.0 0.0 3.58197 1.34508 0.15372
2 0.2683 0.5669 3.58197 1.34508 0.15372
4 0.5403 1.1236 3.58197 1.34508 0.15372
6 0.7902 1.6654 3.58197 1.34508 0.15372
8 1.0187 2.1912 3.58197 1.34508 0.15372
10 1.2354 2.7038 3.58197 1.34508 0.15372
12 1.4457 3.2064 3.58197 1.34508 0.15372
14 1.6523 3.7016 3.58197 1.34508 0.15372
16 1.8565 4.1914 3.58197 1.34508 0.15372
18 2.0593 4.6773 3.58197 1.34508 0.15372
20 2.2612 5.1602 3.58197 1.34508 0.15372
Invariants and error norms for single solitary wave, 6, 0.3, 0.1, 0.01, 1,c h t
40,0 100o xx
Table 5.5 Accuracy test for problem 5.4 Time
1C 2C 3C
0 (Analytical) 11.467698 14.629277 22.880432
2 11.467698 14.624259 22.860365
4 11.467698 14.619226 22.840279
6 11.467699 14.614169 22.820069
8 11.467700 14.606821 22.787857
10 11.467700 14.603687 22.771773
12 11.467699 14.603056 22.775766
14 11.467699 14.598059 22.756029
16 11.467700 14.593048 22.736127
18 11.467700 14.588061 22.716289
20 11.467701 14.583089 22.696510
20 (Khalifa et al. [75]) 11.4677 14.6292 22.8809
Invariants for two solitary waves, 1 2 1 2 0 2506, 4, 1, 25, 55, 0.2, 0.025, xc c x x h t
Chapter 5 MRLW equation
112
Table 5.6 Accuracy test for problem 5.4 Present method ADM (Khalifa et al. [76])
Time 1C 2C 3C 1C 2C 3C
0 6.345403 0.592825 0.00548537 6.34540 0.592825 0.00548538
0.2 6.345374 0.592825 0.00548536 6.34532 0.592789 0.00548414
0.4 6.345377 0.592825 0.00548536 6.34508 0.592683 0.00548068
0.6 6.345380 0.592825 0.00548537 6.34466 0.592511 0.00547517
0.8 6.345383 0.592825 0.00548537 6.34408 0.592280 0.00546787
1.0 6.345385 0.592825 0.00548537 6.34333 0.591998 0.00545912
1.2 6.345388 0.592825 0.00548537 6.34241 0.591678 0.00544929
1.4 6.345390 0.592825 0.00548537 6.34132 0.591334 0.00543879
1.6 6.345392 0.592825 0.00548537 6.34007 0.590985 0.00542807
1.8 6.345394 0.592825 0.00548537 6.33864 0.590648 0.00541753
2 6.345395 0.592825 0.00548537 6.33705 0.590348 0.00540758
Invariants for interaction of two solitary waves, 1 2 1 26, 0.03, 0.01, 18, 58,c c x x 40 180x
Chapter 5 MRLW equation
113
Table 5.7 Invariant quantities for problem 5.5 Time
1C 2C 3C
0 14.980099 15.837528 23.008136
5 14.980105 15.824928 22.957891
10 14.980109 15.807025 22.877972
15 14.980106 15.807032 22.885947
20 14.980106 15.795022 22.837454
25 14.980107 15.782840 22.788852
30 14.980107 15.770634 22.740419
35 14.980108 15.758480 22.692279
40 14.980108 15.746389 22.644448
45 14.968030 15.734374 22.596591
45 (Khalifa et al.
[75])
13.7043 15.6563 22.9303
Invariants for three solitary waves, 1 2 3 1 2 3 , 0.2, 0.025,4, 1, 0.25, 15, 45, 60 h tc c c x x x
0 250x
Chapter 5 MRLW equation
114
Table 5.8 Invariant quantities for problem 5.5 Present method ADM (Khalifa et al. [76])
Time 1C 2C 3C 1C 2C 3C
0 9.517705 0.904129 0.00786304 9.51769 0.904129 0.00786310
0.1 9.517580 0.904130 0.00786272 9.51770 0.904193 0.00786515
0.2 9.517585 0.904130 0.00786286 9.51771 0.904383 0.00787064
0.3 9.517590 0.904130 0.00786293 9.51772 0.904700 0.00787956
0.4 9.517594 0.904130 0.00786294 9.51772 0.905144 0.00789191
0.5 9.517598 0.904130 0.00786296 9.51773 0.905715 0.00790768
0.6 9.517602 0.904130 0.00786297 9.51774 0.906413 0.00792688
0.7 9.517606 0.904130 0.00786298 9.51775 0.907239 0.00794952
0.8 9.517610 0.904130 0.00786299 9.51776 0.908192 0.00797559
0.9 9.517613 0.904130 0.00786300 9.51776 0.909270 0.00800510
1.0 9.517616 0.904130 0.00786300 9.51777 0.910476 0.00803807
Invariants for interaction of three solitary waves, 1 2 3 16, 0.03, 0.02, 0.01, 18,c c c x 2 348, 88, 40 180x x x
Chapter 5 MRLW equation
115
Table 5.9 Invariant quantities for problem 5.6 Time
1C 2C 3C
0.1 0 1.77245 1.37865 0.76089 3 1.77246 1.37877 0.76158 6 1.77246 1.37867 0.76139 9 1.77246 1.37857 0.76125 12 1.77246 1.37846 0.76111 15 1.77246 1.37836 0.76097 0.04 0 1.77245 1.30345 0.83609 3 1.77264 1.30454 0.84638 6 1.77240 1.30191 0.83980 9 1.77236 1.30110 0.83961 12 1.77218 1.29905 0.83504 15 1.77216 1.29843 0.83549 0.015 0 1.77245 1.27211 0.86743 3 1.77244 1.27112 0.87222 6 1.77236 1.27003 0.87212 9 1.77224 1.26859 0.87036 12 1.77189 1.26478 0.85872 15 1.77222 1.26423 0.85410 0.01 0 1.77245 1.26585 0.87369 3 1.77238 1.26583 0.89749 6 1.77167 1.26155 0.89028 9 1.77027 1.25148 0.85771 12 1.76957 1.24724 0.85094 15 1.76864 1.24121 0.83689 Conservative quantities for MRLW equation using Mawellian initial condition
Chapter 5 MRLW equation
116
-40 -30 -20 -10 0 10 20 30 40 50 600
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
x
U
t=6
t=4
t=2
t=0
t=10
t=8
Fig. 5.1(a): Single solitary wave solution at 0, 2, 4, 6,8,10t
0 10 20 30 40 50 60 70 80 90 100-0.2
0
0.2
0.4
0.6
0.8
1
1.2
t=0 t=2 t=5 t=10
Fig. 5.2(a): Single solitary wave solution
Chapter 5 MRLW equation
117
-40 -30 -20 -10 0 10 20 30 40 50 60-2
-1
0
1
2
3
4x 10
-6
x
Err
or
Fig. 5.1(b): Error graph at 10t
0 10 20 30 40 50 60 70 80 90 100-0.0025
-0.002
-0.0015
-0.001
-0.0005
0
0.0005
0.001
0.0015
0.002
0.0025
X
Err
or
Fig. 5.2(b): Error graph at 10t
Chapter 5 MRLW equation
118
0 10 20 30 40 50 60 70 80 90 100-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
x
U
t=0 t=5 t=10 t=15 t=20
Fig. 5.3(a): Single solitary wave solution method 5.2.1
0 10 20 30 40 50 60 70 80 90 100-2.5
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
2.5x 10
-5
Fig. 5.3(b): Error graph at 20t
Chapter 5 MRLW equation
119
0 10 20 30 40 50 60 70 80 90 1000
0.1
0.2
0.3
0.4
0.5
t=0 t=6 t=12 t=20
Fig. 5.4(a): Single solitary wave solution method 5.2.2
0 10 20 30 40 50 60 70 80 90 100-0.000025
-0.00002
-0.000015
-0.00001
-0.000005
0
0.000005
0.00001
0.000015
0.00002
0.000025
x
Err
or
Fig. 5.4(b): Error graph at 20t
Chapter 5 MRLW equation
120
0 50 100 150 200 250-1
0
1
2
0 50 100 150 200 250-1
0
1
2
0 50 100 150 200 250-1
0
1
2
0 50 100 150 200 250-1
0
1
2
0 50 100 150 200 250-1
0
1
2
0 50 100 150 200 250-1
0
1
2
t=0 t=4
t=8 t=10
t=14 t=20
Fig 5.5: Interaction of two solitary waves at selected times
0 50 100 150 200 250-1
0
1
2
0 50 100 150 200 250-1
0
1
2
0 50 100 150 200 250-1
0
1
2
0 50 100 150 200 250-1
0
1
2
0 50 100 150 200 250-1
0
1
2
0 50 100 150 200 250-1
0
1
2
t=0
t=8
t=20
t=5
t=40
t=15
Fig. 5.6: Interaction of three solitary waves at selected times
Chapter 5 MRLW equation
121
0.00 20.00 40.00 60.00 80.00 100.00-0.50
0.00
0.50
1.00
1.50
0 .00 20.00 40.00 60.00 80.00 100.00-0.50
0.00
0.50
1.00
1.50
0.00 20.00 40.00 60.00 80.00 100.00-0.05
0.00
0.50
1.00
1.50
0 .00 20.00 40.00 60.00 80.00 100.00-0.50
0.00
0.50
1.00
1.50
= 0.1 =0.04
=0.015 =0.01
Fig. 5.7: The Maxwellian initial condition at 14.5t
Chapter 5 MRLW equation
122
5.5 Conclusion
Collocation method based on quartic and quintic B-splines is applied to investigate
propagation of nonlinear dispersive solitary waves of the MRLW equation. The method
is validated by choosing six test problems from literature. The accuracy of the method
is examined through 2 ,L L error norms and the invariant quantities 1 2 3, ,C C C . It has
been observed that the error norms are sufficiently small and the invariants are almost
constant during the simulation. The obtained results are in agreement with some earlier
results from literature. Linear stability analysis and numerical tests proved that the new
schemes are unconditionally stable.
***************************************************
Chapter 6
Collocation Method using B-Splines for
Numerical Solution of
Kuramoto-Sivashinsky Equation
6.1 Introduction
The Kuramoto-Sivashinsky (KS) equation is given by
0,t x xx xxxxu uu u u (6.1)
subject to the following boundary conditions:
1 2, , , , 0 ,u a t g t u b t g t t (6.2)
and initial condition
,0 , .u x f x a x b (6.3)
The parameters and are some arbitrary constants. KS equation is a canonical
nonlinear evolution equation having various applications, e.g., long waves on thin
films, long waves on the interface between two viscous fluids (Hopper and Grimshaw
[48]) , unstable drift waves in plasmas, reaction diffusion systems (Kuramoto and
Tsuzuki [83]) and flame front instability (Sivashinsky [133]). It also describes the
fluctuations of the position of a flame front, the motion of a fluid going down a vertical
wall, or a spatially uniform oscillating chemical reaction in a homogeneous medium
(Conte [31]). Its well- posedness was established in (Tadmor [141]).
A variety of numerical and approximate methods are available in the numerical
literature to solve KS equation. Some of those are given in the references (Abdel-
Gawad and Abdusalam [1], Hyman and Nicolaenko [50], Hyman et al. [51], Jolly et al.
[66], Kevrekidis et al. [73], Khater and Temsah [78], Lopez-Marcos [84], Manickam et
al. [87], Papageorgiou and Smyrlis [99], Smyrlis and Papageorgiou [134], Xu and Shu
[160]). Exact solution through the balance of the dispersive and dissipative effects has
been investigated within (Abdel-Hamid [2]). For more details about the exact methods,
Chapter 6 Kuramoto-Sivashinsky equation
124
we refer to (Baldwin et al. [17], Parkes and Duffy [100], Wazwaz [158]).
We solve Eqs. (6.1)-(6.3) by using quartic and quintic B-Spline collocation methods
and splitting techniques. The organization of this chapter is as follows. In Section 6.2,
the numerical methods are presented. The numerical results are given in section 6.3 to
examine performance of the new methods and section 6.4 includes conclusion.
6.2 The B-Spline Collocation Methods
In this section we construct the numerical methods for approximating solution of
boundary value problem (6.1)-(6.3). The spatial domain [ , ]a b is divided into N
subintervals of equal length using the knots mx a mh , 0,1,...,m N , 0 , Nx a x b
and .b a
hN
6.2.1 Quintic B-Spline Collocation Method 1
We use the collocation method with quintic B-splines given by Eq. (5.11) as basis and
derive some consistency relations, which lead us to find numerical solution of the
problem under consideration. To get approximate solution we proceed as follows. Let
2
2
, ,N
m mm
U x t t B x
(6.4)
be the approximate solution, where m t are time dependent unknown real
coefficients and mB x are fifth-degree B-spline functions . From Eqs. (5.11) and (6.4),
the values of U and its space derivatives at the knots mx can be obtained as
2 1 1 2
2 1 1 2
2 1 1 22
2 1 1 24
26 66 26 ,
510 10 ,
202 6 2 ,
1204 6 4 .
m m m m m m
m m m m m
m m m m m m
m m m m m m
U
Uh
Uh
Uh
(6.5)
Application of weighted scheme with 12 to Eq. (6.1) yields
Chapter 6 Kuramoto-Sivashinsky equation
125
1 1 11
0,2 2 2
n n n n n nn n
x x xx xx xxxx xxxxUU UU U U U UU U
t
(6.6)
where t is time step and the superscripts n and n+1 denote the time step labels. The
nonlinear term in Eq. (6.6) is approximated by Taylor series
1 1 1 .n n n n n n n
x x x xUU U U U U U U (6.7)
At the nth time step, we denote mU and its space derivatives at the knots mx by the
following expressions:
1 2 1 1 2
2 2 1 1 2
3 2 1 1 22
4 2 1 1 24
26 66 26 ,
510 10 ,
202 6 2 ,
1204 6 4 .
n n n n nm m m m m m
n n n nm m m m m
n n n n nm m m m m m
n n n n nm m m m m m
H
Hh
Hh
Hh
(6.8)
Using the knots , 0,1,.., ,mx m N as collocation points, we obtain the following
recurrence relation at point mx form Eqs. (6.5)-(6.8):
4
1 3 4
1 1 1 1 11 2 2 1 3 4 1 5 2 2 m m m
n n n n nm m m m m m m m m m h H t H Ha a a a a
(6.9)
where
3 21 0 1
3 22 0 1
23 0
3 24 0 1
3 25 0 1
5 20 120 ,
26 50 40 480 ,
66 120 720 ,
26 50 40 480 ,
5 20 120 ,
m m m
m m m
m m
m m m
m m m
a H h H t h t t
a H h H t h t t
a H h t t
a H h H t h t t
a H h H t h t t
40 2 2 and 0,1,..., .m mH h tH m N
The Eq. (6.9) relates parameters at adjacent time levels and gives 1N equations in
5N unknowns , 2, 1,..., 2i i N . In order to get a unique solution the parameters
2 1 1 2, , ,N N are eliminated from Eq. (6.9). Using Eqs. (6.2) and (6.5), the
following values of the parameters are obtained:
Chapter 6 Kuramoto-Sivashinsky equation
126
12 0 1 2
11 0 1 2
21 2 1
22 2 1
5165 65 9,
4 2 4 833 9 1
,8 4 8 161 9 33
,8 4 8 16
59 65 165.
4 2 4 8
N N N N
N N N N
g
g
g
g
(6.10)
Elimination of the above parameters from Eq. (6.9) yields a five-diagonal linear system
of (N+1) equations in (N+1) unknown parameters , 0,1,..., .m m N After calculating
the initial parameters 0m , this linear system can be solved by the penta-diagonal solver
successively for , 1, 2,..., .nm n The approximate solution ,U x t will then be
obtained from Eq. (6.5).
The initial parameters 0m are determined with the help of derivatives at boundaries
and initial conditions:
0 0 0 0 0
2 1 1 2
0 0 0 00 2 1 1 2 0
0 0 0 0 00 2 1 0 1 2 02
0 0 0 02 1 1 2
0 0 02 12
,0 ,0 26 66 26
5,0 10 10 ,
20,0 2 6 2 ,
,
5,0 10 10 ,
20,0 2 6
m m m m m m m m
N N N N N N
N N N N
U x u x f x
U x uh
U x uh
U x uh
U xh
0 01 22 ,
0,1,..., .
N N Nu
m N
(6.11)
Eq. (6.11) consists of a five-diagonal linear system of 5N equations in
5N unknowns 0 , 2, 1,..., 2,m m N which can also be solved by penta-diagonal
solver.
6.2.2 Quintic B-Spline Collocation Method 1I
We use splitting technique in space for the numerical solution of the KS equation.
Splitting methods in time and in space are used to get economy and simplicity in the
technique. For that purpose we introduce a new variable defined by xxu v and split
Chapter 6 Kuramoto-Sivashinsky equation
127
Eq. (6.1) as follows:
0,t x xxu uu v v (6.12)
0,xxu v (6.13)
subject to the following boundary conditions:
1 2, , , , , 0, , 0, 0 ,u a t g t u b t g t v a t v b t t (6.14)
and initial conditions
,0 , ,0 , .u x f x v x f x a x b (6.15)
The approximations ,U V to the exact solutions u and v are given by
2 2
2 2
, , , ,N N
m m m mm m
U x t t B x V x t t B x
(6.16)
where ,m mt t are time dependent parameters to be determined and mB x are
fifth-degree B-spline functions.
The values of U, V and their space derivatives at the knots mx are obtained from
Eqs. (5.11) and (6.16), in the following manner
2 1 1 2
2 1 1 2
2 1 1 2
2 1 1 2
2 1 1 22
26 66 26 ,
510 10 ,
26 66 26 ,
510 10 ,
202 6 2 .
m m m m m m
m m m m m
m m m m m m
m m m m m
m m m m m m
U
Uh
V
Vh
Vh
(6.17)
We apply the implicit midpoint approximation rule for the unknown parameters in Eqs.
(6.12)-(6.13) and discretize the time derivative in Eq. (6.12) by finite-difference forward
formula at two adjacent time levels and 1n n to get:
1 11 1
1 1
0,2 2 2
0.2 2
n n n nn n n nx x xx xx
n n n nxx xx
UU UU V VU U V V
t
U U V V
(6.18)
At the nth time step, the approximate solutions ,m mU V and their space derivatives at the
knots mx are denoted by the following expressions:
Chapter 6 Kuramoto-Sivashinsky equation
128
1 2 1 1 2
2 2 1 1 2
3 2 1 1 22
1 2 1 1 2
2 2 1 1 2
3 22
26 66 26 ,
510 10 ,
202 6 2 ,
26 66 26 ,
510 10 ,
202
n n n n nm m m m m m
n n n nm m m m m
n n n n nm m m m m m
n n n n nm m m m m m
n n n nm m m m m
nm m
H
Hh
Hh
K
Kh
Kh
1 1 26 2 .n n n nm m m m
(6.19)
Let , 0,1,.., ,mx m N be the collocation points in the interval , .a b Using Eqs. (6.7),
(6.17) and (6.19) in Eq. (6.18), the following system of coupled equations is obtained:
1 1 1 1
1 2 2 1 3 4 1
1 1 1 1 2
5 2 1 2 2 1 3 1 1 3
1 1
2 1 1 2
1 1 1 1 1
2 1 1 2
2 1 1
2 1
2
20 40 120 40 20
26 66
n n n n
m m m m m m m m
n n n n
m m m m m m m m m m m
n n
m m m m
n n n n n
m m m m m
n n
m m
b b b b
b c c c h H t K K
c c
h
2
1 31 1 1
1 2
1 0 1 2 0 1 3 0
4 0 1 5 0 1
2 2 2
1 2 3
20
26
where
5 , 26 50 , 66 ,
26 50 , 5 ,
20 , 26 40 , 66 120 ,
with 2
m mn n n
m m m
m m m m m m m m
m m m m m m
m m m
m m
h K H
b H hH t b H hH t b H
b H hH t b H hH t
c th t c th t c th t
H h tH
2 and 0,1,..., .m N
(6.20)
Eq. (6.20) contains 2 2N equations in 2 10N unknowns. To obtain a unique solution
to this system, we eliminate the parameters 2 1 1 2 2 1 1 2, , , , , , ,N N N N from
it. The values of 2 1 1 2, , ,N N are already given in Eq. (6.10) whereas the
remaining ones are calculated and listed as follows:
2 0 1 2
1 0 1 2
15 35 ,2 23 3 1 ,4 2 4
(6.21)
1 2 1
2 2 1
1 3 3 ,4 2 4
3 1 55 .2 2
NN N N
NN N N
Elimination of above parameters from Eq. (6.20) gives a linear system of (2N+2)
Chapter 6 Kuramoto-Sivashinsky equation
129
equations in (2N+2) unknown parameters , , 0,1,..., ,i i i N which can be solved
successively to find , , 1, 2,...,n ni i n provided we find the initial parameters 0 0,m m
first. Consequently the approximate solution ,U x t will be obtained from Eq. (6.17).
We find initial parameters under the following constraints:
0 0 0 0
0 0 0 0
,0 ,0 , ,0 ,0 ,
,0 , ,0 ,
,0 , ,0
,0 , ,0 ,
,0 , ,0 ,
0,1,..., .
m m m m m m
N N N N
N N N N
U x u x f x V x u x f x
U x f x V x f x
U x f x V x f x
U x f x V x f x
U x f x V x f x
m N
(6.22)
6.2.3 Quartic B-Spline Collocation Method
Using Eq. (4.4) the approximate solutions are given by
1 1
2 2
, , , ,N N
m m m mm m
U x t t B x V x t t B x
(6.23)
where ,m mt t are time dependent parameters to be determined and mB x are
fourth-degree B-spline functions. Once again splitting method is used to obtain the
results given in the Eqs. (6.12)-(6.15) and (6.18). Eqs. (4.4) and (6.23) provide the
following values of ,U V and their space derivatives at the collocation points:
2 1 1
2 1 1
2 1 12
2 1 1
2 1 1
2 1 12
11 11 ,
43 3 ,
12,
11 11 ,
43 3 ,
12.
m m m m m
m m m m m
m m m m m
m m m m m
m m m m m
m m m m m
U
Uh
Uh
V
Vh
Vh
(6.24)
For the n-th time level the variables ,m mU V and their space derivatives at the knots mx
are denoted by:
Chapter 6 Kuramoto-Sivashinsky equation
130
1 2 1 1 2 2 1 1
3 2 1 12
1 2 1 1
2 2 1 1
3 2 1 12
411 11 , 3 3 ,
12,
11 11 ,
43 3 ,
12.
n n n n n n n nm m m m m m m m m m
n n n nm m m m m
n n n nm m m m m
n n n nm m m m m
n n n nm m m m m
L Lh
Lh
M
Mh
Mh
(6.25)
Substitution of Eqs. (6.24)-(6.25) into Eq. (6.18) gives rise to the following system of
coupled equations:
1 1 1 11 2 2 1 3 4 1 2
1 1 31 1 1 11 2 2 1 2 1 1
1 1 1 12 1 1 2
1 32 1 1 1 12 1 1
2
12 12 12 12
11 11
n n n nm m m m m m m m
m m mn n n nm m m m m m m m
n n n nm m m m
m mn n n nm m m m
d d d dh L t M M
e e e e
h M Lh
(6.26)
where
1 0 1 2 0 1 3 0 1
2 24 0 1 1 2
20 2
4 , 11 12 , 11 12 ,
4 , 12 , 11 12 ,
with 2 and 0,1,..., .
m m m m m m m m m
m m m m m
m m
d L hL t d L hL t d L hL t
d L hL t e th t e th t
L h tL m N
Eq. (6.26) consists of 2 2N equations in 2 8N unknowns. Imposition of boundary
conditions (6.14) gives the following relations:
12 0 1
11 0 1
1 2 1 2
2 0 1
1 0 1
1 2 1
333 7,
4 4 87 1
,4 4 8
11 11 ,
3 1,
2 21 1
,2 2
11 11 .
N N N N
N N N N
g
g
g
(6.27)
Elimination of above parameters from Eq. (6.26) with the help of Eq. (6.27) yields
2 2 2 2N N linear system of unknown parameters , , 0,1,..., ,i i i N which
can be solved successively to find , , 1, 2,...,.n ni i n In order to initiate the solution, we
Chapter 6 Kuramoto-Sivashinsky equation
131
first find the initial parameters 0 0,m m which are obtained with the help of Eq. (6.22).
Consequently the approximate solution ,U x t can be obtained from Eq. (6.24).
Following the procedure adopted in the previous chapters the stability of the
methods can be established.
6.3 Numerical validation
To demonstrate the effectiveness of the numerical methods we consider some test
problems. The accuracy of the schemes is measured in terms of the following discrete
error norms L and 2L .
Problem 6.1 For the sake of comparison with an earlier work (Xu and Shu [160]), we
Consider the following KS equation
0t x xx xxxxu uu u u (6.28)
subject to the boundary conditions
3
0 0
3
0 0
30,
30,
15 119 tanh 30 11tanh 30 ,
19 19
15 119 tanh 30 11tanh 30 ,
19 19
u t
u t
c k ct x k ct x
c k ct x k ct x
(6.29)
and the initial condition
30 0
15 11,0 9 tanh 11tanh .
19 19u x c k x x k x x
(6.30)
The theoretical solution for this problem is given by
30 0
15 11, 9 tanh 11tanh .
19 19u x t c k x ct x k x ct x
(6.31)
We take 00.1, 1/ 2 11/19, 10.c k x Table 6.1 contains the 2L and L error
norms at time 1.t In Table 6.2 we re-produce the results reported in local
discontinuous Galerkin (LDG) method (Xu and Shu [160]) for the sake of comparison
with pk elements, (k=0,1,2). It is clear from the comparison of the Tables 6.1 and 6.2
that our methods 6.2.2 and 6.2.3 give less maximum absolute errors than (Xu and Shu
[160]) , however the L2 errors are comparable. Method 6.2.1 can be compared for
0,1k . Fig. 6.1 shows the regular shock profile for the above problem. From the
figure and computed results, it is observed that the moving shock is resolved very well.
Chapter 6 Kuramoto-Sivashinsky equation
132
Problem 6.2 Consider the following KS equation;
0t x xx xxxxu uu u u (6.32)
with the exact solution
30 0
15, 3 tanh tanh
19 19u x t c k x ct x k x ct x (6.33)
and the boundary conditions;
1 250, , 50, , 0,u t f t u t f t t (6.34)
where if t are obtained from Eq. (6.35). The initial condition is given by
30 0
15 11,0 9 tanh 11tanh .
19 19u x c k x x k x x
(6.35)
We take 00.2, 1/ 2 19, 10c k x and compare our results with (Xu and Shu [160]).
Table 6.3 contains the 2L and L error norms at time 1.t Table 6.4 contains the results
for LDG method (Xu and Shu [160]) with pk elements, (k=0,1,2) . Comparison of
tables 6.3 and 6.4 provides similar conclusion as discussed in problem 6.1. Fig. 6.2
shows the regular shock profile for the above problem. From the figure and numerical
computation it is again observed that the moving shock is resolved very well.
Problem 6.3 We finally consider the KS equation;
0t x xx xxxxu uu u u (6.36)
with the Gaussian initial condition
2,0 exp ,u x x (6.37)
and the boundary conditions;
30, 30, 0.u t u t (6.38)
Here the simulations are done upto 30.t Fig. 6.3 shows the solutions at 0,5,20,30t
and illustrates the chaotic solution for the above problem. From the figure it is observed
that the numerical results are convergent for the chaotic nature.
Chapter 6 Kuramoto-Sivashinsky equation
133
Table 6.1 Accuracy test for problem 6.1 Methods N L∞ error L2 error
Method 6.2.1 40 1.56×10-1 3.58×10-1
80 7.03×10-2 1.30×10-1
160 1.86×10-2 3.37×10-2
320 4.69×10-3 8.48×10-3
Method 6.2.2 40 1.44×10-1 3.01×10-1
80 3.12×10-3 5.59×10-3
160 1.44×10-4 2.69×10-4
320 1.95×10-5 3.32×10-5
Method 6.2.3 80 7.94×10-3 1.38×10-2
160 4.37×10-4 8.20×10-4
320 4.29×10-5 6.81×10-5
640 2.44×10-5 3.72×10-5
00.1, 11/19, 10 at 11/ 2c k x t
Table 6.2 Accuracy test for problem 6.1 (Xu and Shu [160]) Methods N L∞ error L2 error
P0 40 1.37 2.38×10-1
80 8.81×10-1 1.58×10-1
160 5.21×10-1 9.41×10-2
320 2.91×10-1 5.24×10-2
P1 40 6.64×10-1 6.08×10-2
80 1.82×10-1 1.49×10-2
160 4.64×10-2 3.78×10-3
320 1.19×10-2 9.57×10-4
P2 40 1.49×10-1 9.15×10-3
80 1.73×10-2 1.06×10-3
160 2.43×10-3 1.32×10-4
320 3.05×10-4 1.65×10-5
00.1, 11/19, 10 at 11/ 2c k x t
Chapter 6 Kuramoto-Sivashinsky equation
134
Table 6.3 Accuracy test for problem 6.2
Methods N L∞ error L2 error
Method 6.2.1 40 1.42×10-4 3.44×10-4
80 3.06×10-5 7.41×10-5
160 7.36×10-6 1.78×10-5
320 1.81×10-6 4.39×10-5
Method 6.2.2 40 3.22×10-5 7.46×10-5
80 1.45×10-6 3.41×10-6
160 6.91×10-8 1.76×10-7
320 7.00×10-8 1.87×10-7
Method 6.2.3 40 6.72×10-5 1.60×10-4
80 4.06×10-6 9.77×10-6
160 3.18×10-7 7.19×10-7
320 8.95×10-8 2.16×10-7
00.2, , 10 at 11/ 2 19c k x t
Table 6.4 Accuracy test for problem 6.2 (Xu and Shu [160]) Methods N L∞ error L2 error
P0 40 7.56×10-2 1.33×10-2
80 3.87×10-2 6.69×10-3
160 1.95×10-2 3.35×10-3
320 9.80×10-3 1.68×10-3
P1 40 1.04×10-2 1.29×10-3
80 2.59×10-3 3.26×10-4
160 6.54×10-4 8.17×10-5
320 1.64×10-4 2.05×10-5
P2 40 7.93×10-4 6.48×10-5
80 1.06×10-4 8.04×10-6
160 1.34×10-5 1.00×10-6
320 1.67×10-6 1.25×10-7
01/ 2 190.2, , 10 at 1c k x t
Chapter 6 Kuramoto-Sivashinsky equation
135
-30 -20 -10 0 10 20 303
4
5
6
7
x
u
t=0
-30 -20 -10 0 10 20 303
4
5
6
7
x
u
t=2
-30 -20 -10 0 10 20 303
4
5
6
7
x
u
t=5
-30 -20 -10 0 10 20 303
4
5
6
7
x
u
t=6
Fig 6.1: The shock profile wave propagation of the KS Eq.(6.28) with initial condition
(6.30), 05, 1/ 2 11/19 12, 160 in [ 30,30],c k x N
-50 -40 -30 -20 -10 0 10 20 30 40 504.6
4.7
4.8
4.9
5
5.1
5.2
5.3
5.4
x
U
t=0
-50 -40 -30 -20 -10 0 10 20 30 40 504.6
4.7
4.8
4.9
5
5.1
5.2
5.3
5.4
x
U
t=12
-50 -40 -30 -20 -10 0 10 20 30 40 504.6
4.7
4.8
4.9
5
5.1
5.2
5.3
5.4
x
U
t=5
-50 -40 -30 -20 -10 0 10 20 30 40 504.6
4.7
4.8
4.9
5
5.1
5.2
5.3
5.4
x
U
t=10
Fig 6.2: The shock profile wave propagation of the KS Eq. (6.32) with initial condition
(6.35), 05, , 25, 200 in 1/ 2 19 [ 50,50]c k x N
Chapter 6 Kuramoto-Sivashinsky equation
136
-30 -20 -10 0 10 20 30-0.5
0
0.5
1
1.5
-30 -20 -10 0 10 20 30-0.5
0
0.5
1
1.5
-30 -20 -10 0 10 20 30-4
-2
0
2
4
-30 -20 -10 0 10 20 30-4
-2
0
2
4
t=0 t=5
t=20 t=30
-30 -20 -10 0 10 20 30
0
5
10
15
20
25
30
-4
-2
0
2
4
t
x
U
Fig 6.3: The chaotic solution of the KS Eq. (6.36) with Gaussian initial condition (6.37) and boundary conditions
(6.38), 320N in [ 30, 30]
Chapter 6 Kuramoto-Sivashinsky equation
137
6.4 Conclusion
Quartic and quintic B-spline collocation methods are used to develop a class of
numerical methods for solving KS equation. Using splitting technique, the equation is
reduced to a problem of second order in space. These methods are computationally
efficient and the algorithm can easily be implemented on a computer. Theses methods
are compared with local discontinuous Galarkin method given in (Xu and Shu [160])
and seem to be effective and accurate.
***************************************************
Future Work
138
Future work
The research presented in the thesis is focused on the use of non-polynomial splins and
B-spline functions to obtain numerical solution of BVPs in ODEs and collocation
method for initial-BVPs in PDEs. This investigation has spawned a number of open
research problems. We will figure out some of them, and further investigation in
specified directions will certainly lead to the improvement and generalization of the
exiting algorithms designed for the numerical solution of initial and BVPs in ordinary
as well as partial differential equations.
(i). Non-polynomial spline functions based algorithms are relatively new methods
and are providing alternative ways of tackling the unilateral problems coupled
with the theory of variational inequalities. This approach can be extended to
nonlinear obstacle problem. Detail analysis of the methods both analytically and
numerically, will constitute an immediate and interesting subject of future
study.
(ii). The use of non-polynomial spline functions for the solution of BVPs can be
extended to eight and higher-order general as well as special linear and
nonlinear BVPs. Such types of problems have variety of applications in science
and engineering.
(iii). Non-polynomial spline functions can be used for the solution of singular BVPs
of third and fourth-orders.
(iv). The work done so far on the application of polynomial spline functions in
developing algorithms for the initial-value problems can be replaced by their
counter parts non-polynomial spline functions to improve the accuracy and
generalize the algorithms.
(v). The Literature on the use of non-polynomial splines for the numerical solution
PDEs is very limited. Hence this is another area open for future investigations.
(vi). High degree B-spline collocation methods can be used for the numerical
solution of evolution problems.
(vii). Numerical methods for the solution of generalized regularized long wave and
equal width wave equations can be investigated using quartic and quintic B-
splines.
Future Work
139
(viii). Numerical solution of the modified equations studied in this thesis can be
extended through the use of higher degree B-splines.
(ix). Numerical algorithms for solution of two dimensional PDEs can be developed
using collocation method based on B-splines.
(x). B-splines of various degrees can be used to find numerical solution of
Sivashinsky and Kawahara equation.
***************************************************
References
140
References
[1] H. I. Abdel-Gawad, H. A. Abdusalam (2001): Approximate solutions of the Kuramoto-Sivashinsky equation for boundary value problems and chaos. Chaos, Solitons and Fractals 12, 2039-2050.
[2] B. Abdel-Hamid (1999): An exact solution to the Kuramoto-Sivashinsky equation. Phys. Lett. A 263, 338-340.
[3] K. O. Abdulloev, H. Bogolubsky, V. G. Makhankov (1967): One more example of inelastic soliton interaction. Phys. Lett. A 56, 427-428.
[4] R. P. Agarwal (1986): Boundary-value problems for higher-order differential equations. World Scientific. Singapore.
[5] R. P. Agarwal, D. O'Regan (2009): Ordinary and partial differential equations with special functions, Fourier series and boundary value problems. Springer. New York.
[6] J. H. Ahlberg, E. N. Nilson (1963): Convergence properties of the spline fit. SIAM Journal 11, 95-104.
[7] J. H. Ahlberg, E. N. Nilson, J. L. Walsh (1967): The theory of splines and their applications. Acadamic press Inc. New York.
[8] G. Akram, S. S. Siddiqi (2006): Solution of linear sixth-order boundary-value problems using non-polynomial spline technique. Appl. Math. Comput. 1, 708-720.
[9] E. A. Al-Said (2001): The use of cubic splines in the numerical solution of a system of second-order boundary-value problems. Comput. Math. Applics 42, 861-869.
[10] E. A. Al-Said (2005): Quartic splines solution for a system of second-order boundary-value problems. Appl. Math. Comput. 166, 254-264.
[11] E. L. Albasiny, W. D. Hoskins (1969): Cubic spline solution of two point boundary-value problems. Comuter J., 151-153.
[12] A. Ali (2009): Modified Regularized Long Wave Equation, pp. 160, GIK Institute of Engineering Sciences and Technology, Topi, Swabi.
[13] A. H. A. Ali, A. A. Soliman, K. R. Raslan (2007): Solitons solution for nonlinear partial differntial equatins by cosine-function method. Phys. Lett. A 368, 299-304.
[14] T. Aziz, A. Khan (2002): Quintic spline approach to the solution of a singularity-perturbed boundary value problems. J. Optim. Theory 112, 517-527.
[15] T. Aziz, A. Khan (2002): A spline method for second-order singularly perturbed boundary-value problems. J. Comput. Appl. Math 147, 445-452.
References
141
[16] T. Aziz, A. Khan, I. Khan (2008): Quintic splines method for second-order boundary value problems. J. Comput. Math. 85(5), 735-743.
[17] D. Baldwin, U. Goktas, W. Hereman, L. Hong, R. S. Martino, J. S. Miller (2004): Symbolic computation of exact solutions expressible in hyperbolic and elliptic functions for nonlinear PDEs. J. Symbolic Comput. 37, 669-705.
[18] P. Baldwin (1987): Asymptotic estimates of the eigenvalues of a sixth-order boundary-value problem obtained by using global-phase intergral methods. Philos. Trans. R. Soc., Lond., Ser. A 322, 281-305.
[19] P. Baldwin (1987): A localized instability in a Bernard layer. Appl. Anal. 24, 117-156.
[20] T. B. Benjamin, J. L. Bona, J. J. Mahony (1972): Model equations for long waves in nonlinear dispersive systems. Philos. Trans. R. Soc., Lond., Ser. A 272, 47-78.
[21] W. G. Bickely (1968): Piecewise cubic interpolation and two-point boundary-value problems. Computer J. 11, 202-208.
[22] I. P. Boglaev (1981): A variational difference scheme for a boundary-value problem with a small parameter in the highest derivative. U.S.S.R. Comput. Math. Phys. 21(4), 71-81.
[23] J. L. Bona, P. J. Bryant (1973): A mathematical model for long waves generated by wave makers in nonlinear dispersive systems. Proc. Cambridge Philos. Soc. 73, 391-405.
[24] A. Boutayeb, E. H. Twizell (1992): Numerical methods for the solution of special sixth-order boundary-value problems. Int. J. Comput. Math. 45, 207-223.
[25] G. Brikhoff, H. Garabedian (1960): Smoth surface interpolation. J. Math. and Phys. 39, 258-268.
[26] H. N. Caglar, S. H. Caglar, E. H. Twizell (1998): The numerical solution of third-order boundary-value problems with fourth-degree B-spline functions. J. Comput. Math. 71, 373-381.
[27] S. Chandrasekhar (1981): Hydrodynamic and hydromagnetic stability. Dover Books. New York.
[28] M. M. Chawala (1977): A sixth-order tri-diagonal finite-difference method for nonlinear two-point boundary value problems. BIT 17, 128-137.
[29] M. M. Chawala, C. P. Katti (1979): Finite difference methods for two-point boundary-value problems involving higher-order differential equations. BIT 19, 27-33.
[30] Q. Chen, G. Wang (2003): A class of Bézier-like curves. Comput. Aided Geom. Design 20, 29-39.
References
142
[31] R. Conte (2003): Exact solutions of nonlinear partial differential equations by singularity analysis, Lecture notes in Physics, Springer. 1-83.
[32] J. Crank, R. S. Gupta (1972): A method for solving moving boundary-value problems in heat flows using cubic splines or polynomials. J. Inst. Maths. Applics. 10, 296-304.
[33] I. Dag, B. Saka, D. Irk (2004): Application of cubic B-splines for numerical solution of the RLW equation. Appl. Math. Comput. 159, 373-389.
[34] I. Dag, B. Saka, D. Irk (2006): Galerkin method for the numeical solution of the RLW equation using quintic B-splines. J. Comput. Appl. Math. 190, 532-547.
[35] C. deBoor (2001): A practical guide to splines. Springer-Verlag. New York.
[36] E. P. Doolan, J. J. H. Miller, W. H. A. Schilders (1980): Uniform numerical methods for problems with initial and boundary layers. Boole press. Dublin.
[37] J. C. Eilbeck, G. R. McGuire (1975): Numerical study of RLW equation 1: numerical methods. J. Compt. Phys 19, 43-57.
[38] A. Esen (2006): A lumped Galerkin method for the numerical solution of the modified equal width wave equation using quadratic B-splines. Int. J. Comput. Math. 83(5-6), 449-459.
[39] A. Esen, S. Kutluay (2008): Solitary wave solutions of the modified equal width wave equation. Comm. Nonlinear Science and Numer. Simul. 13, 1538-1546.
[40] D. J. Evans, K. R. Raslan (2005): Solitary waves for the generalized equal width (GEW) equation. Int. J. Comput. Math. 82(4), 445-455.
[41] M. E. Gamel, J. R. Cannon, J. Latour, A. I. Zayed (2003): Sinc-Galerkin method for solving linear sixth-order boundary-value problems. Math. Comput. 73, 1325-1343.
[42] L. R. T. Gardner, G. Gardner, T. Geyikli (1994): The boundary forced MKdV equation. J. Comput. Phys. 11, 5-12.
[43] L. R. T. Gardner, G. A. Gardner, F. A. Ayub, N. K. Amein (1997): Approximations of solitary waves of the MRLW equation by B-spline, finite element. Arab. J. Sci. Eng. 22, 183-193.
[44] G. A. Glatzmaier (1985): Numerical simulations of stellar convective dynamics III: at the base of the convection zone. Geophys. Astrophys. Fluid Dyn. 31, 137-150.
[45] B. Y. Gou, W. M. Cao (1988): The fourier pseudospectral method with a restrain operator for the RLW equation. J. Compt. Phys 74, 110-126.
[46] S. Haq, S. Islam, A. Ali (2008): A Numerical Meshfree Technique for the Solution of the MEW Equation. Comp. Modell. Engin. Sciences 38(1), 1-27.
References
143
[47] P. Henrici (1962): Discrete variable methods in ordinary differential equations. John Wiley. New York.
[48] P. Hopper, R. Grimshaw (1985): Nonlinear instability at the interface between two vacous fluids. Phys. Fluids 28, 37-45.
[49] W. D. Hoskins, P. J. Ponzo (1972): Some properties of a class of band matrices. Math. Comput. 26(118), 390-400.
[50] J. M. Hyman, B. Nicolaenko (1986): The Kuramoto-Sivashinsky equations, a bridge between PDEs and dynamical systems. Physica D 18, 113-126.
[51] J. M. Hyman, B. Nicolaenko, S. Zaleski (1986): Order and complexity in the Kuramoto-Sivashinsky model of turbulent interfaces. Physica D 23, 265-292.
[52] E. Isaacson, H. B. Keller (1966): Analysis of numerical methods. John Wiley and Sons. New York.
[53] S. Islam, S. Haq, A. Ali (2009): A meshfree method for the numerical solution of RLW equation. J. Comput. Appl. Math 223(2), 997-1012.
[54] S. Islam, M. A. Khan, S. I. A. Tirmizi, E. H. Twizell (2005): A non-polynomial spline approach to the solution of a system of third-order boundary-value problems using non-polynomial splines. Appl. Math. Comput. 168(1), 152-163.
[55] S. Islam, M. I. Noor, S. I. A. Tirmizi, M. A. Khan (2006): Quadratic non-polynomial spline approach to the solution of a system of second-order boundary value problems. Appl. Math. Comput. 179, 153-160.
[56] S. Islam, S. I. A. Tirmizi (2006): Non-polynomial spline approach to the solution of a system of second-order boundary-value problems. Appl. Math. Comput. 173, 1208-1218.
[57] S. Islam, S. I. A. Tirmizi (2006): A smooth approximation for the solution of special non-linear third-order boundary-value problems based on non-polynomial splines. J. Comput. Math. 83(4), 397-407.
[58] S. Islam, S. I. A. Tirmizi, S. Ashraf (2006): A class of methods based on non-polynomial spline functions for the solution of a special fourth-order boundary-value problems. Appl. Math. Comput. 174, 1169-1180.
[59] S. Islam, S. I. A. Tirmizi, F. Haq (2007): Quartic non-polynomial splines approach to the solution of a system of second-order boundary-value problems. Int J. High Performance Computing Applications 21(1), 42-49.
[60] S. Islam, S. I. A. Tirmizi, F. Haq, M. A. Khan (2008): Non-Polynomial splines approach to the solution of sixth-order boundary-value problems. Appl. Math. Comput. 195, 270-284.
[61] S. Islam, S. I. A. Tirmizi, F. Haq, S. K. Taseer (2008): Family of numerical methods based on non-polynomial splines for solution of contact problems. Comm. Nonlinear Science and Numer. Simul. 13, 1448-1460.
References
144
[62] S. Islam, S. I. A. Tirmizi, M. A. Khan (2006): A numerical method based on polynomial sextic spline functions for the solution of special fifth-order boundary-value problems. Appl. Math. Comput. 181, 356-361.
[63] M. K. Jain (1984): Numerical solution of differential equations. Wiley Eastern. New Delhi.
[64] M. K. Jain, T. Aziz (1981): Spline function approximation for differential equations. Methods in App. Mech. and Engg. 26, 129-143.
[65] M. K. Jain, T. Aziz (1983): Cubic spline solution of two-point boundary-value problem with significant first derivatives. Comp. Methods in App. Mech. and Engg. 39, 83-91.
[66] M. S. Jolly, I. G. Kevrekides, E. S. Titi (1990): Approximate inertial manifolds for the Kuramoto-Sivashinsky equation: Analysis and computations. Physica D 44, 38-60.
[67] D. J. Jones (1973): Solution of Troesch's and other two-point boundary value problems by shooting techniques. J. Compt. Phys. 12, 429-434.
[68] L. Junfeng (2007): He's variational iteration method for the modified equal width equation, In press. Chaos, Solitons and Fractals.
[69] M. K. Kadalbajoo, R. K. Bawa (1996): Variable-mesh difference scheme for singularly-perturbed boundary-value problems using splines. J. Optim. Theory 9, 405-416.
[70] M. K. Kadalbajoo, K. C. Patidar (2002): A survey of numerical techniques for solving singularly-perturbed ordinary differential equations. Appl. Math. Comput. 130, 457-510.
[71] D. Kaya (2004): A numerical simulation of solitary wave solutions of the generalized regularized long wave equation. Appl. Math.Comput. 149, 833-841.
[72] D. Kaya, S. M. El-Sayed (2003): An application of the decomposition method for the generalized KDV and RLW equations. Chaos, Solitons and Fractals 17, 869-877.
[73] J. B. Kevrekidis, B. Nicolaenko, J. C. Scovel (1990): Back in the saddle again: A computer assisted study of Kuramoto-Sivashinsky equaton. J. Appl. Math. 50, 760-790.
[74] A. K. Khalifa, K. R. Raslan, H. M. Alzubaidi (2007): A finite difference scheme for the MRLW and solitary wave interactions. Appl. Math. Comput. 189, 346-354.
[75] A. K. Khalifa, K. R. Raslan, H. M. Alzubaidi (2008).A collocation method with cubic B- splines for solving the MRLW equation. J. Comput. Appl. Math. 212, 406-418.
References
145
[76] A. K. Khalifa, K. R. Raslan, H. M. Alzubaidi (2008): Numerical study using ADM for the modified regularized long wave equation. Appl. Math. Modell. 32, 2962-2972.
[77] A. Khan, I. Khan, T. Aziz (2006): Sextic spline solution of a singularly perturbed boundary-value problems. Appl. Math. Comput. 181, 432-439.
[78] A. H. Khater, R. S. Temsah (2008): Numerical solutions of the Generalized Kuramoto-Sivashinsky equation by Chebyshev spectral collocation methods. Computers Math. Applic. 56(6), 1465-1472.
[79] M. Kubicek, V. Hlavacek (1975): Solution of Troesh's two-point boundary-value problem by shooting techniques. J. Compt. Phys. 17, 95-101.
[80] M. Kumar, H. K. Mishra, P. Singh (2009): Boundary value approch for a class of linear singularly perturbed two-point boundary-value problems. Advances Engg. Soft. 40, 298-304.
[81] M. Kumar, N. Singh (2009): A collection of computational techniques for solving singular boundary-value problems. Advances Engg. Soft. 40, 288-297.
[82] M. Kumar, P. K. Srivastava (2009): Computational Techniques for Solving Differential Equations by Cubic, Quintic, and Sextic Spline. Comp. Methods in Engg. Sci. and Mech. 10, 108-115.
[83] Y. Kuramoto, T. Tsuzuki (1976): Persistent propagation of concentration waves in dissipative media far from thermal equilibrium. Prog. Theor. Phys. 55, 356.
[84] M. A. Lopez-Marcos (1994): Numerical analysis of pseudospectral methods for the Kuramoto-Sivashinsky equation. IMA J. Numer. Anal. 14, 233-242.
[85] F. R. Loscalzo, T. D. Talbot (1967): Spline function approximation for solution of ordinary differential equations. SIAM J. Numer. Anal. 4, 433-445.
[86] D. H. Maclaren (1958): Formula for fitting a spline curve through a set of points. Boeing Appl. Math. Report No. 2.
[87] A. V. Manickam, K. M. Moudgalya, A. K. Pani (1998): Second order splitting combined with orthogonal cubic spline collocation method for the kuramoto-Sivashinsky equation. Compuers math. Applic. 35, 5-25.
[88] J. J. H. Miller, E. O'Riordan, G. I. Shiskin (1996): Fitted numerical methods for singular perturbation problems. World Scientific. Singapore.
[89] A. R. Mitchell, D. F. Griffiths (1980): The finite difference equations in partial differential equations. John Wiley & Sons. New York.
[90] R. K. Mohanty, N. Jha (2005): A class of variable mesh spline in compression methods for singularly perturbed two-point singular boundary-value problems. Appl. Math. Comput. 168, 704-716.
[91] P. J. Morrison, J. D. Meiss, J. R. Carey (1984): Scattering of RLW solitary waves. Physica D 11, 324-336.
References
146
[92] K. W. Morton (1996): Numerical solution of convection-diffusion problems. Chapman & Hall. London.
[93] S. Natesan, N. Ramanujam (1999): Improvement of numerical solution of self-adjoint singular perturbation problems by incorporation of asymptotic approximations. Appl. Math. Comput. 98, 119-137.
[94] Niijima (1980): On a three point difference scheme for a singular perturbation problem without a first derivative term I. Mem. Numer. Math. 7, 1-10.
[95] M. A. Noor, A. K. Khalifa (1987): Cubic splines collocation methods for unilateral problems. Int. J. Eng. Sci. 25, 1527-1530.
[96] M. A. Noor, S. I. A. Tirmizi (1988): Finite difference techniques for solving obstacle problems. Appl. Math. Lett. 1, 267-271.
[97] E. O'Riordan, M. Stynes (1986): A uniformly accurate finite element method for a singularly perturbed one-dimensional reaction diffusion problem. Math. Comput. 47, 555-570.
[98] P. J. Olver (1979): Euler operators and conservation laws of the BBM equation. Math. Proc. Comb. Phil. Soc. 85, 143-159.
[99] D. S. Papageorgiou, Y. S. Smyrlis (1991): The route to chaos for the Kuramoto-Sivashinsky equation. Theor. Comput. Fluid Dynam. 3, 15-42.
[100] E. J. Parkes, B. R. Duffy (1996): An automated tanh-function method for finding solitary wave solutions to non-linear evolution equations. Comput. Phys. Commun 98, 288.
[101] D. H. Pergrine (1966): Calculations of the Development of an Undular Bore. J. Fluid Mech. 25(2), 321-330.
[102] P. M. Prenter (1975): Splines and Variational Methods. Wiley. New York.
[103] M. A. Ramadan, T. S. El-Danaf (2005): Numerical treatment for the modified burgers' equation. Math. Comput. Simul. 70, 90-98.
[104] M. A. Ramadan, T. S. El-Danaf, F. E. I. A. Alaal (2007): Application of the Non-polynomial spline approach to the solution of the Burgers' equation. The Open Applied Math. J. 1, 15-20.
[105] M. A. Ramadan, I. F. Lashien, W. K. Zahra (2009): Quintic non-polynomial spline solutions for fourth order two-point boundary value problem. Comm. Nonlinear Science and Numer. Simul. 14, 1105-1114.
[106] M. A. Ramadan, M. A. I. F. Lashien, W. K. Zahra (2008): A class of methods based on a septic non-polynomial spline function for the solution of sixth-order two-point boundary-value problems. J. Comput. Math. 85(5), 759-770.
[107] J. I. Ramos (2007): Solitary wave interactions of the GRLW equation. Chaos, Solitons and Fractals 33, 479-491.
References
147
[108] J. Rashidinia, M. Ghasemi, Z. Mahmoodi (2007): Spline approach to the solution of a singular-perturbed boundary-value problems. Appl. Math. Comput. 189, 72-78.
[109] J. Rashidinia, R. Jalilian (2007): Non-polynomial spline for solution of boundary-value problems in plate deflection theory. J. Comput. Math. 84(10), 1483-1494.
[110] J. Rashidinia, R. Jalilian, K. Farajeyan (2008): Spline approximate solution of eight-order boundary-value problems. J. Comput. Math. 00, 1-15.
[111] J. Rashidinia, R. Mahmoodi (2008): Non-polynomial cubic spline methods for the solution of parabolic equations. J. Comput. Math. 85(5), 843-850.
[112] J. Rashidinia, Z. Mahmoodi (2007): Non-polynomial spline solution of a singulaly-perturbed boundary-value problems. Int J. Contemp. Math. Sci. 32(2), 1581-1586.
[113] K. R. Raslan (2004): A computational method for the equal width equation. J. Comput. Math. 81(1), 63-72.
[114] K. R. Raslan (2006): Collocation method using cubic B-spline for the generalised equal width equation. Int. J. Simul. Process Modell. 1(2), 37-44.
[115] S. G. Rubin, P. K. Khosla (1976): Higher order numerical solutions using cubic splines. AIAAJ 14, 851-858.
[116] B. Saka (2007): Algorithms for numerical solution of the modified equal width wave equation using collocation method. Math. Comp. Modl. 45, 1096-1117.
[117] B. Saka, I. Dag (2007): Quartic B-spline collocation method to the numerical solutions of the Burgers' equation. Chaos, Solitons & Fractals 32, 1125-1137.
[118] B. Saka, I. Dag (2008): A numerical study of Burgers' equation. J. Franklin Inst. 345, 328-348.
[119] B. Saka, I. Dag, A. Dogan (2004): Galarkin method for the numerical solution of the RLW equation using quadratic B-Splines. Int. J. Computer Math. 81(6), 727-739.
[120] S. S. Sastry (1976): Finite-difference approximations to one dimensional parabolic equations using cubic spline technique. J. Comput. Appl. Math. 2, 23-26.
[121] I. J. Schoenberg (1946): Contributions to the problems of approximation of equidistant data by analytic functions. Quart. Appl. Math. 4, 45-99 and 112-141.
[122] I. J. Schoenberg (1958): Spline functions, convex curves and mechanical quadrature. Bull. Am. Math. Soc. 64, 352-357.
[123] L. L. Schumaker (1981): Spline fuctions: Basic theory. John Wiley and Son. New York.
References
148
[124] E. V. Shikin, A. I. Plis (1995): Handbook on splines for the user. CRC press. New York.
[125] G. I. Shishkin (1983): A difference scheme on a non-uniform mesh for a differential equation with a small parameter in the highest derivative. U.S.S.R. Comput. Math. Phys. 23(3), 59-66.
[126] S. S. Siddiqi, G. Akram (2006): Solution of fifth-order boundary-value problems using non-polynomial spline technique. Appl. Math. Comput. 175, 1574-1581.
[127] S. S. Siddiqi, G. Akram (2007): Numerical solution of a system of fourth order boundary-value problems using cubic non-polynomial spline method. Appl. Math. Comput.
190, 652-661.
[128] S. S. Siddiqi, G. Akram (2007): Solution of 10th-order boundary-value problems using non-polynomial spline technique. Appl. Math. Comput. 190, 641-651.
[129] S. S. Siddiqi, G. Akram (2007): Solution of the system of fourth-order boundary-value problems using non-polynomial spline technique. Appl. Math. Comput. 185, 128-135.
[130] S. S. Siddiqi, G. Akram (2008): Solutions of 12th-order boundary-value problems using non-polynomial spline technique. Appl. Math. Comput. 199, 559-571.
[131] S. S. siddiqi, G. Akram, S. A. Malik (2007): Non-polynomial sixtic spline method for the solution along with convergence of linear special case fifth-order two-point boundary value problems. Appl. Math. Comput. 190, 532-541.
[132] S. S. Siddiqi, E. H. Twizell (1996): Spline solutions of linear sixth-order boundary-value problems. Int. J. Comput. Math. 45, 295-304.
[133] G. I. Sivashinsky (1983): Instabilities, pattern-formation and turbulance in flames. Ann. Rev. Fluid Mech. 15, 179-199.
[134] Y. S. Smyrlis, D. T. Papageorgiou (1996): Computer assisted study of strange attractors of the Kuramoto-Sivashinsky equation. Z. Angew. Math. Mech (ZAAM) 76(2), 57-60.
[135] A. A. Soliman (2005): Numerical simulation of the generalized regularized long wave equation by He's variational iteration method. Mathematics and computers in Simulation 70, 119-124.
[136] H. Spath (1995): One dimensional spline interpolation algorithms. A. K. Peters Wellesley. Massachusetts.
[137] E. Suli, D. F. Mayers (2003): An introduction to numerical analysis. Press syndicate of Univeristy of Cambridge. Cambridge.
[138] K. Surla, D. Herceg, L. Cvetkovic (1991): A family of exponential spline difference schemes. Review of Research, Faculty of Science, Mathematics Series, University of Novi Sad 19, 12-23.
References
149
[139] K. Surla, M. Stojanovic (1988): Solving singularly-perturbed boundary-value problems by splines in tension. J. Comput. Appl. Math 24, 355-363.
[140] K. Surla, V. Vukoslavcevic (1995): A spline difference scheme for boundary-value problems with a small parameter. Review of Research, Faculty of Science, Mathematics Series, University of Novi Sad 25, 159-166.
[141] E. Tadmor (1986): The well-posedness of the Kuramoto-Sivashinsky equation. SIAM J. Math. Anal. 17, 884-893.
[142] S. I. A. Tirmizi (1984): Numerical methods for bounary-value problems with applications to the wave equation, pp. 175: Department of Mathematics and Statistics, Brunel University, England.
[143] S. I. A. Tirmizi, F. Haq, S. Islam (2008): Non-polynomial spline solution of singularly perturbed boundary-value problems. Appl. Math. Comput. 196, 6-16.
[144] S. I. A. Tirmizi, E. H. Twizell (1986): A fourth order multiderivative method for linear second order boundary value problems. J. Comput. Math. 20, 145-155.
[145] S. I. A. Tirmizi, E. H. Twizell (2002): Higher order finite-difference methods for nonlinear second-order two-point boundary-value problems. Appl. Math. Lett. 15, 897-902.
[146] J. Toomre, J. R. Zahn, J. Latour, E. A. Spiegel (1976): Stellar convection theory II: single-mode study of the second convection zone in A-type stars. Astrophys. J. 207, 545-563.
[147] E. H. Twizell (1984): Computational methods for partial differential equations. Ellis Horwood Limited. England.
[148] E. H. Twizell (1988): A second-order convergent method for sixth-order boundary-value problems, in R.P. Agarwal, Y.M Chow, S.J Wilson (Eds.), Singapore. Numerical Math. Cirkhhauser Verlag, Basel, 495-506.
[149] E. H. Twizell, Boutayeb (1990): Numerical methods for the solution of special and general sixth-order boundary-value problems, with applications to Benard layer eigenvalue problems. Proc. R. Soc. Lond. A. 431, 433-450.
[150] E. H. Twizell, S. I. A. Tirmizi (1987): Multiderivative methods for non-linear second-order boundary value problems. J. Comput. Appl. Math 17, 299-307.
[151] R. A. Usmani (1971): An o(h6) finite difference analogue for the numerical solution of a two point boundary-value problem. J. Inst. Maths. Applics. 8, 335-343.
[152] R. A. Usmani (1992): The use of quartic spline in the numerical solution of fourth-order boundary-value problem. J. Comput. Appl. Math 44, 187-199.
[153] R. A. Usmani, M. Sakai (1984): Quartic spline solution for two-point boundary-value problems involving third-order differential equation. J. Math. and Phys. Sc., 365-380.
References
150
[154] R. A. Usmani, S. A. Warsi (1980): Quintic spline solutions of boundary-value problems. Comput. Math with Appl. 6, 197-203.
[155] M. G. Van-Daele, V. Berghe, H. D. Meyer (1994): A smooth approximation for the solution of a fourth-order boundary-value problem based on non-polynomial splines. J. Comput. Appl. Math 51, 383-394.
[156] V. Vukoslavcevic, K. Surla (1996): Finite element method for solving self-adjoint singularly perturbed boundary-value problems. Math. Montisnigri 7, 86-89.
[157] A. M. Wazwaz (2001): The numerical solution of sixth-order boundary-value problems by modified decomposition method. Appl. Math. Comput. 118, 311-325.
[158] A. M. Wazwaz (2006): New solitary wave solutions to the Kuramoto-Sivshinsky and the Kawahara equations. Appl. Math. Comput. 182, 1642-1650.
[159] A. M. Wazwaz (2006): The tanh and sine-cosine methods for a reliable treatment of the modified equal width equation and its invariants. Comm. Nonlinear Science and Numer. Simul. 11, 148-160.
[160] Y. Xu, C.-W. Shu (2006): Local discontinous Galerkin methods for the KS equations and the Ito-type coupled KDV equations. Comput. Methods Appl. Mech. Engg. 195, 3430-3447.
[161] S. I. Zaki (2000): A quintic B-spline finite elements scheme for the KdVB equation. Comp. Methods in App. Mech. and Engg. 188, 121-134.
[162] S. I. Zaki (2000): Solitary wave interactions for the modified equal width equation. Comput. Phys. Commun. 126, 219-231.
[163] L. Zhang (2005): A finite difference scheme for generalized long wave equation. Appl. Math. Comput. 168, 962-972.