Numerical Heat Transfer, Part B: Fundamentals

An International Journal of Computation and Methodology

ISSN: 1040-7790 (Print) 1521-0626 (Online) Journal homepage: https://www.tandfonline.com/loi/unhb20

Keller-Box shooting method and its application to

nanofluid flow over convectively heated sheet with
stability and convergence

Yasir Nawaz & Muhammad Shoaib Arif

To cite this article: Yasir Nawaz & Muhammad Shoaib Arif (2019) Keller-Box shooting
method and its application to nanofluid flow over convectively heated sheet with stability
and convergence, Numerical Heat Transfer, Part B: Fundamentals, 76:3, 152-180, DOI:
10.1080/10407790.2019.1644924

To link to this article: https://doi.org/10.1080/10407790.2019.1644924

Published online: 31 Jul 2019.

Submit your article to this journal

Article views: 139

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=unhb20
NUMERICAL HEAT TRANSFER, PART B: FUNDAMENTALS
2019, VOL. 76, NO. 3, 152–180
https://doi.org/10.1080/10407790.2019.1644924

Keller-Box shooting method and its application to nanofluid

flow over convectively heated sheet with stability and
convergence
Yasir Nawaz and Muhammad Shoaib Arif
Department of Mathematics, Air University Islamabad, Pakistan

ABSTRACT ARTICLE HISTORY

The main aim of the present contribution is to present a coupling of Received 17 April 2019
Keller-Box method using Jacobi and Gauss–Seidel iterative methods with Accepted 12 July 2019
shooting approach and also this method is implemented on nanofluid
flow problem. Present method of Keller-Box shooting method can be con-
sidered as an explicit approach whereas the standard Keller-Box method is
an implicit method. Previously constructed nanofluid flow models are
extended with exothermic/endothermic chemical reactions and the flow is
considered over stretching, rotating, porous disk which is convectively
heated from its bottom with the hot liquid. Similarity transformations are
utilized to reduce the set of nonlinear partial differential equations into
nonlinear ordinary differential equations with an assumption. Presently
modified Keller-Box shooting method using Jacobi and Gauss–Seidel itera-
tive methods is applied to investigate MHD nanofluid flow problem subject
to Dirichlet, Neumann, and Robin’s boundary conditions. Von Neumann
stability criterion is adopted to check the stability of the present method
using Gauss–Seidel iterative method. The bounds for maximum errors are
found for Jacobi and Gauss–Seidel iterative methods and convergence con-
ditions are given for the case of Gauss–Seidel iterative method. The
obtained results from presently developed modified Keller-Box shooting
method are in good agreement with those obtained by Matlab built in
solver “bvp4c” in case of wall temperature gradient and obtained results
are in good agreement with those obtained by Runge–Kutta (4, 5) shoot-
ing method in case of skin friction coefficient.

1. Introduction
Partial differential equations arise in mathematical modeling of particular type of physical phe-
nomena. The mathematical modeling of the fluid flow phenomena coupled with heat and mass
transfer gives the set of partial differential equations. The modeling of fluid flow problems can be
performed by Navier Stokes equations that may contain both the convective and diffusion parts.
The convective part of Navier Stokes equations is nonlinear if acceleration is described as material
derivative which can be formed into nonlinear terms consisting of velocities along the coordinate
axis and derivative of velocities with respect to space variables. The fluid flow problem combined
with heat and mass transfer can be studied with different forces, heat sources and chemical

CONTACT Yasir Nawaz [email protected], [email protected] Department of Mathematics, Air University

Islamabad, PAF Complex E-9, Islamabad 44000, Pakistan.
Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/unhb.
This article was originally published with errors, which have now been corrected in the online version. Please see Correction
(http://dx.doi.org/10.1080/10407790.2020.1717911)
ß 2019 Taylor & Francis Group, LLC
 NUMERICAL HEAT TRANSFER, PART B: FUNDAMENTALS 153

reactions. Present case of the heat and mass transfer deals with exothermic and endothermic
chemical reactions. The exothermic chemical reactions are reactions in which energy realizes and
endothermic chemical reaction are those chemical reactions which absorbs energy. The energy
can be used or can be gained from construction and destruction of the bonds. Abdul Maleque [1]
studied the flow problem combined with heat and mass transfer using exothermic and endother-
mic chemical reactions. A similarity technique has also been used for reducing the nonlinear par-
tial differential equations into set of linear and nonlinear ordinary differential equations. The
solutions of the system of differential equations were found using Nachtsheim-Swigert [2] tech-
nique in which Runge–Kutta of order six methods has been utilized after finding the missing ini-
tial conditions. Swalmeh et al. [3] applied the Keller-Box for finding the solutions of boundary
value problems. The problem was studied with nanoparticles Copper (Cu) and alumina (Al2O3)
in water base Micro-polar fluid using natural convection on solid sphere. Mustafa [4] applied the
Keller-Box method and optimal homotopy analysis method for axisymmetric nanofluid flow over
nonlinearly radially stretching sheet with the use of zero mass flux boundary condition at the
wall and found that nanoparticles volume fraction is negative near to the sheet. It is also men-
tioned in his study that the equations due to linearly stretching sheet can be obtained from non-
linearly stretching sheet with particular value of the parameter. It can also be noted that the cor-
rect discretization procedure is mentioned in the work of Mustafa [4] for discretizing the ordin-
ary differential equations using Keller-Box method. The work for the flow problems using
convectively heated sheets can be seen in [5–7] whereas flow problems over rotating disks can be
seen in [8–10]. To consider the flow on convectively heated sheet, the sheet is placed on the hot
fluid and the fluid on top of the disk is heated from the bottom fluid of the disk.
Modified Arrhenius function is a temperature dependent chemical reaction and this function
has been used in [11,12] with the flow problems. The temperature dependent chemical reaction
can be studied with the mass transfer equation and the behavior of the concentration profile or
nanoparticles volume fraction profile can be discussed with the variations of the different parame-
ters used in the Arrhenius function. The constant of proportionality which is used in the
Modified Arrhenius function is considered due to the increase in temperature that yields increase
in the rate of reaction. The applications of the modified Arrhenius function are geothermal or oil
reservoir engineering, coating of metallic objects and glasses. Modified Arrhenius reaction with
nanoparticles can be studied with the nanofluid flows and the nanofluid flows are those flows in
which the nano-sized particles are added in the fluid to enhance the thermal properties of the
base fluid.
Different kind of numerical methods can be implemented for solving the flow problems. From
these, the stability of the explicit scheme is discussed in [13] for studying the unsteady nanofluid
flow under nonlinear thermal radiation subject to bidirectional stretching rotational oscillating
sheet with cubic autocatalysis chemical reaction effects. The model of the flow problem using
continuity, momentum, temperature, and concentration equations has been presented under
some assumptions. The stability procedure is made using Von Neuman stability criteria. The Von
Neuman stability criteria with trivial manipulations yielded the system in matrix-vector form and
the conditions on eigen values of the matrix gave the stability conditions. It should be noted that
the stability conditions were found for the system of nonlinear partial differential equations.
In [14], generalized finite difference method was adopted for solving second-order parabolic
partial differential equations and convergence conditions have been found using the condition on
common ratio of geometric series for sum of infinite number of terms. The convergence condi-
tions using explicit and implicit schemes have been found for the nonlinear partial differential
equations. Also, the consistency of the scheme has also been presented in their work which can
be implied from convergence of the sum of infinite terms of geometric series.
Among the numerical techniques for solving problems, the Keller-Box method is an implicit
method for which set of differential equations are reduced to the system of first-order differential
154 Y. NAWAZ AND M. SHOAIB ARIF

equations. The equations are discretized using central differences and in the next step, the discre-
tized difference equations are linearized using Newton’s method and solving the tri-diagonal
block matrices system gives the solutions of the set of differential equations. The method is
second order accurate and unconditionally stable. Application of standard Keller-Box can be seen
in the work done in Cebeci and Bradshaw [15]. In [16], an improved finite difference method is
proposed with the development of compact correction term and it is constructed by the coupling
of two numerical methods. Some more work on time-fractional diffusion equation on non-uni-
form grids and coupling problem in incompressible fluid flow can be seen in [17] and [18],
respectively.
Similar to the standard Keller-Box method, present method requires the reduction of the system of
differential equations into system of first-order differential equations. The discretization of the system
of the first-order differential equations is performed by the standard Keller-Box method. The differ-
ence equation can be directly solved by Jacobi or Gauss–Seidel iterative methods with shooting
approach. So, instead of applying implicit Keller-Box method, kind of explicit Keller-Box method can
be applied for solving numerically ordinary as well as partial differential equations. The discretized
equations can be linearized using quasi-linearization and then Jacobi or Gauss–Seidel iterative meth-
ods can be implemented. To apply the Jacobi or Gauss–Seidel iterative method, initial conditions are
required for solving each first differential equation using Jacobi or Gauss–Seidel iterative methods. So,
to fulfill this requirement, shooting method can be employed for finding missing initial conditions.
The present procedure of applying the Keller-Box shooting method is to implement the Matlab built-
in solver “fsolve” for finding the missing initial conditions. Matlab solver “fsolve” utilizes the optimiza-
tion algorithms for solving equations. So the present method of solving differential equations needs
two iterative methods the one for solving equations or missing initial conditions and other for solving
difference equations. Results obtained by present Keller-Box shooting method is more similar to
Matlab solver “bvp4c” then Runge–Kutta shooting method, for the case when the three methods are
applied for finding the numerical values of wall temperature gradient with variations of different
parameters. The Runge–Kutta shooting method is formed using two Matlab solvers in which one
solver is used for finding the missing initial conditions and it can be called as “fsovle” and other
Matlab solver is “ode45” which can be considered for solving the differential equations using
Runge–Kutta method (4, 5) method.
Present work deals with the discretization procedure using Keller-Box shooting method and
also procedure of the stability is presented when Gauss–Seidel iterative method is implemented
after discretization of nonlinear differential equations using standard Keller-Box method. The sta-
bility is performed using Von Neumann stability analysis and linearization process of nonlinear
differential equations into considered linear differential equations is performed using quasi-linear-
ization procedure. The bounds for maximum errors are found using Jacobi and Gauss–Seidel
iterative method and convergence conditions are obtained using Gauss–Seidel iterative method.
Ten first-order differential equations are discretized, stability is shown for the system of ten equa-
tions. First six eigen values are came out to be zero and the last four eigen values are nonzero.
The cases for stable and unstable solutions are shown using graphs in Cartesian and polar coord-
inate axis.

2. Problem formulation
Present problem is concerned with the incompressible laminar nanofluid flow with heat and mass
transfer. The coordinate system is cylindrical with rh
plane horizontal and z-axis is perpendicu-
lar to the plane. Let the stretching rate of the disk is a and rotational velocity is X and stretching
and rotating of the disk are responsible to generate the flow. Let u be the component of velocity
in r direction, v be component of velocity in h direction and w be the component of velocity in
z-direction. The flow is axisymmetric and the fluid is placed on the porous disk in z>0 space.
 NUMERICAL HEAT TRANSFER, PART B: FUNDAMENTALS 155

The disk is rotating, stretching in its axial direction and it is convectively heating below with the
fluid placed on other side of the disk having constant temperature Tf . The flow is subjected
through the magnetic field and chemically reaction with activation energy. Effect of the chemical
reaction on temperature of the fluid is also considered through exothermic/endothermic chemical
reactions. The governing equations of the present flow problem can be expressed as
@u u @w
þ þ ¼0 (1)
@r
r @z 
@u v2 @u 1 @P @ 2 u 1 @u u @2u tf rB20
u
þw ¼
þ tf þ
þ
u
u (2)
@r r @z q @r @r2 r @r r2 @z2 K qf


@v uv @v @ 2 v 1 @v v @2v tf rB20
u þ þ w ¼ tf þ
þ
v
v (3)
@r r @z @r2 r @r r2 @z2 K qf


@w @w 1 @P @ 2 w 1 @w @ 2 w
u þw ¼
þ tf þ þ (4)
@r @z q @z @r2 r @r @z2



2
2 !!
@T @T @ 2 T 1 @T @ 2 T ðqcÞp @T @C @T @C DT @T @T
u þ w ¼ am þ þ þ DB þ þ þ
@r @z @r2 r @r @z2 ð Þf
qc @r @r @z @z T 1 @r @z

m

T
Ea
þ bk2r ðC
C1 Þ exp (5)
T1 jT



@C @C @ 2 C 1 @C @ 2 C DT @ 2 T 1 @T @ 2 T
u þ w ¼ DB þ þ þ þ þ
@r @z @r2 r @r @z2 T @r2 r @r @z2

m
1  (6)
T
Ea

kr ðC
C1 Þ
2
exp
T1 jT
Subject to the below mentioned boundary conditions
@T  
u ¼ Uw ¼ ar; v ¼ rX; w ¼ 0; kf ¼ hf T
Tf ; C ¼ Cf when z ! 0 (7)
@z
u ! 0; v ! 0; w ¼ 0; T ! 0; C ! 0whenz ! 1 (8)
where T represents the temperature of the fluid, C represents the nanoparticles volume fraction,
b is used for exothermic/endothermic chemical reaction. When b ¼ 1 reaction is exothermic and
for b ¼
1 the reaction is endothermic, the parameter Ea is used for activation energy, j is the
Bolzmann constant, k2r is the reaction rate, DB is the Brownian diffusion coefficient, DT represents
the thermophoretic diffusion coefficient, tf represents the density of the fluid, K  is the perme-
able parameter, a represents the thermal diffusivity, kf represents the thermal conductivity and hf
represents the convective heat transfer coefficient.
Utilizing of the below mentioned similarity transformations [8] for Eqs. (1)–(6) with boundary
conditions (7)–(8)
pffiffiffiffiffiffiffiffiffiffi
u ¼ rXf 0 ðgÞ; v ¼ rXg ðgÞ; w ¼
2Xtf f ðgÞ; P
P1 ¼
2qtf XpðgÞ
sffiffiffiffiffiffi
T
T1 C
C1 2X (9)
h¼ ;/¼ ;g¼ z;
Tf
T1 Cf
C1 tf

reduces the partial differential Eqs. (2)–(6) to the following ordinary differential equations.
1 2 1
f 000 þ ff 00
f 0 þ g 2
ðM þ K Þf 0 ¼ 0 (10)
2 2
g 00 þ fg 0
f 0 g
ðM þ K Þg ¼ 0 (11)
p0
f 00
ff 0 ¼ 0 (12)
156 Y. NAWAZ AND M. SHOAIB ARIF

 
00 0 0 0 02 E
n
h þ Pr f h þ Nbh / þ Nth þ brð1 þ dhÞ exp
/ ¼0 (13)
1 þ dh


Nt 00 E
/00 þ Scf /0 þ
n
h þ Scrð1 þ dhÞ exp
/¼0 (14)
Nb 1 þ dh
Eqs. (10)–(14) are reduced ordinary differential equations with the assumption that the ratio
between difference in nanoparticle volume fraction and difference in temperature is a constant
number 1, that is
Cf
C1
¼1 (15)
Tf
T1
The following boundary conditions can be achieved after utilization of transformations (9) on
the boundary conditions (7)–(8) corresponding to the partial differential Eqs. (1)–(6)
f ¼ 0; f 0 ¼ c; g ¼ 1; h0 ¼
cð1
hÞ; Nb/0 þ Nth0 ¼ 0 when g ¼ 0 (16)
f 0 ! 0; g ! 0; p ! 0; h ! 0; / ! 0 when g ! 1 (17)
where Pr represents the Prandtl number, Sc represents the Schmidt number, c represents the
stretching-strength parameter, Nb and Nt represent the Brownian motion and thermophoresis
parameters, respectively, M represents the Hartman number, K represents the permeability par-
ameter, c represents the Biot number, r represents the reaction rate parameter, E represents the
activation energy parameter, d represents the temperature difference parameter, n represents the
fitted rate constant and these parameters are defined as

rf B20 tf s ðqcÞp DB ðCf
C1 Þ ðqcÞp DT ðTf
T1 Þ

M¼ ; K ¼  ; c ¼ ; Nb ¼ ; Nt ¼ ;
2qf X 2K X X ðqcÞf tf ðqcÞf T1 tf
rffiffiffiffiffiffi
hf tf k2 Tf
T1 Ea tf tf
c¼ ;r¼ r;d¼ ;E¼ ; Pr ¼ ; Sc ¼
kf 2X a T1 jT1 af DB

The skin friction coefficient and local Nusselt number which are the quantities of physical interest
are defined below [19]
Skin-friction coefficient
!
1=2 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Xr2 2 2
Cf ¼ f 00 ð0Þ þ g 0 ð0Þ (18)
2tf

Local Nusselt number

sffiffiffiffiffiffiffiffiffiffi
2Xr2 0
Nur ¼
h ð0 Þ (19)
tf

3. Keller-Box shooting method

Let the nth order differential equation
 


uðmÞ ¼ h g; u; u0 ; :::; uðm
1Þ ; where g 2 X (20)
Subject to the n boundary conditions
 
B u; u0 ; :::; uðm
1Þ ¼ 0 where g 2 C (21)
 NUMERICAL HEAT TRANSFER, PART B: FUNDAMENTALS 157

where the domain is represented by the symbol X
and boundary of the domain is represented by
the symbol C. The boundary operator B can be a function of uðmÞ which is the case of second
order velocity slip in fluid flows for the value of m ¼ 3 [20].
The procedure of the Keller-Box method is required to reduce the nth order differential
Eq. (20) into system of first order differential equation, which are mentioned below
u0 ¼ u1 ; u0 1 ¼ u2 ; :::; u0 n
1 ¼ h1 ðg; u; u1 ; :::; un
1 Þ (22)
Discretize the Eq. (22) using standard Keller-Box method and apply the Jacobi method that
gives
9
ukþ1
uki
1 >
>
i
¼ ðu1 Þi
1=2
k
>
>
h >
>
>
>
ðu1 Þi
ðu1 Þi
1
kþ1 k >
>
¼ ðu2 Þi
1=2
k =
h (23)
.. >
>
>
>
. >
>
  >
>
kþ1
ð n
1 Þi
u
k
ð n
1 Þi
1
u >
>
¼ h2 gi
1=2 ; ui
1=2 ; ðu1 Þi
1=2 ; :::; ðun
1 Þi
1=2 ;
k k k k
h
Equation (23) are treated explicitly due to only one unknown in each equation. To precede the
present method with explicit procedure, some of the conditions are assumed and their values are
found using the Matlab solver “fsolve”. The Matlab solver “fsolve” can be replaced with any algo-
rithm which can be used for finding the missing initial conditions.

4. Application
To apply standard Keller-Box method to discretize the Eqs. (10)–(14), it is required to reduce the
Eqs.(10)–(14) into system of first-order differential equations
9
f 0 ¼ f1 ; >
>
=
f 0 1 ¼ f2 ; (24)
1 1 >
>
f 0 2 þ ff2
f12 þ g 2
ðM þ K Þf1 ¼ 0 ;
2 2
g 0 ¼ g1
(25)
g 0 1 þ fg1
f1 g
ðM þ K Þg ¼ 0
p0
f2
ff1 ¼ 0 (26)
0 9
h ¼ h1 =


dh1   n E (27)
ð Þ
þ Pr f h1 þ Nbh1 Ø1 þ Nth1 þ br 1 þ dh exp

2
Ø ¼ 0;
dg 1 þ dh
9
/0 ¼ /1 =


0 Nt 0 E (28)
/ ¼ 0;
n
/ 1 þ Scf /1 þ h 1
Scrð1 þ dhÞ exp

Nb 1 þ dh
Present procedure of applying Keller-Box shooting method is an explicit one that requires the ini-
tial conditions. In this procedure, one Dirichlet type initial condition for each equation in the system
of Eqs. (24)–(28) is required. So, there should be ten Dirichlet type initial conditions corresponding to
each first-order differential equation in (24)–(28). Since, three of the given initial conditions are
Dirichlet type, so remaining seven initial conditions are assumed further in present contribution and
to this task, Matlab solver “fsolve” will be considered for finding missing initial conditions.
158 Y. NAWAZ AND M. SHOAIB ARIF

The Eqs. (24)–(28) are discretized now using standard Keller-Box method with Jacobi iterative
method subject to the given and assumed initial conditions
9
fikþ1
fik
1 1  k k

=
þ ðf1 Þi þ ðf1 Þi
1 ¼ 0
h 2 (29)
;
f0 ¼ 0
kþ1
9
ðf1 Þikþ1
ðf1 Þki
1 1  k k
 >
=
þ ðf 2 Þi þ ðf 2 Þi
1 ¼ 0
h 2 (30)
>
;
ðf1 Þ0kþ1 ¼ c
ðf2 Þkþ1
ðf2 Þki
1 1  k   1 k 2 1  2 9
>
i k k
þ fi þ fi
1 ðf2 Þi þ ðf2 Þi
1
k
ðf1 Þi þ ðf1 Þi
1 þ gi þ gi
1 >
k k k >
>
>
h 4 8 8 =
1  

ðM þ K Þ ðf 1 Þi þ ðf 1 Þi
1 ¼ 0
k k
>
>
2 >
>
kþ1 >
;
ðf 2 Þ0 ¼ x 1
(31)
9
gikþ1
gik
1 1  
=
þ ðg1 Þi þ ðg1 Þi
1 ¼ 0
k k
h 2 (32)
;
g0kþ1 ¼ 1 9
ðg1 Þi
ðg1 Þi
1 1  k
kþ1 k        =
1 1
þ fi þ fik
1 ðg1 Þki þ ðg1 Þki
1
ðf1 Þki þ ðf1 Þki
1 gik þ gik
1
ðM þ K Þ gik þ gik
1 ¼ 0
h 4 4 2
;
ðg1 Þ0 ¼ x2
kþ1

(33)
9
pikþ1
pki
1 1  1  
=

ðf2 Þki þ ðf2 Þki
1
fik þ fik
1 ðf1 Þki þ ðf1 Þki
1 ¼ 0
h 2 4 (34)
;
p0kþ1 ¼ x3 9
  >
hkþ1
hki
1 1 k k =
i
þ ðh1 Þi þ ðh1 Þi
1 ¼ 0
h 2 (35)
>
;
h0kþ1 ¼ x4

9
       >
ðh1 Þkþ1
ðh1 Þi
1 Pr  k
k
k k k k k k >
>
i
þ fi þ fik
1 ðh1 Þi þ ðh1 Þi
1 þ Nb ðh1 Þi þ ðh1 Þi
1 ð/1 Þki þ ð/1 Þki
1 þ Nt ðh1 Þi þ ðh1 Þi
1 >
>
h 4 >
>
  n 0 1   >
=
E 1
þbr 1 þ d 2 hi þ hi
1
1 k k
exp @
 A /i þ /i
1 ¼ 0
k k
1 2 >
>
1 þ hki þ hki
1 >
>
2 >
>
>
>
kþ1 ;
ðh1 Þ0 ¼ x5
(36)
9
  >
/ikþ1 1
/ki =
ð/1 Þki þ ð/1 Þki
1 ¼ 0
þ
h2 (37)
>
;
/0kþ1 ¼ x6
  9
ð/1 Þikþ1
ð/1 Þki
1 Sc  k
k k
Nt ðh1 Þi
ðh1 Þi
1 >
>
þ k k
f þ fi
1 ð/1 Þi þ ð/1 Þi
1 þ
k
>
>
h 4 i Nb h >
>
  n 0 1   >
=
E 1
þScr 1 þ d 2 hi þ hi
1
1 k k
exp @
 A /i þ /i
1 ¼ 0
k k
(38)
1 2 >
>
1 þ hki þ hki
1 >
>
2 >
>
>
;
kþ1
ð/ 1 Þ0 ¼ x 7
 NUMERICAL HEAT TRANSFER, PART B: FUNDAMENTALS 159

Superscripts of above used symbols represent the iteration number for Jacobi iteration method and
subscripts of each symbol are used to represent the present and past nodes. Since the initial conditions
for each equation is not given, so the first iteration will be started using the assumed conditions which
will be considered as an initial guess. The initial guess is given for finding the values of unknown initial
conditions and these values are found using the Matlab solver “fsolve.” The constraints for finding the
unknown values are the boundary conditions given in Eqs. (16) and (17).
The iterations are started using Jacobi/Gauss–Seidel iterative method(s) that needs initial
guesses and Dirichlet type initial conditions on each first node of each equation. The value of
each dependent variable on each proceeded node will be found explicitly using Jacobi/
Gauss–Seidel iterative method. The solution on each node will be updated except at first node
until the stopping criterion is met. The stopping criterion used to stop the iterations of Jacobi/
Gauss–Seidel iteration method is

ukþ1 2

uk 2
< (39)

where u can be considered as the vector for each dependent variable having the length equal to
the total number of nodes. When the stopping condition is met, the iterations for Jacobi/
Gauss–Seidel iterative method will be stopped and the values of each entry of the following vector
will be found for the last iteration of Jacobi iterative method.
h   i
kþ1 kþ1
fgkþ1
max
; g ;
kþ1 kþ1
p
gmax gmax ; ð h Þ
1 0 þ c 1
h kþ1
0 ; hkþ1
gmax ; Nbð / Þ
1 0
kþ1
þ Nt ð h Þ
1 0 ; /kþ1
gmax
(40)

where gmax represents the right end point of the boundary and subscript “0” represents first start-
ing point of the grid, that is, at zero in the present case. The Matlab solver “fsolve” will get the
constraints of vector (40) which should be satisfied in the sense that each entry of the vector (40)
should be approximately or exactly equal to zero. The iterations of Matlab solver “fsolve” will
continue and assumed initial conditions will be updated and for each updated initial condition,
Keller-Box method using Jacobi/Gauss–Seidel iterative method will find the solutions, then the
constraint vector (40) will be updated and Matlab solver “fsolve” will update the assumed initial
conditions. The iterations for Jacobi/Gauss–Seidel iterative method will remain continue until the
stopping criterion is met and the whole iterative procedure from start to till end of Jacobi/
Gauss–Seidel iterative method will be run every time when Matlab solver “fsolve” will compute
its each iteration. The Matlab solver “fsolve” will be stopped if the default or given criterion of
the solver is met.

5. Stability
Stability procedure of the Keller-Box shooting method is started from the reduction of the non-
linear differential equations into linear differential equations and this task is performed using
Quasi-linearization technique applied to Eqs. (24)–(28). Some equations are already linear, so
those are not required to linearize and the linearization process is applied to nonlinear equations.
So apply the Quasi-Linearization technique to nonlinear equations in Eqs. (24)–(28) results the
following equations
df2   1 2 1
þ f f 2 þ f f2
f 1 þ M þ K f1 þ g g ¼
f 1 þ g 2 þ f 2 f (41)
dg 2 2
dg1  
þ f g 1 þ f g1
g f1
f 1 þ M þ K g ¼
f g 1
f 1 g (42)
dg
dp

f2
f f1
f f 1 ¼
f 1 f (43)
dg
160 Y. NAWAZ AND M. SHOAIB ARIF

1 dh1
þ f h 1 þ f h1 þ NbØ 1 h1 þ 2Nth 1 h1 þ bQØ þ Nbh 1 Ø1 þ bd1 Øh
Pr dg (44)
2
þ f h 1 þ NbØ 1 h 1 þ bd1 Øh
¼ Nth 1
dØ1   Nt  
þ Ø 1 f þf Ø1
QØ
Ød1 h
Pr f h 1 þf h1 þNbØ 1 h1 þ2Nth 1 h1 þbQØþNbh 1 Ø1 þbd1 Ø 1 h
dg Nb
  Nt  2

¼Sc Ø 1 f
ØQh þ Pr Nth 1 þf h 1 þNbØ 1 h 1 þbd1 Øh
Nb
(45)

where


n E
Q¼ rð1þdh Þ exp

1þdh
nQd QE
d1 ¼ þ
1þdh ð1þdh Þ2
The total number of considered equations are ten, and this system of linear and linearized equa-
tions form a system that can be expressed as the matrix-vector equation with consideration of
only homogenous systems.
du
¼Au (46)
dx
Where u ¼ ½ f ; f1 ; f2 ; g; g1 ; p; h; h1 ; /; /1 t
And A is a 10x10 matrix with the following entries:
A12 ¼ A23 ¼ A45 ¼ A78 ¼ A910 ¼ 1; (47)
A31 ¼ −f 2 ; A32 ¼ f 1 þ M þ K; A33 ¼ −f ; A34 ¼ −g ; (48)
A51 ¼ −g 1 ; A52 ¼ g ; A54 ¼ f 1 þ M þ K; A55 ¼ −f ; (49)
A61 ¼ f 1 ; A62 ¼ f ; A63 ¼ 1; (50)
 
A81 ¼ −Prh 1 ; A87 ¼ −bPrd1 /; A88 ¼ −Pr f þ Nb/ 1 þ 2Nth 1 ;
(51)
A89 ¼ −PrQ; A810 ¼ −PrNbh 1 ;

Nt Nt
A101 ¼ −/ 1 þ Prh 1 ; A107 ¼ /d1 þ Prbd1 /;
Nb Nb (52)
Nt   Nt
A108 ¼ Pr f þ Nb/ 1 þ 2Nth 1 ; A109 ¼ Q þ PrbQ; A1010 ¼ −f þ NtPrh 1
Nb Nb
and remaining entries of the matrix A are zero. Where


n E
ð Þ
Q ¼ r 1 þ dh exp −
1 þ dh
nQd QE
d1 ¼ þ
1 þ dh ð1 þ dh Þ2
By applying the Keller-Box method using the Gauss-Seidel iterative method, the resulting
equation is
 NUMERICAL HEAT TRANSFER, PART B: FUNDAMENTALS 161

uikþ1 −ukþ1 1 
i−1
¼ A uki þ ukþ1
i−1 (53)
h 2
kþ1 −ði−1ÞIw
By substituting uikþ1 ¼ U kþ1 e−iIw , ukþ1
i−1 ¼ U e , uki ¼ U k e−iIw into Eq. (53), the result is
the following equation:

U kþ1 e−iIw −U kþ1 e−ði−1ÞIw 1 

¼ A U k e−iIw þ U kþ1 e−ði−1ÞIw (54)
h 2
Dividing by e−iIw on both sides of the last equation gives

U kþ1 −U kþ1 e−Iw 1

¼ A ðU k þ U kþ1 e−Iw Þ (55)
h 2
This implies


−Iw h −Iw h
I:D−e I:D−A e U kþ1 ¼ A U k (56)
2 2
Where I:D is an identity matrix of order 10x10. The last equation implies

−1
U kþ1 −Iw h −Iw h
G ¼ k ¼ I:D−e I:D−A e A (57)
U 2 2
where G is an amplification matrix. Using the stability condition, the following inequality can be
constructed:

−1
−Iw h h
1−e −kmax e−Iw kmax 1
2 2

Where kmax is maximum eigen value of the matrix A. This implies

h h
jkmax j  1−e−Iw −kmax e−Iw
2 2
h
 2 þ jkmax j
2
This implies
h h
jkmax j  2 þ jkmax j
2 2
The last inequality implies
02
Thus the Keller-Box method using Gauss-Seidel iterative method is unconditionally stable for
the considered ordinary differential equations
Residuals of Eqs. (10)–(14) can be found by finding one extra derivative from the numerical
available solutions of each equation. Since the discretization procedure of each equation can pro-
vide the numerical derivatives that are one less the order of the differential Eqs. (10)–(14). So, for
the requirement of one extra derivative for finding the numerical residuals of the Eqs. (10)–(14),
second order forward, backward and central differencing formulas are implemented on the solu-
tions obtained by applying the Keller-Box shooting method. The procedure of finding the final
residual of all the Eqs. (10)–(14) is briefly described as under.
162 Y. NAWAZ AND M. SHOAIB ARIF

Figure 1. Comparison of Keller-Box Shooting Method (K.B.S.M) with Matlab solver “bvp4c” and Maximum absolute
of residuals.

Consider the following system of first-order equations at the ith location

9
fi0 ¼ f1i >
>
=
f1i0 ¼ f2i (58)
1 1 >
>
f2i0 þ f2i0 þ fi f2i
f1i2 þ gi2
ðM þ K Þf1i ¼ 0 ;
2 2
The third-equation contains an unknown term f20 that cannot be found from the previous two
equations, but the available data from the third equation from (58) can be used to find the
numerical derivative. Therefore, at the ith grid point, the residual of Eq. (10) can be found in the
following manner
1 2 1 2
R1 ¼ f2i0 þ fi f2i
f þ g
ðM þ K Þf1i (59)
2 1i 2 i
Similarly, one extra derivative of each Eqs. (11)–(14) can be found by using second-order for-
ward and backward difference formulas at the boundary nodes and second order central differ-
encing formula at the central nodes. The final residual can be determined by choosing maximum
absolute value of the found residuals values of the five equations, i:e:
Res ¼ maxjRl j for l ¼ 1; 2; 3; 4; 5 (60)
Figure 1 shows the norm of the maximum of the absolute of the residual error (60) over the
number of iterations for each Eqs. (29)–(38) using Gauss–Seidel iterative method in the first fig-
ure from right. The first figure from left in Figure 1 shows the comparison of residuals (60) from
the present coupling scheme of Keller-Box shooting method and Matlab solver “bvp4c” using 100
grid points in ½0; 9. Residuals of Eqs. (10)–(14) are calculated by finding one more derivative for
each dependent variable of each equation. Second order accurate forward and backward formulae
have been implemented in both ends of the domain and central difference formula has been uti-
lized for the central nodes for Keller-Box shooting method. Fifth-order forward and backward
differencing formulas are implemented at the end nodes and sixth-order central differencing
 NUMERICAL HEAT TRANSFER, PART B: FUNDAMENTALS 163

formulas are implemented in the central grid points to find more accurate residuals using Matlab
solver bvp4c.

6. Error analysis

Theorem 6.1. The maximum error attained by the Keller-Box shooting method using Jacobi itera-
tive method for Eqs. (29)–(38) is bounded subject to the assumption of the differentiability of the
followings functions,
 
H 1 ¼ H 1 ðf ; f1 ; f2 ; g Þ; H2 ¼ H2 f; f 1 ; g; g1 ; H3 ¼ H3 ðf; f 1 ; f 2 Þ;
H4 ¼ H4 ðf; h; h1 ; Ø; Ø1 Þ; H 5 ¼ H 5 ðf ; h; h1 ; Ø; Ø1 Þ

Proof. To prove the maximum point wise error of the present scheme for the Eqs. (29)–(38), choose
the first three equations for finding the bound for maximum error. Discretize Eq. (24) using Keller-
box method with Jacobi iterative method gives
fikþ1
fik
1 ðf1 Þkþ1
ðf1 Þki
1
þ ðf1 Þki
1 = ¼ 0; i
þ ðf2 Þki
1 = ¼ 0;
h 2 h 2

kþ1 k
ðf 2 Þi
ðf 2 Þi
1
þ ðH1 Þki
1 =2 ¼ 0
h
and let the exact schemes are given as
fiE
fiE
1 ðf1 ÞEi
ðf1 ÞEi
1
þ ðf1 ÞEi
1 = ¼ 0; þ ðf2 ÞEi
1 = ¼ 0;
h 2 h 2

ðf2 ÞEi
ðf2 ÞEi
1

þ ðH 1 ÞEi
1 =2 ¼ 0
h
Let the error between exact and approximate solutions of any point on the grid is given by
k E k E
ðe1 Þi ¼ fi
fi ; ðe2 Þi ¼ ðf1 Þi
ðf1 Þi ; ðe3 Þi ¼ ðf2 Þi
ðf2 Þi ; ðe4 Þi ¼ gi
gi
k k E k k k k E

By applying the Mean value theorem on the function H 1 gives

   
k
H 1 fik ; ðf1 Þki ; ðf2 Þki ; gik
H 1 fiE ; ðf1 ÞEi ; ðf2 ÞEi ; giE ¼ ð^e 1 Þi :rH 1 ðc1 ; c2 ; c3 ; c4 Þ

where c1 ¼ fik þ e1 ðe3 Þki ; c2 ¼ ðhf1 Þki þ e2 ðe3 Þki ; c3 ¼ ðf2iÞki þ e3 ðe3 Þki ; c4 ¼ gik þ e4 ðe3 Þki and
k
e1 ; e2 ; e3 ; e4 2 ½0; 1, and ð^e 1 Þi ¼ ðe1 Þki ; ðe2 Þki ; ðe3 Þki ; ðe4 Þki .
The convergence error equations can be described as
kþ1
ðe 1 Þi ¼ ðe1 Þki
1 þ hðe2 Þki
1 = (61)
2
kþ1
ðe 2 Þi ¼ ðe2 Þki
1 þ hðe3 Þki
1 = (62)
2
k
ðe 3 Þi
kþ1
¼ ðe3 Þki
1 þ hð^e 1 Þi
1 =2 :rH 1 (63)
The following inequalities can be implied from above Eqs. (61)–(63)

ðe 1 Þ i
kþ1
 ðe1 Þki
1 þ h ðe2 Þki
1 = (64)
2

ðe 2 Þi
kþ1
 ðe2 Þki
1 þ h ðe3 Þki
1 = (65)
2

k
kþ1
ðe 3 Þi  ðe3 Þki
1 þ h ð^e 1 Þi
1 =2 :rH 1 (66)
164 Y. NAWAZ AND M. SHOAIB ARIF
h i
1 2 3 4
Let rH 1 ¼ H 1 ; H 1 ; H 1 ; H 1 , then the inequality (66) can be expressed by
X
4
j
X
4
j
kþ1
ðe3 Þi  ðe3 Þki
1 þ h ðej Þi
1 =2 H 1  ðe3 Þi
1 þ h
k k k
ðej Þi
1 =2 H 1 (67)
j¼1 j¼1
h
Let ðe1 Þk ¼ maxi¼1;:::;N ðe1 Þki , ðe2 Þk ¼ maxi¼1;:::;N ðe2 Þki , ðe Þk ¼ max maxi¼1;:::;N ðe1 Þki ;
maxi¼1;:::;N ðe2 Þki ; maxi¼1;:::;N ðe3 Þki ; maxi¼1;:::;N ðe4 Þki  where N denotes the total number of nodes.
The inequalities (64), (65), and (67) can be expressed as
 1 Hð h 2 Þ
e1kþ1  ek1 þ hek2 þ M (68)
 2 Hð h 2 Þ
e2kþ1  ek2 þ hek3 þ M (69)
0 1
X4
e kþ1  @1 þ 4h H 1 Ae k þ M
j
 3 Hð h 2 Þ (70)
j¼1

Substitution of k ¼ 0 in last inequality

0 (70) results1the following inequality
X4
e 1  @1 þ 4h H 1 Ae 0 þ M
j
 3 Hð h 2 Þ (71)
j¼1

When k ¼ 1 in (70) gives

0 1 0 12
X
4 X
4
e 2  @1 þ 4h
j
 3 HðhÞ  @1 þ 4h
H 1 Ae 1 þ M H1 A e0
j

j¼1 j¼1
0 0 11
X
4
þ @1 þ @1 þ 4h  3 Hðh2 Þ
H 1 AAM
j

j¼1

Thus, the following inequality can be formed

0 1k 0 0 1 0 1k
1 1
X 4
B X4 X
4
C  ð 2Þ
e k  @1 þ 4h H 1 A e 0 þ @1 þ @1 þ 4h H 1 A þ    þ@1 þ 4h H 1 A AM
j j j
3H h
j¼1 j¼1 j¼1

(72)
Using the formula for the sum of the first k terms of geometric series yields
0 !k 1
B 1 þ 4h P4
0 1k
1C
j
B H1 C
X 4 B j¼1 C
e k  @1 þ 4h H1 A e0 þ B CM  ð 2Þ
j
B P4 C 3H h
j¼1 @ 4h H
j A
1
j¼1
0 1k 0 1
X
4 X
4
 @1 þ 4h
j
H1 A e þ exp @4kh
0 j
H1 AM3 Hðh2 Þ (73)
j¼1 j¼1
 NUMERICAL HEAT TRANSFER, PART B: FUNDAMENTALS 165

Using the last inequality (73) in inequality (68) yields

00 1k 0 1 1
B X 4 X
4
C  ð 2Þ
ek1  ð1 þ hÞ@@1 þ 4h H 1 A e 0 þ exp @4ðk
1Þh H 1 AM3 HðhÞA þ M
j j
1H h (74)
j¼1 j¼1

Similarly, by using of the inequality (73) in inequality (69) yields

00 1k 0 1 1
B X4 X4
C  ð 2Þ
ek2  ð1 þ hÞ@@1 þ 4h H 1 A e 0 þ exp @4ðk
1Þh H 1 AM3 HðhÞA þ M
j j
2H h (75)
j¼1 j¼1

Thus, the maximum errors bounds for the Eq. (24) using Keller-Box shooting with Jacobi iter-
ations have been found. Similarly the maximum error bound for Eqs. (25) and (26) can be
expressed as
00 1k 0 1 1
B X
4 X4
C
ek4  ð1 þ hÞ@@1 þ 4h H 2 A e þ exp @4ðk
1Þh H 2 AM5 HðhÞA þ M4 Hðh2 Þ
j 0 j

j¼1 j¼1

0 1k 0 1
X
4 X
4
e  @1 þ 4h H 2 A e þ exp @4ðk
1Þh H 2 AM5 Hðh2 Þ
k j 0 j

j¼1 j¼1
0 1k 0 1
X
3 X
3
e k  @1 þ 3h
j
H3 A e 0 þ exp @3ðk
1Þh j
H3 AM6 Hðh2 Þ
j¼1 j¼1

Since, second equation of Eq. (27) is nonlinear, so the error for nonlinear equation can be
proved to be bounded in the similar way. To get more detailed error bound for the nonlinear
equation, Newton method can be implemented to linearize any nonlinear term in the equation.
Instead of linearize the whole equation using Newton method or quasi-linearization technique,
linearize just one or more needed number of terms of the nonlinear equation and rest terms will
be the same as in the original equation. So to find the more detailed error bound which describes
the detailed error bound, linearize the nonlinear term in Eq. (27) which is fg1 . Apply the Newton
method to linearize the term in following manner
fg1 ¼ f g 1 þ f g1
f g 1 (76)
To add the error for f ; pick the first linearized term from Eq. (76) and the remaining terms
would become part of the differentiable function H 2 . So H  2 ¼ f g1
f g
gf1
ðM þ K Þg.
1
Similarly to add the effect of the term f1 on error bound for second equation in system of Eq.
(27), linearize the other nonlinear term using Newton method.
The error bounds for rest of the equations are similar and are given by
00 1k 0 1 1
B X
5 X
5
C
en7  ð1 þ hÞ@@1 þ 5h H 4 A e 0 þ exp @5kh H 4 AM8 HðhÞA þ M7 Hðh2 Þ
j j

j¼1 j¼1

0 1k 0 1
X
5 X
5
e  @1 þ 5h
k j
H4 A e þ exp @5kh
0 j
H4 AM8 Hðh2 Þ
j¼1 j¼1
166 Y. NAWAZ AND M. SHOAIB ARIF

00 1k 0 1 1
B X j
5 X j
5
C
en9  ð1 þ hÞ@@1 þ 5h H 5 A e 0 þ exp @5kh H 5 AM10 HðhÞA þ M9 Hðh2 Þ
j¼1 j¼1
0 1k 0 1
X
5 X
5
e k  @1 þ 5h H 5 A e 0 þ exp @5kh H 5 AM10 Hðh2 Þ
j j
䊏
j¼1 j¼1

Theorem 6.2. The maximum error attained by the Keller-Box shooting method using Gauss–Seidel
iterative method for Eqs. (24)–(28) is bounded subject to the assumption of differentiability of the follow-
ings functions,

H 1 ¼ H 1 ðf ; f1 ; f2 ; g Þ; H 2 ¼ H 2 ðf ; f1 ; g; g1 Þ; H 3 ¼ H 3 ðf ; f1 Þ; H 4 ¼ H 4 ðf ; h; h1 ; Ø; Ø1 Þ;

H 5 ¼ H 5 ðf ; h; h1 ; Ø; Ø1 Þ
and satisfaction of the following inequality
0 1
X
2 h i
j 2 @H 3 @H 3
@ H3
A > 0 where rH 3 ¼ ;
1 2
¼ H 3; H 3 :
j¼1
h @f @f1

Proof. To prove the maximum point wise error bounds for Eqs. (24)–(28) using present scheme
of Keller-Box shooting with Gauss–Seidel iterative method, choose the first three equations in the
first stage of the proof for finding the maximum error bounds.
Discretize Eq. (24) using Keller-box method with Gauss–Seidel iterative method gives
fikþ1
fikþ1

1 1  ðf1 Þikþ1
ðf1 Þikþ1

1 1 k 
þ ðf1 Þki þ ðf1 Þikþ1

1 ¼ 0; þ ð f 2 Þ i þ ð f 2 Þ kþ1
i
1 ¼ 0;
h 2 h 2
 
ðf2 Þkþ1
ðf2 Þikþ1

1 1
i
þ ðH 1 Þki þ ðH 1 Þikþ1
1 ¼ 0
h 2
and let the exact schemes are given by
fiE
fiE
1 1  E  ðf1 ÞEi
ðf1 ÞEi
1 1  E 
þ ðf1 Þi þ ðf1 ÞEi
1 ¼ 0; þ ðf2 Þi þ ðf2 ÞEi
1 ¼ 0;
h 2 h 2
E E
ðf2 Þi
ðf2 Þi
1 1  
þ ðH 1 ÞEi þ ðH 1 ÞEi
1 ¼ 0
h 2
Let the error between exact and approximate solutions of any point on the grid is given by
n E n E
ðe1 Þi ¼ fi
fi ; ðe2 Þi ¼ ðf1 Þi
ðf1 Þi ; ðe3 Þi ¼ ðf2 Þi
ðf2 Þi
n n E n n

By applying the Mean value theorem on the function H 1 gives

   
k
H 1 fik ; ðf1 Þki ; ðf2 Þki ; gik
H 1 fiE ; ðf1 ÞEi ; ðf2 ÞEi ; giE ¼ ð^e 1 Þi :rH 1 ðc1 ; c2 ; c3 ; c4 Þ

where c1 ¼ fik þ e1 ðe3 Þki ; c2 ¼ ðf1 Þki þ e2 ðe3 Þki ; c3 ¼ ðf2 Þki þ e3 ðe3 Þki ; c4 ¼ gik þ e4 ðe3 Þki and e1 ; e2 ;
h i
k
e3 ; e4 2 ½0; 1, and ð^e 1 Þi ¼ ðe1 Þki ; ðe2 Þki ; ðe3 Þki ; ðe4 Þki .
The convergence error equations using Gauss–Seidel iteration method can be described as
h 
ðe1 Þi
ðe1 Þi
1 ¼ ðe2 Þi þ ðe2 Þi
1
kþ1 kþ1 k kþ1
(77)
2
 NUMERICAL HEAT TRANSFER, PART B: FUNDAMENTALS 167

h 
ðe 2 Þi
kþ1

ðe2 Þikþ1

1
¼ ð e 3 Þ
k
i
þ ð e 3 Þ
kþ1
i
1
(78)
2
 
h k kþ1
ðe 3 Þi
kþ1
¼ ðe3 Þikþ1

1
þ ð^e Þ
1 i þ ð^e Þ
3 i
1 :rH 1 (79)
2
Followings inequalities can be derived by using Eqs. (77)–(79)


2
ðe 2 Þi
1  ðe 2 Þi þ ðe1 Þi
1 þ ðe1 Þi
nþ1 n nþ1 nþ1
(80)
h


2
ðe 3 Þi
1  ðe 3 Þi þ ðe2 Þi
1 þ ðe2 Þi
nþ1 n nþ1 nþ1
(81)
h
0 1
X4 X4
 B ej Þki
1 H 1 C
h j j
kþ1
ðe3 Þi þ ðe3 Þkþ1i
1 @ ð ej Þkþ1 H1
ð A (82)
2 j¼1
i
j¼1

Let ðe1 Þk ¼ maxi¼1;:::;N ðe1 Þki , ðe2 Þk ¼ maxi¼1;:::;N ðe2 Þki , ðe3 Þk ¼ maxi¼1;:::;N ðe3 Þki ,

e k ¼ max max ðe1 Þki ; max ðe2 Þki ; max ðe3 Þki ; max ðe4 Þki
i¼1;:::;N i¼1;:::;N i¼1;:::;N i¼1;:::;N

and
h i
@H 1 @H 1 @H 1 @H 1 1 2 3 4
rH 1 ¼ ; ; ; ¼ H 1; H 1; H 1; H 1
@f @f1 @f2 @g
where N is the symbol used for the total number of nodes.
Therefore, the inequalities (80)–(82) imply that
4
ðe 2 Þ
nþ1
 ðe2 Þn þ ðe1 Þnþ1 þ M1 Hðh2 Þ (83)
h
4
ðe 3 Þ
nþ1
 ðe3 Þn þ ðe2 Þnþ1 þ M2 Hðh2 Þ (84)
0 h 1
h X 4 X4
 @ e H 1 A þ M3 Hðh2 Þ
j j
2e kþ1
e kþ1
H1
k
(85)
2 j¼1 j¼1

By using one of the first two inequalities (83) or (84) gives

h
e kþ1  e k þ MHðh2 Þ (86)
h
4
where M ¼ M1 or M2 . By combining the inequalities (85) and (86), it follows that
0 1
2h X4 X 4
@ H 1 Ae k  2he kþ1 H 1 þ MHðh2 Þ
j j
þ 2h (87)
h
4 j¼1 j¼1

This implies
0 1
1 1 X
4
@ þ
j
^ ðh2 Þ  e kþ1
H 1 Ae k þ MH (88)
P
4
j h
4 j¼1
H1
j¼1

!
P
4
j
Let a ¼ P
4
1 1
h
4 þ H1
jH j1 j j¼1
j¼1
168 Y. NAWAZ AND M. SHOAIB ARIF

Then, the inequality (88) becomes for k ¼ 0

^ ðh2 Þ
e 1  ae 0 þ MH (89)
For k ¼ 1,
^ ðh2 Þ  a2 e 0 þ ða þ 1ÞMH
e 2  ae 1 þ MH ^ ðh2 Þ (90)
If this is continue in the same manner, the following inequality can be formed
^ ðh2 Þ
e k  ak e 0 þ ðak
1 þ ::: þ a þ 1ÞMH (91)
Thus for any finite n 2 W
By using the formula for sum of first k terms of geometric series gives


ak
1 ^
e a e þ
k k 0
MHðh2 Þ (92)
a
1
Similarly the bound for the maximum error for the next two Eq. (25) is given by
!
bk
1 
e b e þ
k k 0
MHðh2 Þ (93)
b
1


P4 j
where b ¼ P3 1 1
þ j¼1 H 2
j¼1
jH j2 j h
4
The discretization of the Eq. (26) using Keller-Box with Gauss–Seidel iterative method is given by,

 
pikþ1
pkþ1
i
1 1 kþ1 kþ1 1 kþ1 kþ1

ðf 2 Þi þ ðf 2 Þi
1 þ ðH 3 Þi
1 þ ðH 3 Þi ¼0 (94)
h 2 2

The exact scheme is given as under


pEi
pEi
1 1 1

ðf2 ÞEi þ ðf2 ÞEi
1 þ ðH 3 ÞEi
1 þ ðH 3 ÞEi ¼0 (95)
h 2 2
The error equations using Mean value theorem can be formed as under
ðe6 Þikþ1
ðe6 Þikþ1

1 1 kþ1 k


ðe3 Þi þ ðe3 Þi
1 þ ðe 6 Þi
1 rH 3 þ ðe 6 Þi rH 3 ¼ 0
kþ1 kþ1
(96)
h h 2 i h 1 2i
where e 6 ¼ ½e1 ; e2 ; rH 3 ¼ @H 3 @H 3
@f ; @f1 ¼ H 3 ; H 3

1X 2
j 1X 2
j 1 2
ð ej Þikþ1
ej k H  ðe3 Þikþ1 þ ðe3 Þikþ1 þ ðe6 Þkþ1 þ M6 Hðh2 Þ
2 j¼1 ð Þi 3

1
H 3
1 (97)
2 j¼1 2 h

1 1 1X 2
j 1X 2
j 2
)
kþ1
ðe3 Þi
ðe3 Þikþ1 þ ð ej Þikþ1  ej k H þ ðe6 Þkþ1 þ M6 Hðh2 Þ
2 j¼1 ð Þi 3

1
1
H 3
2 2 2 j¼1 h

1X 2
kþ1 j 1 1 1X 2
j 2
) ð ej Þ

H 3  ð e3 Þ
kþ1
i
þ ð e 3 Þ
kþ1
i
1
þ ð ej Þki H 3 þ ðe6 Þkþ1 þ M6 Hðh2 Þ
2 j¼1 i 1 2 2 2 j¼1 h

1X 2
j 1X 2
j 2
) ð ej Þikþ1  ð Þ
kþ1
þ ej k H þ ðe6 Þkþ1 þ M6 Hðh2 Þ
2 j¼1 ð Þi 3

1
H 3 e3 (98)
2 j¼1 h
k
e ¼ max max ðe1 Þki ; max ðe2 Þki ; max ðe6 Þki
i¼1;:::;N i¼1;:::;N i¼1;:::;N
 NUMERICAL HEAT TRANSFER, PART B: FUNDAMENTALS 169

0 1
X
2
j 2 kþ1 X2
j
@ H 3
Ae H 3 þ M6 Hðh2 Þ
k
 e kþ1 þ e (99)
j¼1
h j¼1


P2 j
Let b2 ¼ j¼1 H3
2
h > 0, then the last inequality implies
1 kþ1 1 k X 2
j M6
Hðh2 Þ
kþ1
e e þ e H3 þ
b2 b2 j¼1 b2
!
akþ1 0 akþ1
1 ^ 1 kX 2
j M6
MHðh2 Þ þ e Hðh2 Þ;
kþ1
e  e þ H3 þ
b2 b2 ða
1Þ b2 j¼1 b2
!
akþ1 0 akþ1
1 1 kX 2
j
 MHðh2 Þ þ e
kþ1
e  e þ H3 ; (100)
b2 b 2 ða
1 Þ b2 j¼1

The similar procedure that was applied to prove the bound of first three equations can be
applied to find the maximum upper bound of error for the system of equations constructed from
the equation of heat transfer equation and the bound can be expressed as
!
ck

1  ðh2 Þ
e  cke þ
k 0
MH (101)
c
1

where e k ¼ max maxi¼1;:::;N ðe1 Þki ; maxi¼1;:::;N ðe7 Þki ;maxi¼1;:::;N ðe8 Þki ; maxi¼1;:::;N ðe9 Þki ; maxi¼1;:::;N


P5 j
ðe10 Þi  and c ¼ P3 j j j h
4 þ 2 j¼1 H 4 .
k 2 2h 5h
5h j¼1
H4

Similarly, the bound for maximum error for last two equations can be derived and the bound
is given by !
lk

1  ðh2 Þ
e k  lke 0 þ MH (102)

 l
1
P5 j
where l ¼ P32 j h2h
4 þ 2
5h
j¼1 H 5 : 䊏
5h
j¼1
j H 5j

Lemma 6.3. The absolute error for initial pressure is bounded and its bound is given by
0 1

 
h ð 2N
1Þ h X2
akþ1
1 ^
nþ1
ðe6 Þ1
kþ1
 ð~e 3 Þfsolve þ @ þ 2h H3
j
A a e þ
kþ1 0
MH hð 2Þ
2 2 j¼1
a
1

under the same assumptions given in Theorem 6.2 and

0 1
1 1 X4
a¼P @ þ H 1 A:
j
4
H
j h
4 j¼1
j¼1 1

Proof. To prove error bound for an initial pressure on the first node, consider the following dif-
ference equation for pressure
ðe6 Þkþ1
ðe6 Þikþ1

1 1 kþ1 k


ðe3 Þi þ ðe3 Þi
1 þ ð^e 6 Þi
1 :rH 3 þ ð^e 6 Þi :rH 3 ¼ 0
i kþ1 kþ1
h 2
h h h kþ1 k
) ðe 6 Þi
1
ðe 6 Þi
kþ1 kþ1

ðe3 Þkþ1i

ðe3 Þikþ1
1
 ð^e 6 Þi
1 :rH 3 þ ð^e 6 Þi :rH 3
2 2 2
170 Y. NAWAZ AND M. SHOAIB ARIF

h h kþ1 h X2
j hX 2
j
) ðe6 Þkþ1
ðe6 Þikþ1
ðe3 Þi
1  e3 þ
kþ1
ð ej Þikþ1 þ ej kþ1 H 3
2 j¼1 ð Þi
i
1
1
H 3
2 2 2 j¼1
kþ1
Let e ¼ max maxi¼2;:::;N ðe1 Þkþ1
i
1
; maxi¼2;:::;N ðe2 Þikþ1

1

kþ1
By using e in above inequality yields
h h kþ1 kþ1
X2
j
ðe6 Þi
1
ðe6 Þi
 þ H3 ;
kþ1 kþ1 kþ1
ð 3 Þi
1
e e 2he (103)
2 2 3 j¼1

Substituting i ¼ 1 in inequality (103) yields,

h h kþ1
X2
j
kþ1
ðe 6 Þ 0
ðe6 Þ1kþ1
kþ1
ðe3 Þ0  e3kþ1 þ 2he H3 ;
2 2 j¼1

when i ¼ 2 in inequality (103) gives

h h kþ1
X2
j
kþ1
ðe 6 Þ 1
ðe6 Þ2kþ1
ðe3 Þ2
kþ1
 e3kþ1 þ 2he H3
2 2 j¼1
..
.
And so on, for i ¼ N, where N denotes the Nth node in the domain, the inequality becomes
h h kþ1 kþ1
X2
j
ðe6 ÞN
1
ðe6 ÞN
ðe3 ÞN
1  e3 þ 2he
kþ1 kþ1 kþ1
H3
2 2 j¼1

Adding the above inequalities and cancel the terms results the following inequality
h hX N
Nh kþ1 kþ1
X2
j
kþ1
ðe 6 Þ 0
ðe6 ÞNkþ1
kþ1
ðe 3 Þ0
ðe3 Þj
1 
kþ1
e3 þ 2he H3 ; (104)
2 2 j¼2 2 j¼1

Since the fix boundary condition or Dirichlet boundary condition is given on the last node for
the Eq. (26) which is the Nth node, so the error at the Nth node should be zero. The error on
the first node of the third equation or error of acceleration of the flow (skin friction coefficient)
on the first node is calculated by the Matlab solver “fsolve,” so that error depends upon the error
of the Matlab solver “fsolve” and let this error is denoted by ðe31 Þfsolve .
Thus the inequality (104) can be expressed as
h 
kþ1
X2
j
nþ1
ðe 6 Þ1  ð~e 3 Þkþ1
fsolve þ ð 2N
1 Þ ekþ1
3 þ 2he H3 ; (105)
2 j¼1

The error term at the Nth node of inequality (105) can be replaced by the maximum error at
any node for the first three Eq. (24) by the Keller box method using Gauss–Seidel iterative
method, so the last inequality (105) implies
0 1

 
h ð 2N
1Þ h X2
akþ1
1 ^
nþ1
ðe6 Þ1
kþ1
 ð~e 3 Þfsolve þ @ þ 2h
j
H3 A a e þ
kþ1 0 ð
MH h 2Þ
2 2 a
1

 j¼1
P j
where a ¼ P4 1 j h
1 4 þ 4j¼1 H 1 䊏
j¼1
jH 1j
 NUMERICAL HEAT TRANSFER, PART B: FUNDAMENTALS 171

7. Convergence conditions
Maximum error bounds for Keller-box method using Jacobi and Gauss–Seidel iterative methods
have been found in Theorems 6.1 and 6.2. Since the standard Keller-Box method is second-order
accurate, so it is consistent and due to this fact the present Jacobi/Gauss–Seidel Keller-Box shoot-
ing method is consistent because the discretization of the problems are performed using standard
Keller-Box method. The convergence conditions for the Keller-box using Gauss–Seidel iterative
method can be proven in the following procedure.
Since the finite terms of geometric series were used for finding the sum of first n terms. To
converge the geometric series, apply the limit when n approaches to larger number, then the sum
of infinite terms of geometric series will converge if the absolute value of the common ratio
remains less than one.
So to get convergence conditions of first three Eq. (24) using Keller-Box shooting method with
Gauss–Seidel iterative method, apply the condition for the convergence of geometric series with infinite
terms 0 1
1 1 X4
@ þ H1 A < 1
j
P4
j h
4 j¼1
H1
j¼1

1 X
4
j
X4
j
þ H1 < H1
h
4 j¼1 j¼1

X
4
j 1 X
4
j
X4
j

H1 < þ H1 < H1
j¼1
h
4 j¼1 j¼1

1 X4
j 1
0< þ2 H1 & <0 (106)
h
4 j¼1
h
4

Similarly, the condition of convergence for Eq. (25) can be derive as

1 X4
j 1
0< þ2 H2 & <0 (107)
h
4 j¼1
h
4

The convergence conditions for pressure equation can be described as

0 1
X2
j 2 1 X 4
1
@ H3
A > 0 & 0 < þ2
j
H2 & < 0; (108)
j¼1
h h
4 j¼1
h
4

The convergence conditions for Eq. (27)

0 1
2 2h 5h X
5
@ þ H4 A < 1
j
P
3
j h
4 2 j¼1
5h H4
j¼1

2h 5h X
5
j 5h X3
j
þ H4 < H4
h
4 2 j¼1 2 j¼1

5h X3
j 2h 5h X
5
j 5h X3
j

H4 < þ H4 < H4
2 j¼1 h
4 2 j¼1 2 j¼1
172 Y. NAWAZ AND M. SHOAIB ARIF

Figure 2. Pressure profile with variation of permeability parameter and semilogy plot for the error in pressure profile.

2h X5
j 2h
0< þ 5h H4 & <0 (108)
h
4 j¼1
h
4

Similarly, for last two Eq. (28) the convergence conditions are
2h X5
j 2h
0< þ 5h H5 & <0 (109)
h
4 j¼1
h
4

8. Results and discussions

The considered problem in the present study is the nanofluid flow problem and the flow is gener-
ated by the linearly stretching rotating porous disk. The flow is considered with the exothermic/
endothermic effects along with the effect of Magnetic field. A numerical scheme of Keller-Box
shooting method is developed and applied to nonlinear ordinary differential equations with stabil-
ity, error analysis and the conditions for convergence are found. The effects of the parameters on
the pressure, temperature, nanoparticle volume fraction profiles, wall temperature gradient and
skin friction coefficient are shown in Figures 2–4 and in Tables 1 and 2, respectively.
It should be noted that the present study deals with the applications of the Keller-box method
explicitly instead of implicitly. Discretization of the system of nonlinear differential equations is
given using standard Keller box method and then Jacobi iterative method and Gauss–Seidel itera-
tive method have been implemented for finding the solutions of the system of difference equa-
tions. To apply the Jacobi iterative and Gauss–Seidel iterative methods, the missing initial
conditions are found using shooting method. In the present study, Matlab solver “fsolve” has
been implemented for finding the missing initial conditions. The solver needs one initial guess
for each missing initial condition and also one initial guess is needed to initial the iterative pro-
cedure using Jacobi and Gauss–Seidel methods. The initial guess for Jacobi and Gauss–Seidel
 NUMERICAL HEAT TRANSFER, PART B: FUNDAMENTALS 173

Figure 3. Temperature profile with variation of thermophoresis parameter and semilogy plot for the error in tempera-
ture profile.

Figure 4. Nanoparticles volume fraction profile with variation of Brownian motion and thermophoresis parameters.

iterative methods is a vector whose length is same as the number of grid points. So, two initial
guesses have been used for starting the iterative procedure of Keller-box shooting method. The
one important point in terms of initializing the iterative process of Jacobi and Gauss–Seidel itera-
tive methods is that an updated initial guess has been considered for the Matlab solver “fsolve.”
The updated initial guess can be obtained by finding the missing initial conditions on small
domain and then use the obtained initial conditions computed on small domain as an initial
guess for the solver “fsolve” and try to find the solution for the extended domain.
174 Y. NAWAZ AND M. SHOAIB ARIF

Table 1. Computations for wall temperature gradient with Nb ¼ 0:1; Sc ¼ 10.

0
h ð0Þ
Nt M c Pr c [8] Matlab “bvp4c” R.K.-(4, 5) shooting Keller-Box shooting
0.1 0.5 0.25 7 0.25 –0.1982 –0.2004 –0.2000 –0.2004
0.3 –0.1965 –0.1989 –0.1985 –0.1989
0.5 –0.1946 –0.1972 –0.1968 –0.1972
0.7 –0.1925 –0.1953 –0.1949 –0.1953
0.5 0 –0.2003 –0.2002 –0.1997 –0.2002
0.5 –0.1946 –0.1971 –0.1968 –0.1972
1.0 –0.1900 –0.1945 –0.1941 –0.1945
2.0 –0.1824 –0.1899 –0.1896 –0.1899
0.5 0.1 10 –0.0919 –0.0923 –0.0922 –0.0923
0.5 –0.3324 –0.3384 –0.3377 –0.3384
1.0 –0.4688 –0.4821 –0.4823 –0.4820
10 –0.6400 –0.6546 –0.6750 –0.6545
0.25 5.0 –0.1858 –0.1892 –0.1886 –0.1892
6.2 –0.1916 –0.1944 –0.1939 –0.1944
8.0 –0.1979 –0.2002 –0.1998 –0.2002
15 –0.2110 –0.2123 –0.2121 –0.2123
7 0 –0.1547 –0.1670 –0.1657 –0.1670
0.5 –0.2080 –0.2091 –0.2089 –0.2091
1.0 –0.2193 –0.2198 –0.2197 –0.2198
2.0 –0.2279 –0.2281 –0.2303 –0.2282

Table 2. Numerical values of skin friction coefficient.

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2 2  2ffi
d f ð0Þ dg ð0Þ
dg 2 þ dg

M c [8] Matlab “bvp4c” R. K.-(4, 5) shooting Keller-box shooting

0 0.25 0.640182 0.668741 0.640193 0.640286
0.5 0.900092 0.804790 0.767043 0.767103
1.0 1.140825 0.933988 0.900098 0.900149
2.0 1.528470 1.161466 1.140833 1.140869
5.0 2.346064 1.681951 1.691510 1.691520
0.25 0 0.655274 0.517755 0.604181 0.604171
0.5 0.930917 0.983128 0.858657 0.858773
1.0 1.415579 1.489132 1.332942 1.333176
2.0 2.882167 2.500949 2.777358 2.778161

The standard Keller-box method is an implicit method coupled with the iterative process using
the Newton method, but the present method has been employed explicitly using the Jacobi and
Gauss–Seidel iterative methods. So the present method does not require any matrix form but it
requires the missing boundary conditions and the remaining procedure of finding the solution is
performed explicitly using Jacobi/Gauss–Seidel iterative methods. It is to be noted that the stabil-
ity of the present method for nonlinear differential equations has been checked by matrix-vector
form of the system of linearized equations, but it requires some kind of maximum values of the
fixed dependent variables on the domain. Since some of the dependent variable were held fixed
due to linearization and the values of these dependent variables are further used to draw the
graphs of the largest eigen values.
Figure 2 shows the variation of the permeability parameter K on the pressure profile and also
it shows the semilogy graph of the error of the pressure on number of iterations, which is calcu-
lated using the following formula
epressure ¼ pkþ1
pk 2

In the drawn graph some values were come out to be zero so those values are not shown in
the semilogy graph on number of iterations. Figure 2 (left) shows that the pressure increases with
 NUMERICAL HEAT TRANSFER, PART B: FUNDAMENTALS 175

the increase of the permeability parameter and Figure 2 (right) shows the semilogy plot over the
number of iterations. Figure 3 shows the variation of the thermophoresis parameter on tempera-
ture profile and also its shows the log of the error calculated using the following formula

eh ¼ hkþ1 2

hk 2
:

Increase in thermophoresis parameter Nt enhances the temperature for both exothermic and
endothermic chemical reactions. Since, increase in thermophoresis parameter Nt increases the
thermophoretic force and this force pushes the particles away from the disk. Since the disk is
convectively heated by the fluid in its bottom and the particles near to the disk get hotter and
increase in thermophoretic force will increase to bring hotter particles from vicinity of the disk to
its surroundings. Figure 4 depicts the variation of the thermophoresis parameter Nt and
Brownian motion parameter Nb on nanoparticle volume fraction profile. Both parameters have
dual effect on nanoparticle volume fraction profile. Since increase in Nt will shift the nanopar-
ticles from vicinity of the plate to its surrounding, so decrease in nanoparticle volume fraction
and after that nanoparticle volume fraction increases with the increase in Nt. The nanoparticle
volume fraction boundary layer thickness decreases with the increase in Brownian motion param-
eter for both exothermic and endothermic chemical reactions which can be seen in Figure 4
(right). Since the plate is convectively heated and increase in Brownian motion parameter will
bring the nanoparticles away from the sheet and this yields the decrease in nanoparticle volume
fraction near to the plate.
Table 1 shows the values of the wall temperature gradient from past research and the results
computed presently by three different methods. One of three methods is the Matlab solver
“bvp4c” which can be used for solving boundary value problems. Second of the three methods is
the shooting method using Matlab solvers “fsolve” and “ode45.” “ode45” is a Matlab solver based
on explicit Runge–Kutta (4, 5) formula and can be used for solving non-stiff initial value prob-
lems. The results in Table 1 using third method are those results which are calculated by pres-
ently modified Keller-box shooting method. Results can be compared up to four decimal places.
Table 1 shows the wall temperature gradient for variations of different parameters. The wall tem-
perature gradient increases with the increase of thermophoresis parameter Nt and this is due to
the increase in temperature with increment of thermophoresis parameter Nt. Increase in wall
temperature gradient implies decrease in Nusselt number. Wall temperature gradient increases
with the growth of Hartman number M, this is due to facts that f decreases with the increase of
Hartman number M and since temperature gradient is negative because temperature decreases
when moving away from the plate, so the term f dh dg increases which can give increase in tempera-
ture with the increase in Hartman number M, so the wall temperature gradient also increases.
Using the boundary condition at g ¼ 0 for temperature gives decrease in wall temperature gradi-
ent because 1
hð0Þ  0 with positive increase in Biot number and this can happen due to the
fact that maximum value of temperature is 1 that occurs at the plate. The maximum value of the
dhðgÞ
temperature can be obtained by substitution of dg g¼0 ¼ 0, which gives hð0Þ ¼ 1 for nonzero
Biot number. So, using the condition on temperature at g ¼ 0 gives decrease in wall temperature
gradient with the enhancement of Biot number. Table 1 shows that wall temperature gradient
decreases with the increase of Prandtl number Pr. Since incompressible Newtonian fluid is con-
sidered in the present case, so increase in Prandtl number means decrease in thermal diffusivity,
and decrease in thermal diffusivity means decrease in thermal conductivity and decrease in ther-
mal conductivity yields low heat transfer, so wall temperature gradient will decrease. Table 1 also
shows that increase in c (which is the ratio of radially stretching to rotation) gives decrease in
wall temperature gradient.
Table 2 shows the values of skin friction coefficient which are obtained from past research and
values are computed using the Matlab solver “bvp4c,” Runge–Kutta shooting method and
176 Y. NAWAZ AND M. SHOAIB ARIF

Table 3: Computations of norms along variation of step size using linearized equations with Pr ¼ 1; β ¼ 1; Sc ¼ 0:9; Nb ¼
0:1; Nt ¼ 0:1; M ¼ 0:4; K ¼ 0:1; c ¼ 1; n ¼ 1; γ ¼ 1; σ ¼ 0:4; δ ¼ 0:5; E ¼ 0:1
Pressure Temperature Nanoparticles volume fraction
h jjejj1 jjejj2 jjejj∞ jjejj1 jjejj2 jjejj∞ jjejj1 jjejj2 jjejj∞
6
=40 1.58e-02 4.69e-03 2.80e-03 1.74e-02 3.9e-03 1.62e-03 1.68e-02 3.61e-03 1.1e-03
6
=100 5.94e-03 1.13e-03 4.58e-04 4.55e-02 6.44e-03 1.75e-03 2.64e-02 3.66e-03 7.35e-04
6
=150 3.86e-03 6.16e-04 2.09e-04 6.91e-02 7.98e-03 1.78e-03 3.99e-02 4.49e-03 7.09e-04
6
=200 2.83e-03 4.10e-04 1.22e-04 9.27e-02 9.27e-03 1.80e-03 5.37e-02 5.22e-03 7.02e-04
6
=250 2.29e-03 3.07e-04 8.20e-05 1.16e-01 1.04e-02 1.81e-03 6.76e-02 5.89e-03 7.0e-04
6
=300 2.06e-03 2.50e-04 6.02e-05 1.40e-01 1.14e-02 1.81e-03 8.14e-02 6.48e-03 6.99e-04
6
=350 1.97e-03 2.17e-04 4.70e-05 1.64e-01 1.24e-02 1.82e-03 9.54e-02 7.03e-03 7.0e-04
6
=400 1.94e-03 1.97e-04 3.84e-05 1.87e-01 1.32e-02 1.82e-03 1.09e-01 7.54e-03 7.09e-04
6
=450 1.96e-03 1.86e-04 3.25e-05 2.10e-01 1.41e-02 1.82e-03 1.23e-01 8.02e-03 7.17e-04

presently implemented Keller-Box shooting method. The comparison between the values of the
three different techniques can be seen and which can also be compared with the values computed
in the past research.
Table 3 shows the values for three kinds of norms computed for the errors. The errors were
found from the difference between the solution obtained by the present scheme of Gauss–Seidel
Keller-Box shooting method and Matlab solver “bvp4c.” It can be seen from Table 3 that three
kinds of norms are decreasing with the decrease of the step size.

9. Conclusion
The standard Keller-Box method using Newton’s iterative procedure is modified with Keller-Box
method using Jacobi and Gauss–Seidel iterative methods and shooting technique has been used
for finding missing initial conditions. The implicit Keller-Box method has been applied explicitly
in the present work. The modified method has been applied for the flow problem on convectively
heated stretching porous rotating disk. The previous nanofluid flow models have been extended
with exothermic/endothermic chemical reactions. The quasi-linearization procedure has been
adopted to reduce the nonlinear boundary value problems into linear boundary value problems
and then stability of the present scheme for the reduced linear boundary value problems using
Gauss–Seidel iterative method has been presented. The bounds for maximum errors have been
found for Keller-Box method with Jacobi and Gauss–Seidel iterative methods. The bound of the
error for initial pressure has been given and convergence conditions have been found using
Gauss–Seidel iterative method. The tables for skin friction coefficient and wall temperature gradi-
ent have been drawn and the values have been computed using three methods including the pre-
sent one. The computed results reveal that the results computed by the present method were
closer to those obtained by the Matlab solver “bvp4c” for wall temperature gradient computed
values and the results of the present method were closer to those obtained by the Runge–Kutta
shooting method for the case of skin friction coefficient. It can be concluded that the present
modified explicitly considered numerical method is an efficient method for solving nonlinear
problems and can be applied for solving linear and nonlinear boundary value problems in differ-
ent areas of science and engineering.

References
[1] K. A. Maleque, “Effects of exothermic/endothermic chemical reactions with arrhenius activation energy on
MHD free convection and mass transfer flow in presence of thermal radiation,” J. Thermodynam., vol.
2013, pp. Article ID 692516, 2013. DOI: 10.1155/2013/692516.
 NUMERICAL HEAT TRANSFER, PART B: FUNDAMENTALS 177

[2] P. R. Nachtsheim, and P. Swigert, “Satisfaction of asymptotic boundary conditions in numerical solution of
system of nonlinear of boundary layer type,” NASA TN-D3004, 1965.
[3] M. Z. Swalmeh, H. T. Alkasasbeh, A. Hussanan, and M. Mamat, “Heat transfer flow of Cu-water and
Al2O3-water micropolar nanofluids about a solid sphere in the presence of natural convection using Keller-
Box method,” Result Phys., vol. 9, pp. 717–724, 2018. DOI: 10.1016/j.rinp.2018.03.033.
[4] M. Mustafa, J. Ahmad Khan, T. Hayat, and A. Alsaedi, “Analytical and numerical solutions for axisymmet-
ric flow of nanofluid due to non-linearly stretching sheet,” Int. J. Nonlin. Mech., vol. 71, pp. 22–29, 2015.
DOI: 10.1016/j.ijnonlinmec.2015.01.005.
[5] B. Mahanthesh, B. J. Gireesha, and R. S. R. Gorla, “Unsteady three-dimensional MHD flow of a nano
Eyring-Powell fluid past a convectively heated stretching sheet in the presence of thermal radiation, viscous
dissipation and Joule heating,” J. Assoc. Arab Univ. Basic Appl. Sci., vol. 23, no. 1, pp. 75–84, 2017. DOI:
10.1016/j.jaubas.2016.05.004.
[6] T. Hayat, Y. Saeed, S. Asad, and A. Alsaedi, “Convective heat and mass transfer in flow by an inclined
stretching cylinder,” J. Mol. Liquid., vol. 220, pp. 573–580, 2016. DOI: 10.1016/j.molliq.2016.03.047.
[7] T. Hayat, S. Qayyum, and A. Alsaedi, “Mechanisms of nonlinear convective flow of Jeffrey nanofluid due
to nonlinear radially stretching sheet with convective conditions and magnetic field,” Result Phys., vol. 7,
pp. 2341–2351, 2017. DOI: 10.1016/j.rinp.2017.06.052.
[8] A. Mushtaq, and M. Mustafa, “Computations for nanofluid flow near a stretchable rotating disk with axial
magnetic field and convective conditions,” Result Phys., vol. 7, pp. 3137–3144, 2017. DOI: 10.1016/j.rinp.
2017.08.031.
[9] M. M. Rashidi, T. Hayat, E. Erfani, S. A. Mohimanian Pour, and A. A. Hendi, “Simultaneous effects of par-
tial slip and thermal-diffusion and diffusion-thermo on steady MHD convective flow due to a rotating
disk,” Commun Nonlinear Sci Numer Simulat, vol. 16, no. 11, pp. 4303–4317, 2011. DOI: 10.1016/j.cnsns.
2011.03.015.
[10] M. Mustafa, and J. A. Khan, “Numerical study of partial slip effects on MHD flow of nanofluids near a
convectively heated stretchable rotating disk,” J. Mol. Liquid., vol. 234, pp. 287–295, 2017. DOI: 10.1016/j.
molliq.2017.03.087.
[11] M. I. Khan, S. Qayyum, T. Hayat, M. I. Khan, A. Alsaedi, and T. A. Khan, “Entropy generation in radiative
motion of tangent hyperbolic nanofluid in presence of activation energy and nonlinear mixed convection,”
Phys. Lett. A., vol. 382, no. 31, pp. 2017–2026, Aug. 2018. DOI: 10.1016/j.physleta.2018.05.021.
[12] M. Dhlamini, P. K. Kameswaran, P. Sibanda, S. Motsa, and H. Mondal, “Activation energy and binary
chemical reaction effects in mixed convective nanofluid flow with convective boundary conditions,” J.
Comput. Design Eng., vol. 6 no. 2, pp. 149–158, 2019, DOI: 10.1016/j.jcde.2018.07.002.
[13] R. Kumar, S. Sood, M. Sheikholeslami, and S. A. Shehzad, “Nonlinear thermal radiation and cubic auto-
catalysis chemical reaction effects on the flow of stretched nanofluid under rotational oscillations,” J.
Colloid Interface Sci., vol. 505, pp. 253–265, 2017. DOI: 10.1016/j.jcis.2017.05.083.
[14] F. Urena, L. Gavete, A. Garcia, J. J. Benito, and A. M. Vargas, “Solving second order non-linear parabolic
PDEs using generalized finite difference method (GFDM),” J. Comput. Appl. Math., vol. 354, pp. 221–241,
2019, DOI: 10.1016/j.cam.2018.02.016.
[15] T. Cebeci, and P. Bradshaw, Physical and Computational Aspects of Convective Heat Transfer. New York:
Springer-Verlag, 1988.
[16] Z. Kun, W. Liangbi, and Z. Yuwen, “Improved finite difference method with a compact correction term for
solving Poisson’s equations,” Num. Heat Transf. B., vol. 70, no. 5, pp. 393–405, 2016. DOI: 10.1080/
10407790.2016.1215715.
[17] Z. Shuying, and F. Xinlong, “A block-centered finite-difference method for time-fractional diffusion equa-
tion on nonuniform Grids,” Num. Heat Transf. B., vol. 69, no. 3, pp. 217–233, 2016. DOI: 10.1080/
10407790.2015.1097101.
[18] M. Zeka, “A novel fully implicit block coupled solution strategy for the ultimate treatment of the velocity–-
pressure coupling problem in incompressible fluid flow,” Num. Heat Transf. B., vol. 69, no. 2, pp. 130–149,
2016. DOI: 10.1080/10407790.2015.1093787.
[19] M. Turkyilmazoglu, and P. Senel, “Heat and mass transfer of the flow due to a rotating rough and porous
disk,” Int. J. Therm. Sci., vol. 63, pp. 146–158, 2013.
[20] M. Kamran, and B. Wiwatanapataphee, “Chemical reaction and Newtonian heating effects on steady con-
vection flow of a micropolar fluid with second order slip at the boundary,” Eur. J. Mech. B Fluid., vol. 71,
pp. 138–150, 2018. DOI: 10.1016/j.euromechflu.2018.04.005.