JIMMA UNIVERSITY COLLEGE OF NATURAL SCIENCES

DEPARTMENT OF MATHEMATICS

NONSTANDARD FINITE DIFFERENCE METHOD FOR

SOLVING SECOND ORDER SINGULARLY PERTURBED

PROBLEM HAVING LARGE DELAY

A Thesis Proposal 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: Mulat Emagne Bekele

Advisor: Habtamu Garoma

Debela(PhD) Co-advisor: Worku

Tilahun Aniley (M.Sc)

February,

2022 Jimma,

Ethiopia
Declaration

I,undersigned, declare that "Nonstandared finite difference method for solving second
order singularly perturbed problem having large delay" is original and it has not been
submitted to any institution elsewhere for the award of any academic degree or like and
that all the sources I have used or quoted have been indicated and acknowledged as
complete references.

Name: Mulat Emagne

Signature....................
Date.........................
Advisor: Habtamu Garoma (PhD)
Signature :....................
Date:..........................

i
Dedication

To my beloved brother Yetsedaw Emagne. I never forget you ever.

ii
Acknowledgment

First of all, I am indebted to my almighty God who gave me long life and helped me to
pass through different up and down to reach this time. Next, my special heartfelt thanks
go to my advisor Dr. Habtamu Garoma and co-advisor Mr. Worku Tilahun for their
constructive and critical comments throughout the preparation of this research work.
Finally, my thanks reach to my lovely wife and my darling families for this ideological
and financial support.

iii
Contents

Declaration i

Dedication ii

Acknowledgment iii

Table of Contents iii

List of Tables vi

List of Figures vi

Abstract viii

Acronyms ix

1 Introduction 1

1.1 Background of the study . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Statement of the problem . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3.1 General objective . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3.2 Specific objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Significance of the study . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

iv
 1.5 Delimitation of the study . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Review of Related Literature 6

2.1 Singularly perturbed problems . . . . . . . . . . . . . . . . . . . . . . . . 6

2.2 Singularly Perturbed Delay Differential Equation . . . . . . . . . . . . . 7
2.3 Recent Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

3 Methodology 10

3.1 Study Site and Period.........................................................................................10

3.2 Study Design.............................................................................................................10
3.3 Source of Information...............................................................................................10
3.4 Mathematical Procedure.....................................................................................10

4 Description of The Method, Result and Discussion 12

4.1 Description of the method..................................................................................12

4.2 Properties of Continuous Solution.....................................................................13
4.3 Formulation of the method.................................................................................16
4.4 Uniform Convergence Analysis................................................................................19
4.5 Numerical Examples and Results.......................................................................23
4.6 Discussion and Conclusion.......................................................................................25

Reference 27

v
List of Tables

4.1 Maximum absolute errors for Example 4.5.1 at number of mesh points N 24
4.2 Comparision of maximum absolute errors and rate of convergence for Ex-
ample 4.5.1 at number of mesh points...............................................................25

vi
List of Figures

4.1 Numerical Solution for Example 4.5.1 at ε = 10−8 and N = 64................25

4.2 Point wise error for our Example 4.5.1 at diffrent mesh points........................25

4.3 ε-uniform convergence of the method using log-log plot for Example 4.5.1. 26

vii
Abstract

In this thesis, nonstandard fitted finite difference method has been presented for the
numerical solution of a second order singularly perturbed problems having large delay.
The behavior of the continuous solution of the problem is studied and shown that it
satisfies the continuous stability estimate and the derivatives are also bounded. The
numerical scheme is developed on a uniform mesh using non standard finite difference
method. To validate the applicability of the method, one model problem is considered for
numerical experimentation for different values perturbation parameter and mesh points.

The method is shown to be ε-uniformly convergent with order of convergence O(h). The

proposed method gives more accurate and ε-uniform numerical result.

viii
Acronyms

* SPDE - Singularly Perturbed Differential Equation.

* SPPs - Singularly Perturbed Problems

* SPDDE -Singularly Perturbed Delay Differential Equation.

* NSFDM -Nonstandard Finite Difference Method

* DDE -Delay Differential Equation

ix
Chapter 1

Introduction

1.1 Background of the study

Numerical analysis plays a significant role and helps us to find an approximate solution
for problems which are difficult to solve analytically. In the field of computational
mathe- matics, numerical method is widely used to solve equations arising in the field of
physics, engineering and other sciences. The design and computation of the numerical
algorithm is one of the mathematical challenges that researchers are facing, but scientists
in the field of computational mathematics are trying to develop numerical methods by
using computers for further application (Burden and Faires, 2011).

An equation involving the derivative of one/more dependent variable(s) with respect

to one/more independent variable(s) is said to be a differential equation (DE). There are
two broad categories of such differential equations, i.e PDE and ODE. If the derivative
of the dependent variable is with respect to a single variable, it is called ordinary
differential equation (ODE). But if the derivative is with respect two/more independent
variables, then the differential equation is called partial differential equation (PDE).

A differential equation in which the highest order derivative term multiplied by a

small positive parameter ε , where 0 < ε 1 , is called singularly perturbed differential

equation (SPDE) and the parameter is called a perturbation parameter. The classification

1
of singularly perturbed problem depends on how the order of the original equation is

affected if one sets ε = 0. If the order is reduced by one, we say that the problem is of
convection-diffusion type and of reaction-diffusion type if the order is reduced by two.
Any system involving a feedback control will almost involve time delays. These arise
because a finite time is required to sense information and then react to it. If we restrict
the class of delay differential equation to a class in which the highest derivative is
multiplied by a small positive parameter and involving at least one delay term, then
it is said to be singularly perturbed delay differential equation. We call delay differential
equations retarded type if the delay argument does not occur in the highest order
derivative term;

otherwise it is known as neutral delay differential equations. As ε tends to (ε → 0), the

solution of problems exhibits interesting behaviors (rapid changes). The region where
these rapid changes occur is called inner region or boundary layer and the region in
which the solution changes regularly is called outer region. The behavior of the solutions
of these types of differential equations depends on the magnitude of the parameters. In
this problem typically there are thin transition layers where the solution varies rapidly
or jumps abruptly, while away from the layers the solution behaves regularly and varies
slowly.

In the recent years, there has been a growing interest in the numerical treatment
of such differential equations. This is due to the versatility of such type of differential
equations in the mathematical modeling of various physical and biological phenomena
such as, population ecology, control theory, viscous elasticity, and materials with thermal
memory, Elsgolt’s and Norkin (1973). Hence in the recent times, many researchers have
been trying to develop numerical methods for solving these problems. For example, An-
dargie and Reddy (2013) presented parameter fitted scheme to solve singularly perturbed
delay differential equations Chakravarthy et al., (2015) presented fitted numerical
scheme to solve singular perturbed delay differential equation, Mickens (2000) presented
a non- standard finite difference schemes to solve singular perturbed delay differential
equation, Pratima and Sharma (2011) presented numerical approximation for a class of
2
singularly

3
perturbed delay differential equations with boundary and interior layer(s). But, still the
accuracy of such numerical methods needs attention, because the treatment of the singu-
lar perturbation is not trivial. Due to this, numerical treatment of singular perturbation
needs improvement. Thus, this study will presents nonstandard finite difference method
for solving second order singularly perturbed problem having large delay.

1.2 Statement of the problem

The numerical treatment of singularly perturbed problems yield major computational
difficulties and the usual numerical methods fail to produce accurate results for all inde-

pendent values of x when ε is very small related to the mesh size h (i.e. ε h) for the

solution singularly perturbation two point boundary value problems Khan and Khandel-
wal (2013). That is there are thin transition layers, where the solution varies rapidly.
The field of delay differential equation (DDE) attracted mathematicians and engineers
due to the following reasons. Firstly, we have to find an appropriate approximation of
the solution at the delayed arguments. Secondly, the algorithm has to take care of the
jump in the discontinuity due to the delay parameter and thirdly, its solution behavior
is very interesting with boundary layers, interior layers and oscillations. 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 fields, where they provide the best simulation of observed
phenomena and hence the numerical approximation of such equations has been growing
more and more. The increasing desire for the numerical solutions to such mathematical
problems, which are more difficult or impossible to solve analytically, has become the
present day scientific research area. Gülsu and Öztürk (2011) present an approximate
solution of the singularly perturbed delay convection-diffusion equation. Chebyshev et
al., (2015) present numerical treatment of singularly perturbed delay convection-
diffusion equation by employing modified upwind finite difference scheme, but they
mainly focuses only on

4
the constant coefficient. Kadalbajoo and Ramesh (2007) 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. It is well known that Taylor’s series expansion
methods for solving a class of second order singularly perturbed delay differential equa-
tions with boundary and interior layer(s) are fail to give accurate results when the delay
is large. Recently, Rai and Sharma (2020) considered singularly perturbed delay
differential equations using fitted mesh method. But, still there is a room to increase the
accuracy. Besides, as far as the researchers’ knowledge is concerned the problem under
considera- tion via nonstandard finite difference method is not yet considered. Hence,
the aim of this thesis is to formulate uniformly convergent non-standard finite difference
methods to solve singularly perturbed problem having large delay. Owing to this, the
present study attempt to answer the following questions:

• How does we construct the nonstandard finite difference method for solving
second order singularly perturbed problem having large delay?

• To what extent the proposed method is convergent?

• To what extent the proposed method approximate the solution?

1.3 Objectives

1.3.1 General objective

The general objective of this study is to develop nonstandard finite difference method for
solving second order singularly perturbed problem having large delay.

1.3.2 Specific objectives

The specific objectives of the study are:

5
 * To formulate the nonstandard finite difference method for solving second order
singularly perturbed problem having large delay.

* To establish the convergence of the present scheme.

* To investigate the accuracy of the proposed method.

1.4 Significance of the study

The results obtained in this study may:

* Help the graduate students to acquire research skills and scientific procedures.

* To introduce the application of numerical methods in different field of studies.

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

1.5 Delimitation of the study

• This study is delimited to nonstandard finite difference method for solving second
order singularly perturbed problem having large delay of the form:

εyJJ(x) + a(x)yJ(x) + b(x)y(x) + c(x)y(x − 1) = f (x), 0 < x < 2, (1.1)

y(x) = φ(x), x ∈ [−1, 0], (1.2)

y(2) = l, (1.3)

where φ(x) is sufficiently smooth on [−1, 0]. For all x ∈ Ω, it is assumed that the
sufficient smooth functions a(x), b(x), c(x) and f (x) satisfy at a(x) > a > 0, b(x) >
b ≥ 0, c(x) ≤ c < 0, and a + b + c > 0 and l is a constant.

6
Chapter 2

Review of Related Literature

2.1 Singularly perturbed problems

Science and technology develops many practical problems, such as the mathematical
boundary layer theory or approximation of solution of various problems described by
differential equations involving small parameters have become increasingly complex and
therefore require the use of asymptotic methods. 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 University. Singularly perturbed problems arise frequently
in applications including geophysical fluid dynamics, oceanic and atmospheric
circulation, chemical reactions, civil engineering, optimal control, etc.

It is well known that the solution of singularly perturbed boundary value problems is
described by slowly and rapidly varying parts. So there are thin transition layers where
the solution can jump suddenly, while away from the layers the solution varies slowly
and behaves regularly Akram and Afia (2013). Many scholars have studied the analytical
and numerical solutions of these problems. Abrahamsson et al.,(1974) solved singularly
perturbed ordinary differential equations using difference approximations. Numerical
treatment of singularly perturbed boundary value problems for higher-order non linear
ordinary differential equations has a great role in fluid dynamics. The development of

7
numerical methods for solving singularly perturbed problems started with methods aimed
at solving ordinary differential equations, an account of which can be found in the first
monograph on this subject by Doolan et al (1980)

2.2 Singularly Perturbed Delay Differential

Equation
Singularly perturbed delay differential equation is an equation in which evolution of sys-
tem at a certain time depends on the rate at an earlier time. The delay in process arises
due to requirement of definite time to sense the instruction and react to it. The delay
differential equation in which the highest derivative is multiplied by perturbation param-
eter is known as perturbed delay differential equation. The delay differential equation
can be classified as retarded delay differential equation and neutral differential equation.
A delay differential equation is said to be of retarded delay differential equation if the
delay argument does not occur in the highest order derivative term, otherwise it is known
as neutral delay differential equations. If we restrict it to a class in which the highest
derivative term is multiplied by a small parameter, then we obtain singularly perturbed
delay differential equation of the retarded type. Frequently, delay differential equations
have been reduced to differential equations with coefficients that depend on the delay
by means Taylor’s series expansions of the terms that involve delay and the resulting
differential equation have been solved either analytically when the coefficients of these
equations are constant or numerically, when they are not. The theory and numerical
solution of singularly perturbed delay differential equation are still at the initial stage.
In the past, only every few people had worked in the area of numerical methods on sin-
gularly perturbed delay differential equations (SPDDEs). Lange and Miura (1994) gave
an asymptotic approximation to solve singularly perturbed second order delay differen-
tial equations. Duressa and Reddy (2013) presented a numerical method that does not
depend on the asymptotic expansion and matching of the coefficients for solving a class
of singularly perturbed delay differential equations with negative shift in the differenti-

8
ated term. Andargie and Reddy (2013) provided a parameter fitted scheme and effect of
small shifts on the boundary layer solution of the problem to solve singularly perturbed
delay differential equations in the differentiated term of second order with left or right
boundary. Accordingly, when the delay parameter is smaller than the perturbation pa-
rameter, the layer behavior is maintained. When the delay argument is sufficiently small,
to tackle the delay term Kadalbajoo and Sharma (2004) used Taylor’s series expansion
and presented an asymptotic as well as numerical approach to solve such type boundary
value problem.

But the existing methods in the literature fail in the case when the delay argument is
bigger one because in this case, the use of Taylor’s series expansion for the term
containing delay may lead to a bad approximation. The numerical treatment of singularly
perturbed problems preserves some major computational difficulties and in recent years a
large number of special purpose methods have been proposed to provide accurate
numerical solutions. This type of problem has been intensively studied analytically and it
is known that its solution generally has boundary layers where the solution varies
rapidly. The outer solution corresponds to the reduced problem, i.e., that obtained
by setting the small perturbation parameter to zero.

2.3 Recent Development

Lange and Miura (1994) gave an asymptotic approximation to solve singularly perturbed
second order delay differential equations. Chakravarthy et al., (2015) deals with the sin-
gularly perturbed boundary value problem for the second order delay differential equa-
tion. Similar boundary value problems are associated with expected first-exit times of
the membrane potential in models of neurons. Chakravarthy et al., (2017) deals with
singularly perturbed boundary value problem for a linear second order delay differential
equation. Kumar and Rao (2020) presented a stabilized central difference method for
the boundary value problem of singularly perturbed differential equations with a large

9
negative shift. The central difference approximations for the derivatives are modified by
re-approximating the error terms, leading to a stabilizing effect. The method is found to
be second order convergent.
As introduced in the literature, most researchers have been tried to find approximate
solution for singularly perturbed differential equations with a large delay, but mainly
focuses on constant coefficients, and some others those who have done for variable co-
efficients did not get more accurate solutions. Owing this, this study presents a more
accurate and convergent numerical method for singularly perturbed differential equations
with a large delay, by using nonstandard finite difference method.

1
Chapter 3

Methodology

3.1 Study Site and Period

The study is conducted at Jimma University, College of Natural Science, Department of
Mathematics from August 2021 to January 2021.

3.2 Study Design

The study is applied both the documentation review and numerical experimentation or
mixed design

3.3 Source of Information

The relevant source of information for this study were books, published articles on rep-
utable journals.

3.4 Mathematical Procedure

In order to achieve the stated objectives, the study followed the following mathematical
procedure:

1
1. Defining (or describing) the problem.

2. Discretizing the solution domain /interval.

3. Constructing nonstandard finite difference scheme for the governing problem and
obtain system of linear equation.

4. Writing an algorithm for the developed schemes.

5. Establishing the stability and convergence analysis of the formulated schemes.

6. Solve the obtained system using Guasian elimination method.

7. Validating the schemes using numerical experiments.

8. Discussing and providing conclusions.

1
Chapter 4

Description of The Method, Result and

Discussion

4.1 Description of the method

From Eq. (1.1) and Eq. (1.2), we have singularly perturbed problem having large delay
of the form:

εyJJ(x) + a(x)yJ(x) + b(x)y(x) + c(x)y(x − 1) = f (x), 0 < x < 2, (4.1)

subject to the interval and boundary conditions,

y(x) = φ(x), −1 ≤ x ≤
0, (4.2)

y(2) = l.

As we observed from Eq. (4.1) and Eq. (4.2), the values of y(x − 1) is known for the
domain Ω1 = (0, 1] and unknown for the domain Ω2 = (1, 2) due to the large delay at
x = 1. So, it impossible to treat the problem throughout the domain (Ω¯ ). Thus, we
have to treat the problem at Ω1 and Ω2 separately.

1
So, Eqs. (4.1)-(4.2) is equivalent to

Ly(x) = R(x), (4.3)

where

 
L1 y(x) = εy JJ (x) + a(x)y J (x) + b(x)y(x), x ∈ Ω1 ,
Ly(x) = (4.4)
 L2 y(x) = εy JJ (x) + a(x)y J (x) + b(x)y(x) + c(x)y(x − 1), x ∈ Ω2 ,

 
f (x) − c(x)φ(x − 1), x ∈ Ω1 ,
R(x) =

f (x), x ∈ Ω2 , (4.5)
with boundary conditions

y(x) = φ(x), x [ 1, 0],
 ∈
−
y(1−) = y(1+), yJ(1−) = yJ(1+), (4.6)


y(2) = l.

4.2 Properties of Continuous Solution

Lemma 4.2.1 (Minimum Principle) Let ψ(x) be any function in X such that ψ(0)
≥ 0, ψ(2) ≥ 0, L1 ψ(x) ≥ 0, ∀x ∈ Ω1 , L2 ψ(x) ≥ 0, ∀x ∈ Ω2 and [ψ J ](1) ≤ 0 then
ψ(x) ≤ 0, ∀x ∈ Ω¯ .

Proof : Define a test function


 1
+ x , x ∈ [0, 1],
8 2
s(x) 3 x
+ , x ∈ [1, 2].
8 4

Note that s(x) > 0, ∀x ∈ Ω¯ , Ls(x) > 0, ∀x ∈ Ω1 ∪ Ω2 , s(0) > 0, s(2) > 0 and [sJ ](1) < 0.
Let µ =
−ψ(x) : x ∈ Ω¯ }. Then, there exists x0 such that ψ(x0) + µs(x0) = 0
max{ s
∈ Ω¯
and ψ(x) + µs(x) ≥ 0, ∀x ∈ Ω¯ .Therefore, the function (ψ + µs) attains its minimum at

1
x = x0. Suppose the lemma does not hold true, then µ > 0.
Case (i): x0 = 0
0 < (ψ + µs)(0) = ψ(0) + µs(0) = 0,

it is a contradiction.
Case (ii): x0 ∈ Ω1

0 < L(ψ + µs)(x0) = ε(ψ + µs)JJ(x0) + a(x0)(ψ + µs)J(x0) + b(x0)(ψ + µs)(x0) ≥ 0,

it is a contradiction.

Case (iii): x0 = 1

0 ≤ [(ψ + µs)J](1) = [ψJ](1) + µ[sJ](1) < 0,

it is a contradiction.
Case (iv): x0 ∈ Ω2

0 < L(ψ + µs)(x0) = ε(ψ + µs)JJ(x0) + a(x0)(ψ + µs)J(x0) + b(x0)(ψ + µs)(x0)

+c(x0)(ψ + µs)(x0 − 1) ≥ 0,

it is a contradiction.

Case (iv): x0 = 2
0 < (ψ + µs)(2) = (ψ + µs)(2) ≤ 0,

it is a contradiction. Hence, the proof of the Lemma.

Lemma 4.2.2 (Stability Result) The solution y(x) of Eqs. (1.1)-(1.2), satisfies the bound

|y(x)| ≤ C max{.y(0)., .y(2)., sup .Ly(x).}, x ∈ Ω.

x∈

Proof : This Lemma can be proved by using Lemma 4.2.1 and the barrier functions

1
 . . . . .
θ (x) = CMs(x) ± y(x),
±
x ∈ Ω, where M = max y(0) , y(2) , supx∈Ω∗ Ly(x)
and s(x) is the test function as in Lemma 4.2.1.

Lemma 4.2.3 Let y(x) be the solution of Eqs. (1.1)-(1.2). Then, we have the
following bounds

|y(k)(x)|Ω∗ ≤ Cε−k, for k = 1, 2, 3.

Proof : To bound yJ(x) on the interval Ω1, we consider

L1y(x) = εyJJ(x) + a(x)yJ(x) + b(x)y(x) = R(x).

Integrating the above equation on both sides, we have

∫ x ∫ x
J J J

ε y (x) − y = [a(x)y(x) − a(0)y(0)] a∫(t)y(t)dt − 0 b(t)y(t)dt

(0) + 0 x

+ 0[f (t) − c(t)φ(t −

Therefore,

J J ∫ x ∫ x
J
εy (0) = εy (x) − [a(x)y(x) − a(0)y(0)] a∫(t)y(t)dt − 0 b(t)y(t)dt
+ 0 x

+ 0[f (t) − c(t)φ(t −

Then by the Mean value theorem, there exits z ∈ (0, ε) such that

|εy J (z)| ≤ C(|y(x)|, |R(x)|, |φ(x)|[−1,0] ) and |εy J (0)| ≤ C(|y(x)| + |R(x)| + |φ(x)|).

Hence,
|εyJ(x)| ≤ C max(|y(x)|, |R(x)|, |φ(x)|).

By a similar argument we can bound yJ(x) on Ω2, as |εyJ(x)| ≤ C. From Eqs. (4.4) and
(4.5), we have
|y(k)(x)|Ω∗ ≤ Cε−k, k = 2, 3.

1
Hence, the proof.

Lemma 4.2.4 Let yε be the solution of (Pε). Then, for k = 0, 1, 2, 3,

−a
| y (k) (x) |≤ C(1 + ε−k exp( x)), ∀x ∈ [0, l].
ε
ε
Proof : For the proof refer Bansal and Sharma (2017).

4.3 Formulation of the method

The theoretical basis of non-standard discrete numerical method is based on the develop-
ment of exact finite difference method. Mickens (2005) presented techniques and rules
for developing non-standard finite difference methods for different problem types. In
Mick- ens’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.

For the problem of the form in Eqs. (1.1)–(1.2), in order to construct exact finite differ-
ence scheme, we follow the procedures used in Bansal and Sharma (2017). Let us
consider the following singularly perturbed differential equation of the form

εyJJ(x) + a(x)yJ(x) + b(x)y(x) = f (x). (4.7)

The constant coefficient homogeneous problems corresponding to Eq. (4.7)

εyJJ(x) + ayJ(x) + by(x) = 0, (4.8)

εyJJ(x) + ayJ(x) = 0, (4.9)

1
where a(x) ≥ a and b(x) ≥ b. Two linear independent solutions of Eq. (4.8) are exp(λ1x)
and exp(λ2x), where
√
−a a2 − . (4.10)
λ1,2 ± 2ε
=

We discretize the domain [0, 1] using uniform mesh length ∆x = h such that, ΩN = {xi =
x0 + ih, 1, 2, ..., N, x0 = 0, xN = 1, h = 1 }, where N denotes the number of mesh points.
N
We denote the approximate solution to y(x) at grid point xi by Yi. Now our main objective
is to calculate a difference equation which
2 has the same general solution as the

differential equation Eq. (4.8) at the grid point xi given by Yi = A1 exp(λ1xi) + A2

exp(λ2xi). Using the theory of difference equations and the procedures used in Bansal
and Sharma (2017), we have
Yi−1 exp(λ1 xi−1 ) exp(λ2 xi−1 )

det = 0. (4.11)
Y i exp(λ1xi)) exp(λ2xi)
 

 Yi+1 exp(λ1xi+1) exp(λ2xi+1)

Simplifying Eq. (4.11), we obtain

√
ah h a2 − 4εb ah
— exp(− )Yi−1 + 2 cosh( )Yi − exp( )Yi+1 = 0, (4.12)
2ε 2ε 2ε

which is an exact difference scheme for Eq. (4.8).

√
h a2−4εb
Since ε → 0, we use an approximation ah
in Eq. (4.12). Hence, multplying
≈ 2
both side of Eq. (4.12) by exp( ah ) and after doing the arithmetic manipulation and
2
rearrangement on Eq. (4.12), for the constant coefficient problem Eq. (4.9), we get

Yi−1 − 2Yi + Yi+1 Yi+1 −

Y
εi +a = 0. (4.13)
hε
(exp( ah ) − 1) h
a ε

hε ha
The denominator function becomes Ψ 2 = a exp ε − 1 . Adopting this denomi-

1
nator function for the variable coefficient problem, we write it as

hε hai
Ψ i2 = exp −1 , (4.14)
a ε

where Ψ 2 is the function of ε, ai and h.

i

Assume that Ω¯ 2N denote partition of [0,2] in to 2N subintervals such that 0 = x0 <

2
x1 < ... < xN = 1 and 1 < xN+1 < xN+2 < ... < x2N = 2 with xi = ih, h =
1
2
= ,
N
i = 0, 1, 2, ..., 2N .
Case 1: Consider Eq. (4.1) on the domain Ω1 = (0, 1) which is given by

  JJ
εy (x) + a(x)y J (x) + b(x)y(x) = f (x) − c(x)φ(x −
1), (4.15)

y0 = y(0) = φ(0).

Undertaking the notation Yi = y(xi) and using the nonstandard finite difference method-
ology of Mickens(1991), for right layer in the domain Ω1 the scheme to solve Eq. 4.15 is
given by

Yi+1 − 2Yi + Yi−1

ε + Yi+1 − Yi i = f − c φ(x i— 1), (4.16)
ψ 2 i +bY i+ 1 i i
a τ
h

where
hcε hpi 3
Ψ2 exp —1 = h 2+ O h ,
i ai cε
= ε

the local truncation term τ1 = hai yiJJ + O(h2) = O(h).

2
Eq. (4.16) can be written as three term recurrence relation as

EiYi−1 + FiYi + GiYi+1 = Hi, i = 1, 2, ..., N, (4.17)

where E ai ai
ε ,F = −2ε
− +b,G = ε and H
i= +
1
=f.
Ψ2 i ψ2 h i i Ψ2 h i i
Case 2: Consider Eq. (4.1) on the domain Ω2 = (1, 2), for right layer in the domain Ω2

2
using the nonstandard finite difference method which is given by

Yi+1 − 2Yi+1 + Yi
ε + Yi+1 − Yi +cY — 1) + =f,
ψ 2 i +b i i
(x
i
τ
1 i

a
Y i
h

Similarly, this equation can be written as

ciYj + EiYi−1 + FiYi + GiYi+1 = Hi, i = N + 1, N + 2, ..., 2N − 1, (4.18)

where Y = y(x — 1), j = 1, 2, ..., N , ε —

pi
, 2ε + ai
+b, G = ε and
E = F =
ψ ψ h ψ
j i i i2 h i i2 i i 2
i

Hi = fi.
Therefore, on the whole domain Ω = [0, 2], the basic schemes to solve Eq. (1.1)-(1.2) are
the schemes given in Eq. (4.17) and Eq. (4.18).together with the local truncation error

of τ1.

4.4 Uniform Convergence Analysis

In this section, we need to show the discrete scheme in Eq. (4.17), satisfy the discrete
minimum principle, uniform stability estimates, and uniform convergence.

Lemma 4.4.1 (Discrete Minimum Principle) Let Yi be any mesh function that satisfies
Y0 ≥ 0, YN ≥ 0 and LN Yi ≤ 0, i = 1, 2, 3, ..., N − 1,then Yi ≥ 0, for i = 0, 1, 2, ..., N.
ε

Proof : The proof is by contradiction. Let j be such that Yj = mini Yi and suppose that
Yj ≤ 0. Clearly, j ∈/ {0, N }. Yj+1 − Yj ≥ 0 and Yj − Yj−1 ≤ 0.
Therefore,

Yj+1 − 2Yj + Yj−1

LN Yj =ε Yj+1 − Yj
+ aj
ε Ψ2 h ,
ε i aj
= 2 j+1 — 2Yj + ) + (Yj+1 — Yj),
h
(Y Ψ Yj−1
i
ε − (Yj −1
= 2 ((Y j+1 —Yj) — Y
Ψ
2
 aj h j+1
)) + (Y —Yj),
≥ 0,

2
where the strict inequality holds if Yj+1 − Yj > 0. This is a contradiction and therefore
Yj ≥ 0. Since j is arbitrary, we have Yi ≥ 0, for i = 0, 1, 2, ..., N .
From the discrete minimum principle we obtain an ε− uniform stability property for the
operator LN .
ε
We provide above the discrete operator LN satisfy the minimum principle. Next we
ε

analyze the uniform convergence analysis.

Using Taylor series expansion, the bound for y(xi−1) and y(xi+1) at xi as

 h
2
h3 4
) = y(x ) − hyJ(x ) + yJJ(x ) − y(3)(x ) + h
y(4)(x ) + O(h5),
y(x i i i i i i
2 3 4
 h
2 3
h (3)
4
h
y(x ) = y(x ) + hy (x ) + J
y (x ) +
JJ
y (x ) + y (x ) + O(h5 ).
(4)
i i i 2 i 3 i 4 i

We obtain the bound for



|D+ D− y(xi )| ≤ C|y JJ (xi )|,
(4.19)
 |y JJ (xi ) − D+ D− y(xi )| ≤ Ch2 |y (4) (xi )|.

Similarly, for the first derivative term

|y J (xi ) − D+ y(xi )| ≤ Ch|y JJ (xi )|, (4.20)

where |y (k) (xi )| = supx ∈(x ,x ) |y (k) (xi )|, k = 2, 3, 4.

i 0

Theorem 4.1 Let the coefficients functions a(x) and the source function f (x) in Eqs.
(1.1)-(1.2) of the domain Ω be sufficiently smooth, so that y(x) ∈ C4[0, 1]. Then, the
discrete solution Yi satisfies

|L (yi − Yi )| ≤ Ch axi
N 1 + sup ! ε
.
x∈(0,1
)
− ε3
exp( )

2
Proof : We consider the truncation error discretization as

|L N(yi − Yi )| =|L Nyi − L NYi |,

JJ J
D+D−h2 +
≤C|εyi + ai yi − yi + aiD yi}|,
{ε 2
Ψi
≤C|ε(yiJJ − D+D−h2
2
yi) + ai(yiJ − D+ yi )|,
Ψi
2
≤Cε|yiJJ − h JJ

D+ D yi | + Cε|( 2 − 1)D +D− yi | + Ch|yi |,

−
Ψ i
≤Cεh2 |y (4) |+Ch|y JJ | + Ch|y JJ |,
i i
≤Cεh2|y(4)i| + Ch|yJJ|.
i i

2
We used the estimate ε| h − 1| ≤ Ch which can be derived from Eq. (4.14). Indeed,
aih Ψi
define ρ = , ρ ∈ (0, ∞) .Then,
ε
h2 h2 11
ε| − 1| = ε| ha − 1| = ai h| − | =: ai hQ(ρ).
Ψ i2 hε
exp ε
i
−1 exp(ρ) − 1 ρ
a

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 discretization
is bounded as
|LN (yi − Yi )| ≤ Cεh2 |y (4) | + Ch|y JJ |. (4.21)
i i

2
From Eq. (4.21) and boundedness of derivatives of solution in Lemma 4.2.3 , we obtain

i i . −axi .
|LN (y(x ) − Y )| ≤ Cεh2 . 1 + ε−4 exp ε
., . ε .
. −ax i .
+ Ch. 1 + ε−2 exp −axi
., Ch .. ε + ε exp ..,
2 −3
≤ ε
. .
−ax i
+ Ch. 1 + ε−2 exp .,
. ε
exp( −axi ) .
1 + sup ε !
≤ Ch x∈(0,1 ε3
)
,
since ε−3 > ε−2. Most of the time during analysis, one encounters with exponential terms
involving divided by the power function in ε which are always the main cause of worry.
For their careful consideration while proving the ε -uniform convergence, we prove as
follows.

Lemma 4.4.2 For a fixed mesh and for ε → 0, it holds

ε
li m = 0, m = 1, 2, 3,

exp( −axi )
ε→0 1≤i≤N
ε
−a(1−xi)
lim max exp( = 0, m = 1, 2, 3, ...
!
)
ε
ε→0 1≤i≤N −1 εm

where xi = ih, h = 1 , i = 1, 2, ..., N − 1.

N

Proof : Consider the partition [0, 1] := {0 = x0 < x1 < .... < xN−1 < xN = 1} for the
interior grid points, we have

−axi −ax1 −ah

max exp exp exp
ε ε ε
1≤i≤N
−1
= ,
εm εm εm

ε 2 ε
≤ = ,
 ≤
max −a(1 − x i ) −a(1 − x N − 1 ) −ah
1≤i≤N exp exp exp
−1
ε
εm εm εm

2
as x1 = 1 − xN−1 = h.
Then, by the application of L’Hospital’s rule m times gives

−ah
exp ε
lim = lim = 0.
rm
m! = r= lim
ε−→0 exp(ahr)
r= −→∞ (ah) −→∞ m exp(ahr)
1 1

ε m
ε ε

Hence, the proof is completed.

Theorem 4.2 Under the hypothesis of boundness of discrete solution (i.e., it satisfies

the discrete minimum principle), Lemma 4.4.2 and Theorem 4.1, the discrete solution
satisfy the following bound.

sup max|yi − Yi |≤ CN −1 . (4.22)

0≤ε≤1
i

Proof : Results from boundness of solution, Lemma 4.4.2 and Theorem 4.1 gives the

required estimates. Hence the proof.

4.5 Numerical Examples and Results

In this section, one example is given to illustrate the numerical method discussed above.
The considered problem contain large delay parameter on the reaction term and small
delay parameter on the convection term. The solution of the problem exhibits interior
layer due to the delay parameter and strong left boundary layer due to the small pertur-
bation parameter ε (see Fig 4.1). Fig 4.2 shows, as the number of mesh point increases
(as the mesh size decreases), the absolute error deceases which shows the convergence of
the scheme and Fig 4.3 and Table 4.1 shows, the ε-uniform convergence of our scheme
for h ≥ ε where the classical numerical method fails.

The exact solutions of the test problems are not known. Therefore, we use the double
mesh principle to estimate the error and compute the experiment rate of convergence to

2
the computed solution. For this we put

EN = max |Y N − Y 2N |, (4.23)
ε i i
0≤i≤2N

where Y N and Y 2N are the ith components of the numerical solutions on meshes of N
i
i
and 2N respectively. We compute the uniform error and the rate of convergence as

EN = max EN , andRN
ε
= log2 EN . (4.24)
ε
E2N

The numerical results are presented for the values of the perturbation parameter ε ∈
{ 10−4, 10−8, ..., 10−20}.

Example 4.5.1 Consider the model singularly perturbed boundary value problem

εyJJ(x) + 4(x + 1)yJ(x) − 3y(x) + y(x − 1) = x + 1 x ∈ (0, 1) ∪ (1, 2),

subject to the boundary conditions

y(x) = 1, x ∈ [−1, 0], y(2) = 1

Table 4.1: Maximum absolute errors for Example 4.5.1 at number of mesh points N
ε N=16 N=32 N=64 N=128 N=256
2 −2
2.8610e-03 1.3216e-03 6.3113e-04 3.0778e-04 1.5189e-04
2 −4
4.1609e-03 1.7580e-03 7.5901e-04 3.4577e-04 1.6413e-04
2−6 4.9381e-03 2.4245e-03 1.0777e-03 4.4913e-04 1.9263e-04
2 −8
4.9619e-03 2.5372e-03 1.2765e-03 6.1620e-04 2.7180e-04
2−10 4.9619e-03 2.5374e-03 1.2829e-03 6.4497e-04 3.2178e-04
2 −12
4.9619e-03 2.5374e-03 1.2829e-03 6.4497e-04 3.2178e-04
2 −14
4.9619e-03 2.5374e-03 1.2829e-03 6.4497e-04 3.2178e-04
............
2−28 4.9619e-03 2.5374e-03 1.2829e-03 6.4497e-04 3.2178e-04
EN 4.9619e-03 2.5374e-03 1.2829e-03 6.4497e-04 3.2178e-04
RN 0.9828 0.9915 0.9957 0.9979

2
Table 4.2: Comparision of maximum absolute errors and rate of convergence for Example
4.5.1 at number of mesh points.
ε N=16 N=32 N=64 N=128 N=256

4.9619e-03 2.5374e-03 1.2829e-03 6.4497e-04 3.2178e-04

0.9828 0.9915 0.9957 0.9979

EN 1.15431e-02 6.43596e-03 3.42406e-03 1.75301e-03 8.83454e-04

RN 0.84 0.91 0.97 0.99
1

Numerical solution at N Numerical Solution at 2N

0.9

0.8

0.7

0.6

0.5

0.4

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

x

Figure 4.1: Numerical Solution for Example 4.5.1 at ε = 10−8 and N = 64.
x
3 10−3
N=32
N=64
2.5 N=12
8

1.5

0.5

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x

Figure 4.2: Point wise error for our Example 4.5.1 at diffrent mesh points.

4.6 Discussion and Conclusion

This study introduces non-standard finite difference numerical method for solving singu-
larly perturbed differential equations having large delay. The behavior of the continuous
solution of the problem is studied and shown that it satisfies the continuous stability
2
 −1
10 O(1/N)
=2−2
=2−8
=2−14
=2−20
=2−28

−2
10

−3
10

−4
10
101 102 103
N− Number of mesh points

Figure 4.3: ε-uniform convergence of the method using log-log plot for Example 4.5.1.

estimate and the derivatives of the solution are also bounded. The numerical scheme
is developed on uniform mesh using non-standard finite difference method in the given
differential equation. The stability of the developed numerical method is established
and its uniform convergence is proved. To validate the applicability of the method, one
model problem is considered for numerical experimentation for different values of the
perturbation parameter and mesh points. The numerical results are tabulated in terms
of maximum absolute errors, numerical rate of convergence and uniform errors (see
Table 4.1). Further, behavior of the numerical solution (Fig 4.1), point-wise absolute

errors (Fig 4.2) and the ε-uniform convergence of the method is shown by the log-log

plot (Fig 4.3). The method is shown to be ε-uniformly convergent with order of

convergence O(h). The proposed method gives more accurate, stable and ε-uniform
numerical result.

3
Reference
Abrahamsson, L. R., Keller, H. B., and Kreiss, H. O. (1974). Difference approxima-
tions for singular perturbations of systems of ordinary differential equations. Nu-
merische Mathematik, 22(5), 367-391.

Akram, G. and Afia, N. (2013a). (2013b). Solution of a fourth order singularly perturbed
boundary value problem using quintic spline. Middle-east journal of specific re-
search,15(2):, 302-311.
Andargie, A., and Reddy, Y. N. (2013). Parameter fitted scheme for singularly per-
turbed delay differential equations. International Journal of Applied Science and
Engineering, 11(4), 361-373.
Bansal, K., and Sharma, K. K. (2017). Parameter uniform numerical scheme for time de-
pendent singularly perturbed convection-diffusion-reaction problems with general
shift arguments. Numerical Algorithms, 75(1), 113-145.
Burden, R. L. and Fairs, J.D.(2011). Numerical analysis, 9th edition, Young stown
univeristy.
Chakravarthy P. P, Kumar S.D., Rao R. N. and Ghate D.P. . (2015). A fitted numerical
scheme for second order singularly perturbed delay differntial equations via cubic
spline in comparession,. Advances in Difference Equations, 1-14.

Chebyshev-gauss grid,and phaneendra, (2015). Numerical treatment of singularly per-

turbed delay differential equations, IOSR Journal of Mathematics (IOSRJM),1(2),
01-07.
Doolan, E.P.,Miller, J.J.H., Schilders, W.H.A. (1980). Uniform Numerical Methods for
problems with initial and boundary layers, Boole Press,Dublin.
Elsgolt’s L. E. and Norkin S. B. (1973). Introduction to the Theory and Applications of
Differntial quations with Deviating Arguments, Academic Press New york,.
Friedrichs, K.O., and Wasow, W. . (1946). Singular Perturbations of Non-linear Oscilla-
tions,. Duke.Mathematical Journal,13, 367-381.

3
Gemechis File and Reddy Y.N. (2013). Computational method for solving singularly
perturbed delay differential equations with negative shift,. International Journal
of Applied Science and Engineering,11 (1), 101-113.
Gülsu M. and Öztürk Y. (2011). Approximate solution of the singular-perturbation prob-
lem on chebyshev-gauss grid, American Journal of Computational Mathematics,
1, 209-218.
Kadalbajoo M. K. and Sharma K.K. (2004). Numerical analysis of singularly perturbed
delay differntial equations with layer behavior, Appl.Math.Comput.157, 11-28.
Kadalbajoo. K. and Ramesh V.P. (2007). Numerical methods on Shishkin mesh for
singularly perturbed delay diffential equations with a grid adaptation strategy,.
Applied Mathematics and computation,188, 1816-1831..
Khan,A and Khandelwal, P. (2013). Khan,A and Khandelwal, P. (2013). Non-
polynomial sextic spline solution of singularly perturbed boundery value
plems,rob . Interna- tional Journal of computer Mathematics.
Kumar and Rao. (2020). stabilized central difference method for the boundary
valueprob- lem of singularly perturbed differential equations with a large
negative shift.
Lange C. G. and Miura R. M. (1994). Singular perturbation analysis of boundary-value
problems for differential-difference equations very small shifts with layer behavior,.
SIAMJ.Appl.Math.54, 249-272.

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

schemes. World Scientific.
Mickens . R.E. (2000). Applications of Nonstandard Finite Difference Schemes,. World
Scintific,Singapore.
Pratima R. and SharmaK. K. . (2011). Numerical analysis of singularly perturbed delay
differential turning point problem, . Applied Mathematics and Computation,218,
3483-3498.

Rai, P., & Sharma, K. K. (2020). Numerical approximation for a class of singularly per-
turbed delay differential equations with boundary and interior layer (s).
3
Numerical

3
Algorithms, 85(1), 305-328.