Fractional Calculus and Applied Analysis

https://doi.org/10.1007/s13540-025-00423-3

ORIGINAL PAPER

A computationally efficient numerical scheme for the

solution of fourth-order time fractional partial
integro-differential equation

Mukesh Kumar Rawani1 · Amit K. Verma2

Received: 16 August 2024 / Revised: 9 May 2025 / Accepted: 12 May 2025

© Diogenes Co.Ltd 2025

Abstract
This study introduces an efficient computational method for solving a nonlinear
fourth-order time-fractional partial integro-differential equation with a weakly sin-
gular kernel. The following steps are followed to implement the proposed method.
First, we approximate the time-fractional derivative using the nonstandard finite dif-
ference (NSFD) scheme and the integral term with the product trapezoidal formula. By
discretizing the time derivative and integral term, the time-dependent partial integro-
differential equation is transformed into an ordinary differential equation (ODE) at
each time level. Next, to discretize spatial derivatives, we apply the Haar series
approximation and the quasilinearization technique to linearize the nonlinear term.
This process requires only a one-dimensional Haar approximation with a significantly
smaller number of Haar coefficients. The proposed method provides flexibility in
selecting different denominator functions and also presents high accuracy for large
step sizes. Additionally, the stability and convergence of the method are thoroughly
analyzed. Finally, some test examples are provided to evaluate the effectiveness and
accuracy of the proposed technique.

Keywords Haar wavelets · Quasilinearization · Partial integro-differential equation ·

Fourth-order · Nonstandard finite difference scheme

Mathematics Subject Classification 65N12 · 65M50 · 35R09

B Amit K. Verma
[email protected]
Mukesh Kumar Rawani
[email protected]

1 Department of Mathematics, Bhagalpur College of Engineering, Bhagalpur, Bihar 813210, India

2 Department of Mathematics, Indian Institute of Technology Patna, Patna, Bihar 801106, India

123
 M. Kumar, A. K. Verma

1 Introduction

Fractional differential and integro-differential equations have become increasingly

popular in recent years due to their ability to produce more precise and realistic models
for a wide range of scientific and engineering problems. The fractional operator is an
effective tool for describing the memory and hereditary properties of the dynamical
systems. This memory effect helps researchers to find a more complete picture of
real-world phenomena. With the rapid advancement and extensive application of the
mathematical theory of these differential equations, a growing number of significant
scientific investigations have been presented. The best way to solve any physical
problem governed by fractional differential and integro-differential equations is to
obtain an analytical solution. Unfortunately, in the majority of the cases, it is not
possible to find their analytical solutions due to the complexity and nonlinear nature
of the problems. However, numerical methods are developed to provide approximate
solutions.
In the present study, we propose to analyse the following time-fractional partial
integro-differential equation (TFPIDE) with a weakly singular kernel, given by
 t
C γ νd
0 Dt u(x, t) − μu x x (x, t) − (t − s)δ−1 u x x (x, s)ds + u x x x x (x, t)
Γ (δ) 0
 
+ f u(x, t), u x (x, t) = k(x, t), (x, t) ∈ ΩT , (1.1)

where
ΩT = (0, 1) × (0, T ], (1.2)
with the boundary conditions (BCs)

u(0, t) = g0 (t), u(1, t) = g1 (t), (1.3)

u x x (0, t) = h 0 (t), u x x (1, t) = h 1 (t), t > 0, (1.4)

and initial condition (IC)

u(x, 0) = φ(x), x ∈ [0, 1], (1.5)

μ, νd are non negative constant and γ , δ ∈ (0, 1). The function f is Lipschitz contin-
uous and the functions φ(x), g0 (t), g1 (t), h 0 (t), h 1 (t) are sufficiently smooth. Also,
  t 1 ∂u(x,s)
C γ
1
Γ (1−γ ) 0 (t−s)γ ∂s ds, 0 < γ < 1,
0 Dt u(x, t) = ∂u(x,t) (1.6)
∂t , γ = 1,

represents the Caputo fractional time derivative, where Γ () stands for gamma func-
tion. Eq. (1.1) can be found in different areas of engineering and science such as bridge
slabs, airplane wings, window glasses and floor systems [23, 26, 37, 50]. The main
advantage of using Caputo sense time fractional derivative is that it allows the utiliza-
tion of physically interpretable initial conditions involving classical derivatives [26].
In addition, its derivative for a constant is zero.

123
A computationally efficient numerical scheme for the solution…

The main difficulties in addressing nonlinear TFPIDEs, both analytically and

numerically, arise from various factors, including drastic fluctuations in the solution
due to kernel singularities, nonlinearity, and the historical nature of the time-fractional
derivative.
Recently, partial integro-differential equations have been solved by several methods
such as finite difference method [36], radial basis function method [4], Haar wavelet
method [31]. Fourth-order TFPIDEs with a weakly singular kernel have been studied
by several methods such as the Sinc-Galerkin method [28], finite element method [18],
finite difference method [45], orthogonal spline collocation method [27], spectral
Galerkin method [11], homotopy analysis method [5]. Several other methods can be
found in the publications [46–48] and the references therein. Most of the numerical
techniques developed in the literature are for linear fourth-order TFPIDEs.
The most commonly used numerical techniques for solving physical problems that
are modelled in PDEs include finite difference methods (FDM), finite element methods
(FEM), finite volume methods (FVM), spectral methods and wavelet methods. Solu-
tions for many nonlinear physical problems involve local phenomena and interactions
across multiple scales [16, 20] (e.g., shock formation, hurricanes and turbulence, par-
ticularly atmospheric turbulence, where motion occurs over a continuous range of
length scales). A characteristic feature of such phenomena is that complex behaviour
occurs in a small region of space and is possibly intermittent in time. This makes it
challenging to solve numerically using the techniques stated above. Wavelet bases are
a good way to represent such solutions because they have nice properties like compact
support (a locality in space) and vanishing moment (a locality in scale). It has been
observed that FDM, FVM and FEM have good spatial localization but poor accu-
racy while the spectral methods have good accuracy but poor spatial localization. The
wavelet based numerical techniques appear to combine the advantages of the spectral
method, FDM, FVM, and FEM. The fundamental advantage of wavelet methods is
that the general functions are represented in terms of simpler building blocks at various
scales and places.
Recently, several wavelet families have been used to solve differential and integro-
differential equations numerically. Among all the known orthogonal wavelet families,
only Shannon and Haar wavelets can be represented in closed form. Additionally,
analytical integration and differentiation of these wavelets are quite simple compared
to other wavelets (see [10, 15]). When compared to other wavelets, Haar wavelets
are the simplest among all wavelet families because they are composed of piece-
wise constant functions and offer highly valuable features such as compact support,
orthogonality, and ease of application. Furthermore, they can be integrated analytically
multiple times. Since Haar matrices contain a large number of zeros, Haar wavelet-
based numerical techniques are significantly faster than others. The Haar wavelet
collocation method (HWCM) has been used to solve several differential equations
numerically (see [3, 9, 13, 14, 35, 42, 43] and the references therein). Its extensions to
fractional differential equations (FDEs) have been discussed in the publications [8, 12,
31, 39, 44] and the references therein. Higher-order Haar wavelet collocation methods,
as an improvement of HWCM, have been introduced in the publications [7, 29] and
the references therein. In most of the numerical techniques for time-dependent PDEs,

123
 M. Kumar, A. K. Verma

two-dimensional Haar approximations are used that require a large number of Haar
coefficients to be determined.
Fractional derivatives, as non-local operators, require extensive operations and
memory storage when discretized using classical approaches. Sometimes, such classi-
cal methods fail to accurately reflect the non-local nature of these operators. Standard
discretization techniques may introduce significant numerical artifacts, leading to inac-
curate approximations and unstable solutions. Achieving accurate numerical solutions
often requires exceedingly small step sizes, which is computationally expensive and
time-consuming. The nonstandard finite difference (NSFD) technique is well-known
for producing highly accurate results with larger step sizes and also offers flexibility
in choosing different denominator functions.
In this work, we implement the HWCM coupled with the NSFD scheme to solve
the nonlinear TFPIDE given in Eqs. (1.1)–(1.5). Combining the NSFD and HWCM
aims to provide a more precise solution to the time-dependent problem using a
one-dimensional Haar series approximation. The most significant advantage of this
technique is that better results can be obtained by choosing a suitable denominator
function. It also provides high accuracy for large temporal step sizes, whereas tradi-
tional methods often require extremely small step sizes. In the solution procedure, we
first discretize the time-fractional derivative and the integral term using the NSFD L1
formula and the product trapezoidal method, respectively, and convert the TFPIDE
into an ODE in the spatial variable at each time level. We then implement the Haar
wavelet series approximation for the spatial derivatives, which requires only a one-
dimensional Haar wavelet approximation with a significantly smaller number of Haar
coefficients to be determined. To deal with the nonlinearity, we use the quasilineariza-
tion technique. We also present the convergence and stability analysis of the developed
scheme.
The main features that attract researchers to finding the numerical solutions using
HWCM in combination with the NSFD method are:

(i) The derivation of the method requires only a very simple algorithm and analytical
integration.
(ii) The sparsity of the transform matrices and the minimal number of wavelet coeffi-
cients result in low computation cost.
(iii) It produces precise results for large step sizes in both space and time.
(iv) The proposed method is highly feasible for boundary value problems, because, the
BCs are directly incorporated during the construction of the algorithm.

The content of this article is divided into different sections: Section 2 introduces
some introductory notions for the Haar wavelets (HWs) and the nonstandard finite
difference (NSFD) scheme. In Section 3, we illustrate the method for the TFPIDE
defined in Eqs. (1.1)–(1.5). Sections 4 and 5 illustrate the convergence and stability
of the technique, respectively. Section 6 is devoted to demonstrating the effectiveness
of the developed technique by presenting several tables and figures. The conclusion
of this work is presented in Section 7.

123
A computationally efficient numerical scheme for the solution…

2 Preliminary

2.1 The NSFD scheme

Consider the differential equation

du
= f (u, t). (2.1)
dt

The most traditional way of discretizing Eq. (2.1) is finite difference model, given by

u j+1 − u j
= f (u j , t j ), (2.2)
Δt

where u j is the discrete analogue of the continuous function u(t) at t j = jΔt, Δt is

the step size. Eq. (2.2) is obtained by the following replacement

du u j+1 − u j
→ , (2.3)
dt Δt

which is based on the traditional definition of derivative

du u(t + Δt) − u(t)

= lim . (2.4)
dt Δt→0 Δt

One of the major difficulty with the use of traditional finite difference technique is the
occurance of instabilities in the solution of the difference equation [21]. One possibility
to construct more general definition of first order derivative of u(t) is

du u(t + Δt) − u(t)

= lim , (2.5)
dt Δt→0 ϕ(Δt)

where ϕ(Δt) is called denominator function and has the property

ϕ(Δt) = Δt + O(Δt 2 ). (2.6)

Thus, the generalized discrete model of Eq. (2.1) is

u j+1 − u j
= f (u j , t j ). (2.7)
ϕ(Δt)

Based on the above discussion, R. E. Mickens [21, 22] introduced the concept of
NSFD scheme. A discrete representation of a derivative is known as the NSFD scheme
if it meets at least one of the rules listed below.
u −u
(a) In the approximation du dt ≈ , Δt is replaced by the function ϕ(Δt) which
j+1 j
Δt
satisfies the property (2.6) as Δt → 0.

123
 M. Kumar, A. K. Verma

(b) Nonlinear terms are represented in a nonlocal way, i.e.,

u j+1 (u j+1 )x ≈ u j (u j+1 )x + u j+1 (u j )x , (u j+1 )2 ≈ u j+1 u j . (2.8)

To construct the denominator function, let us consider the standard differential

equation
du
= −u. (2.9)
dt
A simple discretization by Euler-forward difference method is

u j+1 − u j
= −u j . (2.10)
Δt
This can also be written as
u j+1 = (1 − Δt)u j . (2.11)
It is well known that for Δt > 1, Eq. (2.11) leads to instabilities (see [21]). However,
if we use the fact that 1 − Δt = e−Δt + O(Δt 2 ), then for sufficiently small Δt, we
can make the replacement 1 − Δt → e−Δt or Δt → 1 − e−Δt , and as a consequence,
we rewrite Eq. (2.10) as
u j+1 − u j
= −u j . (2.12)
1 − e−Δt
Thus, the NSFD scheme for Eq. (2.9) is given by Eq. (2.12) with ϕ(Δt) = 1 − e−Δt .
Different types of denominator functions (DFs) [21, 22, 30, 32, 40, 41] are

eλΔt − 1
1 − e−Δt , , sin Δt, tan Δt, tanh Δt, sinh Δt, etc.
λ
The key benefits of applying the NSFD scheme are:
(i) A suitable NSFD scheme significantly reduces or eliminates numerical instabilities
produced by the standard finite difference (SFD) scheme.
(ii) The NSFD scheme preserves specific characteristics of the solutions, such as
boundedness, monotonicity, and positivity.
(iii) It allows the use of a variety of denominator functions (DFs), depending on the
nature of the problem.
(iv) High precision can be achieved with large step sizes, whereas traditional methods
often require extremely small step sizes.

2.2 Haar wavelets

Here, we illustrate the basic concepts regarding HWs which are needed in this work.
Let Δx = 1/2M be the width of the subintervals of [0, 1], where M = 2 J , J is the
maximum level of resolutions of the Haar function. The functions [1]

1 for x ∈ [0, 1),
1 (x) = (2.13)
0 othersise,

123
A computationally efficient numerical scheme for the solution…

⎧
⎪
⎨1 for x ∈ [0, 0.5),
2 (x) = −1 for x ∈ [0.5, 1), (2.14)
⎪
⎩
0 otherwise,

are known as the Haar scaling function and mother wavelet, respectively. Addition-
ally, two new parameters κ = 0, 1, · · · , m − 1 and m = 2 j , j = 0, 1, · · · , J are
introduced, which are referred to as the dilation and translation parameters, respec-
tively. The other functions in the Haar wavelet family are generated by dilation and
translation of the mother wavelet and are given by

⎧
⎪
⎨1 for x ∈ [β1 , β2 ),
i (x) = −1 for x ∈ [β2 , β3 ), (2.15)
⎪
⎩
0 otherwise,

where
β p = ξ Δx(2κ + p − 1), ξ = M/m, p = 1, 2, 3. (2.16)

The index of i (x) can be obtained by i = 1 + κ + m. The υ th time integration of

i (x) is given by
 x
1
Ti,υ (x) = (x − z)υ−1 i (z)dz,
(υ − 1)! 0
i = 1, 2, · · · , 2M, υ = 1, 2, · · · , n, n ∈ N, (2.17)

which is required for solving the differential equations of n th -order. Using Eqs. (2.13)–
(2.17), the integral Ti,υ (x) can be calculated analytically which are given by

xυ
T1,υ (x) = , (2.18)
υ!
⎧
⎪
⎪ 0, for x < β1 ,
⎨(x − β )υ /υ!,
⎪
for x ∈ [β1 , β2 ),
1
Ti,υ (x) =  
⎪
⎪
⎪ (x − β1 )υ − 2(x − β2 )υ /υ!, for x ∈ [β2 , β3 ),
⎩ 
(x − β1 )υ − 2(x − β2 )υ + (x − β3 )υ /υ!, for x ≥ β3 .
(2.19)

3 Method description

3.1 Discretization of time fractional derivative and integral term

Let N be a positive integer and denote the time step by Δt = T /N such that
t = Δt,  = 0, 1, 2, · · · , N . In view of the L1 discretization formula [33, 49],

123
 M. Kumar, A. K. Verma

we approximate the fractional time derivative as

∂ γ u(t) (Δt)−γ
−1 
= Θηγ u(t−η ) − u(t−1−η ) + O(Δt), (3.1)
∂t γ t Γ (2 − γ )
η=0

γ
where Θη = (η + 1)1−γ − η1−γ . Replacing Δt by ϕ(Δt), we get NSFD L1 dis-
cretization formula. To approximate the integral term, we apply the product trapezoidal
rule [17, 36], given by
 t
(t − s)δ−1 u x x (x, s)ds
0
−1  tη+1 
t+1 − s s − t
≈ s δ−1 u x x (x, t−η ) + u x x (x, t−η−1 )
Δt Δt
η=0 tη
−1
1 δ
= tη+1 u x x (x, t−η−1 ) − tηδ u x x (x, t−η )
δ
η=0
−1  tη+1
1 u x x (x, t−η ) − u x x (x, t−η−1 )
+ sδ ds
δ Δt
η=0 tη
 t  t1
1 1  1 1 
= tδ − s δ ds u x x (x, t0 ) − t0δ − s δ ds u x x (x, t )
δ Δt t−1 δ Δt t0
−1   tη+1  tη 
1 1
+ s δ ds − s δ ds u x x (x, t−η ). (3.2)
δ Δt tη tη−1
η=1

A simple calculation of the Eq. (3.2) provides

 t
 (Δt)δ  δ δ+1 − ( − 1)δ+1 
(t − s)δ−1 u x x (x, s)ds ≈  − u x x (x, t0 )
0 δ δ+1
−1
1     
+ u x x (x, t ) + (η + 1)δ+1 − ηδ+1 − (ηδ+1 − (η − 1)δ+1 u x x (x, t−η ) .
δ+1
η=1
(3.3)

Now, discretizing time fractional derivative using Eq. (3.1) and taking θ -weighted [2]
to the Eq. (1.1), we obtain

(Δt)−γ (Δt)−γ

 
u +1 − u − Θηγ u −η+1 − u −η − (1 − θ )μ(u  )x x
Γ (2 − γ ) Γ (2 − γ )
η=1
 t
(1 − θ )νd
−θ μ(u +1 )x x − (t − s)δ−1 u x x (x, s)ds
Γ (δ) 0

123
A computationally efficient numerical scheme for the solution…

 t+1
θ νd
− (t+1 − s)δ−1 u x x (x, s)ds + (1 − θ )(u  )x x x x
Γ (δ) 0
   
+θ (u +1 )x x x x + (1 − θ ) f u  , (u  )x + θ f u +1 , (u +1 )x
= (1 − θ )k(x, t ) + θ k(x, t+1 ), (3.4)

with IC
u 0 (x) = φ(x), x ∈ [0, 1], (3.5)
and BCs

u +1 (0) = g0 (t+1 ), u +1 (1) = g1 (t+1 ), (3.6)

(u +1 )x x (0) = h 0 (t+1 ), (u +1 )x x (1) = h 1 (t+1 ), for  = 0, 1, 2, · · · , N − 1.
(3.7)

Now, we replace Δt by ϕ(Δt) and take θ = 21 , in Eq. (3.4), we obtain

 −γ  −γ
ϕ(Δt) ϕ(Δt) 
 
u +1 − u − Θηγ u −η+1 − u −η
Γ (2 − γ ) Γ (2 − γ )
η=1
 t+1
μ  νd
− (u +1 )x x + (u  )x x − (t+1 − s)δ−1 u x x (x, s)ds
2 2Γ (δ) 0
 t  1 
+ (t − s)δ−1 u x x (x, s)ds + (u  )x x x x + (u +1 )x x x x
0 2
1     1 
+ f u +1 , (u +1 )x + f u  , (u  )x = k(x, t+1 ) + k(x, t ) , (3.8)
2 2

with IC (3.5) and BCs (3.6)–(3.7).

Now, applying formula (3.3) into the Eq. (3.8), we get

1 1  1  
(u +1 )x x x x − μ + H (u +1 )x x + Gu +1 + f u +1 , (u +1 )x
2 2 2
1  1   μ 1
= k(x, t+1 ) + k(x, t ) − f u  , (u  )x + (u  )x x − (u  )x x x x + G u 
2 2 2 2

γ  H H b + b 
− Θη u −η+1 − u −η + (u  )x x + (δ + 1) C − −1 u x x (x, t0 )
2 2 1+δ
η=1
−1 
H     
+ bη − bη−1 u x x (x, t−η ) + bη − bη−1 u x x (x, t−η+1 ) , (3.9)
2
η=1 η=1

where

(Δt)δ νd (ϕ(Δt))−γ
H= ,G = , bη = (η + 1)δ+1 − ηδ+1 ,
Γ (δ)δ(δ + 1) Γ (2 − γ )
Cη = ηδ + (η + 1)δ+1 , η = 0, 1, 2, · · · ,

123
 M. Kumar, A. K. Verma

with IC (3.5) and BCs (3.6)–(3.7).

At each time level, Eq. (3.9) is a nonlinear ODE. Applying quasilinearization tech-
nique [6] for the nonlinear term, we have
 +1 r +1  
+1
f u r+1 , (u +1 )x = f (ur+1 )x (u r+1 )x − (u r+1 )x
  
+1
+ f ur+1 u r+1 − u r+1 + f u r+1 , (u r+1 )x , (3.10)

where r represents the quasilinearization iterative index. From Eq. (3.10) and Eq. (3.9),
we get

1 r +1  1  r +1 1  1
+1 +1
(u +1 )x x x x + G + f ur u +1 − μ + H (u r+1 )x x + f (ur )x (u r+1 )x
2 2 +1 2 2 +1
1  1  1
= k(x, t+1 ) + k(x, t ) + μ + H (u r+1 )x x − (u r+1 )x x x x + G u r+1
2 2 2
  H
γ  +1 +1  b + b  r +1
− Θη u r−η+1 − u r−η + (δ + 1) C − −1 u x x (x, t0 )
2 1+δ
η=1
−1 
H     
+ bη − bη−1 u rx+1
x (x, t−η ) + bη − bη−1 u rx+1
x (x, t−η+1 )
2
η=1 η=1
1   1   1 1
− f u r+1 , (u r+1 )x − f u r+1 , (u r+1 )x + u r+1 f ur + (u r+1 )x f (ur )x ,(3.11)
2 2 2 +1 2 +1

with IC
u r0+1 (x) = φ(x), x ∈ [0, 1], (3.12)
and BCs

+1 +1
u r+1 (0) = g0 (t+1 ), u r+1 (1) = g1 (t+1 ), (3.13)
+1 +1
(u r+1 )x x (0) = h 0 (t+1 ), (u r+1 )x x (1) = h 1 (t+1 ), for  = 0, 1, 2, · · · , N − 1.
(3.14)

3.2 Spatial approximation

We use the Haar wavelets finite series approximation for highest order derivative
+1
(u r+1 )x x x x (x) as

2M
+1
(u r+1 )x x x x (x) = ci i (x). (3.15)
i=1

Double integration of Eq. (3.15) from 0 to x and using BCs provide

2M
+1  
(u r+1 )x x x (x) = h 1 (t+1 ) − h 0 (t+1 ) − ci Ti,2 (1) − Ti,1 (x) , (3.16)
i=1

123
A computationally efficient numerical scheme for the solution…

+1  
(u r+1 )x x (x) = h 0 (t+1 ) + h 1 (t+1 ) − h 0 (t+1 ) x
2M
 
− ci xTi,2 (1) − Ti,2 (x) . (3.17)
i=1

Again, integrating (3.17) twice from 0 to x and inserting BCs (3.13), we get

+1  1 x2 1 
(u r+1 )x (x) = g1 (t+1 ) − g0 (t+1 ) + x − h 0 (t+1 ) + − h 1 (t+1 )
2 2 6
 2M  x2 1 
−h 0 (t+1 ) − ci Ti,4 (1) − Ti,3 (x) + − Ti,2 (1) ,(3.18)
2 6
i=1

+1   x2 x
u r+1 (x) = g0 (t+1 ) + g1 (t+1 ) − g0 (t+1 ) x + − h 0 (t+1 )
2 2
x3 x
+ − (h 1 (t+1 ) − h 0 (t+1 ))
6 6
2M  x3 x 
− ci xTi,4 (1) − Ti,4 (x) + − Ti,2 (1) . (3.19)
6 6
i=1

Now, substituting Eqs. (3.15)–(3.19) into the Eq. (3.11), we get

2M 1 1   1
ci i (x) + μ + H xTi,2 (1) − Ti,2 (x) − f (wr )x Ti,4 (1) − Ti,3 (x)
2 2 2 +1
i=1
x2 1   1  x3 x 
+ − Ti,2 (1) − G + f wr xTi,4 (1) − Ti,4 (x) + − Ti,2 (1)
2 6 2 +1 6 6
   μ
1 γ  
= k(x, t+1 ) + k(x, t ) + G u r+1 − +1
Θη u r−η+1 +1
− u r−η + (u r+1 )x x
2 2
η=1
1 H H b + b  r +1
− (u r+1 )x x x x + (u r+1 )x x + (δ + 1) C − −1 u x x (x, t0 )
2 2 2 1+δ
−1 
H     
+ bη − bη−1 u rx+1
x (x, t−η ) + bη − bη−1 u rx+1
x (x, t−η+1 )
2
η=1 η=1
1   1   1 1
− f u r+1 , (u r+1 )x − f u r+1 , (u r+1 )x + u r+1 f ur + (u r+1 )x f (ur )x
2 2 2 +1 2 +1
1    
+ μ + H h 0 (t+1 ) + x h 1 (t+1 ) − h 0 (t+1 ) − g0 (t+1 ) + x g1 (t+1 )
2
 x2 x x3 x  
−g0 (t+1 ) + − h 0 (t+1 ) + − h 1 (t+1 ) − h 0 (t+1 )
2 2 6 6
1  1 1
G + f ur − g1 (t+1 ) − g0 (t+1 ) + (x − )h 0 (t+1 )
2 +1 2 2
x2 1  
+ − h 1 (t+1 ) − h 0 (t+1 ) f (ur )x . (3.20)
2 6 +1

123
 M. Kumar, A. K. Verma

j−0.5
Now, inserting the collocation points x j , j = 1(1)2M, where x j = 2M , into the
Eq. (3.20), we obtain

AC = b, (3.21)

where A = [a j,i ]2M×2M is square matrix, C and b are column vectors of length 2M.
Also,

1 1   1
a j,i = i (x j ) + μ + H x j Ti,2 (1) − Ti,2 (x j ) − f (wr )x Ti,4 (1) − Ti,3 (x j )
2 2 2 +1

x 2j 1   1  x 3j xj  
+ − Ti,2 (1) − G + f wr x j Ti,4 (1) − Ti,4 (x j ) + − Ti,2 (1) ,
2 6 2 +1 6 6
(3.22)
  
1 γ  +1 +1 
bj = k(x j , t+1 ) + k(x j , t ) + G u r+1 − Θη u r−η+1 − u r−η
2
η=1
μ 1 H b + b  r +1
+ (u r+1 )x x − (u r+1 )x x x x + (δ + 1) C − −1 u x x (x j , t0 )
2 2 2 1+δ
−1 
H     
+ bη − bη−1 u rx+1
x (x j , t−η ) + bη − bη−1 u rx+1
x (x j , t−η+1 )
2
η=1 η=1
H 1   1   1
+ (u r+1 )x x − f u r+1 , (u r+1 )x − f u r+1 , (u r+1 )x + u r+1 f ur
2 2 2 2 +1
1 r 1   
+ (u +1 )x f (ur )x + μ + H h 0 (t+1 ) + x j h 1 (t+1 ) − h 0 (t+1 )
2 +1 2
  x 2j xj  x 3j xj 
− g0 (t+1 ) + x j g1 (t+1 ) − g0 (t+1 ) + − h 0 (t+1 ) + −
2 2 6 6
 1  1
(h 1 (t+1 ) − h 0 (t+1 )) G + f ur − g1 (t+1 ) − g0 (t+1 )
2 +1 2
1 
xj2
1
+(x j − )h 0 (t+1 ) + − (h 1 (t+1 ) − h 0 (t+1 )) f (ur )x . (3.23)
2 2 6 +1

At each time level, we solve the system (3.21) to obtain ci and determine the numerical
solution by substituting ci into Eq. (3.19).

4 Convergence analysis

Eq. (3.19) has the asymptotic expression given by

  x3 x  
u(x) = g0 (t+1 ) + g1 (t+1 ) − g0 (t+1 ) x + − h 1 (t+1 ) − h 0 (t+1 )
6 6
x 2 x  ∞  x3 x 
+ − h 0 (t+1 ) − ci xTi,4 (1) − Ti,4 (x) + − Ti,2 (1) .
2 2 6 6
i=1
(4.1)

123
A computationally efficient numerical scheme for the solution…

d4u
Theorem 1 Assume w(x) ∈ L 2 (R), where w(x) = dx4
is continuous and
dw(x)
∃ϑ : dx< ϑ, ∀x ∈ [0, 1]. Then E J (u) 2 = u E x. (x) − u N um. (x) 2 → 0 as
J → ∞. Also, the convergence is of order two.

Proof The proof is similar to the convergence result of [19].

5 Stability

Here, we address the stability of the current scheme. The matrix form of the Eqs.
(3.15)–(3.19) by inserting the collocation points are given by:

(U+1 )x x x x = Hs C+1 , (5.1)

(U+1 )x x x = Ps C+1 + (P̃s )+1 , (5.2)
(U+1 )x x = Qs C+1 + (Q̃s )+1 , (5.3)
(U+1 )x = Rs C+1 + (R̃s )+1 , (5.4)
U+1 = Zs C+1 + (Z̃s )+1 , (5.5)

where C+1 represents the column vector of the Haar coefficient at ( + 1)th time
level and the column vectors U+1 , (U+1 )x , (U+1 )x x , (U+1 )x x x , (U+1 )x x x x rep-
resent the approximated solutions and their derivatives. Hs = [h j,i ]2M×2M , Ps =
[ p j,i ]2M×2M , Qs = [q j,i ]2M×2M , Rs = [r j,i ]2M×2M , Zs = [z j,i ]2M×2M are the
interpolation matrix of (U+1 )x x x x , (U+1 )x x x , (U+1)x x , (U+1 )x , U+1 , respectively.
We have h j,i = i (x j ), p j,i = Ti,1 (x j ) − Ti,2 (1) , q j,i = Ti,2 (x j ) − x j Ti,2 (1),
 x2 
r j,i = Ti,3 (x j ) − Ti,4 (1) − 2j − 16 Ti,2 (1), z j,i = Ti,4 (x j ) − x j Ti,4 (1) −
 x 3j xj 
6 − 6 Ti,2 (1). The boundary terms are (Q̃s )+1 = [(q̃+1 ) j,i ]2M×1 , (P̃s )+1 =
[( p̃+1 ) j,i ]2M×1 , (R̃s )+1 = [(r̃+1 ) j,i ]2M×1 , (Z̃s)+1 = [(z̃ +1 ) j,i ]2M×1
 such that
( p̃+1 ) j,i = h 1 (t+1 ) − h 0 (t+1 ), (q̃+1 ) j,i = x j h 1 (t+1 ) − h 0 (t+1 ) + h 0 (t+1 ),
   x2  
(r̃+1 ) j,i = g1 (t+1 ) − g0 (t+1 ) + x j − 21 h 0 (t+1 ) + 2j − 16 h 1 (t+1 ) − h 0 (t+1 )
   x2 x   x3
and (z̃ +1 ) j,i = x j g1 (t+1 ) − g0 (t+1 ) + g0 (t+1 ) + 2j − 2j h 0 (t+1 ) + 6j −
xj 
6 (h 1 (t+1 ) − h 0 (t+1 )). Using Eqs. (5.1)–(5.5), the matrix form of the Eq. (3.11) is
given by

MC+1 = N C + K, (5.6)

where
1 1 1  1 
M= h j,i − (μ + H)q j,i + G + f ur+1 z j,i + f (ur+1 )x r j,i ,
2 2 2 2 2M×2M
1 1 1 
N = (μ + H)q j,i − h j,i + Gz j,i − X j,i ,
2 2 2 2M×2M

123
 M. Kumar, A. K. Verma

1  
 +1 +1 
K= k(x j , t+1 ) + k(x j , t ) − G Θηγ u r−η+1 − u r−η
2
η=1
H b−1 + b  r +1 1   1
+ (δ + 1) C − u x x (x j , t0 ) − f u r+1 , (u r+1 )x − X̃ j,i
2 1+δ 2 2
H  
  −1
  r +1 
+ bη − bη−1 u rx+1
x (x j , t −η+1 ) + bη − bη−1 u xx (x j , t −η )
2
η=1 η=1
1 1 1  1 
+ u r+1 f ur+1 + (u r+1 )x f (ur+1 )x + μ + H (q̃+1 ) j,i + μ + H (q̃ ) j,i
2 2 2 2
 1  1 
− G + f ur+1 (z̃ +1 ) j,i + G(z̃  ) j,i − f (ur+1 )x (r̃+1 ) j,i .
2 2 2M×1
⎧
⎪(c − 1)z j,i , for f (u, u x ) = u(u − 1),
⎨
X j,i = (c − 1)2 z j,i , for f (u, u x ) = −u(1 − u)(u − 1),
⎪
⎩
cr j,i , for f (u, u x ) = uu x ,
⎧
⎪
⎨(c − 1)(z̃  ) j,i , for f (u, u x ) = u(u − 1),
X̃ j,i = (c − 1)2 (z̃  ) j,i , for f (u, u x ) = −u(1 − u)(u − 1),
⎪
⎩
c(r̃ ) j,i , for f (u, u x ) = uu x ,
c = u r+1 is treated as constant locally.

Substituting Eq. (5.5) into the Eq. (5.6), we get

 
M Zs−1 U+1 − Zs−1 (Z̃s )+1 = N Zs−1 U − Zs−1 (Z̃s ) + K. (5.7)

Simplifying Eq. (5.7), we get

U+1 = Zs M−1 N Zs−1 U + (Z̃s )+1 − Zs M−1 N Zs−1 (Z̃s )

+Zs M−1 K. (5.8)

The relation (5.8) permits refinement in time for the present scheme. Let the exact
solution is denoted by U +1 , then

U +1 = Zs M−1 N Zs−1 U  + (Z̃s )+1 − Zs M−1 N Zs−1 (Z̃s )

+Zs M−1 K. (5.9)

Subtracting Eq. (5.8) from Eq. (5.9), we obtain

E+1 = T E (5.10)

where T = Zs M−1 N Zs−1 is the amplification matrix and E+1 = U +1 −U+1 is the
error. By Lax-Richtmyer criterion, the scheme will be stable if there exists a constant
K , independent of , Δt and Δx such that T  ≤ K , and any choice of matrix

123
A computationally efficient numerical scheme for the solution…

norm (since this only effects the constant K ). Let S be the non-singular matrix of eigen
vectors corresponding to different eigen values λi of the matrix T and therefore T S =
S D, where D = diag(λ1 , λ2 , ..., λ2M ). Hence, T = S DS −1 , T 2 = S DS −1 S DS −1 =
S D 2 S −1 and T  = S D  S −1 . Therefore, T  ≤ S D  S −1 . But 1, 2 and
∞ norm of the diagonal matrix D  = diag(λ1 , λ2 , ..., λ2M ) are max|λi |. Hence,
i
max|λi | ≤ 1 is the sufficient condition for T  ≤ K (see [24, 34]). For different test
i
examples, it has been computationally validated in Tables 4, 8, 12 and 16.

6 Numerical experiment

The current section focuses on testing the accuracy and efficiency of the developed
scheme. We provide maximum absolute error norm ( e ∞ ) and root mean square error
norm ( e 2 ), defined as


 2M
e ∞ = max | u Ej x. − u Nj um. |, e 2 = Δx | u Ej x. − u Nj um. |2 ,
j
j=0

where u E x. , u N um. denote the exact solution and numerical solution, respectively.
Furthermore, we compute the convergence rate by the formula [25]

log[E u (2M)/E u (4M)]

Rate of convergence (Rc ) = , (6.1)
log(2)

where E u (−) denotes any of the error norms. For the whole computations, we have
used iterations up to r = 3 for the quasilinearization. We perform the computations
with Mathematica 11.3 and list CPU timings which are in seconds. We have used a
personal computer with 64-bit Intel (R) Core(TM) i5-7200U CPU and 4.00 GB of
RAM.

6.1 Test example 1

We consider TFPIDE given in Eq. (1.1) with f = u(u − 1) and

2t 2−γ sin π x t 2+δ sin π x

k(x, t) = + μπ 2 t 2 sin π x + 2π 2 νd + t 2 π 4 sin π x
Γ (3 − γ ) Γ (3 + δ)
+t 4 sin2 π x − t 2 sin π x, (6.2)

then the function

u(x, t) = t 2 sin π x, (6.3)

represents the exact solution. By inserting t = 0 and x = 0, x = 1 into the Eq. (6.3),
we obtain IC and BCs, respectively.

123
 M. Kumar, A. K. Verma

Table 1 Error norms at T = 0.5 with distinct DFs for νd = 1, μ = 1, J = 5, Δt = 0.01 of the example 6.1

ϕ(Δt) δ = 3/4, γ = 1/4 δ = 1/2, γ = 1/2 δ = 1/4, γ = 3/4

e ∞ e 2 e ∞ e 2 e ∞ e 2

Δt 8.00539E-05 5.66243E-05 8.86458E-05 6.27016E-05 8.28826E-05 5.86251E-05

sinh Δt 8.00398E-05 5.66143E-05 8.86062E-05 6.26735E-05 8.28018E-05 5.85680E-05
tan Δt 8.00257E-05 5.66043E-05 8.85665E-05 6.26455E-05 8.27211E-05 5.81090E-05
e2π Δt −1 5.34316E-05 3.77936E-05 1.40542E-05 9.94094E-05 6.83751E-05 4.83635E-05
2π

Table 2 Error norms with Rc at T = 1 for distinct J , Δt = 0.001, νd = 1 = μ of the example 6.1

δ γ J e ∞ Rc e 2 Rc

3/4 1/4 1 2.03883E-02 1.56048E-02

2 5.40357E-03 1.91576 3.89588E-03 2.00197
3 1.37039E-03 1.97932 9.73743E-04 2.02233
4 3.44384E-04 1.99250 2.43820E-04 1.99772
1/2 1/2 1 2.00966E-02 1.53815E-02
2 5.31319E-03 1.91930 3.83072E-03 2.00551
3 1.33844E-03 1.98902 9.51038E-04 2.01004
4 3.27561E-04 2.03071 2.31910E-04 2.03593
1/4 3/4 1 1.97600E-02 1.51239E-02
2 5.20694E-03 1.92407 3.75412E-03 2.01028
3 1.29911E-03 2.00291 9.23092E-04 2.02392
4 3.05568E-04 2.08795 2.16338E-04 2.09318

In Table 1, we illustrate the accuracy of the method for different DFs and different
pairs of δ and γ with the parameters J = 5, Δt = 0.01, μ = 1, νd = 1 at T = 0.5.
Table 1 shows that the presented technique has better accuracy with ϕ(Δt) = (e2π Δt −
1)/2π in comparison to tan Δt, sinh Δt and Δt. For the remaining calculations of the
test example 6.1, we use ϕ(Δt) = (e2π Δt −1)/2π. In Table 2, we illustrate the accuracy
of the scheme for different pairs of δ and γ with Δt = 0.001, νd = 1, μ = 1 and for
distinct J at T = 1. We obtain that the accuracy increases when J is increased. Also,
Rc of the scheme is tending to two. In Table 3, we illustrate the convergence behaviour
of the method for different pairs of δ and γ and distinct Δt with μ = 1, J = 5, νd = 1
at T = 1. It has been observed from Table 4 that the spectral radius of T is smaller
than one. Figure 1 represents the convergence behaviour of the method at T = 0.5 for
different J and for the parameters νd = 1, μ = 1, Δt = 0.005 and for different pairs
of δ and γ . We see that the accuracy increases when J is increased. Figure 2 presents
the error norms in loglog scale to visualize the order of convergence of the method for
the problem 6.1.

123
A computationally efficient numerical scheme for the solution…

Table 3 Error norms at T = 1 for the example 6.1 with distinct step sizes Δt and the parameters J = 5,
μ = 1 = νd

Δt δ = 3/4, γ = 1/4 δ = 1/2, γ = 1/2 δ = 1/4, γ = 3/4 CPU time

e ∞ e 2 e ∞ e 2 e ∞ e 2

1/20 1.34806E-04 9.53552E-04 4.79064E-04 3.38866E-04 1.30753E-03 9.24880E-03 418.781

1/40 1.07684E-04 7.61701E-05 1.96125E-04 1.38728E-04 6.14395E-04 4.34592E-03 1087.55
1/80 9.63052E-05 6.81214E-05 5.46549E-05 3.86601E-05 2.64039E-04 1.86768E-04 3098.22

Table 4 Spectral radius for T at T = 1 with μ = 1 = νd , distinct J and Δt for the example 6.1

J δ = 3/4, γ = 1/4 δ = 1/2, γ = 1/2 δ = 1/4, γ = 3/4

Δt = 0.01 Δt = 0.001 Δt = 0.01 Δt = 0.001 Δt = 0.01 Δt = 0.001

2 0.999954 0.999204 0.999857 0.999545 0.999565 0.997517

3 0.999997 0.999995 0.999991 0.999971 0.999972 0.999844

Fig. 1 Absolute errors with δ = 4/5, γ = 1/5 (left) δ = 1/5, γ = 4/5 (right) of the example 6.1 for
distinct J and μ = 1 = νd , Δt = 0.005 at T = 0.5

Fig. 2 Loglog plots for e 2 (left) and e ∞ (right) error norms presented in Table 2 of the problem 6.1

123
 M. Kumar, A. K. Verma

Table 5 Error norms at T = 1 of the example 6.2 with distinct DFs for νd = 1, μ = 1, J = 4, Δt = 0.01

ϕ(Δt) δ = 3/4, γ = 1/4 δ = 1/2, γ = 1/2 δ = 1/4, γ = 3/4

e ∞ e 2 e ∞ e 2 e ∞ e 2

Δt 4.54311E-05 3.21801E-05 5.07840E-05 3.59708E-05 4.70658E-05 3.33380E-05

sinh Δt 4.54185E-05 3.21712E-05 5.07513E-05 3.59476E-05 4.70033E-05 3.32938E-05
tan Δt 4.54060E-05 3.21623E-05 5.07186E-05 3.59244E-05 4.69409E-05 3.32495E-05
eπ Δt −1 3.36125E-05 2.38115E-05 2.00001E-05 1.41729E-05 1.16213E-05 8.21671E-06
π

6.2 Test example 2

We examine TFPIDE given in Eq. (1.1) with f = −u(1 − u)(u − 1) and

6t 3−γ (1 − x) sin x  
k(x, t) = + μ(1 + t 3 ) (1 − x) sin x + 2 cos x
Γ (4 − γ )
 tδ 6t δ+3 
+νd (1 − x) sin x + 2 cos x + + (1 + t 3 )(1 − x) sin x
Γ (1 + δ) Γ (4 + δ)
2 
1 − (1 + t 3 )(1 − x) sin x + (1 + t 3 ) (1 − x) sin x + 4 cos x , (6.4)

then the exact solution is

u(x, t) = (1 + t 3 )(1 − x) sin x. (6.5)

The IC and BCs are given by inserting t = 0 and x = 0, x = 1 into the Eq. (6.5).
In Table 5, we illustrate the accuracy of the method for distinct DFs and for different
pairs of δ and γ with the parameters J = 4, Δt = 0.01, μ = 1, νd = 1 at T = 1.
Table 5 shows that the presented technique has better accuracy with ϕ(Δt) = (eπ Δt −
1)/π in comparison to tan Δt, sinh Δt and Δt. For the remaining calculations of
the test example 6.2, we use ϕ(Δt) = (eπ Δt − 1)/π. In Table 6, we describe the
accuracy of the scheme at T = 1 for several pairs of δ and γ with Δt = 0.001 and
νd = 1, μ = 1, and for distinct J . We obtain that the accuracy increases when J is
increased. Additionally, we observe that the first four rows do not exhibit the expected
spatial discretization order because the time discretization step size is not sufficiently
small. To reflect the true spatial convergence rate, the time step must be sufficiently
small so that it does not dominate the overall error. In Table 7, we demonstrate the
convergence behaviour of the method for different pairs of δ and γ and distinct Δt with
μ = 1, J = 5, νd = 1 at T = 1. It has been observed from Table 8 that spectral radius
of T is smaller than one. Figure 3 demonstrates the convergence behaviour of the
technique for different J at T = 0.5 with the parameters νd = 1, μ = 1, Δt = 0.005
and for different pairs of δ and γ . We see that the accuracy increases when J is
increased. Figure 4 presents the error norms in loglog scale to visualize the order of
convergence of the method for the problem 6.2.

123
A computationally efficient numerical scheme for the solution…

Table 6 Error norms with Rc at T = 1 of the example 6.2 for distinct J , Δt = 0.001, and νd = 1 = μ

δ γ J e ∞ Rc e 2 Rc

3/4 1/4 1 2.74522E-04 2.08991E-04

2 7.71979E-05 1.83029 5.56353E-05 1.90936
3 2.18406E-05 1.82154 1.55449E-05 1.83955
4 7.64359E-06 1.51469 5.42194E-06 1.51956
1/2 1/2 1 2.71239E-04 2.06481E-04
2 7.53852E-05 1.84721 5.43291E-05 1.92621
3 2.64822E-05 1.87990 1.45798E-05 1.89775
4 6.40456E-06 2.04785 4.54461E-06 1.68174
1/4 3/4 1 2.66613E-04 2.02945E-04
2 7.20811E-05 1.88705 5.19483E-05 1.96593
3 1.75914E-05 2.03474 1.25260E-05 2.05215
4 3.62272E-06 2.27972 2.57486E-06 2.28235

Table 7 Error norms at T = 1 for the example 6.2 with J = 5, μ = 1 = νd and distinct step sizes Δt

Δt δ = 3/4, γ = 1/4 δ = 1/2, γ = 1/2 δ = 1/4, γ = 3/4 CPU time

e ∞ e 2 e ∞ e 2 e ∞ e 2

1/20 1.50721E-04 1.06677E-04 7.00564E-05 4.96089E-05 1.05670E-04 7.47280E-05 607.76

1/40 7.41942E-05 5.25059E-05 3.71299E-05 2.62833E-05 4.68585E-05 3.31421E-05 1461.4
1/80 3.73435E-05 2.64265E-05 2.00291E-05 1.41765E-05 2.00831E-05 1.42044E-05 3915.5

Table 8 Spectral radius for T at T = 1 with μ = 1 = νd , distinct J and Δt for the example 6.2

J δ = 3/4, γ = 1/4 δ = 1/2, γ = 1/2 δ = 1/4, γ = 3/4

Δt = 0.01 Δt = 0.001 Δt = 0.01 Δt = 0.001 Δt = 0.01 Δt = 0.001

2 0.999955 0.999236 0.999859 0.999548 0.999564 0.997518

3 0.999997 0.999995 0.999991 0.999971 0.999972 0.999844

Fig. 3 Absolute errors of the example 6.2 for distinct J , Δt = 0.005 and μ = 1 = νd at T = 0.5 with
δ = 1/10, γ = 9/10 (left) and δ = 9/10, γ = 1/10 (right)

123
 M. Kumar, A. K. Verma

Fig. 4 Loglog plots for e 2 (left) and e ∞ (right) error norms presented in Table 6 of the problem 6.2

6.3 Test example 3

Now, we examine TFPIDE given in Eq. (1.1) with f = uu x and

t 1−γ sin π x t δ sin π x t δ+1 sin π x 

k(x, t) = + μπ 2 (t + 1) sin π x + π 2 νd +
Γ (2 − γ ) Γ (1 + δ) Γ (2 + δ)
+(t + 1)2 π sin π x cos π x + (t + 1)π 4 sin π x, (6.6)

then the exact solution is

u(x, t) = (t + 1) sin π x. (6.7)

The IC and BCs are given by inserting t = 0 and x = 0, x = 1 into the Eq. (6.7),
respectively.
In Table 9, we present the accuracy of the scheme for several DFs and for distinct
pairs of δ and γ with the parameters J = 4, Δt = 0.01, μ = 1, νd = 1 at T = 1.
Table 9 shows that the presented technique has better accuracy with ϕ(Δt) = (e2π Δt −
2

1)/2π in comparison to tan Δt, sinh Δt and Δt. For remaining calculations of the
2

test example 6.3, we use ϕ(Δt) = (e2π Δt − 1)/2π 2 . In Table 10, we describe the
2

accuracy of the scheme at T = 0.5 for different pairs of δ and γ for distinct J
and μ = 1, νd = 1, Δt = 0.001. We obtain that the accuracy increases when J is
increased. Also, Rc of the scheme is tending to two. In Table 11, we demonstrate the
convergence behaviour of the method for various combinations of δ and γ and for
distinct Δt with μ = 1, J = 5, νd = 1 at T = 1. It has been shown in the Table 12
that the spectral radius of T is smaller than one. Figure 5 demonstrates accuracy for
distinct J at T = 1 with the parameters νd = 1, μ = 1, Δt = 0.01 and for different
pairs of δ and γ . We see that the accuracy increases when J is increased. Figure 6
presents the error norms in loglog scale to visualize the order of convergence of the
method for the problem 6.3.

123
A computationally efficient numerical scheme for the solution…

Table 9 Error norms at T = 1 with νd = 1, μ = 1, J = 4, Δt = 0.01, different DFs for the example 6.3

ϕ(Δt) δ = 3/4, γ = 1/4 δ = 1/2, γ = 1/2 δ = 1/4, γ = 3/4

e ∞ e 2 e ∞ e 2 e ∞ e 2

Δt 7.02490E-04 4.97151E-04 6.87876E-04 4.86809E-04 6.74572E-04 4.77395E-04

sinh Δt 6.87795E-04 4.86752E-04 6.87795E-04 4.86752E-04 6.74455E-04 4.77312E-04
tan Δt 6.87715E-04 4.86695E-04 6.87715E-04 4.86695E-04 6.74338E-04 4.77230E-04
2
e2π Δt −1 4.68161E-04 3.31317E-04 2.15526E-04 1.52527E-04 3.52339E-06 2.49361E-06
2π 2

Table 10 Error norms with Rc at T = 0.5 of the example 6.3 for distinct J with Δt = 0.001, μ = 1 = νd

δ γ J e ∞ Rc e 2 Rc

3/4 1/4 1 3.06000E-02 2.33730E-02

2 8.09271E-03 1.91842 5.82774E-03 2.00383
3 2.04260E-03 1.98612 1.45051E-03 2.00637
4 5.04738E-04 2.01679 3.57235E-04 2.02161
1/2 1/2 1 3.02209E-02 2.30837E-02
2 7.96993E-03 1.99781 5.73969E-03 2.00782
3 1.99550E-03 2.06405 1.41710E-03 2.01802
4 4.77210E-04 2.06405 3.37752E-04 2.06890
1/4 3/4 1 2.98741E-02 2.28189E-02
2 7.85282E-03 1.92761 5.65538E-03 2.01253
3 1.94613E-03 2.01260 1.38201E-03 2.03285
4 5.04738E-04 1.94700 3.57235E-04 1.95182

Table 11 Error norms with μ = 1 = νd , J = 5 and distinct step sizes Δt at T = 1 for the example 6.3

Δt δ = 3/4, γ = 1/4 δ = 1/2, γ = 1/2 δ = 1/4, γ = 3/4 CPU time

e ∞ e 2 e ∞ e 2 e ∞ e 2

1/20 1.09794E-03 7.76461E-04 2.17542E-03 1.53845E-03 2.92661E-03 2.06969E-03 581.56

1/40 3.76473E-04 2.66241E-04 9.48525E-04 6.70795E-04 1.45201E-03 1.02686E-03 1423.2
1/80 8.53966E-05 6.03927E-05 3.95724E-04 2.79855E-04 6.64835E-04 4.70168E-04 3815.5

Table 12 Spectral radius for T at T = 0.5 with μ = 1 = νd , distinct J and Δt for the example 6.3

J δ = 3/4, γ = 1/4 δ = 1/2, γ = 1/2 δ = 1/4, γ = 3/4

Δt = 0.01 Δt = 0.001 Δt = 0.01 Δt = 0.001 Δt = 0.01 Δt = 0.001

2 0.999966 0.999928 0.999867 0.999550 0.999591 0.997532

3 0.999998 0.999995 0.999992 0.999972 0.999974 0.999845

123
 M. Kumar, A. K. Verma

Fig. 5 Absolute errors for distinct J at time T = 1 for δ = 2/5, γ = 3/5 (left) and δ = 3/5, γ = 2/5
(right) with μ = 1 = νd , Δt = 0.01 for the example 6.3

Fig. 6 Loglog plots for e 2 (left) and e ∞ (right) error norms presented in Table 10 of the problem 6.3

6.4 Test example 4

Now, to show the supremacy of the proposed method, we consider nonlinear TFPIDE
given in Eq. (1.1) with μ = 0, f = uu x and

2t γ t γ +1 2
k(x, t) = −(2 + t) sin 2π x + π sin 4π x 1 − −
Γ (γ + 1) Γ (γ + 2)
2t γ t γ +1 
+16π 4 sin 2π x 1 − − + 4νd π 2 t δ sin 2π x
Γ (γ + 1) Γ (γ + 2)
1 2t γ t γ +1 
− − , (6.8)
Γ (δ + 1) Γ (γ + δ + 1) Γ (γ + δ + 2)

then the exact solution is

2t γ t γ +1 
u(x, t) = 1 − − sin 2π x. (6.9)
Γ (γ + 1) Γ (γ + 2)

The IC and BCs are given by inserting t = 0 and x = 0, x = 1 into the Eq. (6.9),
respectively.

123
A computationally efficient numerical scheme for the solution…

Table 13 Error norms at T = 1 with νd = 1, J = 4, Δt = 0.01, different DFs for the example 6.4

ϕ(Δt) δ = 3/4, γ = 1/4 δ = 1/2, γ = 1/2 δ = 1/4, γ = 3/4

e ∞ e 2 e ∞ e 2 e ∞ e 2

Δt 2.98714E-03 2.12259E-03 3.08071E-03 2.18871E-03 2.81724E-03 2.00156E-03

sinh Δt 2.98764E-03 2.12258E-03 3.08069E-03 2.18870E-03 2.81722E-03 2.00154E-03
tan Δt 2.98763E-03 2.12258E-03 3.08067E-03 2.18869E-03 2.81720E-03 2.00152E-03
2
e2π Δt −1 2.92537E-03 2.10475E-03 2.99318E-03 2.12653E-03 2.68211E-03 1.90554E-03
2π 2

Table 14 Error norms at time T = 1 for different J and νd = 1, Δt = 0.002 of the problem 6.4

γ δ J e ∞ Rc e 2 Rc e ∞ [38] e 2 [38](Δt = 1/1024)

0.35 0.25 1 1.39193E-01 1.39063E-01 — —

2 4.78135E-02 1.54159 3.65817E-02 1.92654 2.23109E-01 —
3 1.27538E-02 1.90649 9.19320E-03 1.99248 5.34549E-02 3.77984E-02
4 3.22889E-03 1.98182 2.29398E-03 2.00271 1.32174E-02 9.34611E-03
0.50 0.50 1 1.35191E-01 1.35065E-01 — —
2 4.64855E-02 1.54014 3.55661E-02 1.92507 2.16504E-01 —
3 1.23993E-02 1.90652 8.93768E-03 1.99253 5.18561E-02 3.66678E-02
4 3.13528E-03 1.98359 2.22749E-03 2.00448 1.28126E-02 9.05986E-03

Table 15 Error norms at T = 0.5 for the example 6.4 with J = 5, νd = 1 and distinct step sizes Δt

Δt δ = 3/4, γ = 1/4 δ = 1/2, γ = 1/2 δ = 1/4, γ = 3/4

e ∞ e 2 e ∞ e 2 e ∞ e 2

1/10 7.92576E-04 5.61095E-04 7.94278E-04 5.62305E-04 1.26488E-03 8.95474E-04

1/20 3.19896E-04 2.26467E-04 2.71337E-04 1.92092E-04 2.25164E-04 1.59405E-04
1/40 8.03169E-05 5.68594E-05 1.77076E-04 1.25361E-04 1.58856E-04 1.12462E-05

In Table 13, we present the accuracy of the scheme for several DFs and for distinct
pairs of δ and γ with the parameters J = 4, Δt = 0.01, νd = 1 at T = 1. Table 13
shows that the presented technique has better accuracy with ϕ(Δt) = (e2π Δt −
2

1)/2π in comparison to tan Δt, sinh Δt and Δt. For remaining calculations of the
2

test example 6.4, we use ϕ(Δt) = (e2π Δt − 1)/2π 2 . In Table 14, we describe the
2

accuracy of the scheme and compare the results with the existing results at T = 1 for
different pairs of δ and γ , for distinct J and νd = 1, Δt = 0.002. We obtain that the
accuracy increases when J is increased. Also, the proposed method provides better
accuracy compare to difference scheme presented in [38]. Also, Rc of the scheme is
tending to two. In Table 15, we demonstrate the convergence behaviour of the method
for various combinations of δ and γ and for distinct Δt with J = 5, νd = 1 at T = 0.5.
It has been shown in the Table 16 that the spectral radius of T is smaller than one.

123
 M. Kumar, A. K. Verma

Table 16 Spectral radius for T at T = 1 with νd = 1, distinct J and Δt for the example 6.4

J δ = 3/4, γ = 1/4 δ = 1/2, γ = 1/2 δ = 1/4, γ = 3/4

Δt = 0.01 Δt = 0.002 Δt = 0.01 Δt = 0.002 Δt = 0.01 Δt = 0.002

2 0.999991 0.999957 0.999874 0.999686 0.999591 0.998541

3 0.999998 0.999997 0.999992 0.999981 0.999974 0.999908

7 Conclusion

In the present article, we have established a computational technique for the fourth-
order TFPIDE based on the Haar wavelet and NSFD L1 formula. The proposed method
for solving such types of TFPIDE is new in the literature which allows the selection
of a wide range of DFs. It has been observed that the DF of the form ϕ(Δt) =
(eλΔt − 1)/λ, λ is a parameter, provides better accuracy compare to Δt, sinh Δt,
tan Δt. The convergence and stability of the method are also discussed. The presented
scheme has second-order spatial convergence. Loglog plots of the error norms also
confirm the convergence order of the method. It has been noted that as the resolution
J increases, both e ∞ and e 2 error norms decrease, as expected. When time step
sizes get smaller, the accuracy of the scheme improves. For various combinations
of δ and γ , it has been reported that the Rc of the scheme tends to two. Numerical
demonstrations show that the current method for computing numerical solutions is
very effective and accurate.

Declarations
Conflict of interest The authors declare that they have no conflict of interest.

References
1. Amin, R., Nazir, S., García-Magariño, I.: Efficient sustainable algorithm for numerical solution of
nonlinear delay Fredholm-Volterra integral equations via Haar wavelet for dense sensor networks in
emerging telecommunications. Trans. Emerg. Telec. Techn. 33(2), e3877 (2020). https://doi.org/10.
1002/ett.3877
2. Ashpazzadeh, E., Han, B., Lakestani, M.: Biorthogonal multiwavelets on the interval for numerical
solutions of Burgers’ equation. J. Comput. Appl. Math. 317, 510–534 (2017). https://doi.org/10.1016/
j.cam.2016.11.045
3. Avazzadeh, Z., Heydari, M.: Haar wavelet method for solving nonlinear age-structured population
models. Int. J. Biomath. 10, 1750114 (2017). https://doi.org/10.1142/S1793524517501145
4. Avazzadeh, Z., Heydari, M., Chen, W., Loghmani, G.B.: Smooth solution of partial integro-differential
equations using radial basis functions. J. Appl. Anal. Comput. 4(2), 115–127 (2014). https://doi.org/
10.11948/2014005
5. Baleanu, D., Darzi, R., Agheli, B.: New study of weakly singular kernel fractional fourth-order partial
integro-differential equations based on the optimum q-homotopic analysis method. J. Comput. Appl.
Math. 320, 193–201 (2017). https://doi.org/10.1016/j.cam.2017.01.032
6. Bellman, R.E., Kalaba, R.E.: Quasilinearization and Nonlinear Boundary-Value Problems. American
Elsevier Publishing Company, New York (1965)

123
A computationally efficient numerical scheme for the solution…

7. Bulut, F., Oruç, Ö., Esen, A.: Higher order Haar wavelet method integrated with strang splitting for
solving regularized long wave equation. Math. Comput. Simul. 197, 277–290 (2022). https://doi.org/
10.1016/j.matcom.2022.02.006
8. Cattani, C.: Haar wavelets based technique in evolution problems. In: Proceedings-Estonian Academy
of Sciences Physics Mathematics, vol. 53(1), pp. 45–63. American Institute of Physics (2004)
9. Chen, C.F., Hsiao, C.H.: Haar wavelet method for solving lumped and distributed-parameter systems.
IEE Proc. Cont. Theory and App. 144(1), 87–94 (1997). https://doi.org/10.1049/ip-cta:19970702
10. Daubechles, I.: Orthonormal bases of compactly supported wavelets. Commun. Pure Appl. Math.
41(7), 909–996 (1988). https://doi.org/10.1002/cpa.3160410705
11. Fakhar-Izadi, F.: Fully spectral-Galerkin method for the one-and two-dimensional fourth-order time-
fractional partial integro-differential equations with a weakly singular kernel. Numer. Methods Partial
Differ. Equ. 38(2), 160–176 (2020). https://doi.org/10.1002/num.22634
12. Heydari, M.H., Hooshmandasl, M.R., Ghaini, F.M., Cattani, C.: Wavelets method for the time fractional
diffusion-wave equation. Phys. Lett. A 379(3), 71–76 (2015). https://doi.org/10.1016/j.physleta.2014.
11.012
13. Heydari, M., Avazzadeh, Z., Hosseinzadeh, N.: Haar wavelet method for solving high-order differential
equations with multi-point boundary conditions. J. Appl. Comput. Mech. 8(2), 528–544 (2022). https://
doi.org/10.22055/JACM.2020.31860.1935
14. Lepik, U.: Solving PDEs with the aid of two-dimensional Haar wavelets. Comput. Math. Appl. 61(7),
1873–1879 (2011). https://doi.org/10.1016/j.camwa.2011.02.016
15. Lepik, U., Hein, H.: Haar Wavelets: With Applications. Springer, Switzerland (2014)
16. Li, B., Chen, X.: Wavelet-based numerical analysis: a review and classification. Finite Elem. Anal.
Des. 81, 14–31 (2014). https://doi.org/10.1016/j.finel.2013.11.001
17. Linz, P.: Analytical and Numerical Methods for Volterra Equations. SIAM, Philadelphia (1985)
18. Liu, Y., Du, Y., Li, H., He, S., Gao, W.: Finite difference/finite element method for a nonlinear
time-fractional fourth-order reaction-diffusion problem. Comput. Math. Appl. 70(4), 573–591 (2015).
https://doi.org/10.1016/j.camwa.2015.05.015
19. Majak, J., Shvartsman, B.S., Kirs, M., Pohlak, M., Herranen, H.: Convergence theorem for the Haar
wavelet based discretization method. Mech. Compos. Mater. 126, 227–232 (2015). https://doi.org/10.
1016/j.compstruct.2015.02.050
20. Mehra, M.: Wavelets and differential equations: a short review. In: AIP Conference Proceedings, vol.
1146(1), pp. 241–252. American Institute of Physics (2009). https://doi.org/10.1063/1.3183545
21. Mickens, R.E.: Nonstandard Finite Difference Models of Differential Equations. World Scientific,
Singapore (1993)
22. Mickens, R.E.: Calculation of denominator functions for nonstandard finite difference schemes for
differential equations satisfying a positivity condition. Numer. Methods Partial Differ. Equ. 23(3),
672–691 (2007). https://doi.org/10.1002/num.20198
23. Miller, R.K.: An integro-differential equation for rigid heat conductors with memory. J. Math. Anal.
Appl. 66(2), 313–332 (1978). https://doi.org/10.1016/0022-247X(78)90234-2
24. Morton, K.W.: Stability of finite difference approximations to a diffusion-convection equation. Int. J.
Numer. Meth. Eng. 15(5), 5677–683 (1980). https://doi.org/10.1002/nme.1620150505
25. Pervaiz, N., Aziz, I.: Haar wavelet approximation for the solution of cubic nonlinear Schrodinger equa-
tions. Phys. A: Stat. Mech. Appl. 545, 123738 (2020). https://doi.org/10.1016/j.physa.2019.123738
26. Podlubny, I.: Fractional Differential Equations: An Introduction to Fractional Derivatives, Fractional
Differential Equations, to Methods of Their Solution and Some of Their Applications. Academic Press,
New York (1998)
27. Qiao, L., Wang, Z., Xu, D.: An alternating direction implicit orthogonal spline collocation method for
the two dimensional multi-term time fractional integro-differential equation. Appl. Numer. Math. 151,
199–212 (2020). https://doi.org/10.1016/j.apnum.2020.01.003
28. Qiu, W., Xu, D., Guo, J.: The Crank-Nicolson-type Sinc-Galerkin method for the fourth-order partial
integro-differential equation with a weakly singular kernel. Appl. Numer. Math. 159, 239–258 (2021).
https://doi.org/10.1016/j.apnum.2020.09.011
29. Ratas, M., Salupere, A., Majak, J.: Solving nonlinear PDEs using the higher order Haar wavelet method
on nonuniform and adaptive grids. Math. Model. Anal. 26(1), 147–169 (2021). https://doi.org/10.3846/
mma.2021.12920
30. Rawani, M.K., Verma, L., Verma, A.K., Agarwal, R.P.: On a weakly L-stable time integration formula
coupled with nonstandard finite difference scheme with application to nonlinear parabolic partial

123
 M. Kumar, A. K. Verma

differential equations. Math. Methods Appl. Sci. 45(3), 1276–1298 (2022). https://doi.org/10.1002/
mma.7853
31. Rawani, M.K., Verma, A.K., Cattani, C.: A novel hybrid approach for computing numerical solution of
the time-fractional nonlinear one and two-dimensional partial integro-differential equation. Commun.
Nonlinear Sci. Numer. Simul. 118, 106986 (2023). https://doi.org/10.1016/j.cnsns.2022.106986
32. Rawani, M.K., Verma, A.K., Verma, L.: Numerical treatment of Burgers’ equation based on weakly
L-stable generalized time integration formula with the NSFD scheme. Appl. Math. Comput. 467(2),
128485 (2024). https://doi.org/10.1016/j.amc.2023.128485
33. Shen, S., Liu, F., Chen, J., Turner, I., Anh, V.: Numerical techniques for the variable order time
fractional diffusion equation. Appl. Math. Comput. 218(22), 10861–10870 (2012). https://doi.org/10.
1016/j.amc.2012.04.047
34. Smith, G.D.: Numerical Solution of Partial Differential Equations: Finite Difference Methods. Oxford
University Press, New York (1985)
35. Swati, Singh, K., Verma, A.K., Singh, M.: Higher order Emden-Fowler type equations via uniform
Haar Wavelet resolution technique. J. Comput. Appl. Math. 376, 112836 (2020). https://doi.org/10.
1016/j.cam.2020.112836
36. Tang, T.: A finite difference scheme for partial integro-differential equations with a weakly singular
kernel. Appl. Numer. Math. 11(4), 309–319 (1993). https://doi.org/10.1016/0168-9274(93)90012-G
37. Tariq, H., Akram, G.: Quintic spline technique for time fractional fourth-order partial differential
equation. Numer. Methods Partial Differ. Equ. 33(2), 445–466 (2017). https://doi.org/10.1002/num.
22088
38. Tian, Q., Zhang, H., Yang, X., Jiang, X.: An implicit difference scheme for the fourth-order nonlinear
non-local PIDEs with a weakly singular kernel. Comput. Appl. Math. 41(7), 328 (2022). https://doi.
org/10.1007/s40314-022-02040-9
39. Ur Rehman, M., Khan, R.A.: Numerical solutions to initial and boundary value problems for linear
fractional partial differential equations. Appl. Math. Model. 37(7), 5233–5244 (2013). https://doi.org/
10.1016/j.apm.2012.10.045
40. Verma, A.K., Kayenat, S.: On the stability of Micken’s type NSFD schemes for generalized Burgers’
Fisher equation. J. Differ. Equ. Appl. 25(12), 1706–1737 (2019). https://doi.org/10.1002/mma.7853
41. Verma, A.K., Rawani, M.K.: Numerical solutions of generalized Rosenau-KDV-RLW equation by
using Haar wavelet collocation approach coupled with nonstandard finite difference scheme and quasi-
linearization. Numer. Methods Partial Differ. Equ. 39(2), 1085–1107 (2023). https://doi.org/10.1002/
num.22925
42. Verma, A.K., Rawani, M.K., Agarwal, R.P.: A novel approach to compute the numerical solution of
variable coefficient fractional Burgers’ equation with delay. J. Appl. Comput. Mech. 7(3), 1550–1564
(2021). https://doi.org/10.1515/fca-2017-0017
43. Verma, A.K., Rawani, M.K., Cattani, C.: A numerical scheme for a class of generalized Burgers’
equation based on Haar wavelet nonstandard finite difference method. Appl. Numer. Math. 168, 41–54
(2021). https://doi.org/10.1016/j.apnum.2021.05.019
44. Wu, J.L.: A wavelet operational method for solving fractional partial differential equations numerically.
Appl. Math. Comput. 214(1), 31–40 (2009). https://doi.org/10.1016/j.amc.2009.03.066
45. Xu, D., Qiu, W., Guo, J.: A compact finite difference scheme for the fourth-order time-fractional
integro-differential equation with a weakly singular kernel. Numer. Methods Partial Differ. Equ. 36(2),
439–458 (2020). https://doi.org/10.1002/num.22436
46. Yang, X., Xu, D., Zhang, H.: Quasi-wavelet based numerical method for fourth-order partial integro-
differential equations with a weakly singular kernel. Int. J. Comput. Math. 88(15), 3236–3254 (2011).
https://doi.org/10.1080/00207160.2011.587003
47. Yang, X., Xu, D., Zhang, H.: Crank-Nicolson/quasi-wavelets method for solving fourth order partial
integro-differential equation with a weakly singular kernel. J. Comput. Phys. 234, 317–329 (2013).
https://doi.org/10.1016/j.jcp.2012.09.037
48. Zhang, H., Han, X., Yang, X.: Quintic B-spline collocation method for fourth order partial integro-
differential equations with a weakly singular kernel. Appl. Math. Comput. 219(12), 6565–6575 (2013).
https://doi.org/10.1016/j.amc.2013.01.012
49. Zhang, Y.: A finite difference method for fractional partial differential equation. Appl. Math. Comput.
215(2), 524–529 (2009). https://doi.org/10.1016/j.amc.2009.05.018
50. Zhao, Z., Jin, X.Q., Lin, M.M.: Preconditioned iterative methods for space-time fractional advection-
diffusion equations. J. Comput. Phys. 319, 266–279 (2016). https://doi.org/10.1016/j.jcp.2016.05.021

123
A computationally efficient numerical scheme for the solution…

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations.

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under
a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted
manuscript version of this article is solely governed by the terms of such publishing agreement and applicable
law.

123