NON STANDARD FINITE DIFFERENCE METHOD FOR SINGULARLY

PERTURBED BOUNDARY VALUE PROBLEMS WITH NEGATIVE SHIFT

PARAMETER

JIMMA UNIVERSITY COLLEGE OF NATURAL SCIENCES

DEPARTMENT OF MATHEMATICS

A Thesis Submitted to the Department of Mathematics Jimma University in Partial

Fulfillment of the Requirements for the Degree of Master of Science in Mathematics.

(Numerical Analysis)

By: Adamu Mokonen Gebre

Supervisor: Prof. Gemechis File Duressa
Co-Supervisor: Habtamu Garoma Debela (PhD)

February, 2022

Jimma, Ethiopia
 Graduate Thesis Ownership Agreement This Thesis is a property of Jimma Univer-
sity, an institution that awarded MSc degree to the graduate student and funded its
research cost fully or partly. The researcher work was accomplished under the closed
support and supervision of assigned university’s academic staff. It is therefore strictly
forbidden to publish, modify, or communicate to put at the disposal of third party the en-
tire document or any part thereof without the common consent the research supervisor(s)
and the graduate student. Disregarding this agreement would lead to accountability ac-
cording to the Jimma University’s research and publication misconduct policy. Article
1.7 of the university’s document for “guidelines and procedures for research, February
2022.”
Name of graduate students: Adamu Mokonen Signature ——- Date——–
Name research supervisor(s): Prof. Gemechis File Signature———- Date——-
Title of the thesis: a nonstandard finite difference method for solving singularly perturbed
boundary value problems with negative shift parameter. Degree awarded: MSc

i
Declaration

I here by declare that the work which is being presented in this thesis entitled “Non
standard finite difference method for solving singularly perturbed boundary
value problem with negative shift parameter” in partial fulfillment of the require-
ment for the degree of Masters of Science in Mathematics, submitted to Jimma University,
department of Mathematics is my original work and it has not been submitted for the
award of any academic degree or the like in any other institution or university, and that
all the sources I have used or quoted have been indicated and acknowledged as complete
references.
Name: Adamu Mokonen
Signature: . . . . . . . . . . . . . . . . . . . . . . . . . . .
Date: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
The work has been done under supervision of:
Name: Prof. Gemechis File
Signature: . . . . . . . . . . . . . . . . . . . . . . . . . . .
Date: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

ii
Acknowledgment

First of all, I would like to thank my omnipotent GOD for His good will and giving
me life to do this thesis. Next to this, it is of great pleasure and proud privilege to express
my deepestand sincere gratitude to my advisor Prof. Gemechis File. I am grateful for his
indispensable encouragement, professionalism, valuable guidance, motivation, persistent
help, and constructive suggestions throughout the period of my thesis work. Without
his active guidance and good natured,it would have not been possible to complete this
work. Then, my heartfelt gratitude goes to my co-advisor Habtamu Garoma (PhD), for
his effective cooperation and great care. I also, express my sincere thank with gratitude
to my wife Gadise Begna,My mother Tejitu Dugassa, and other my family members for
their extended support.

iii
Contents

Declaration ii

Acknowledgment iii

Table of Contents iv

Acronyms vi

List of Tables vi

List of Figures viii

Abstract ix

1 INTRODUCTION 1
1.1 Background of the Study . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Statement of the problem . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Objectives of the study . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3.1 General Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3.2 Specific Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Significance of the study . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 Delimitation of the study . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

iv
2 RIVIEW OF RELATED LITERATURE 6
2.1 Singular perturbation Theory . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Singularly Perturbed Delay Differential Equation . . . . . . . . . . . . . 7
2.3 Boundary Value Problem . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.4 Non standard Finite Difference Method . . . . . . . . . . . . . . . . . . . 8

3 METHODOLOGY 9
3.1 Study Area and Period . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.2 Study Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.3 Source of Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.4 Mathematical Procedure of the study . . . . . . . . . . . . . . . . . . . . 10

4 DESCRIPTION OF THE METHODS, EXAMPLES AND RESULTS 11

4.1 Description of the method . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.2 Properties of continuous function . . . . . . . . . . . . . . . . . . . . . . 13
4.3 Formulation of the Numerical Method . . . . . . . . . . . . . . . . . . . 14
4.4 Uniform convergence analysis . . . . . . . . . . . . . . . . . . . . . . . . 18
4.5 Numerical Example and Results . . . . . . . . . . . . . . . . . . . . . . . 21

5 Discussion,Conclusion and Scope for future Work 26

5.1 Discussion and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5.2 Scope for future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

v
Acronyms

* SPPs- Singularly Perturbed Problems.

* DDE – Delay differential equation

* SPDE – Singularly perturbed differential equation

* SPDDE - Singularly perturbed delay differential equation

* SPDCDE - Singularly perturbed delay convection diffusion equation

* BVPs- Boundary Value Problems.

* NSFDM- Non Standard Finite Difference Method.

vi
List of Tables

4.1 Maximum absolute errors and rate of convergent for different values
of ε, δ = 0.5ε and number of mesh size, N for Example 1. . . . . . . 23

4.2 Comparison of maximum absolute errors and ε=0.1 for Example 1

at number of mesh points N . . . . . . . . . . . . . . . . . . . . . . . 23

4.3 Maximum absolute errors and rate of convergent for different values
of ε, δ = 0.5ε and number of mesh size, N for Example 2. . . . . . . 23

4.4 Comparison of maximum absolute errors and ε=0.1 for Example 2

at number of mesh points N . . . . . . . . . . . . . . . . . . . . . . . 23

4.5 Maximum absolute errors and rate of convergent for different values
of ε, δ = 0.5ε and number of mesh size, N for Example 3. . . . . . . 24

4.6 Comparison of maximum absolute errors and ε=0.1 for Example 3

at number of mesh points N . . . . . . . . . . . . . . . . . . . . . . . 24

vii
List of Figures

4.1 The behavior of Numerical Solution at ε = 10−8 and different values

of N for Example 1 and Example 2 respectively. . . . . . . . . . . 24

4.2 Point wise absolute error plot at ε = 10−8 and different values of N
for Example 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

viii
Abstract

In this thesis, we consider singularly perturbed differential equation containing neg-

ative shift parameter on the convection term. The considered problem exhibits
boundary layer on the left or right side of the domain, depending on the sign of the
coefficient of convective term. The terms with the negative shift treated using Tay-
lor’s approximation. The resulting singularly perturbed boundary value problem
is solved using the technique of non-standard finite difference method. The formu-
lated scheme converges uniformly with order of convergence O(h). The theoretical
finding is validated using numerical examples and observed to be more accurate
than the results in the literature.

ix
Chapter 1

INTRODUCTION

1.1 Background of the Study

Numerical analysis is a branch of mathematics concerned with theoretical foun-

dations of numerical algorithms for the solution of problems arising in scientific
applications, Wasow (1942).
The problems in which the highest order derivative is multiplied by a small positive
parameter are known to be singularly perturbed problems and the parameter is
known as the perturbation parameter. Depending on the solution behavior of the
problem in the limiting case when perturbation parameter goes to zero, such type
of problems are classified into two classes, namely; (i) regularly perturbed problems
(ii) singularly perturbed problems. If the solution of the original problem tends to
the solution of the reduced problem (i.e., the problem which is obtained by putting
ε = 0 in the original problem) as the perturbation parameter tends to zero, the
problem is known as regularly perturbed otherwise, it is known as singularly per-
turbed.
The classification of singularly perturbed higher order problems depends on
how the order of the original equation is affected when small positive parameter ε
is multiplying the highest derivative occurring in the differential equation. If the

1
order is reduced by one, we say that the problem is of convection-diffusion type
and reaction-diffusion type if the order is reduced by two. Singularly perturbed
differential equations with negative shift are special cases of functional differential
equations, where the evolution of a system at a certain time, depends on the present
state of the system as well as the state of the system at an earlier time.
In general, a singularly perturbed differential equation with negative shift pa-
rameter is an ordinary differential equation in which the highest derivative is mul-
tiplied by a small parameter and involving at least one delay term. In recent
years, many researchers have tried to develop different numerical methods for solv-
ing singularly perturbed delay differential equations. For examples, finite differ-
ence method of various orders and approaches ( Phaneendra and Soujanya, 2014;
Gemechis et al., 2017; Gashu et al., 2018), Galerkin method (Swamy et al., 2016),
and Differential quadrature method are presented for solving singularly perturbed
delay differential equations. However, the issue of accuracy and convergence of the
scheme still needs attention and improvement. In this thesis, we present a stable
and convergent method and more accurate than the stated methods for solving
singularly perturbed delay convection-diffusion equations of the type under consid-
eration.

1.2 Statement of the problem

Any differential equation in which the highest order derivative is multiplied by a

small positive parameter and containing at least one negative/positive shift param-
eter is known as a singularly perturbed differential-difference equation. Such types
of problem have a variety of applications in the mathematical modeling of various
physical and biological phenomena. However, the computation of its solution has
been a great challenge and has been of great importance due to the versatility of
such equations in the mathematical modeling of processes in various application

2
fields. The numerical treatment of such problems presents severe difficulties that
have to be addressed to ensure accurate numerical solutions as (Roos et al.,1996)
states that, the accuracy of the problem increased by increasing the resolution of
the grid, which might be impractical in some cases like higher dimensions. A vari-
ety of different numerical approaches have been suggested in an attempt to obtain
accurate and reliable schemes for the treatment of boundary value problems of sin-
gularly perturbed differential-difference equations with a small negative shift in the
convection term (Gadisa and File, 2019). They also tried to discuss the effect of
small shifts on the solution profile of the problem.
Recently, Duressa (2021) was presented the problem under consideration using
exponential fitted operator method. But, still there is a room to increase the accu-
racy. Owing to this, the present study attempt to answer the following questions:
1. How does this study was describe the numerical method for singularly perturbed
boundary value problem with negative shift parameter?
2. To what extent the proposed method converges?
3. To what extent the present method approximate the exact solution?

1.3 Objectives of the study

1.3.1 General Objective

The general objective of this study is to develop non- standard finite difference
method for solving singularly perturbed boundary value problem with negative shift
parameter

1.3.2 Specific Objectives

The specific objectives of the present study are:

3
 – To formulate non-standard finite difference method for solving singularly per-
turbed boundary value problem with negative shift parameter.

– To establish the convergence of the scheme.

– To investigate the accuracy of the scheme

1.4 Significance of the study

The results obtained in this research may

– Serve as a reference material for scholars who works on this area.

– Give an idea about the application of numerical methods in different field of

studies.

– Help the graduate students to acquire research skills and scientific procedure.

– Provide a numerical method for solving singularly perturbed convection diffu-

sion equation with negative shift parameter.

1.5 Delimitation of the study

The singularly perturbed delay differential equations perhaps arise in variety

of applied mathematics that contributes for the advancement of science and tech-
nology. Though, singularly perturbed delay differential equations are vast topics
and have many applications in the real world, this study is delimited to singularly
perturbed delay convection-diffusion equation of the form :

00 0
εy (x) + a(x)y (x − δ) + b(x)y(x) = f (x), x ∈ (0, 1), (1.5.1)

4
with the interval and boundary conditions

y(x) = φ(x), −δ ≤ x ≤ 0, y(1) = γ, (1.5.2)

with small perturbation parameter, 0 < ε  1, δ < ε and δ is small negative shift
parameter; a(x), b(x), φ(x) and f (x) are bounded smooth functions in (0, 1) and γ
is given constant.
Further, the study is delimited to non standard finite difference method for solving
singularly perturbed convection-diffusion equations with negative shift parameter,
though there are varieties of methods for solving the problems under the study.

5
Chapter 2

RIVIEW OF RELATED
LITERATURE

2.1 Singular perturbation Theory

The term “singular perturbations” was first used by (Friedrichs and Wasow,
1946) in a paper presented at a seminar on non-linear vibrations at New York Uni-
versity. Perturbation theory is a vast collection of mathematical methods used to
obtain approximate solution to problems that have no closed form of analytical
solution. Perturbation problems depend on a small positive parameter(s). These
parameters affect the problem in such a way that the solution varies rapidly in
some region of the problem domain and slowly in other parts. The study of many
theoretical and applied problems in science and technology leads to boundary value
problems for singularly perturbed differential equations that have a multi-scale char-
acter.perturbation theory is a subject, which studies the effect of small parameter
in the mathematical model problems in ordinary differential equations. In mathe-
matics, more precisely in perturbation theory, a singular perturbation problem is
a problem containing a small parameter that cannot be approximated by setting

6
the parameter value to zero. The boundary value problems for ordinary differential
equations in which one or more small positive parameter(s) multiplying the second
derivative(s) are known as singularly perturbed problems.

2.2 Singularly Perturbed Delay Differential Equa-

tion

Delay differential equations (DDEs) model problems where there is after effect
affecting the variable of the problem as compared to differential equations which
model the problem to current conditions. DDEs is said to be retarded type if the
delay argument does not occur in the highest order derivative term, otherwise it
is known as neutral DDEs. A singularly perturbed delay differential equations is
differential equations in which its highest order derivative is multiplied by small
perturbation parameter and having delay parameter(s) on the terms different from
the highest order derivative, Gopalsamy(2013). Singularly perturbed DDEs arise
in the mathematical modeling of various physical phenomena.

2.3 Boundary Value Problem

A boundary value problem is a problem, typically an ordinary or partial dif-

ferential equation that has values assigned on physical boundary of the domain in
which the problem is specified. A boundary value problem for a given differential
equation consists of finding a solution of the given differential equation subject to
a given set of boundary conditions, Kumar( 2012). Finding the numerical solution
of a boundary value problem is more difficult than that of corresponding initial
value problem. There is a wide class of asymptotic expansion methods available for
solving the two small parameters singularly perturbed boundary value problems.

7
But there can be difficulties in applying these asymptotic expansion methods, such
as finding the appropriate asymptotic expansions in the inner and outer regions,
which are not routine exercises but require skill, insight and experimentation.

2.4 Non standard Finite Difference Method

Non-standard finite difference schemes (NSFD) have emerged as an alternative

method for solving a wide range of problems whose mathematical models involve
algebraic, differential and biological models as well as chaotic systems, Mickens
(2005). These techniques have many advantages over classical techniques and pro-
vide an efficient numerical solution. In fact, the non-standard finite difference
method is an extension of the standard finite difference method. Non-standard
schemes as introduced by Mickens, (1990) are used to resolve some of the issues
related to numerical instabilities. Furthermore, Mickens, (2005,2000,1999) intro-
duced certain rules for obtaining the best difference equations.
Mickens, (1994) mentions in details about these construction rules in his refer-
ence book Non standard Finite Difference Models of Differential Equations. Non-
standard method is more stable than the standard finite methods and the domain
of h for stability in the non-standard is larger than those of the standard method. If
the denominator functions are chosen in appropriate form the non-standard meth-
ods produce better results,YOghoubi( 2015)

8
Chapter 3

METHODOLOGY

3.1 Study Area and Period

The study was conducted in Jimma University, department of Mathematics

from August 2020 to February 2022.

3.2 Study Design

The study was employed mixed-design (documentary review and experiment).

3.3 Source of Information

The relevant sources of information for this study are books, published articles
on reputable journal and related studies from internet services.

9
3.4 Mathematical Procedure of the study

In order to achieve the above mentioned objectives, the study was follows the
following steps:
1. Defining( or describing) the problems.
2.Discretizing the solution domain.
3.Formulating the numerical scheme for the governing problem under consideration.
4. Establishing the stability and convergence analysis of the scheme.
5. Write MATLAB code for the obtained numerical scheme.
6. Validating the scheme using numerical experment
7. Presenting the results in tables and graphs.

10
Chapter 4

DESCRIPTION OF THE
METHODS, EXAMPLES AND
RESULTS

4.1 Description of the method

In this section, the description of second order finite difference methods and
their stability and convergence analysis is discussed. Consider singularly perturbed
delay convection – diffusion equation of the standard form:

00 0
εy (x) + a(x)y (x − δ) + b(x)y(x) = f (x), x ∈ (0, 1), (4.1.1)

subject to the interval boundary conditions

y(x) = φ(x), −δ ≤ x ≤ 0, y(1) = γ, (4.1.2)

with small perturbation parameter,0 < ε  1 and δ is small delay parameter.

The functions a(x), b(x) and f (x) are assumed to be sufficiently smooth with a(x) ≥

11
a > 0 and b(x) ≥ b > 0 for x ∈ [0, 1]. When the shift parameter δ is smaller than ε
the use of Taylor’s series expansion for the term containing shift argument is valid
,Tian (2002). In this work the case when δ < ε is considered. Thus , to approximate
the term with delay parameter, Taylor’s series expansion is applied as follows:

y 0 (x − δ) = y 0 (x) − δy 00 (x) + O(δ 2 ). (4.1.3)

Now substituting Eq.(4.1.3) in to Eq.(4.1.1),

we get
εy 00 (x) + a(x)y 0 (x) − a(x)δy 00 (x) + b(x)y(x) = f (x), x ∈ (0, 1).
We obtain an asymptotically equivalent singularly perturbed two point boundary
value problem of the form:

00 0
cε y (x) + a(x)y (x) + b(x)y(x) = f (x), (4.1.4)

subject to the boundary conditions

y(0) = φ(0), y(1) = γ, (4.1.5)

where cε = ε − δa(x),and assumed to be positive throughout the interval [0, 1] since

δ is smaller than ε, the effect of the value of the truncated term with O(δ 2 ) is
negligible.
Hence, the solution of the asymptotically equivalent problem is equivalent to that
of the original problem. Further, its error bound is also equivalent to that of the
original problem .

12
4.2 Properties of continuous function

The following lemmas are necessary for the existence and uniqueness of the
solution and for the problem to be well-posed .
Lemma 1 (Continuous minimum principle)
Assume that v(x) ∈ C 2 (Ω̄) be any function satisfying v(0) ≥ 0, v(l) ≥ 0 and
Lv(x) ≤ 0, ∀x ∈ Ω = (0, l).then v(x) > 0, ∀x ∈ Ω̄ = [0, l].
Proof : Let x∗ be such that v(x∗ ) = minx∈[0,l] v(x) and assume that v(x∗ ) < 0.
Clearly x∗ ∈
/ {0, l}, therefore v 0 (x∗ ) = 0 and v 00 (x∗ ) ≥ 0. Moreover, Lv(x∗ ) =
εv 00 (x∗ ) + a(x∗ )v 0 (x∗ ) ≥ 0, which is a contradiction. It follows that v(x∗ ) > 0 and
thus v(x) ≥ 0, ∀x ∈ [0, l].

The uniqueness of the solution is implied by this minimum principle. Its ex-
istence follows trivially (as for linear problems, the uniqueness of the solution im-
plies its existence). This principle is now applied to prove that the solution of
Eqs.(4.1.4) − (4.1.5) is bounded.
The following lemma shows the bound for the derivatives of the solution.
Lemma 2 (Boundedness of the solution) Let u(x) be the solution of the Eqs.
(4.1.1) − (4.1.2), then we obtain the bound
||f ||
|u(x)| ≤ b
+ {max |φ|, |γ|} , for b(x) ≥ b > 0,
where b is lower bound of b(x).
Proof : Defining barrier function
||f ||
ϑ± (x, t) as u± (x, t) = |u(x)| ≤ b
+ {max |φ|, |γ|} ± y(x) and applying the maxi-
mum principle , we obtain the required bound. At the boundary points.
||f ||
y± (0) = b
+ {max |φ|, |γ|} ± y(0) ≥ 0,
||f ||
y± (1) = b
+ {max |φ|, |γ|} ± y(1) ≥ 0,
on the differential operator
0
Lϑ± (x) = cε ϑ± (x) + a(x)ϑ± (x) + b(x)ϑ± (x)
=cε (0 ± u (x) + a(x)(0 ± u (x) + b(x)( ||Lu||
00 0
b
+ {max |φ|, |γ|} ± y(x))

13
 =b(x)( ||Lu||
b
+ {max |φ|, |γ|}) ± f (x))
=0, since b(x) ≥ b > 0,
which implies
L± ϑ(x) ≥ 0.
Hence, by maximum principle we obtain
ϑ(x) ≥ 0, ∀x ∈ Ω
. Lemma 3
Let yε be the solution of (Pε ). Then, for k = 0, 1, 2, 3,

−a
| yε(k) (x) |≤ C(1 + ε−k exp( x)), ∀x ∈ [0, l].
ε

Proof: For the proof refer (Woldaregay and Duressa, 2020).

4.3 Formulation of the Numerical Method

The theoretical basis of non-standard discrete numerical method is based on the

development of exact finite difference method. Mickens(2005) presented techniques
and rules for developing non-standard finite difference methods for different problem
types. In Mickens’s rules, to develop a discrete scheme, denominator function for
the discrete derivatives must be expressed in terms of more complicated functions of
step sizes than those used in the standard procedures. These complicated functions
constitute a general property of the schemes, which is useful while designing reliable
schemes for such problems.we consider separately for left and right boundary layer
problems and develop individual schemes for each. First let us consider the right
boundary layer problem.
Case (1): Right boundary layer problems
For the problem of the form in (4.3.4) − (4.3.5) , in order to construct exact finite
difference scheme we follow the procedures of Bansal and Sharma, (2017).

14
Consider the constant coefficient sub equations from (4.3.4) − (4.4.5) as

00 0
cε y (x) + a(x)y (x) + by(x) = 0, (4.3.6)

00 0
cε y (x) + a(x)y (x) = 0, (4.3.7)

where a(x) ≥ a and b(x) ≥ b

Thus, Eq. (4.3.6) has two independent solutions namely exp(λ1 x) and exp(λ2 x)
√
−a± (a)2 −4cε b
with λ1 ,2 = 2cε
.
We discretize the domain (0, 1), using uniform mesh length x = h such that
1
ω1N = {xi = xo + ih, 1, 2, ........, N, xo = 0, xN = 1, h = N
},
where N is the number of mesh points. We denote Yi as the approximate solution
of y(x) at mesh point xi .
The target is to calculate a difference equation which has the same general so-
lution as the differential equation in (4.3.7) has at the mesh pointxi is given by
Yi = A1 exp(ω1 xi )+ A2 exp(ω2 xi ).
Using the theory of difference equations for second order linear difference equations
in (Bansal and Sharma, 2017), we obtain

 
 Yi−1 exp(λ1 xi−1 ) exp(λ2 xi−1 ) 
 
det 
 Yi exp(λ1 xi ) exp(λ2 xi )  = 0,
 (4.3.8)
 
Yi+1 exp(λ1 xi+1 ) exp(λ2 xi+1 )

Substituting the values of λ1 andλ2 gives

p
−ah (a)2 − 4cε b ah
− exp( )Yi−1 + 2cosh(h )Yi − exp( )Yi+1 , (4.3.9)
2cε 2cε 2cε

is an exact difference scheme for Eq. (4.3.7). For cε → 0, we use the approximation
√ 2
(a) −4cε b ah
h 2cε
) ≈ 2cε
in (4.3.9).

15
 ah
Multiplying both sides by exp( 2c ε
) and simplifying, we obtain

ah
Yi−1 − 2Yi + Yi+1 = (exp( ) − 1)(Yi − Yi−1 ), (4.3.10)
cε

Rearranging Eq.(4.3.10) , we obtain

Yi−1 − 2Yi + Yi+1 Yi − Yi−1

cε hcε ah
)+a = 0. (4.3.11)
a
(exp( cε
) − 1 h

The required denominator function for second derivative discretezation becomes

hcε ah
ωR = (exp( ) − 1). (4.3.12)
a cε

Adopting ω R for the variable coefficient problem we write as

hcε ah(xi )
ωiR = (exp( ) − 1) (4.3.13)
a(xi ) cε

Using the denominator function ωiR in to the scheme (4.3.4), the difference scheme
becomes

[yi+1 − 2yi + yi−1 ] [yi − yi−1 ]

LR
ω Y i ≡ cε R
+ a(xi ) + b(xi )yi = f (xi ), i = 1, 2, 3, .., N − 1.
ωi h
(4.3.14)
This can be written as three term recurrence relation of the form:

Ei yi−1 + Fi yi + Gi yi+i = Hi , i = 1, 2, 3........, N − 1, (4.3.15)

cε ai ai
where Ei = ωiR
− h
, Fi = − 2c
ωR
ε
+ h
+ bi , Gi = cε
ωiR
, and Hi = fi .
i

Case (2): Left boundary layer problems

In this case −a(x) ≤ −a < 0 in(4.3.4) − (4.3.5), we consider the constant coefficient
sub- equations from (4.3.4) − (4.3.5) as

16
 00 0
cε y (x) + a(x)y (x) + βy(x) = 0, (4.3.16)

00 0
cε y (x) + a(x)y (x) = 0, (4.3.17)

where b(x) ≥ b,
Thus, Eq. (4.3.17) has two independent solutions namely exp(λ1 x) and exp(λ2 x)
√
a∓ (a)2 −4cε b
with λ1 ,2 = 2cε

We discretize the domain (0, 1), using uniform mesh length x = h such that
1
ω1N = {xi = xo + ih, 1, 2, ........, N, xo = 0, xN = 1, h = N
},
where N is the number of mesh intervals. We denote Yi as the approximate solution
of y(x) at mesh point xi .
The target is to calculate a difference equation which has the same general solution
as the differential equation in (4.3.17) whose solution at the mesh point xi is given
by Yi = A1 exp(ω1 xi ) + A2 exp(ω2 xi ) .
Using the theory of difference equations for second order linear difference equations
in (Bansal and Sharma, 2017) , we obtain

 
 Yi−1 exp(λ1 xi−1 ) exp(λ2 xi−1 ) 
 
det 
 Yi exp(λ1 xi ) exp(λ2 xi )  = 0,
 (4.3.18)
 
Yi+1 exp(λ1 xi+1 ) exp(λ2 xi+1 )

Substituting the values of λ1 and λ2 and simplfying , we obtain

p
ah (a)2 − 4cε b −ah
− exp( )Yi−1 + 2cosh(h )Yi − exp( )Yi+1 (4.3.19)
2cε 2cε 2cε

is an exact difference scheme for Eq. (4.3.15). For cε → 0 , we use the approximation
√ 2
(a) −4cε b ah
h 2cε
) ≈ 2c ε
. After doing the arithmetic adjustment, we obtain

Yi−1 − 2Yi + Yi+1 Yi + 1 − Yi

cε hcε ah
)+a = 0, (4.3.20)
a
(1 − exp( cε ) h

17
The denominator function becomes
hcε
ωL = a
(1 − exp( ah
cε
)) .
adopting it for the variable coefficient problem , we write as

hcε ah(xi )
ωiL = (1 − exp)( ). (4.3.21)
a(xi ) cε

The required finite difference schemes becomes

[yi+1 − 2yi + yi−1 ] [yi+1 − yi ]

LLω Yi ≡ cε L
+ a(xi ) + b(xi )yi = f (xi ), i = 1, 2, 3, .., N − 1,
ωi h
(4.3.22)
This can be written as three term recurrencerelation as of the form:

Ei yi−1 + Fi yi + Gi yi+i = Hi , i = 1, 2, 3........, N − 1, (4.3.23)

cε ai ai
where Ei = ωiL
, Fi = − 2c
ωL
ε
− h
+ bi , Gi = cε
ωiL
+ h
and Hi = fi .
i

Therefore, on the whole domain, [0, 1], the basic schemes to solve Eq.(4.1.4) and
Eq. (4.1.5) are the scheme given in Eq. (4.3.15) and Eq. (4.3.23) using Thomas
algorithm.

4.4 Uniform convergence analysis

In this section,we need to show the discrete scheme satisfying the discrete min-
imum principle and uniform convergence.
Lemma 4 : (Discrete Minimum Principle) Let vi be any mush function that satis-
fies v0 ≥ 0, vN ≥ 0 and Lh vi ≤ 0 ,i = 1, 2, ........N − 1,then vi ≥ 0, i = 0, 1, 2.....N
proof : The proof is obtained by contradiction.Let f be such that vj = minvi and
suppose that vj < 0.clearly, j ∈
/ {0, N }, vj+1 − vj ≥ 0 and vj − vj−1 ≤ 0 Therefore
cε aj
Lh vj = (v
ωiR j+1
− 2vj + vj−1 ) + h
(vj+1 − vj ) − bj vj

18
 aj
= ωcRε [(vj+1 − vj ) − (vj − vj−1 ) + h
(vj+1 − vj ) − bj vj ≥ 0
i

where the strict inequality holds if vj+1 − vj > 0. This is a contradiction and there-
fore vj ≥ 0. since j is arbitrary,we havevi ≥ 0 i = 1, 2, 3..., N
We proved above that the discrete operator Lh satisfy the minimum principle.
Next we analyze the uniform convergence analysis.Let us define the forward,backward
and cenral finite difference operators as:
vj+1 −vj vj −vj−1 2 D+ vj −D− vj
D+ vj = h
,D− vj = h
,δ vj = D+ D− vj = h

Lemma 5 For a fixed mesh and for cε → 0, it holds

!
exp( −ax
cε
i
)
lim max = 0, m = 1, 2, 3, ...
cε →0 1≤i≤N −1 cm
ε

exp( −a(1−x i)
!
cε
)
lim max = 0, m = 1, 2, 3, ...
cε →0 1≤i≤N −1 cm
ε

1
where xi = ih, h = N
,i = 1, 2, ..., N − 1.
proof : Consider the partition [0, 1] := {0 = x0 < x1 < .... < xN −1 < xN = 1} for
the interior grid points, we have

     
−axi −ax1 −ah
exp exp exp
cε cε cε
max m
≤ m
= m
,
1≤i≤N −1 cε cε cε
     
−a(1 − xi ) −a(1 − xN −1 ) −ah
exp exp exp
cε cε cε
max m
≤ m
= m
,
1≤i≤N −1 cε cε cε

as x1 = 1 − xN −1 = h.
Then, by the application of L’Hospital’s rule m times gives
 
−ah
exp
cε rm m!
lim = lim = lim = 0.
cε −→0 cm
ε r= c −→∞ exp(ahr)
1
ε
1
ε
m
r= c −→∞ (ah) exp(ahr)

Hence, the proof is completed.

Theorem 1: Let the coefficients functions a(x) and the source function f (x) in Eqs.

19
(4.4.17) − (4.4.18) of the domain Ω be sufficiently smooth, so that y(x) ∈ C 4 [0, 1].
Then, the discrete solution Yi satisfies
!!
exp( −ax
cε
i
)
|LN (yi − Yi )| ≤ Ch 1 + sup .
x∈(0,1) c3ε

proof: We consider the truncation error discretization as

|LN (yi − Yi )| =|LN yi − LN Yi |,

D+ D− h2
≤C|cε yi00 + ai yi0 − {cε R
yi + ai D+ yi }|,
ωi
D+ D− h2
≤C|cε (yi00 − R
yi ) + ai (yi0 − D+ yi )|,
ωi
h2
≤Ccε |yi00 − D+ D− yi | + Ccε |( R − 1)D+ D− yi | + Ch|yi00 |,
ωi
(4)
≤Ccε h2 |yi |+Ch|yi00 | + Ch|yi00 |,
(4)
≤Ccε h2 |yi | + Ch|yi00 |.

2 ai h
We used the estimate cε | ωhR − 1| ≤ Ch . Indeed, define ρ = , ρ ∈ (0, ∞)
i cε
.Then,
h2 1 1
cε | R
− 1| = ai h| − | =: ai hQ(ρ).
ωi exp(ρ) − 1 ρ

By simplifying and writing explicitly, we obtain

exp(ρ) − ρ − 1
Q(ρ) = ,
ρ(exp(ρ) − 1)

and we obtain the limit is bounded as

1
lim Q(ρ) = , lim Q(ρ) = 0.
ρ−→0 2 ρ−→∞

Hence, for all ρ ∈ (0, ∞) , we have Q(ρ)C. So, the error estimate in the discretiza-
tion is bounded as
(4)
|LN (yi − Yi )|Ccε h2 |yi | + Ch|yi00 |. (4.4.24)

20
From Eq. (4.4.24) and boundedness of derivatives of solution , we obtain

  
−axi
N
||L (y(xi ) − Yi )|| ≤ Ccε h 2
1+ c−4
exp
ε ,
cε
  
−2 −axi
+ Ch 1 + cε exp ,
cε
  
2 −3 −axi
≤ Ch cε + cε exp ,
cε
  
−2 −axi
+ Ch 1 + cε exp ,
cε
!!
exp( −axcε
i
)
≤ Ch 1 + sup 3
,
x∈(0,1) c ε

since c−3 −2
ε > cε .

Theorem 2: Under the hypothesis of boundness of discrete solution (i.e., it satisfies

the discrete minimum principle), Lemma 5 and Theorem 1, the discrete solution
satisfies the following bound.

sup max|yi − Yi |≤ CN −1 . (4.4.25)

0≤ε≤1 i

Proof: Results from boundness of solution, Lemma 5 and Theorem 1 gives the
required estimates. Hence the proof.

4.5 Numerical Example and Results

To validate the established theoretical results, we perform numerical experiments

using the model problems of the form in Eqs.(4.1.1) − (4.1.2).
Example 1: 
 εy 00 (x) − exp(x)y 0 (x − δ) − xy(x) = 0,


 y(x) = 1, −δ ≤ x ≤ 0, y(1) = 1.


21
Example 2:

 εy 00 (x) − (1 + x)y 0 (x − δ) − exp(−x)y(x) = 1,


 y(x) = 1, −δ ≤ x ≤ 0, y(1) = 1.


Example 3: 
 εy 00 (x) + y 0 (x − δ) + y(x) = 0,


 y(x) = 1, −δ ≤ x ≤ 0, y(1) = 1.


Having yj ≡ yjN (the approximated solution obtained via fitted operator finite
difference method) for different values of h and ε, the maximum errors. Since the
exact solution is not available, the maximum errors (denoted by EεN ) are evaluated
using the double mesh principle for fitted operator finite difference methods using
formula
EεN := max |yjN − y2j
2N
|.
0≤j≤n

Further, we tabulate the ε- uniform error

E N = max EεN .
0<ε≤1

The numerical rate of convergence are computed as

log(EεN ) − log(Eε2N )
rεN := .
log(2)

and the ε- uniform rate of convergence is computed using

log(E N ) − log(E 2N )
RN = .
log(2)

22
Table 4.1: Maximum absolute errors and rate of convergent for different values of ε,
δ = 0.5ε and number of mesh size, N for Example 1.
ε N=16 N=32 N= 64 N= 128 N= 256
−2
10 1.0815e-03 2.9684 e-04 7.6113e-05 1.9154e-05 4.7962e-06
10−4 2.0031e-03 1.0180 e-03 5.1327e-04 2.5766e-04 2.2891e-04
10−6 2.0031e-03 1.0180 e-03 5.1327e-04 2.5766e-04 2.2891e-04
10−8 2.0031e-03 1.0180 e-03 5.1327e-04 2.5766e-04 2.2891e-04
10−10 2.0031e-03 1.0180 e-03 5.1327e-04 2.5766e-04 2.2891e-04

EN 2.0031e-03 1.0180 e-03 5.1327e-04 2.5766e-04 2.2891e-04

RN 0.9765 0.9879 0.9942 0.9991

Table 4.2: Comparison of maximum absolute errors and ε=0.1 for Example 1 at number
of mesh points N .
δ N=8 N=32 N=128 N=8 N=32 N=128
Present M Duressa(2021)
0.03 5.1683e-04 3.2995e-05 2.0653e-06 1.8773e-03 1.247e-04 7.8243e-06
0.05 4.2381e-04 2.6999e-05 1.6891e-06 1.5524e-03 1.0187e-04 6.3843e-06
0.07 3.5805e-04 2.6663e-05 1.4172e-06 1.3187e-03 8.5539e-05 5.3586e-06
0.09 3.0779e-04 1.9404e-05 1.2137e-06 1.1409e-03 7.3473e-05 4.5998e-06

Table 4.3: Maximum absolute errors and rate of convergent for different values of ε,
δ = 0.5ε and number of mesh size, N for Example 2.
ε 16 32 64 128 256
10−2 4.7914e-03 1.3185 e-03 3.3984e-04 8.5532e-05 2.1428e-05
10−4 8.6731e-03 4.4133 e-03 2.2256e-03 1.1175e-03 5.5925e-04
10−6 8.6731e-03 4.4133 e-03 2.2256e-03 1.1175e-03 5.5925e-04
10−8 8.6731e-03 4.4133 e-03 2.2256e-03 1.1175e-03 5.5925e-04
10−10 8.6731e-03 4.4133 e-03 2.2256e-03 1.1175e-03 5.5925e-04

EN 8.6731e-03 4.4133 e-03 2.2256e-03 1.1175e-03 5.5925e-04

RN 0.9747 0.9877 0.9939 0.9987

Table 4.4: Comparison of maximum absolute errors and ε=0.1 for Example 2 at number
of mesh points N .
δ N=8 N=32 N=128 N=8 N=32 N=128
Present M Duressa (2021)
0.03 2.3518e-03 1.4946e-04 9.3492e-06 7.8120e-03 5.1772e-04 3.2466e-05
0.05 1.9088e-03 1.2077e-04 7.5557e-06 6.4652e-03 4.2158e-04 2.6415e-05
0.07 1.5765e-03 1.0057e-04 6.2894e-06 5.4621e-03 3.5314e-04 2.2121e-05
0.09 1.3529e-03 8.5462e-05 5.3440e-06 4.6929e-03 3.0211e-04 1.8924e-05

23
Table 4.5: Maximum absolute errors and rate of convergent for different values of ε,
δ = 0.5ε and number of mesh size, N for Example 3.
ε 16 32 64 128 256
−2
10 1.7887e-02 6.6606 e-03 1.8827e-03 4.9114e-04 1.2389e-04
10−4 1.9521e-02 1.0186 e-02 5.2007e-03 2.6274e-03 1.3205e-03
10−6 1.9521e-02 1.0186 e-02 5.2007e-03 2.6274e-03 1.3205e-03
10−8 1.9521e-02 1.0186 e-02 5.2007e-03 2.6274e-03 1.3205e-03
10−10 1.9521e-02 1.0186 e-02 5.2007e-03 2.6274e-03 1.3205e-03

EN 1.9521e-02 1.0186 e-02 5.2007e-03 2.6274e-03 1.3205e-03

RN 1.0001 1.0000 1.0000 1.0001

Table 4.6: Comparison of maximum absolute errors and ε=0.1 for Example 3 at number
of mesh points N .
δ N=8 N=32 N=128 N=8 N=32 N=128
Present M Duressa (2021)
0.03 6.7217e-03 4.3369e-04 2.7141e-05 2.4155e-02 1.6227e-03 1.0172e-04
0.05 1.0288e-02 6.5201e-04 4.0814e-05 3.4122e-02 2.4758e-03 1.5568e-04
0.07 1.7230e-02 1.1693e-03 7.3343e-05 5.3192e-02 4.4352e-03 2.8433e-04
0.09 3.2744e-02 3.6524e-03 2.4050e-04 7.0034e-02 1.2727e-02 9.4159e-04

1 1
δ=0.00 δ=0.00
0.98 δ=0.03 0.9 δ=0.03
δ=0.05 δ=0.05
0.96
δ=0.07 0.8 δ=0.07
0.94
0.7
0.92
Solution
Solution

0.9 0.6

0.88 0.5

0.86
0.4
0.84
0.3
0.82

0.8 0.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x x

Figure 4.1: The behavior of Numerical Solution at ε = 10−8 and different values of N for
Example 1 and Example 2 respectively.

24
 2.5
δ=0.00
δ=0.03
δ=0.05
δ=0.07

2
Solution

1.5

1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x

Figure 4.2: Point wise absolute error plot at ε = 10−8 and different values of N for
Example 3.

25
Chapter 5

Discussion,Conclusion and Scope for

future Work

5.1 Discussion and Conclusion

In this thesis, we consider three examples exhibiting boundary layer. Example

3 exhibit left boundary layer and Examples 1 and 2 exhibit right boundary layer. In
the computed solutions we used the perturbation parameter ε very small compared
to the number of mesh points N. For each Examples, we computed the maximum
absolute error, parameter uniform error and uniform rate of convergence. In Tables
4.1, 4.3 and 4.5 one can observe that the maximum absolute error is independent
of the perturbation parameter as goes small. This means that, as the perturbation
parameter goes small, the maximum absolute error of the scheme is bounded and it
becomes uniformly convergent. On the last two rows of these tables the parameter
uniform error and the parameter uniform rate of convergence are given. In Table
4.2,4.4 and 4.6 we give the comparison of the obtained result with the result given
in Example 1,Example 2 and Example 3 respectively paper (Duressa,2021). As
one can see, the obtained result is more accurate than the one in (Duressa,2021).

26
For left boundary layer problems, one can observe from Figure 4.2 for Example
3 as the values of the delay parameters increases the size of the boundary layer
decreases. For the case of the right boundary layer problems as the values of the
delay parameter increases the size of the boundary layer increases as it is seen on
Figure 4.1 for Example 1 and 2.

5.2 Scope for future Work

In this thesis, non standard finite difference method for solving singularly per-
turbed boundary value problem with negative shift parameter is introduced.

27
References

1. Andargie, A., Reddy, Y. N. (2013). Parameter fitted scheme for singularly

perturbed delay differential equations. International Journal of Applied Sci-
ence and Engineering, 11(4), 361-373.

2. Duressa, G. F. (2021). Novel approach to solve singularly perturbed boundary

value problems with negative shift parameter. Heliyon, 7(7), e07497.

3. File, G., Gadisa, G., Aga, T., Reddy, Y. N. (2017). Numerical solution of
singularly perturbed delay reaction-diffusion equations with layer or oscillatory
behaviour. American Journal of Numerical Analysis, 5(1), 1-10.

4. Friedrichs, K. O., Wasow, W. R. (1946). Singular perturbations of non-linear

oscillations. Duke Mathematical Journal, 13(3), 367-381.

5. Gadisa, G., File, G., Aga, T. (2018). Fourth order numerical method for
singularly perturbed delay differential equations. International Journal of Ap-
plied Science and Engineering, 15(1), 17-32.

6. Gadissa, G., File, G. (2019). Fitted fourth order scheme for singularly per-
turbed delay convection-diffusion equations. Ethiopian Journal of Education
and Sciences, 14(2), 102-118.

7. H. Tian, The exponential asymptotic stability of singularly perturbed delay

differential equations with a bounded lag, J. Math. Anal. Appl., 270(2002),
143149.

8. K. Bansal and K.K. Sharma. Parameter uniform numerical scheme for time

28
 dependent singularly perturbed convection-diffusion-reaction problems with
general shift arguments. Numerical Algorithms 75(1): 113-145, 2017.

9. K. Gopalsamy, Stability and Oscillations in Delay Differential Equations of

Population Dynamics, Springer Science Business Media; 2013.

10. Kumar, D., Kadalbajoo, M. K. (2012). Numerical treatment of singularly

perturbed delay differential equations using B-Spline collocation method on
Shishkin mesh. JNAIAM, 7(3-4), 73-90.

11. Mickens, R. E. (2000). Applications of nonstandard finite difference schemes.

World Scientific.

12. Mickens, R. E. (2005). Advances in the applications of nonstandard finite

difference schemes. World Scientific.

13. Mickens, R. E. (1994). Nonstandard finite difference models of differential

equations. world scientific.

14. Mickens, J. R.E. (1990). Difference Equations:Theory and Applications. Van

Nostrand Reinhold New York.

15. Roos, H.G., Stynes, M. And Tobiska, L. Methods For Singularly Perturbed
Differential Equations: Convection-Diffusion And Flow Problems. Springer
Verlag Vol 22(1996).Pp.134-145.

16. Roussel, M. R. (2018). The Mackey-Glass models, 40 years later. Biomath

Communications, 5(2), 140-158.Mickens, J. R.E. (1990). Difference Equa-
tions:Theory and Applications. Van Nostrand Reinhold New York.

17. Sangeetha, G., Mahesh, G., Phaneendra, K. (2014). Numerical Approach for
Differential-Difference Equations with Layer Behaviour. Communications in
Mathematics and Applications, 10(4), 851-863.

18. Swamy, D. K., Phaneendra, K., Reddy, Y. N. (2016). Solution of singularly

perturbed differentialdifference equations with mixed shifts using Galerkin
method with exponential fitting. Chinese Journal of Mathematics, 2016.

29
19. Wasow, W. R. (1942). On Boundary Layer Problems In The Theory Of Ordi-
nary Differential Equation. New York University.

20. Woldaregay, M. M., Duressa, G. F. (2020). Robust Numerical Scheme for

Solving Singularly Perturbed Differential Equations Involving Small Delays.
Applied Mathematics E-Notes.

21. Yaghoubi, A. R., Najafi, H. S. (2015). Comparison between standard and non-
standard finite difference methods for solving first and second order ordinary
differential equations. International Journal of Applied Mathematics Research,
4(2), 316.

30