AIMS Mathematics, 8(x): xxx–xxx

DOI:10.3934/math.2024xxx
Received: January 2024
Revised: February 2024
Accepted: March 2024
https://www.aimspress.com/journal/Math Published: May 2024

Type of article
FINITE ELEMENT METHOD FOR SINGULARLY PERTURBED
DELAY DIFFERENTIAL EQUATION OF REACTION DIFFUSION
TYPE

G.Lakshmi1 and L.S.Senthilkumar2, *

1
Department of Mathematics, SRM Institute of Science and Technology
2
Department of Mathematics, SRM Institute of Science and Technology

* Correspondence: [email protected]

Abstract: To develop the numerical scheme for singularly perturbed delay differential equations of
reaction diffusion problems using finite element method. A differential equation with a small positive
parameter multiplying at the highest derivative term subject to boundary conditions belongs to a class
of problems known as singular perturbation problems. Singularly perturbed boundary value problems
appear in many branches of applied mathematics. The solution of singular perturbation problems has
non-uniform behaviour. A subclass of these equations consists of singularly perturbed ordinary differ-
ential equations with a delay. Such type of equations arise frequently in the mathematical modelling
of various practical phenomena, for example, in modelling of automatic system, population dynamics,
nonlinear optics etc.. Asymptotic numerical method and streamline diffusion finite element method
are presented for solving second order singularly perturbed delay differential equations of reaction
diffusion type.
Keywords: Singular perturbation, Delay differential equation, Shishkin Mesh
Mathematics Subject Classification: xxx, xxx

1. Introduction

A differential equations depend on a small positive parameter ε and whose solutions(or their
derivatives)approach disconinous limit as ε approaches zero, such problems are said to be singularly
perturbed, where ε is considered as perturbation parameter.
In the Third International Congress of Mathematics held in Heidelberg in 1904, Prandtl presented
a paper on his invention of the concept of boundary layer. Prandtl’s revolutionary boundary layer
theory proclaimed the critical role of viscosity in determining the flows of common fluids like water
and air. He through his research proved that viscous forces of the flow can be determined by splitting
 2

the domain into two regions. The flow passing through an airfoil the viscous forces can be confined
within the thin regions, called as the boundary layers. But the flow passing through a sphere the viscous
effects need not be confined to the thin layers as the frictional effects are dramatic and it remains in the
outer region. After Prandtl’s invention of boundary layer theory, Wasow along with Friedrichs used for
the first time the term singular perturbation which was a significant contribution in this field.
Singularly perturbed differential equations play a vital role in real life problems. In the mathemat-
ical modeling of various practical phenomenon, certain forms of equations often occur such as in the
modeling of the human pupil-light reflex [15]. One can find many applications of such type of equa-
tions in [8, 12, 13]. in all branches of applied mathematics, it would be common to note the occurrence
of singular perturbation problems. A wide range of techniques for the solution of these problems are
available in [14]. In all these techniques, either the problem is divided into inner region problem and
outer region problem or expressing the inner and outer solutions as asymptotic expansions.
For arbitrary-shaped field problems in space, the finite element method (FEM) provides an effective
numerical process that can be used to establish the solution. Essentially, using a finite element ansatz
and the weighted residuals method, FEM converts the differential equation into an algebraic system of
equations. Numerous specialist techniques were created to solve convection-diffusion issues numeri-
cally. Streamline diffusion finite element method is one of the most popular approaches, Hughes and
Brooks presented this.
Authors Roos et.al. discussed various numerical methods for ODEs and PDEs. Till 2008, they
covered the most of the works in the literature. They discussed behavior of solutions and various
numerical methods which include Finite Difference Methods (FDM), Finite Element Methods and
Finite Volume Methods. Earlier, uniformly convergent numerical methods for singularly perturbed
problems (SPPs) were considered by Doolan et.al. [1], Miller et.al [2], Farrell et.al [3]. First order
initial value problems for SPPs (single and system of equations) and second order boundary value
problems subject to various boundary conditions are discussed in [1]. Miller et.al [2] showed interest
towards some uniformly convergent numerical methods which include fitted FDM, fitted mesh FEMs
and Schwarz methods for linear ordinary and partial singularly perturbed differential equations. Farrell
et.al presented robust layer resolving methods in [3] for a broad class of mathematical problems related
to linear convection diffusion problems subject to different types of boundary conditions, nonlinear
reaction-diffusion problems. The authors Kadalbajoo and Bawa [4] and Kadalbajoo and Patidar [5]
took some spline methods as a tool to solve numerically some SPPs.
In [10], a higher-order numerical method is presented for solving the singularly perturbed delay
differential equations (SPDDEs). Twin boundary layers or oscillatory behavior is exhibited by such
kind of equations that also bear a delay parameter on reaction term.
Remarkable approaches have been made to develop some of the numerical techniques for SPPs
like Initial Value Techniques (IVT) [16]. Boundary Value Techniques (BVT) for solving this type of
problems can be had from [17]. Some more methods like Schwarz Booster Method , Shooting Method,
Asymptotic Method are introduced to solve same type of problems.
FEMs for solving singularly perturbed both reaction and convection diffusion problems are pre-
sented in the following:
Ramesh Babu and Ramanujam [6] studied second order ordinary differential equations with Neu-
mann boundary conditions and discontinuous source term. Using SDFEM, a parametric uniform error
bound for the solution is obtained. Roos and Zarin [9] studied a singularly perturbed convection-

AIMS Mathematics Volume 8, Issue x, xxx–xxx

 3

diffusion problem with two small parameters and the problem is solved using the SDFEM on a Shishkin
(S-type) mesh.

2. Statement of the Problem

we find the solution y(x) for reaction-diffusion delay differential equation.

Find y ∈ τ such that
−εy′′ (x) + a(x)y(x) + b(x)y(x − 1) = f (x), x ∈ Ω
y(x) = ϕ(x) = 0, x ∈ [−1, 0], y(2) = L,
where a(x) ≥ α1 > α >0, b(x) ≥ β0 ≥ 0 , Ω = (0, 2) and a,b and f are sufficiently differential functions
on Ω and ϕ is sufficiently differentiable on [-1,0]

2.1. Reduced Problem

By setting ϵ = 0 in the above equation and replacing y by y0 , we get the following equation,
y0 (x)a(x) + y0 (x − 1)b(x) = f (x), x ∈ Ω
y0 (x) = ϕ(x) = 0, x ∈ [−1, 0], y0 (2) = L,
here a(x) ≥ α1 > α >0, b(x) ≥ β0 ≥ 0 , Ω = (0, 2) and a,b,f are sufficient differential functions on Ω and
ϕ is sufficiently differentiable on [-1,0]

2.2. Equivalent Problem

The above problem can also be written as
Pε y(x) := −ϵy′′∗ (x) + a(x)y∗ (x) = f ∗ (x)

 f (x) − c(x)ϕ(x − 1), x ∈ Ω− , Ω− = (0, 1)

where f ∗ (x) = 

 f (x) − c(x)y(x − 1), x ∈ Ω+ , Ω+ = (1, 2)

Ω=[0,2]
y∗ (0)=ϕ(0),y∗ (1−)=y∗ (1+)
y1∗ (1−)=y1∗ (1+)=y∗ (2) = l

2.3. Shishkin Mesh

The solution y∗ will exhibit boundary layers at x=0 and x=2 and interior layer at x=1. It is obvious
that the problem has a unique solution in H 2 (Ω) and thus in C 1 (Ω) ∩ C 2 (Ω− ∪ Ω+ ) If N is a positive
even number greater than 8,and
ϵ −d + 1 ϵ
r r
d
σ1 = min , 2 InN, σ2 = min ,2 InN (2.1)
4 γ 4 γ
q
First we shall assume σ1 = σ2 = 2 γϵ InN, as otherwise N −1 is exponentially small compared toϵ. Our
mesh will be equidistant on Ω s for which
Ω s = (σ1 , −σ1 + d) ∪ (σ2 + d, −σ2+1 ) (2.2)

AIMS Mathematics Volume 8, Issue x, xxx–xxx

 4

and graded on Ω0 where

Ω0 = (0, σ1 ) ∪ (d − σ1 , d) ∪ (d, d + σ2 ) ∪ (1 − σ2 , 1) (2.3)

The transition points that we select are

xN/8.0 = σ1 , x3N/8.0 = −σ1 + d, xN/2 = d, x5N/8.0 = σ2 + d, x7N/8.0 = −σ2 + 1
Shishkin mesh:
Ψ1 (t) = e−8tInN
Ψ2 (t) = e−4(1−2t)InN
Ψ3 (t) = e−4(2t−1)InN
Ψ4 (t) = e−8(1−t)InN
Here, four mesh generating functions are required because to the unique layers. ϕ1 , ϕ2 , ϕ3 and ϕ4 ,they
have the following characteristics and are all continuous and piecewise continuously differentiable:ϕ1
and ϕ3 - consistently increasing, ϕ2 and ϕ4 - consistently decreasing functions and
ϕ1 (0) = 0, ϕ1 (1/8) = InN
ϕ2 (3/8.0) = InN, ϕ2 (1/2.0) = 0
ϕ3 (3/8.0) = 0, ϕ3 (5/8.0) = InN
ϕ4 (7/8.0) = InN, ϕ4 (1.0) = 0
The mesh
q points are



2 γϵ ϕ1 (ti ), i = 0, ...N/8.0


σ1 + N4 (d − 2σ1 )(i − N8 ), i = N/8.0 + 1, ...3N/8.0





 q
ϵ

+d ϕ (t ), i = 3N/8.0 + 1, ...N/2.0

− 2


γ 2 i

xi = 

 q
+d + 2 γϵ ϕ3 (ti ), i = N/2.0 + 1, ...5N/8.0







d + σ2 + N4 (1 − d − 2σ2 )(i − 5N ), i = 5N/8.0 + 1, ...7N/8.0



8


 q
1 − 2 ϵ ϕ4 (ti ), i = 7N/8.0 + 1, ...N




γ
where ti = i/N.
We now define new functions Ψ1 (r), Ψ2 (r), Ψ3 (r) and Ψ4 (r) closely related to ϕ1 , ϕ2 , ϕ3 and ϕ4 by
ϕi = −InΨi , i=1,2,3,4.
Decomposing the solution y∗ into its smooth portion (m) and layer part (n) is helpful for our analysis.
For our analysis it is useful to have a decomposition of the soultion y∗ in the smooth part v and the
layer part w.
Theorem 2.1. For each k, 0 ≤ k ≤ 4, it holds
|v( k)(x)| ≤C, x ∈ Ω− ∪ Ω+
Cε−k/2 e1 (x), x ∈ Ω−


|wk (x)| ≤ 
Cε−k/2 e2 (x), x ∈ Ω+


3. Weak Formulation

Now we obtain the weak formulation for the equation,

−εy′′∗ (x) + a(x)y∗ (x) = f∗ (x) (3.1)

AIMS Mathematics Volume 8, Issue x, xxx–xxx

 5
Z Z Z
−ε y′′∗ (x)v + a(x)y∗ (x)v = f∗ (x)v (3.2)
Z Z Z
−ε[vy∗ (x) − y′∗ v′ dx] + a(x)y∗ (x)v = f∗ (x)v (3.3)
Z Z Z
ε y′∗ v′ dx] + a(x)y∗ (x)v = f∗ (x)v (3.4)

Standard weak formulation is,

A(y∗ , v) := ε < y′∗ , v′ > + < a(x)y∗ (x), v >=< f∗ , v > (3.5)

when v ∈ H01 (Ω) deontes the usual Sobolev space and ¡.,.¿ is the inner product on L2 (Ω).
N
We denote (Ω∈ ) = x0 , x0 , ...x0 to be the set of mesh points xi for some positive integer N.
We assign hi = xi − xi−1

3.1. Discretization of Weak Formulation

Now the discrete problem is, find yh ∈ Vh such that
Ah (y∗h , vh ) = fn∗ (vh ), vh ∈ Vh
where
Ah (y∗h , vh ) =< εy′∗ , v′h > + < a(x)y∗ (x), vh > (3.6)

fn∗ (vh ) =< f∗ n, vh > (3.7)

and Vh be the space of piecewise linear functions with the basis {Φi (x)}
x−xi 1
 hi − , x ∈ [xi− 1 , xi ]

when Φi (x)= 

 xi+1 −x , x ∈ [xi , xi+ 1 ]

hi+1
An (y∗h , ϕi ) = ε y′∗ ϕ′i dx + a(x)y∗ ϕi dx
R R

Let y∗ h = y∗i ϕi
P
Hence we obtain the following difference scheme
Z xi Z xi+1 Z xi+1 Z xi
y∗i−1 P + y∗i Q + y∗i+1 R = f∗ Φi dx + f∗ Φi dx + f∗ Φi dx + f∗ Φi dx
xi−1 xi xi xi−1

where, P = [− hεi + h4i ai−1/2 ]

ε
R = [− hi+1 + hi+1 a
4 i+1/2
]

4. Error Estimates in General Setup

In this section, the estimates of interpolation error, projection error and consistency error are given.
Hence it is proved that the method is consistent FEM, as the consistency error vanishes.
A general problem of the form A(y∗ , v) = f ∗ (v), ∀v ∈ V is discretized by:
find y∗h ∈ Vh ⊊ Vsuch that Ah (y∗h , vh ) = fh∗ (vh ), ∀vh ∈ Vh .
I
Since the continuous problem has a unique solution y∗ its interpolant y∗ ∈ Vh is well defined.

AIMS Mathematics Volume 8, Issue x, xxx–xxx

 6

Define a biorthogonal basis of Vh with respect to Ah to be the set of functions {λ j }Nj=0 which satisfy
Ah (ϕi , λ j ) = δi j , i = 0, . . . , N.
The function vh ∈ Vh can be uniquely expressed as vh = i=0 Ah (vh , λi )ϕi .
PN
Ah (v, λi )ϕi .
PN
Define a projection operator [11] P : V → Vh such that Pv := i=0
Further, for a consistent method we have Py∗ = y∗h . The error y∗ − y∗h can be determined from the
following y∗ − y∗h = y∗ − y∗ + P(y∗ − y∗ ) + Py∗ − y∗h .
I I

Let K = Py∗ − y∗h , we shall call this as the consistency error as it vanishes in case of consistent FEM.

4.1. Interpolation Error

Following the arguments of [7] one can prove the error convergence.

 [7] On S-type mesh, for τ0 ≥ 1 the interpolation error is bounded above by |y (x) −
∗
Theorem 4.1.
I N −2 (log N)2 , x ∈ Ω0


y∗ (x)| ≤ C 
N −2 ,
 x ∈ Ωs.

4.2. Projection Error

From [7], the error representation, the projection error at the mesh points is
I I
P(y∗ − y∗ )(xi ) = Ah (y∗ − y∗ , λi )(xi ), xi ∈ Ω̄εN .

By the definition of the bilinear form Ah and the procedure adopted in [7], one can easily prove that
I ′
|Ah (y∗ − y∗ , λi )| ≤ CN −2 max |ψ |2 .

4.3. Consistency Error

We now estimate consistency error [7]. As K = Py∗ − y∗h ,

N
X N
X
K= (Ah − A)(y∗ , λi )ϕi + ( f ∗ − fh∗ )(λi )ϕi .
i=0 i=0

Let fh,k = hk fk . Then for a fixed point xi ∈ Ω̄εN ,

K(xi ) =(Ah − A)(y∗ , λi ) + ( f ∗ − fh∗ )(λi )

N−1
X Z 2 Z 2
= h¯k bk yk λk −
∗ i
by λ dx +
∗ i
(by∗ )I λi dx
i=1 0 0
Z 2 Z 2
− (by∗ )I λi dx + f λi dx
0 0
Z 2 Z 2
− ( f )I λi dx + ( f )I λi dx
0 0
Z 2 Z 2
− cy0 (x − 1)λ dx +
i
(cy0 (x − 1))I λi dx
1 1

AIMS Mathematics Volume 8, Issue x, xxx–xxx

 7

Z 2 N
X X
− (cy0 (x − 1))I λi dx + ck y0 (xk − 1)λik − fh,k λik .
1 k=N/2+1
N−1
Z 2
Z 2 
X X
Let K (xi ) = ( f ) λ dx − fh,k λik by λ dx − h¯k bk yk λk 
∗ I i
 ∗ i ∗ i

− 
0 0 i=1
Z N

 2 X 
−  (cy0 (x − 1))I λi dx − ck y0 (xk − 1)λik  .
1 k=N/2+1
Z 2 Z 2
Then, K(xi ) =K (xi ) +
∗
((by ) − (by ))λ +
∗ I ∗ i
( f − f I )λi dx
0 0
Z 2
+ ((cy0 (x − 1))I − (cy0 (x − 1)))λi dx.
1

Now for our convenience, we denote

XN Z xk N
X
< S , λ >h =
I i
S I λi − S h,k λik ,
k=1 xk−1 k=1
K (xi ) =< f , λ >h − < (by ) , λi > − < (cy0 (x − 1))I , λi > .
∗ I i ∗ I

Now, by trapezoidal rule

Z 2 N−1
X N−1 Z
X xk N−1
X
| < f , λ >h | =
I i
f λ −
I i
fh,k λik = f λ −
I i
fh,k λik
0 k=1 k=1 xk−1 k=1
N−1
X ′ ′
≤CN −2 | f (εk )|λik ≤ CN −2 || f ||L∞ (Ω)||λi || ≤ CN −2 .
k=1

Let y∗ = v + w, then
< (by∗ )I , λi >h =< (bv)I , λi >h + < (bw)I , λi >h
N−1
X ′
| < (bv)I , λi >h | ≤ CN −2 ||b|| ||v ||∞,Ω∗ λik ≤ CN −2 ||λi ||∞,Ω∗ ≤ CN −2 .
k=1

Similarly we can prove that | < (bw)I , λi >h | ≤ CN −2

Now, | < (cy0 (x − 1))I , λi >h | = | < cgI , λi > | ≤ CN −2 ||c|| k=1 ||g ||∞,Ω∗ λik ≤ CN −2 , where
PN−1 ′
g = y0 (x − 1).
I
|K ∗ (xi )| ≤| < f I , λi > | + | < (by∗ ), λi >h | + | < (cy0 (x − 1))I , λi >h | ≤ CN −2 .
Z 2 Z 2
I
|K(xi )| ≤|K (xi )| +
∗
| f − f |λ dx +
I i
|by∗ − by∗ |λi dx
0 0
Z 2
+ |cy0 (x − 1)I − cy0 (x − 1)|λi dx.
1

Z 2  
( f − f I )λi dx ≤C || f − f I ||L∞ (Ω̄) + N −2 || f ||L′ (Ω̄) ||λi ||L′ (Ω) ≤ CN −2 .
0

AIMS Mathematics Volume 8, Issue x, xxx–xxx

 8

Figure 1. Numerical solution of the problem stated in Example .

Similar concepts can be applied to the remaining two integrals. Hence |K(xi )| ≤ CN −2 .
Therefore,

I I
|y∗ (xi ) − y∗h (xi )| ≤|y∗ (xi ) − y∗ (xi )| + |P(y∗ − y∗ )(xi )| + |(Py∗ − y∗h )(xi )|
≤CN −2 (log N)2 + CN −2 + CN −2 ≤ CN −2 (log N)2 .

5. Numerical Illustration

Example 5.1. Consider the problem −εy′′ (x) + a(x)y(x) + c(x)y(x − 1) = f (x)
with a(x)=5 c(x)=1 and f(x)=2

6. Conclusion

In this dissertation, we study class of second order delay differential equations of reaction diffusion
type that are singularly perturbed. Using both zeroth order and first order asymptotics, the perturbation
technique constructs an estimate of the solution’s zeroth order asymptotic expansion. Almost second
order convergence is achieved in maximum norm by the present initial value method being provided
ε ≤ CN −1 . The graph represents numerical solution of given example problem. The exact solution
of problem is obtained and compared with numerical solution. The double mesh principle is applied
when it is extremely difficult to find the exact solution. in that case, we double the number of mesh
points in a double mesh and compare those results.

AIMS Mathematics Volume 8, Issue x, xxx–xxx

 9

Use of AI tools declaration

Declaring that have not used Artificial Intelligence (AI) tools in the creation of this article.

References

1. E.P Doolan, J.J.H. Miller, W.H.A Schilders, ., Uniform numerical methods for problems with initial
and boundary layers, Boole Press, 1980.
2. J.J.H. Miller, E. O’Riordan, G.I. Shishkin, Fitted numerical methods for singular perturbation
problems Singapore : World Scientific,2012
3. P.A. Farrell, A.F. Hegarty, J.J.H.Miller, E. O’Riordan, and Shishkin, G.I.,Robust computational
techniques for boundary layers Boca Raton: Chapman Hall/CRC, 2000
4. M.K.Kadalbajoo, R.K. Bawa Cubic spline method for a class of non-linear singularly-perturbed
boundary-value problems J. Optim. Theory Appl. 76(1993), 415–428
5. M.K.Kadalbajoo, K.C. Patidar Variable mesh spline in compression for the numerical solution of
singular perturbation problems Int. J. Comput. Math. 80(2003), 83–93
6. A.Ramesh Babu, N.Ramanujam An Sdfem for a convection-diffusion problem with Neumann
boundary condition and discontinuous term J.Appl.Math.and Informatics 28(2010),31–48
7. A.Ramesh Babu, N.Ramanujam The Sdfem for singularly perturbed convection-diffusion prob-
lem with discontinuous source term arising in chemical reactor theory Int.J.Comput.Math.
88(2011),1664–1680
8. D.Arsalan A new error evaluation for singularly perturbed problem with multi point boundary
condition Computational Methods for Differential Equations 8(2)(2020), 236–250
9. H-G. RoosThe SDFEM for a convection-diffusion problem with two small parameters Comput.
Methods Appl. Math. 3(2003), 443–458
10. G.G. Kiltu, G.G. Duressa, T.A. Bullo, Journal of Ocean Engineering and Science 6(2021), 285–
291
11. H-G. Roos, H.Zarin The streamline-diffusion method for a convection-diffusion problem with a
point source J. Comput. Appl. Math. 150(2003), 109–128
12. T.A.Bullo, G.F.Duressa, G.A.Degla Accelerated fitted operator finite difference method for sin-
gularly perturbed parabolic reaction-diffusion problems Computational Methods for Differential
Equations 9(3) (2021), 886–898
13. Mahboubeh, Molavi-Arabshahi Application of cubic b-spline collocationmethod for reaction dif-
fusion fisher’s equations Computational Methods for Differential Equations 9(1) (2021), 22–35
14. J.Kevorkian , J.D.Cole Perturbation Methods in Applied Mathematics Verlag: Springer, 1981
15. A. Longtin,and J. G. Milton Complex oscillations in the human pupil light reflex with mixed and
delayed feedback, Math. Biosci., 90 (1988), 183–199.
16. M.Kumar, S.Tiwari An initial value technique to solve third order reaction-diffusin singularly per-
turbed boundary value problems Int.J.Comput.Math. 89(2012), 2345–2352

AIMS Mathematics Volume 8, Issue x, xxx–xxx

 10

17. J.Jayakumar, N.Ramanujam A numerical method for singular perturbation problems arising in
chemical reactor theory Computers and Mathematics with Applications, 27(1994), 83–99

© 2024 the Author(s), licensee AIMS Press. This

is an open access article distributed under the
terms of the Creative Commons Attribution License
(https://creativecommons.org/licenses/by/4.0)

AIMS Mathematics Volume 8, Issue x, xxx–xxx