Вы находитесь на странице: 1из 42

Applied Mathematics and

Computation
Manuscript Draft

Manuscript Number: AMC-D-17-04192

Title: Application of the Bivariate Chebyshev Spectral Collocation Quasi-


linearisation Method for Non-Similar Boundary Layer Equations

Article Type: Full Length Article

Keywords: Boundary layer; Unsteady flow; Non-linear; Non-similarity; MHD;


BI-SQLM.
*Manuscript
Click here to download Manuscript: Makibelo_Harley_AMC2.pdf Click here to view linked References

Application of the Bivariate Chebyshev Spectral Collocation


Quasi-linearisation Method for Non-Similar Boundary Layer
Equations

V. Makibelo1 , Charis Harley∗


DST/NRF Centre of Excellence in the Mathematical and Statistical Sciences
School of Computer Science and Applied Mathematics
University of the Witwatersrand
Private Bag 3, Johannesburg, Wits 2050, South Africa

Abstract

In this paper the bivariate spectral quasi-linearisation method (BI-SQLM) will be used to solve
non-similar boundary layer equations. Three models are considered: 1) unsteady boundary layer
flow adjacent to a permeable stretching surface in a porous medium, 2) thermal radiation effects on
magnetohydrodynamics (MHD) forced convection flow adjacent to a non-isothermal wedge in the
presence of a heat source or sink, and 3) unsteady laminar MHD flow near a forward stagnation
point of an impulsively rotating and translating sphere in the presence of buoyancy forces. Numerical
simulations are presented which allow us to examine the effects of the different parametric quantities.
Convergence and error analyses are conducted to validate the resolutions obtained.
Keywords:
Boundary layer, Unsteady flow, Non-linear, Non-similarity, MHD, BI-SQLM.

1. Introduction

Due to the extensive range of applications in engineering and industry for which the study of non-
similar boundary layer equations are important, these models have gained much interest [1]. The
non-similarity of boundary layers may stem from a variation in free stream velocity, the effect of
5 suction or injection of fluid at the wall, wall temperature and buoyancy force effect; in some cases

∗ Corresponding author
Email address: Charis.Harley@wits.ac.za or Charis.Harley@gmail.com (Charis Harley)
1 Email: vuyelwa.makibelo@students.wits.ac.za

Preprint submitted to Journal of LATEX Templates November 14, 2017


a combination of these effects may occur to cause a flow to be characterised as non-similar [1].
Models describing non-similar flows were more prominently featured in texts after Ludwig Prandtl
introduced boundary layer theory. Initially the investigation of such flows could only be done experi-
mentally or by means of the analytical solution of relevant models [2].The development of numerical
10 methods and scientific computing has provided new avenues for the investigation of models de-
scribing non-similar flows. Traditional methods which have been applied to examine non-similar
boundary layer flows are: i. analytic methods, for instance a series expansion was carried out by
Howarth (1938), Gertler (1957) and Machin (1961); ii. approximate integral methods were used by
Karman (1921) and Rosenhead (1963), where the velocity and temperature profiles were estimated
15 and experimental relationships incorporated into integral parameters; iii. numerical techniques such
as finite difference or finite element methods as originally employed by Spalding (1977) and Keller
(1978). The latter have gained popularity among researchers, however these numerical methods
have some drawbacks such as being tedious and computationally costly for solving systems of equa-
tions or maintaining a high level of accuracy [2].
20

Non-similar boundary layer equations have been found to be more arduous to solve than simi-
lar boundary layer equations, but over the years significant progress has been made by various
researchers. In Chen’s approach [3] non-similar equations were reduced to ordinary differential
equations (ODEs) using perturbation theory; the ODEs are then solved using the shooting method
25 [4]. Nakamara [4] formulated a numerical technique to solve non-similar boundary layer equations
by means of successive substitution iterations of finite difference schemes. The main idea was to
reduce a third-order similar boundary layer equation to a second-order boundary value problem
which is solved by a tridiagonal matrix solver. This was first applied when considering the natural
convection analysis for a non-Newtonian fluid using the power law model and also using bound-
30 ary layer equations for micropolar fluids [4]. The method developed by Nakamara was proven to
be successful upon similar boundary layer equations. The derivation of the difference equations is
substantially simpler than that of Cebecci and Bradshaw who set up their difference schemes by
means of the Keller-box method [5, 6]. The technique of Nakamara for unsteady similar equations
was also made applicable to non-similar boundary layer equations. Kafoussias and Williams further
35 extended the method of Nakamara; their technique is based on the common difference method with
central differencing tridiagonal matrix manipulation and an iterative procedure [4, 7].

2
Recently Hayat et al. [8] used the homotopy analysis method (HAM) to solve unsteady three
dimensional magnetohydrodynamic (MHD) flow and mass transfer in porous media. This tech-
40 nique has been widely used by researchers with similar non-linear problems [9]. HAM is based
on homotopy in topology and overcomes the preceding restrictions and limitations of perturbation
techniques. The benefit of HAM is the independence of physical parameters and adjustments of
the convergence region, but it is time consuming and very tedious to attain higher-order approxi-
mations [10]. Soon after Hayat et al. [6] Motsa et al. [11, 12] made use of the spectral relaxation
45 method (SRM) and the spectral quasi-linearisation method (SQLM) to solve unsteady boundary
layer problems which have shown to be more precise, easier to implement and computationally effi-
cient. Spectral techniques are advantageous as they are efficient and have a high degree of accuracy
without being computationally taxing. As such, we will employ the BI-SQLM to attain numerical
solutions to certain non-similar boundary layer equations, extending the work conducted by Motsa
50 et al. [13]. Three models are considered where two are decoupled non-linear differential equations
and the remaining model consists of coupled non-linear differential equations - these are discussed in
section 3. In sections 4 to 6 approximate solutions attained through the BI-SQLM will be shown as
well as convergence and error analyses for each model respectively; comparisons to results obtained
by other researchers will also be presented. Concluding remarks are made in section 7.

55 2. Nomenclature

The following set of quantities will be used throughout the paper. Most of the quantities will become
non-dimensional quantities, parameters or constants.

Symbol Description

Bo Externally imposed magnetic field in the y-direction


cp Specific heat constant pressure
Cf Local skin-friction coefficient
Ec Eckert number
f Stream function

3
fo Wall mass transfer parameter
k Thermal conductivity
K Mean radiation absorption coeficient
m Pressure gradient parameter
NR Thermal radiation parameter
Nux Local Nusselt number
Pr Prandtl number
qr Radiative heat flux
Qo Heat generation or absorption coefficient
Rex Local Reynolds number
T Temperature
u Velocity component in the x-direction
U∞ Potential flow or free stream velocity
v Velocity in the y-direction
vo Suction or injection velocity
x Coordinate along the wedge surface
y Coordinate normal to the wedge surface

Greek Symbol
α Thermal diffusivity or bouyancy parameter
β Angle factor of the wedge
η Pseudosimilarity variable
ξ Magnetic variable
ω Total angle of the wedge
φ Dimensionless heat generation or absorption coefficient
θ Dimensionless temperature
ρ Density of fluid
ν Kinematic viscosity
σ Electrical conductivity

σ Stefan-Boltzmann constant

4
ψ Stream function
λ Porosity parameter or rotation parameter
60

Subscripts w Condition at the wall


∞ Condition at free stream

3. Models

3.1. Model I: Unsteady boundary layer flow adjacent to a permeable stretching surface in a porous
medium

The flow of a viscous fluid over a stretching sheet plays a significant role in engineering applications
65 such as the extrusion of plastic sheets, boundary layer flows along a liquid film, condensation pro-
cesses, a cooling metallic plate in a cooling bath and applications in glass and polymer industries
[14]. Sakiadis [15] conducted a numerical study of boundary layer flows over a stretched surface
moving with constant velocity and the work was validated by Tsou et al. [16] where a combined
analytical and experimental study was done on the flow and heat transfer of a continuous moving
70 plate. Soon after several researchers further expanded the study by including other physical fea-
tures such as suction, injection, a magnetic field, the rotation of a disk, heat transfer analysis and
viscoelasticity of a liquid.

The effect of suction or injection on boundary flows due to the stretching of the wall was analysed
75 by Crane [17], where exact solutions were obtained. Banks [18] discussed non-similarity solutions
of boundary layer equations for a stretching wall analytically and numerically. Banks and Zaturska
[19] focused on boundary layer flow adjacent to the stretching wall and attained analytical solutions
for certain eigenvalues and eigenfunctions. In research conducted by Grubka and Bobba [20], the
heat transfer characteristics of a continuous stretching surface with variable temperature was ex-
80 plored, where the power law temperature distribution was solved in terms of Kummer functions and
various closed form analytical solutions were found; Chen and Char [21] conducted similar work.
Similarity solutions for heat transfer characteristics of a continuous stretching plate were reported
by Ali [22]. Erickson et al. [23] looked into the flow of a moving continuous plate with suction and
injection for cases where the transverse velocity profile is non-zero at the surface; results were ac-

5
85 quired numerically through the fourth-order Runge-Kutta formula, the iteration method of Hartree
and numerical integration. Nazar et al. [24], Liao and Xu [25] and Liao [26] produced results for
unsteady boundary flow equations by means of transformations found by Williams and Rhyne [27]
and the Keller-box numerical method.

90 In this section an unsteady boundary layer flow adjacent to a permeable a stretching surface in
a porous medium will be solved using the BI-SQLM. This problem was previously analysed by
Ali and Mehmood using HAM [28]. An incompressible hydrodynamic viscous fluid bounded by an
infinite plate situated at y = 0 is considered.

Figure 1: Visual representation of Model I.

Assumptions
95 1. The upper half space of the plate is porous and the fluid lies in this porous region at y > 0.
2. The plate is a highly elastic membrane and is uniformly porous so that the suction or injection
is possible.
3. Two equal and opposite forces are applied along the x-axis so that the wall is stretched while
keeping the origin fixed.

6
The following equations describe the boundary layer flow
∂u ∂v
+ = 0, (3.1.1)
∂x ∂y

∂u ∂u ∂u ∂ 2 u νφ
+u +v =ν 2 − u, (3.1.2)
∂t ∂x ∂y ∂y k
subject to the boundary conditions (for t > 0)

u = ax, v = −V0∗ at y = 0, and

u −→ 0 as y −→ ∞, (3.1.3)

where a > 0 is a constant; the initial condition for (t ≤ 0) is given by

u = v = 0, (3.1.4)

at all points (x, y). The similarity transformation is written as


r
a p
η= y, ψ = aνξxf (η, ξ), ξ = 1 − e−τ , τ = at, (3.1.5)
νξ
where ψ denotes the stream function. Therefore, the governing system (3.1.1) - (3.1.4) transforms
into
2 !
∂3f ∂2f ∂2f ∂2f

1 ∂f ∂f
+ (1 − ξ)η +ξ f 2 − − ξ(1 − ξ) − λξ = 0, (3.1.6)
∂η 3 2 ∂η 2 ∂η ∂η ∂η∂ξ ∂η

subject to the following boundary conditions



∂f ∂f
f (η, 0) = V0 , and , (3.1.7)
∂η η=0 ∂η η=∞

100 where λ = νφ/ak is the porosity parameter and V0 = V0∗ / aνξ is the dimensionless suction/injection
velocity. Here V0 > 0 corresponds to uniform suction whereas V0 < 0 corresponds to uniform injec-
tion at the porous wall.

3.2. Model II: Thermal radiation effects on magnetohydrodynamics (MHD) forced convection flow
adjacent to a non-isothermal wedge in the presence of heat source or sink
105 Thermal radiation plays an important role on flow and heat transfer processes such as the design
of many advanced energy conversion systems functioning at a great temperature. In 1998 Raptis

7
[29] looked at the impact of thermal radiation on the flow of a micropolar fluid past a continuously
moving plate. Upon the two-dimensional flow Raptis approximated the temperature with the Tay-
lor series to order four and neglected higher order terms. Transformations were then applied to
110 get a system of three equations that were solved numerically. Similarly, in the same year, Raptis
examined two-dimensional steady free convection flow through a porous medium in the presence
of thermal radiation [30]. Further work was conducted by Yih (1999) [31], who considered MHD
forced convection flow adjacent to a non-isothermal wedge with the effects of viscous dissipation
and stress; the problem was solved numerically using the Keller-box method. Chamkha et al. [32]
115 generalized the work of Yih [31] by employing an implicit finite difference method.

In this section we will employ the BI-SQLM for the solution of the non-linear problem that was dis-
cussed by Chamkha et al. [32]. A steady, laminar, two-dimensional hydromagnetic boundary layer
flow and heat transfer over a permeable non-isothermal wedge in the presence of thermal radiation
and heat generation or absorption effects is examined.

Figure 2: Visual representation of Model II.


120

Assumptions
1. Fluid suction or injection is enforced at the surface of the wedge.

8
2. The applied magnetic field is uniform and is in the direction normal to the surface.
3. Fluid properties are constant.
125 4. The Reynolds number is small, so that the magnetic field can be ignored.
5. The electric field does not exist.

The equations governing this boundary-layer forced convection flow are established through the
balance laws of mass, linear momentum and energy. They are written as

∂u ∂v
+ = 0, (3.2.1)
∂x ∂y

∂u ∂u ∂2u dU∞ σBo2


u +v = ν 2 + U∞ + (U∞ − u) , (3.2.2)
∂x ∂y ∂y dx ρ

2
∂2T

∂T ∂T Qo 1 ∂qr ν ∂u
u + =α 2 + (T − T∞ ) + +
∂x ∂y ∂y ρcp ρcp ∂y cp ∂y

σBo2 σBo2 u2
 
u dU∞
− U∞ + U∞ + . (3.2.3)
cp dx ρ ρcp
The physics of the problem depicts the following boundary conditions

y = 0 : u = 0, v = −vo , T = Tw (x) = T∞ + Ax2m , (3.2.4)

y −→ ∞ : u −→ U∞ = Cxm , T −→ T∞ , (3.2.5)

where x and y are coordinates measured along and normal to the surface respectively; u and v are
the velocity components in the x and y directions respectively. In addition the radioactive heat flux
qr is employed according to the Rosseland approximation such that

4σ∗ ∂T 4
qr = − . (3.2.6)
3K ∂y

The fluid-phase temperature within the fluid flow is assumed to be sufficiently small so that T 4 may
be defined as a linear function of temperature. This is done by expanding T 4 in a Taylor series
about the free stream temperature T∞ and abridging higher-order terms to give

T 4 = 4T∞
3 4
T − 3T∞ . (3.2.7)

9
By substituting (3.2.6) and (3.2.7) in (3.2.3), the following is attained
2
∂2T 3
∂2T

∂T ∂T Qo 16σ ∗ T∞ ν ∂u
u + =α 2 + (T − T∞ ) − +
∂x ∂y ∂y ρcp 3Kρcp ∂y 2 cp ∂y

σBo2 σBo2
 
u dU∞
− U∞ + U∞ + . (3.2.8)
cp dx ρ ρcp
The stream function is expressed as
∂ψ ∂ψ
u= and v = − . (3.2.9)
∂y ∂x
These transformations along with the following dimensionless variables
1/2
σBo2 x

y U∞ x
ξ= , η= , (3.2.10)
ρU∞ x ν

ψ T − T∞
f (ξ, η) = and θ(ξ, η) = (3.2.11)
(U∞ xν)1/2 Tw − T∞
are substituted into equations (3.2.1), (3.2.2) and (3.2.8). The following non-similar equations are
then obtained
∂f 0
   
1+m ∂f
f 000 + f f 00 + m 1 − (f 0 )2 + ξ(1 − f 0 ) = (1 − m)ξ f 0 − f 00

, (3.2.12)
2 ξ ∂ξ

 
(1 + NR ) 00 1+m
f θ0 − 2mf 0 θ + φξθ + Ec (f 00 )2 − (m + ξ)f 0 + ξ(f 0 )2

θ +
Pr 2
 
∂θ ∂f
= (1 − m)ξ f 0 − θ0 , (3.2.13)
∂ξ ∂ξ
where
ν 16ρ∗ T∞
3
Qo C
Pr = , NR = , φ= , Ec = , (3.2.14)
α 3Kk σcp Bo2 cp A
are defined to be the Prandtl number, thermal radiation parameter, dimensionless heat generation or
absorption coefficient and the Eckert number respectively. The corresponding boundary conditions
are
 
0 2
η=0: f = 0, f= fo ξ 1/2 , θ = 1, (3.2.15)
1+m
η −→ ∞ : f 0 −→ 1, θ −→ 0. (3.2.16)

where fo is the dimensionless suction or injection parameter.

10
3.3. Model III: Unsteady laminar MHD flow near forward stagnation point of an impulsively rotating
and translating sphere in presence of bouyancy forces

130 The study of flow and heat transfer on rotating bodies has importance in may engineering appli-
cations such as projectile motion, re-entry missile design of rotating machinery and fiber coating.
Rajasekan and Palekar [33] used the shooting method to obtain solutions to a model describing a
mixed convection flow about a rotating sphere. Then Hatzikostantinou [34] looked into the conse-
quences of viscous dissipation on forced2 and combined3 convection of unsteady incompressible flow
135 about a rotating sphere; this was done through the utilization of finite difference approximations.
Ece [35] analysed initial boundary layer flow past a translating and spinning rotational symmetric
body and acquired first-order solutions analytically and second-order results numerically. HAM was
applied to MHD flow near a rotational sphere with buoyancy forces by Dinarvand et al. [36].

140 In this work the BI-SQLM will be employed for the solution of this model. The unsteady lam-
inar incompressible boundary layer flow close to the front stagnation point of a rotating sphere
where magnetic fields and buoyancy forces are present is considered. Before t = 0, the sphere is
immobile and plunged in an ambient fluid with surface temperature T∞ which is tantamount to
that of a rotating fluid. At time t = 0, an abrupt motion is transferred to the ambient fluid and the
145 sphere is incontinently rotated with constant angular velocity Ω. An unvarying magnetic field B is
applied in the z direction [36].

Assumptions
1. The Reynolds number Rm = µ0 σV L  1, where µ0 is the magnetic permeability, σ the
electrical conductivity, V and L are the characteristic velocity and length respectively.
150 2. The induced magnetic field is neglected.
3. The wall and free stream temperature are taken as coefficients.
4. The dissipation terms, Ohmic heating and surface curvature, are disregarded in the locality
of the stagnation point.

2 Forced convection is when fluid is forced to flow over a surface or in a tube, induced by external factors such as
a pump or fan.
3 Combined convection is a combination of natural and forced convection, where natural convection is when a fluid

moves by itself due to temperature.

11
Figure 3: Visual representation of Model III.

5. The fluid has constant thermophysical properties except for those that cause density changes
155 that generate the buoyancy force, under the Boussinesq approximation.
6. The effect of the buoyancy-induced stream-wise pressure gradient terms on the flow and tem-
perature fields are inconsequential.

The boundary layer equations governing the axisymmetric flow are [35]

∂(ux) ∂(wx)
+ = 0, (3.3.1)
∂x ∂z
∂u ∂u ∂u v 2 due ∂2u x σB 2
+u +w − = ue + v 2 + ḡβ (T − T∞ ) − (u − ue ) , (3.3.2)
∂t ∂x ∂z x dx ∂z R ρ
∂v ∂v ∂v uv ∂ 2 w σB 2
+u +w + =v 2 − v, (3.3.3)
∂t ∂x ∂z x ∂z ρ
∂T ∂T ∂T k ∂2T
+u +w = , (3.3.4)
∂t ∂x ∂z ρcp ∂z 2

12
where x is the distance along the meridian from the forward stagnation point, y represents the
distance in the direction of rotation, z is the distance normal to the surface; u, v and w are the
velocity components along x, y and z directions accordingly. B is the magnetic field, Ω denotes
the angular velocity of the sphere, with β being the coefficient of thermal expansion. R is defined
to be the raduis of the sphere, ḡ is referred to as the acceleration due to gravity, k is the thermal
conductivity and the subscripts e, w and ∞ represent conditions of the edge of the boundary layer
on a free stream respectively. The initial conditions are

u(x, z, t) = v(x, z, t) = w(x, z, t) = 0, (3.3.5)

T (x, z, t) = T∞ , for t < 0 (3.3.6)

and the corresponding boundary conditions are as follows

u(x, 0, t) = 0, v(x, 0, t) = Ωx, T (x, 0, t) = Tw , (3.3.7)

u(x, ∞, t) = ue (x), v(x, ∞, t) = 0, T (x, ∞, t) = T∞ . (3.3.8)

The transformations [26]


∂f (η, ξ) 1 1
u(x, z, t) = ax , v(x, z, t) = Ωxg(η, ξ) and w(x, z, t) = −(2av) 2 ξ 2 f (η, ξ) (3.3.9)
∂η
transform equations (3.3.1) - (3.3.4) into a system of non-linear partial differential equations [37]

2 !
∂3f ∂2f ∂2f
  
1 1 ∂f 2 1 ∂f
+ η (1 − ξ) + ξf + ξ 1− + λg + Mξ 1 −
∂η 3 4 ∂η 2 ∂η 2 2 ∂η 2 ∂η
1 1 ∂2f
+ αξθ − ξ (1 − ξ) = 0, (3.3.10)
2 2 ∂ξ∂η
∂2g
 
1 ∂g ∂g ∂f 1 1 ∂g
+ η (1 − ξ) + ξ f − g − M ξg − ξ (1 − ξ) = 0, (3.3.11)
∂η 2 4 ∂η ∂η ∂η 2 2 ∂ξ
∂2θ 1 ∂θ ∂θ 1 ∂θ
+ P rη (1 − ξ) + P rξf − P rξ (1 − ξ) = 0, (3.3.12)
∂η 2 4 ∂η ∂η 2 ∂ξ
subject to the following boundary conditions

∂f (η, ξ) ∂f (η, ξ)
f (0, ξ) = 0, = 0, = 1, (3.3.13)
∂η η=0 ∂η η=∞

g(0, ξ) = 1, g(∞, ξ) = 0, (3.3.14)

θ(0, ξ) = 1 and θ(∞, ξ) = 0. (3.3.15)

13
Here η is the transformed variable, ξ represents dimensionless time, ∂f /∂η and g are the dimen-
160 sionless velocity components along the x and y directions respectively, θ denotes the dimensionless
temperature, a is the velocity gradient at the edge of the boundary layer, whilst P r is the Prandtl
number, GrR and ReR are the Grashof and Reynolds number in that given order. Lastly α represents
the buoyancy parameter.

4. Model I

165 4.1. Determining an initial guess for f (η, ξ)

An initial approximation is ascertained using equations (3.3.6) - (3.3.7), by letting ξ = 0 (τ = 0)


then the non-linear partial differential equation (3.3.6) simplfies to
∂3f 1 ∂2f
3
+ η 2 =0 (4.1.1)
∂η 2 ∂η
and the boundary conditions become

f (0, 0) = V0 , f 0 (0, 0) = 1, f 0 (∞, 0) = 0. (4.1.2)

Solving (4.1.1) we obtain


  2 
η 2 −η
f (η, 0) = ηerfc +√ 1 − exp + V0 (4.1.3)
2 π 4
as the initial solution for (3.3.6), where erfc (η) is the complementary error function delineated as
Z ∞
2
erfc (η) = √ exp (−z 2 ) dz.
π η

4.2. Application of the BI-SQLM

The solution of (3.3.6) is approximated by the bivariate Lagrange interpolation polynomial of the
form
Nη Nη
X X
f (η, ξ) ≈ f (ηi , ξj )Li (η)Lj (ξ), (4.2.1)
i=0 j=0

which interpolates f (η, ξ) at selected points in the η direction. It is convenient to transform the
interval ξ ∈ [0, Lξ ] to ζ ∈ [−1, 1], then choose Chebyshev-Gauss-Lobatto points as collocation points
as follows
   
πi πj
ηi = cos , i = 0, 1, 2..., Nη , ζj = cos , j = 0, 1, 2..., Nξ . (4.2.2)
Nη Nξ

14
Approximating a non-linear operator H using linear terms of the Taylor series expansion we obtain

n
X ∂H  (k) 
H[f, f 0 , ..., f (n) ] ≈ H[fr , fr0 , ..., fr(n) ] + f − f (k)
r , (4.2.3)
k=0
∂f (k) r+1
where the primes denote differentiation with respect to η. Differentiating H with respect to
0 00 000
f, f , f , f , the following coefficients are acquired
∂H
a0,r (η, ξ) = = ξfr00 , (4.2.4)
∂f
∂H
a1,r (η, ξ) = = −2ξfr0 − λξ (4.2.5)
∂f 0
∂H 1
a2,r (η, ξ) = 00
= (1 − ξ)η + ξfr0 , (4.2.6)
∂f 2
∂H
a3,r (η, ξ) = = 1. (4.2.7)
∂f 000
The coefficients (4.2.4) - (4.2.7) are then substituted into (4.2.3) to get

000 00 00
a3,r (η, ξ)fr+1 + a2,r (η, ξ)fr+1 + a1,r (η, ξ)fr+1 + a0,r (η, ξ)fr+1

0
∂fr+1
−ξ(1 − ξ) = Rr (η, ξ), (4.2.8)
∂ξ
where
Rr (η, ξ) = ξ(fr00 fr − (fr0 )2 ).

The derivative of f with respect to ξ is described as



∂f 0
X
= di,j Fj0 (η), i = 1, ..., Nξ . (4.2.9)
∂ξ ξ=ξi j=0

Employing the collocation at ξi upon (4.2.8) we get

(i) 000 (i) 00 00(i) (i)


a3,r (η, ξ)Fr+1,i (η) + a2,r (η, ξ)Fr+1,i (η) + a1,r (η, ξ)Fr+1,i (η) + a0,r (η, ξ)Fr+1,i (η)

X
− 2ξi (1 − ξi ) di,j Fj0 (η) = Rr (η, ξ). (4.2.10)
j=0

The derivatives with respect to η are characterised in terms of the Chebyshev differentiation matrix


dp Fr+1,i

X
= Dp Fr+1,i , (4.2.11)
dξ p η=ηj k=0

15
where D = (2/Lη ) Dj,k and p is the order of the derivative. Substituting (4.2.11) into (4.2.10) gives

Nξ −1
(i)
X
(i)
A Fr+1,i − 2ξi (1 − ξi ) di,j DFr+1,j = R1 . (4.2.12)
j=0

The boundary conditions are expressed as


N
X N
X
D0,k fk = 0, fr+1 (ηN ) = V0 and DN,k fk = 1. (4.2.13)
k=0 k=0

Solving for F in (4.2.12) gives the solution to equation (3.1.6).

4.3. Results and analysis

In this section we unpack the results, which are presented graphically. The graphs are plotted for
170 dimensionless velocity components f and f 0 for varying parametric quantities. It is important to
note that the initial solution (4.1.3) was used to arrive to the final solutions presented in the figures
below. We also tested the initial guess Ali and Mehmood [28] utilized in their work and similar re-
sults were established. The velocity f 0 decreases by increasing the porosity parameter λ, see Figure

1.2 1.4
λ=0 λ=0
1 λ=2 λ=2
λ=5
1.2 λ=5
0.8 λ = 10 λ = 10
1
0.6
f0

0.4 0.8

0.2
0.6
0

−0.2 0.4
0 5 10 15 20 0 5 10 15 20
η η

Figure 4: Distribution of velocity f 0 at several values of the porosity parameter λ for V0 = 0.5 (left) and the impact
of the porous medium on the velocity component f where V0 = 0.5 (right).

4. The consequence of the porosity parameter λ on the velocity component f is similar to that of
175 f 0 . An increase in λ causes a diminution in the velocity of the fluid.

In Figure 5 (left) f 0 diminishes for all the values of suction (V0 > 0) for λ = 0.5. The impact

16
1.2 1.2
V0 = 0 V0 = 0
1 V0 = 1 1 V0 = −1
V0 = 2 V0 = −2
0.8 0.8
V0 = 3 V0 = −3
0.6 0.6
f0

f0
0.4 0.4

0.2 0.2

0 0

−0.2 −0.2
0 5 10 15 20 0 5 10 15 20
η η

Figure 5: The effect of suction (left) and injection (right) on velocity f 0 at the stretching wall for λ = 0.5.

of injection on f 0 is converse to that of suction in Figure 5 (right) which is to be expected. The


boundary layer thickness reduces by increasing the suction velocity at the wall or by raising the
180 porosity of the medium, but as the porosity further increases the injection at the stretching wall
the boundary layer thickness increases.

1.4 1

1.2 0.8

1 0.6
V0 = 0
0.8 0.4 V0 = −0.1
V0 = 0
f

V0 = −0.2
0.6 V0 = 0.1 0.2
V0 = −0.3
V0 = 0.2
0.4 0
V0 = 0.3
0.2 −0.2

0 −0.4
0 5 10 15 20 0 5 10 15 20
η η

Figure 6: Velocity profile f for different values of suction parameter V0 (left) and injection parameter V0 (right)
when λ = 0.5.

As suction increases the velocity f upsurges at the stretching wall and becomes steady far from
the plate; the effect of an increasing injection decreases the velocity f , see Figure 6.

17
185 4.4. Convergence and error analysis
A convergence test is done to see whether the numerical solution attained using the BI-SQLM is
consistent to some error order; we also provide a simulation of the residual errors - see Figure 4.4
below. Convergence with respect to η and ξ is met after five iterations with an elapsed time of
2.87105 seconds and the error is around 10−10 . Convergence was also tested using the initial guess
190 Ali and Mehmood [28] employed in their paper and convergence was reached after five iterations; the
elapsed time was 4.569710 seconds with a similar error or approximately 10−10 . Hence, the initial
solution employed in this work is more efficient for the solution of the model via the BI-SQLM.
0 η 0 η
10 10
19.7815 19.7815
18.9101 18.9101
16.6913 16.6913
−5 −5
10 Ef 0 10
Ef

−10 −10
10 10

−15 −15
10 10
0 5 10 15 20 0 5 10 15 20
Number of iterations Number of iterations

Figure 7: Convergence and residual error of f (left) and f 0 (right) with respect to η.

5. Model II

5.1. Determining initial guesses for f (η, ξ) and θ(η, ξ)


Deriving the initial solution for f (η, ξ), we let ξ = 0 from (3.2.12) giving
 
000 1+m
f0 + f0 f000 + m(1 − (f00 )2 ) = 0, (5.1.1)
2
where f0 = f0 (η) and boundary conditions (3.2.15) - (3.2.16) are transformed to

f00 (0) = 0, f0 (0) = 0, f00 (∞) = 0, η ∈ [0, ∞). (5.1.2)

Equation (5.1.1) is solved numerically by applying the spectral quasi-linearisation method (SQLM)
for non-linear ordinary differential equations. The initial resolution for (5.1.1) is

f0 (η) = η − exp(−η) − 1, (5.1.3)

18
which satisfies the newly transformed boundary conditions (5.1.2). Applying the SQLM to (5.1.1),
we let
 
1+m
F0 [η, f0 (η), f00 (η), f000 (η), f0000 (η)] f0000 f0 f000 + m 1 − (f00 )2 ,

= + η ∈ [0, ∞). (5.1.4)
2

Thus the SQLM scheme becomes

a03,r f0000r+1 (η) + a02,r f000r+1 (η) + a01,r f00 r+1 (η) + a00,r f0r+1 (η) = Rf0r (η), (5.1.5)

where coefficients are defined to be

∂F0
a03,r = = 1,
∂f0000
 
∂F0 1+m
a02,r = = f0r ,
∂f000 2
∂F0
a01,r = = −2mf00 r ,
∂f00
 
∂F0 1+m
a00,r = = f0000r ,
∂f0 2

and
 
1+m
Rf0r (η) = f0r f000r − m − m(f00 r )2 .
2

Employing spectral collocation on (5.1.5) gives

A0 F0r+1 = Rf0r , (5.1.6)

with
A0 = D3 + a02,r D2 + a01,r D + a00,r (5.1.7)

and
a0p,r = [a0p,r (η0 ) , a0p,r (η1 ) , a0p,r (η2 ) , ..., a0p,r (ηN ) ]T

are diagonal matrices. The boundary conditions are represented as follows


N
X N
X
D0,k f0k = 1, f0r+1 (ηN ) = 0 and DN,k f0k = 0. (5.1.8)
k=0 k=0

19
The initial solution for f is obtained by solving (5.1.6). Similarly the initial guess for θ is determined
by letting ξ = 0 so that equation (3.2.13) is represented as
   
1 + NR 1+m 
2

θ000 + f0 θ00 − 2mf00 θ0 = −Ec (f000 ) − mf00 , η ∈ [0, ∞), (5.1.9)
Pr 2

where θ0 = θ0 (η) and the boundary conditions are as follows

θ0 (η) = 1 and θ0 (∞) = 0. (5.1.10)

In order to apply the spectral collocation method upon (5.1.9), {η0 , η1 , ..., ηN −1 , ηN } are represented
as grid points. The function θ0 (η) is approximated by an interpolating polynomial, formulated in
terms of values of θ at each collocaton point. Thus

θ0 = {θ00 , θ01 , ..., θ0N −1 θ0N }T

is the vector of unknowns. The derivatives at collocation points are expressed as

d2 θ0

dθ0
= Dθ0 and = D2 θ0 , i = 0, 1, ..., N. (5.1.11)
dη η=ηi dη 2 η=ηi

Substituting (5.1.11) into (5.1.9) gives

B0 Θ0 = Rθ0 , (5.1.12)

where
   
1 + NR 1+m
B0 = 2
D + f0r D − 2mf00 r + I,
Pr 2
 
2
Rθ0 = −Ec (f000 ) − mf00

and I is the identity matrix of size (Nη + 1) × (Nη + 1); the boundary conditions are

θ0 (η0 ) = 0 and θ0 (ηN ) = 1.

195 Solving (5.1.12), θ0 is then obtained.

5.2. Application of the BI-SQLM

We consider equation (3.2.12) and let H be the non-linear operator of the equation,
0
   
000 1+m 00 0 2
 0 0 ∂f 00 ∂f
H=f + f f + m 1 − (f ) + ξ(1 − f ) − (1 − m)ξ f −f , (5.2.1)
2 ξ ∂ξ

20
where H = H[η, f (η), f 0 (η), f 00 (η), f 000 (η)] and f = f (η, ξ). The linearised equation is shown to be

∂fr+1 ∂f 0
000
a3,r (η, ξ)fr+1 00
+ a2,r (η, ξ)fr+1 00
+ a1,r (η, ξ)fr+1 + a0,r (η, ξ)fr+1 + a4,r (η, ξ) + a5,r (η, ξ) r+1
∂ξ ∂ξ

∂fr+1 ∂f 0
−ξ(1 − m)fr00 + ξ(1 − m)fr0 r+1 = Rr (η, ξ), (5.2.2)
∂ξ ∂ξ
where

∂f 0
 
1+m
a0,r (η, ξ) = fr00 , a1,r (η, ξ) = −2mfr0 − ξ − (1 − m)ξ r ,
2 ∂ξ
 
1+m ∂fr
a2,r (η, ξ) = fr + (1 − m)ξ , a3,r (η, ξ) = 1,
2 ∂ξ
a4,r (η, ξ) = (1 − m)ξfr00 , a5,r (η, ξ) = −(1 − m)ξfr0 ,

and

∂f 0
 
1+m ∂fr
Rfr (η) = −m(fr0 )2 + fr fr00 + (1 + m)ξfr00 − (1 − m)ξfr0 r − m − ξ.
2 ∂ξ ∂ξ

Applying the collocation at ξi in (5.2.2) we attain



(i) (i) (i) (i) (i)
X
000 00 00
a3,r (η, ξ)Fr+1,i (η) + a2,r (η, ξ)Fr+1,i (η) + a1,r (η, ξ)Fr+1,i (η) + a0,r (η, ξ)Fr+1,i (η) + a4,r di,j Fj (η)
j=0
Nξ Nξ Nξ
(i)
X X X
+ 2a5,r di,j Fj0 (η) − 2ξi (1 − 00
m)Fr,i di,j Fj (η) + 2ξi (1 − 0
m)Fr,i di,j Fj0 (η) = Rfr (η, ξ),
j=0 j=0 j=0

(5.2.3)

where
Nξ Nξ
∂f 0

∂f X X
= di,j Fj (η) and = di,j Fj0 (η), i = 1, ..., Nξ ,
∂ξ ξ=ξi j=0 ∂ξ ξ=ξi j=0

is the derivative of f and f 0 with respect to ξ. The derivatives with respect to η are expressed as

dp Fr+1,i
X
= Dp Fr+1,i , (5.2.4)
dξ p η=ηj
k=0

where p is the order of the derivative. Substituting (5.2.4) into (5.2.3) gives

(i)
A(i) Fr+1,i = R1 , (5.2.5)

21
where

Ai,i = A(i) , i = 0, 1, ...Nξ − 1,

Ai,j = −(1 − m)ξi di,Nξ D + (1 − m)ξi di,Nξ when i 6= j.

The boundary conditions are depicted as


N
X N
X
D0,k fk = 1, fr+1 (ηN ) = 0 and DN,k fk = 0. (5.2.6)
k=0 k=0

The approximate solution for (3.2.12) is attained by solving for F from the matrix representation
(5.2.5). Thus knowing the solution to f (η, ξ) equation (3.2.13) may be solved. Letting G be a
non-linear operator for (3.2.13), the following coefficients are acquired
 
1+m ∂f
b0,r (η, ξ) = −2mfr + φξ, b1,r (η, ξ) = fr + (1 − m) ξ ,
2 ∂ξ
(1 + NR )
b2,r (η, ξ) = , b3,r (η, ξ) = − (1 − m) ξfr0
Pr

and
 
2 2
Rθr (η, ξ) = −Ec (fr00 ) − (m + ξ) fr0 + ξ (fr0 ) .

Therefore the linearised equation can be expressed as

∂θr+1 00 0
b3,r + b2,r θr+1 + b1,r θr+1 + b0,r θr+1 = Rθr , (5.2.7)
∂ξ

where the derivatives with respect to ξ are respresented as



∂θ X
= di,j Θj (η),
∂ξ ξ=ξi j=0

Applying collocation at ξi we get


Nξ −1
(i) (i) (i) (i) (i) (N )
X
b3,r di,j Θj (η) + b2,r Θ00r+1 (η) + b1,r Θ0r+1 (η) + b0,r Θr+1 = Rθr − b3,rξ di,j ΘNξ (η). (5.2.8)
j=0

The derivatives with respect to η are set in terms of the chebyshev differentiation matrix are

∂Θr+1,i X p
= Di,j Θj (η) = Dp Θr+1,i . (5.2.9)
∂η η=ηi
k=0

22
Substituting (5.2.9) into (5.2.8) gives
Nξ −1
(i) (i)
X
B(i) Θr+1,i + b3,r di,j Θr+1,j = R2 , (5.2.10)
j=0

where

(i)
Bi,i = B(i) + b3,r di,Nξ , i = 0, 1, ..., Nξ − 1,
(i)
Bi,j = b3,r di,Nξ , i 6= j

and I is the identity matrix. The boundary conditions are as follows

θr+1 (η0 ) = 0 and θr+1 (ηN ) = 1. (5.2.11)

Solving for Θ in (5.2.10) gives the approximate solution for θ(η, ξ).

5.3. Results and analysis

The solutions of equations (3.2.12) - (3.2.13) are graphically illustrated. The diagrams show the
200 impact of the assorted parameters on velocity and temperature profiles, as well as the local skin-
friction coefficient and Nusselt number.

In the work Chamkha et al. [32] published, initial guesses for equations (3.2.12) - (3.2.13) were
not specified, hence we made use of the initial solutions that were calculated in the earlier section
205 5.1. We also make comparisons to some results that previous researchers obtained to confirm the
validity of the BI-SQLM employed here.

The effects of suction reduces both hydrodynamic and thermal boundary layers. This means that
there is an increase in fluid velocity and a decrease in fluid temperature. It is anticipated that
210 the opposite effect occurs when injection is imposed, see Figure 8. In Figure 9 the local Nusselt
number and skin-friction coefficient upsurges when f > 0 and reduces when f < 0 over the region
ξ. An increment in suction causes a relatively higher Nusselt number; this indicates that convection
dominates the transfer of thermal energy at the surface considerably.

215 In Figure 10 we observe that as the pressure gradient increases the velocity reaches a plateau

23
1 1 f0 = −1.0
f0 = −0.5
0.8 0.8 f0 = 0.0
f0 = 0.5
0.6 0.6
f0 = 1.0
f0

θ
f0 = −1.0
0.4 0.4
f0 = −0.5
0.2 f0 = 0.0 0.2
f0 = 0.5
0 0
f0 = 1.0

−0.2 −0.2
0 2 4 6 8 10 0 2 4 6 8 10
η η

Figure 8: The influence of f0 (suction and injection) on velocity profiles (left) and temperature profiles (right) for
m = 0.3333, Pr = 0.733, Ec = 0, φ = 0, NR = 0 and ξ = 1.0.

7 4
f0 = −1.0 f0 = −1.0
6 3.5 f0 = −0.5
f0 = −0.5
f0 = 0.0 3 f0 = 0.0
5
f0 = 0.5 f0 = 0.5
2.5
1/2

−1/2
0.5Cf Rex

4 f0 = 1.0 f0 = 1.0
Nux Rex

2
3
1.5
2
1
1 0.5

0 0
0 2 4 6 8 10 0 2 4 6 8 10
ξ ξ

Figure 9: The repercussions of f0 on the skin-friction coefficient (left) and the Nusselt number (right) for m = 0.3333,
Pr = 0.733, Ec = 0, φ = 0, NR = 0 and η = 16.

and the temperature diminishes. This is as one would expect given that an increase in the pressure
gradient is accompanied by an increase in the hydrodynamic boundary layer and a decrease in the
thermal boundary layer. In Figure 11 we note that an accretion of m increases the skin-friction coef-
ficient as well as the Nusselt number. When NR (thermal radiation parameter) rises the conduction
220 effect of the thermal boundary layer increases, as observed in Figure 12. This causes the temper-
ature to rise at every point away from the wedge surface. An increment in NR decreases the local
Nusselt number due to it being negatively related to the value of the temperature profile at the wall.

24
1
m=0
1
m = 0.3333
0.8 m = 0.5
0.8 m = 1.0
0.6
0.6
f0

θ
0.4
0.4 m=0
m = 0.3333 0.2
0.2 m = 0.5
m = 1.0
0
0
0 5 10 15 0 5 10 15
η η

Figure 10: Impact of pressure gradient m on velocity profiles (left) and temperature profiles (right) when Ec =
0.0, f = 0.0, NR = 5.0, Pr = 0.733, φ = 0.0 and ξ = 1.0.

3.5 0.5

3
0.4
2.5
1/2

−1/2

0.3
0.5Cf Rex

2
Nux Rex

1.5
0.2

1 m=0 m=0
m = 0.3333 m = 0.3333
0.1
0.5 m = 0.5 m = 0.5
m = 1.0 m = 1.0
0 0
0 2 4 6 8 10 0 2 4 6 8 10
ξ ξ

Figure 11: The affect of the pressure gradient m upon the skin-friction coefficient (left) and the Nusselt number
(right) when Ec = 0.0, f = 0.0, NR = 5.0, Pr = 0.733, φ = 0.0 and η = 16.

Figure 13 illustrates the changes that are brought about in the temperature profile at ξ = 1.0
225 and ξ = 3.0. The presence of a heat source within the flow has a propensity to increase the thermal
state of the fluid. This is depicted by an increase in the temperature profile at every point away
from the wedge surface as φ increases. On the contrary the presence of a heat sink (φ < 0) in the
flow reduces the temperature of the fluid away from the wall. The increase in heat generation or
the absorption coefficient φ appears to have no effect on the thermal boundary layer for ξ = 1.0.

25
1 0.8
NR = 0
0.7
0.8 NR = 1.0
NR = 5.0 0.6

0.6 NR = 10 0.5

−1/2
Nux Rex
0.4
θ

0.4
0.3
NR = 0
0.2 0.2 NR = 1.0
NR = 5.0
0.1
0 NR = 10
0
0 5 10 15 0 2 4 6 8 10
η ξ

Figure 12: The effect of thermal radiation NR on temperature profiles (left) and the Nusselt number (right), when
Ec = 0.0, f = 0.0, m = 0.3333, Pr = 0.733, φ = 0.0, ξ = 1.0 and η = 16 respectively.

2
1 φ = −1.0 φ = −1.0
φ = −0.5 φ = −0.5
φ=0 1.5 φ=0
0.8
φ = 0.5 φ = 0.5
φ = 1.0 φ = 1.0
0.6
1
θ

0.4

0.5
0.2

0
0
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
η η

Figure 13: The impact of φ (heat generation or absorption) on the temperature profile for Ec = 0.0, f = 0.0, m =
0.3333, NR = 0.0, Pr = 0.733, φ = 0.0.

230 However for ξ = 3.0 a peak is exhibited; the peak value is higher than that of the wall. In Figure
14 it is observed that heat absorption causes the local Nusselt number to be enhanced whereas the
the heat generation diminishes.

Figure 15 allows us to observe that an increasing Eckert number has an incremental influence
235 on the Nusselt number in the presence and in the absence of thermal radiation but more so when
NR = 0. This is contrary to what Yih (1999) [31] obtained in his work. It is shown in Table 1 and

26
4

−1/2
Nux Rex
0 φ = −1.0
φ = −0.5
−1 φ=0
φ = 0.5
−2 φ = 1.0

−3

−4
0 2 4 6 8 10
ξ

Figure 14: The impact of φ on the Nusselt number for Ec = 0.0, f = 0.0, m = 0.3333, NR = 0.0, Pr = 0.733, φ = 0.0
where and η = 16.

1.25
Ec = 1
Ec = 1
Ec = 0.5 1.2
0.55 Ec = 0.5
Ec = 0.1
Ec = 0.1
Ec = 0 1.15 Ec = 0
0.5
1.1
−1/2

−1/2
Nux Rex

Nux Rex

0.45 1.05

1
0.4
0.95
0.35
0.9

0.3 0.85
0 2 4 6 8 10 0 2 4 6 8 10
ξ ξ

Figure 15: The influence of Ec on the Nusselt number for Ec = 0.0, f = 0.0, m = 0.3333, Pr = 0.733, φ = 0.0, η =
16, NR = 0.0 (left) and NR = 10 (right).

2 that the results of the BI-SQLM compare favourably with results obtained by previous authors.
This further validates the credibility and accuracy of the numerical technique applied.

5.4. Convergence and error analysis


240 The application of BI-SQLM upon equations (3.2.12) - (3.2.13) results in convergence with respect
to η and ξ to be reached after five iterations for both f and θ. The elapsed time is 0.3353 seconds
and the error is below 10−11 . Convergence as well as residual error for the derivatives of f and θ
are also shown, akin results are attained.

27
Table 1: Comparison of the values of f 00 (ξ, 0) for m = 1, f0 = 0.

ξ Sparrow et al. (1962) [38] Ariel (1994) [39] Yih (1999) [31] Chamkha et al. (2003) [32] BI-SQLM (Present results)

0 1.231 1.232588 1.232588 1.232710 1.23258766


1 1.584 1.585331 1.585331 1.585480 1.57354438
4 2.345 2.346663 2.346663 2.341002 2.35821589
9 - 3.240950 3.240950 3.240992 3.24342037
25 - 5.147965 5.147964 5.148002 5.15565377
100 - 10.074741 10.074741 10.075001 10.07474113

Table 2: Comparison of the values of −θ(ξ, 0) for different values of Pr and Ec where m = 0, f0 = 0, NR = 0 and
φ = 0.

Watanabe and Pop (1994) [40] Yih (1999) [31] Chamkha et al. (2003) [32] BI-SQLM (Present results)

Pr ξ Ec = 0 Ec = 1 Ec = 0 Ec = 1 Ec = 0 Ec = 1 Ec = 0 Ec = 1

0.733 0.0 0.29755 0.12395 0.297526 0.170272 0.297600 0.170184 0.297526 0.170272
0.5 0.35699 0.20871 0.357022 0.210072 0.357040 0.209858 0.356980 0.210045
1.0 0.38336 0.22857 0.382588 0.228813 0.383191 0.229000 0.382553 0.228786
1.5 0.39959 0.24122 0.398264 0.240798 0.399980 0.241002 0.398239 0.240772
2.0 0.41091 0.25022 0.409168 0.249316 0.409450 0.249307 0.409148 0.249298

1 0.0 0.33206 0.16603 0.332057 0.166027 0.332173 0.166302 0.332057 0.166029


0.5 0.40280 0.20144 0.402864 0.201452 0.403103 0.201630 0.402816 0.201426
1.0 0.43446 0.21727 0.433607 0.216814 0.433901 0.216721 0.433567 0.216787
1.5 0.45413 0.22710 0.452634 0.226323 0.452808 0.226313 0.452604 0.226296
2.0 0.46798 0.23401 0.465987 0.232988 0.466111 0.233102 0.465963 0.232981

6. Model III

245 6.1. Determining initial guesses for f (η, ξ), g(η, ξ) and θ(η, ξ)

Starting with equation (3.3.10), let ξ = 0 then

1
f 000 + ηf 00 = 0, (6.1.1)
4

integrating equation (6.1.1) we obtain the initial guess for f , which is


   2r r
η η 2 2
f (η) = ηerf √ + 2 exp − −2 , (6.1.2)
2 2 8 π π

28
η η
0 0
10 16 10 16
15.4686 15.4686
13.0346 13.0346
−5 −5
10 10
Ef


−10 −10
10 10

−15 −15
10 10

0 5 10 15 20 0 5 10 15 20
Number of iterations Number of iterations

Figure 16: Convergence and residual error of f (left) and θ (right) with respect to η.

η η
0 0
10 16 10 16
15.4686 15.4686
13.0346 13.0346
−5 −5
10 10
Ef 0

E θ0

−10 −10
10 10

−15 −15
10 10

0 5 10 15 20 0 5 10 15 20
Number of iterations Number of iterations

Figure 17: Convergence and residual error of f 0 (left) and θ0 (right) with respect to η.

where Z ∞
2
erf (η) = 1 − √ exp (−z 2 ) dz.
π η
Similarly for equation (3.3.11) we let ξ = 0 to obtain
η
g 00 + g 0 = 0, (6.1.3)
4
Solving equation (6.1.3) through integration the initial guess for g becomes
 
η
g(η) = erfc √ . (6.1.4)
8
Finding the initial resolution for θ, ξ = 0 is imposed upon equation (3.3.12) which gives

P rη 0
θ00 + θ = 0. (6.1.5)
4

29
Making use of integration upon (6.1.5) gives
 
P rη
θ(η) = erfc √ (6.1.6)
8
as the initial guess of θ.

6.2. Application of the BI-SQLM


Beginning with equation (3.3.10), which is equated to a non-linear operator F F ,

1 1  2
 1
f 000 + η (1 − ξ) f 00 + ξf f 00 + ξ 1 − (f 0 ) + λg 2 + M ξ (1 − f 0 )
4 2 2
1 1 ∂f 0
+ αξθ − ξ (1 − ξ) = F F, (6.2.1)
2 2 ∂ξ
where F F = F F [η, f (η), f 0 (η), f 00 (η), f 000 (η)]. The linearised equation becomes
0
000 00 00 ∂fr+1
a0,r (η, ξ)fr+1 + a1,r (η, ξ)fr+1 + a2,r (η, ξ)fr+1 + a3,r (η, ξ)fr+1 + a4,r (η, ξ)
∂ξ
+ a5,r (η, ξ)gr+1 + a6,r (η, ξ)θr+1 = Rf = R1 (6.2.2)

where
η
a0,r (η, ξ) = 1, a1,r (η, ξ) = (1 − ξ) + ξfr ,
4
1
a2,r (η, ξ) = −ξfr − M ξ, a3,r (η, ξ) = ξfr ,
2
1
a4,r (η, ξ) = − ξ (1 − ξ) , a5,r (η, ξ) = ξλgr ,
2
1
a6,r (η, ξ) = αξ,
2
and
1 2 1
Rf = − ξ (fr0 ) + ξfr00 fr − ξ (1 + M ) . (6.2.3)
2 2
Similarly for equation (3.3.11) a non-linear operator GG represents,
1 1 1 ∂g
g 00 + η (1 − ξ) g + ξ (f g − gf 0 ) − M ξg − ξ (1 − ξ) = GG, (6.2.4)
4 2 2 ∂ξ
where GG = GG[η, g(η), g 0 (η), g 00 (η)]. The linear form of GG is,

00 0 ∂gr+1
b0,r (η, ξ)gr+1 + b1,r (η, ξ)gr+1 + b2,r (η, ξ)gr+1 + b3,r (η, ξ)
∂ξ
0
+ b4,r (η, ξ)fr+1 + b5,r (η, ξ)fr+1 = Rg = R2 (6.2.5)

30
and the coefficients are defined as
η
b0,r (η, ξ) = 1, g1,r (η, ξ) = (1 − ξ) + ξfr ,
4
1 1
b2,r (η, ξ) = −ξfr0 − M ξ, b3,r (η, ξ) = − ξ (1 − ξ) ,
2 2
b4,r (η, ξ) = ξgr0 , b5,r (η, ξ) = −ξgr ,

with

Rg = ξ (gr fr0 + fr gr0 ) . (6.2.6)

Equation (3.3.12) is also equated to a non-linear operator ΘΘ,


1 1 ∂θ
θ00 + P rη (1 − ξ) θ0 + P rξf θ0 − P rξ (1 − ξ) = ΘΘ, (6.2.7)
4 2 ∂ξ
where ΘΘ = ΘΘ[η, θ(η), θ0 (η), θ00 (η)]. The linearised equation of (6.2.7) is
00 0 ∂θr+1
c0,r (η, ξ)θr+1 + c1,r (η, ξ)θr+1 + c2,r (η, ξ)θr+1 + c3,r (η, ξ)
∂ξ
+ c4,r (η, ξ)fr+1 = Rθ = R3 , (6.2.8)

where
η
c0,r (η, ξ) = 1, c1,r (η, ξ) = P r (1 − ξ) + ξP rfr ,
4
1
c2,r (η, ξ) = 0, c3,r (η, ξ) = − P rξ (1 − ξ) ,
2
c4,r (η, ξ) = ξP rθr0

and

Rθ = P rξθr0 fr . (6.2.9)

We now set up equations (3.3.10) - (3.3.12) in the form of matrices.


The matrix form for equation (3.3.10) is

(i) (i) (i)
X
A11 Fi + a4,r di,j DFj + A12 Gi + A13 Θi = R1,i , (6.2.10)
j=0

where i = 0, 1, ..., Nξ − 1.
Similarly the matrix notation for equation (3.3.11) is represented as

(i) (i)
X
A21 Fi + A22 Gi + b3,r di,j Gj = R2,i , i = 0, 1, ..., Nξ − 1. (6.2.11)
j=0

31
Lastly for equation (3.3.12) we obtain

(i) (i)
X
A31 Fi + A33 Θi + c3,r di,j Θj = R2,i , i = 0, 1, ..., Nξ − 1. (6.2.12)
j=0

The Bi,i matrices are incorporated as they depict the time direction,

(ii) (i)
B11 = A11 + a4,r di,i D,
(ii) (i)
B12 = A12 ,
(ii) (i)
B13 = A13 ,
(ii) (i)
B21 = A21 ,
(ii) (i)
B22 = A22 + b3,r di,i I,
(ii) (i)
B23 = A23 ,
(ii) (i)
B31 = A31 ,
(ii) (i)
B32 = A32 ,
(ii) (i)
B33 = A33 + c3,r di,i I.

where i = 0, 1, ..., Nξ − 1.
The off-diagonal elements of B are as follows

(ij)
B11 = a4,r di,i D,
(ij)
B12 = 0.I,
(ij)
B13 = 0.I,
(ij)
B21 = 0.I,

(ij)
B22 = b3,r di,i I,
(ii)
B23 = 0.I,
(ii)
B31 = 0.I,
(ii)
B32 = 0.I,
(ii)
B33 = c3,r di,i I.

32
250 where i = 0, 1, ..., Nξ − 1. The matrix representation of the above is given as

    
B11 B12 B13 F0 R1,0
    
B23   G0  =  R2,0 , (6.2.13)
    
 B21 B22
    
B31 B32 B33 Θ0 R3,0
where the boundary conditions for f , g and θ are represented as follows
N
X N
X
D0,k fk = 1, fr+1 (ηN ) = 0, DN,k fk = 0, (6.2.14)
k=0 k=0

gr+1 (η0 ) = 0, gr+1 (ηN ) = 1, (6.2.15)

θr+1 (η0 ) = 0, and θr+1 (ηN ) = 1. (6.2.16)

In solving (6.2.13) solutions for f , g and θ are obtained.

6.3. Results and Analysis

Similarly to model I and II the results are shown using diagrams for various parameters. The initial
guesses calculated in section 5.1 are employed when computing solutions presented in this section.
255 The initial guesses that Dinarvand et al. [36] used and the corresponding results are compared to
the results displayed below; however the elapsed time for convergence is 4.01345 seconds in compari-
son to 3.96527 seconds which we obtained when employing the initial guesses derived in section 6.1.

In Figure 18 - see left - the effect of an increase in the buoyancy parameter α is an increase in
260 f , which means that there is a strong presence of the pressure gradient which accelerates the fluid.
A similar effect is exhibited in the plot on the right of Figure 18 when f is plotted against η.
In Figure 19 variations of M (magnetic parameter) show that f increases as the magnetic parameter
increases. The same is observed when we plot f against η for several values of M .

265 Initially f 0 increases rapidly as α increases; peaks for α = 3 and α = 5 are quite visible and a
substantial presence of the buoyancy force is experienced by the velocity distribution when α = 5

33
1.8 2
α=0 α=0
1.7 α=1 α=1
α=3 1.5
1.6 α=3
α=5 α=5
1.5
1
1.4
f

f
1.3 0.5

1.2
0
1.1

1 −0.5
0 0.2 0.4 0.6 0.8 1 0 0.5 1 1.5 2
ξ η

Figure 18: The influence of α on f plotted against ξ where α = 1, Pr = 0.7 and λ = 1 (left) and the impact of α on
f plotted against η where M = 1, Pr = 0.7 and λ = 1 (right).

1.8 2
M=0 M=0
1.7 M=1 M=1
M=3 1.5 M=3
1.6
M=5 M=5
1.5
1
1.4
f

1.3 0.5

1.2
0
1.1

1 −0.5
0 0.2 0.4 0.6 0.8 1 0 0.5 1 1.5 2
ξ η

Figure 19: The effect of M on f plotted against ξ where α = 1, Pr = 0.7 and λ = 1 (left) and the effect of M on f
plotted against η where α = 1, Pr = 0.7 and λ = 1 (right).

in Figure 20. This is because positive buoyancy forces operate as a favourable pressure gradient
which expedites the fluid in the buoyancy layer. So the higher the value of α the greater the velocity.

270 In Figure 21 (left), the effect of the different values of α display a very minimal increase in g.
For increasing values of M in Figure 21 (right) g decreases. The plot to the left of Figure 22 in-
dicates that initially f 0 , with varying Prandtl’s numbers, is increasing; visible peaks occur for Pr
= 0.7, the greatest being at Pr = 0. This means that either the presence of momentum diffusivity
(kinematic viscosity) is not there or the fluid responds very quickly to a change in temperature.

34
1.2 1.2

1 1
α=0
0.8 α=1 0.8
α=3
0.6 α=5 0.6
f0

f0
0.4 0.4

0.2 0.2 λ=0


λ=1
0 0 λ=3
λ=5
−0.2 −0.2
0 2 4 6 8 10 0 5 10 15 20
η η

Figure 20: The velocity distribution f 0 plotted against η for several values of α where M = 1, Pr = 0.7 and λ = 1
(left) and for different values of λ where M = 1, Pr = 0.7 and α = 1 (right).

1.2 1.2
α=0 M=0
1 α=1 1 M=1
α=3 M=3
0.8 α=5 0.8 M=5

0.6 0.6
g

0.4 0.4

0.2 0.2

0 0

−0.2 −0.2
0 5 10 15 20 0 5 10 15 20
η η

Figure 21: Plots for g for varying values of α where λ= 1, M = 1 and Pr = 1 (left) and the effect of M on g where
α= 1, λ = 1 and Pr = 1 (right).

275 The plot to the right of Figure 22 indicates that Prandtl’s number is the parameter that makes a
visible impact on θ; with rising values of Pr, θ declines. This means that there is a strong presence
of thermal diffusivity.

6.4. Convergence and error analysis

Convergence is obtained after five iterations for solutions of f , g and θ and the respective derivatives.
280 The errors are below 10−10 in each case as depicted in Figure 23.

35
1.4 1.2
Pr = 0
1.2 1 Pr = 0.7
Pr = 2
1 Pr = 5
0.8
0.8
0.6
f0

0.6

θ
0.4
0.4
Pr = 0 0.2
0.2 Pr = 0.7
Pr = 2 0
0
Pr = 5
−0.2 −0.2
0 5 10 15 20 0 5 10 15 20
η η

Figure 22: The influence of Pr upon θ where α= 1, λ = 1 and Pr = 1 (left) and the influence of Prandtl’s number
on the velocity distribution where λ= 1, M = 1 and α = 1 (right).

5 5
10 10
Ef Ef
Eg Eg
0 0
10 Eθ 10 Eθ
E f 0 , E g 0 , E θ0
Ef , Eg , Eθ

−5 −5
10 10

−10 −10
10 10

−15 −15
10 10
0 5 10 15 20 0 5 10 15 20
Number of iterations Number of iterations

Figure 23: Error and convergence of f , g and θ (left) and the derivatives of f , g and θ (right).

7. Concluding remarks

In this research the Chebyshev collocation spectral method was proposed for the solution of decou-
pled and coupled non-simliar boundary layer equations. The BI-SQLM was formulated by conflating
the quasi-linearisation method and Chebyshev spectral collocation method with bivariate Lagrange
285 interpolation. For the research models considered the BI-SQLM was relatively easy to implement,
computationally efficient and solutions reached convergence rapidly. Furthermore it was observed
that the residual errors were close to zero.

36
Model II specifically was compared to the solutions obtained by previous authors. It was found
290 that the BI-SQLM provided solutions which were comparable to those methods employed by other
authors [32]. Initial guesses were calculated for each of the three models considered in this paper.
The initial guesses used by Ali and Mehmood [28] for model I and Dinarvand et al. [36] for model
III were tested on the corresponding models. It was found that the convergence was slightly slower
than the initial guesses derived and employed in this work.

295 Disclosure

This paper is a collaborative effort.

Conflict of Interest

We declare that there is no conflict of interests regarding the publication of this paper.

Acknowledgements

300 The DST-NRF Centre of Excellence in Mathematical and Statistical Sciences (CoE-MaSS) is recog-
nized for its financial support. C. H. acknowledges support from the National Research Foundation,
South Africa, under Grant no. 91114.

References

[1] E. Mureithi and D. Mason, Local non-similarity solutions for a forced free boundary layer flow
305 with viscous dissipation, Mathematical and Computational Applications, 15:558-573, 2010.

[2] M. Asgharinia and M. Heydari,The analysis of the boundary layer flow on walls of ducts of a
training wind tunnel with flow mach of subsonic test, Scientific Research, 3:57-71, 2014.

[3] T. Chen, Parabolic systems: local nonsimilarity method, Handbook of Numerical Heat Transfer,
Wiley Interseience, New York, 1988.

310 [4] S. Nakamura, Iterative finite difference schemes for similar and non-similar boundary layer
equations, Advances in Engineering Software, 21:123-131, 1994.

37
[5] R. Gorla, J. Lee, S. Nakamura and I. Pop, Effects of transverse magnetic field on mixed
convection in wall plume of power law fluids, Engineering and Science, 31(7):1035-45, 1993.

[6] R. Gorla, J. Lee, S. Nakamura and I. Pop, Mixed convection in wall plume of power law fluid,
315 Mechanics, 101:47-58, 1993.

[7] N. Kafoussias and E. Williams, An improved technique to obtain numerical solution of a class
of two-point boundary value similarity problems in fluid mechanics, Numerical Methods Fluids,
17:145-62, 1993.

[8] T. Hayat, M. Mustafa and S. Asghar, Unsteady flow with heat and mass transfer of a third
320 grade fluid over a stretching surface in the presence of chemical reaction, Non-linear Analysis:
Real World Applications, 11:3186-3197, 2010.

[9] H. Xu, S. Liao and I. Pop, Series solutions of unsteady three-dimensional MHD flow and
heat transfer in the boundary layer over an impulsively stretching plate, Mechanics B/Fluids,
26:1527, 2007.

325 [10] Y. Zhao, Z. Lin and S. Liao, A modified homotopy analysis method for solving boundary layer
equations, Applied Mathematics, 4:11-15, 2013.

[11] S. Motsa, P. Dlamini, and M. Khumalo, Spectral relaxation method and spectral quasilineariza-
tion method for solving unsteady boundary layer flow problems, Advances in Mathematical
Physics, 2014.

330 [12] S. Motsa, F. Awad, Z. Makukula and P. Sibanda, The spectral homotopy analysis method
extended to systems of partial differential equations, Applied Analysis, 2014.

[13] S. Motsa and V. Magagula and P. Sibanda, A Bivariate Chebyshev Spectral Collocation Quasi-
linearization Method for Nonlinear Evolution Parabolic Equations, World Scientifc Journal,
2014.

335 [14] S. Liao, A new branch of solutions of boundary layer flows over an impermeable stretched plate,
Heat and Mass Transfer, 48:2529-2539, 2005.

[15] B. Sakiadis, Boundary layer behavior on continuous solid surface. I. Boundary layer equations
for two-dimensional and axisymmetric flow, AIChE, 7:26-8, 1961.

38
[16] F. Tsou, E. Sparrow and R. Glodstein, Flow and heat transfer in the boundary layer on a
340 continuous moving surface, Heat and Mass Transfer, 7:219-35, 1967.

[17] L. Crane, Flow past a stretching plate, ZAMP, 21(4):645-647, 1970.

[18] W. Banks, Similarity solutions of the boundary-layer equations for a stretching wall, Theoretical
and Applied Mechanics, 2:375-392, 1983.

[19] W. Banks and M. Zaturska, Eigensolutions in boundary layer flow adjacent to a stretching
345 wall, IMA J. Applied Mathematics, 36(3):263-273, 1986.

[20] L. Grubka and K. Bobba, Heat transfer characteristics of a continuous, stretching surface with
variable temperature, ASME J. Heat Transfer, 107(1):248-250, 1985.

[21] C. Chen and M. Char, Heat and mass transfer on a continuous stretching surface with suction
or blowing, Mathematical Analysis and Applications, 135:568-580, 1988.

350 [22] M. Ali, Heat transfer characteristics of a continuous stretching surface, Heat and Mass Transfer,
29(4):227-234, 1994.

[23] L. Erickson, L. Fan and V. Fox, Heat and mass transfer on a moving continuous flat plate with
suction or injection, Industrial Chemical Engineering, 15:19-25, 1966.

[24] N. Nazar, N. Amin and I. Pop, Unsteady boundary layer flow due to stretching surface in a
355 rotating fluid, Mechanics Research Communications, 31:121-8, 2004.

[25] S. Liao and H. Xu, Series solutions of unsteady magnetohydrodynamic flows of non-Newtonian
fluids caused by an impulsively stretching plate, Non-Newtonian Fluid Mechanics, 129:46-55,
2005.

[26] S. Liao, An analytic solution of unsteady boundary-layer flows caused by an impulsively stretch-
360 ing plate, Communications in Nonlinear Science and Numerical Simulation, 11:326-39, 2006.

[27] J. Williams and T. Rhyne, Boundary layer developed on a wedge impulsively set into motion,
SIAM J. Applied Mathematics, 38:215-24, 1980.

39
[28] A. Ali and A. Mehmood, Homotopy Analysis of Unsteady Boundary Layer Flow Adjacent to
Permeable Stretching Surface in Porous Medium, Communications in Nonlinear Science and
365 Numerical Simulation, 13:340-349, 2008.

[29] A. Raptis, Flow of a micropolar fluid past a continuously moving plate by the presence of
radiation, Heat and Mass Transfer, 41:2865-2866, 1998a.

[30] A. Raptis, Radiation and free convection flow through a porous medium, Communications in
Heat and Mass Transfer, 25:289-295, 1998b.

370 [31] K. Yih, MHD forced convection flow adjacent to a non-isothermal wedge, Communications in
Heat and Mass Transfer, 26:819-827, 1999.

[32] A. Chamkha, M. Mujtaba, A. Quadri and C. Issa, Thermal Radiation Effects on MHD Forced
Convection Flow Adjacent to a Non-Isothermal Wedge in the Presence of a Heat Source or
Sink, Heat and Mass Transfer, 39:305-312, 2003.

375 [33] R. Rajasekaran and M. Palekar, Mixed convection about a rotating sphere, Heat and Mass
Transfer, 28:959-965, 1985.

[34] H. Hatzikonstantinou, Effects of mixed convection and viscous dissipation on heat transfer
about a porous rotating sphere, ZAMM, 70:457-464, 1990.

[35] M. Ece, The initial boundary layer flow past a translating and spinning rotational symmetric
380 body, Mathematical engineering, 26:415-428, 1992.

[36] S. Dinarvand, A. Doosthoseini, E. Doosthoseini and M. Rashidi, Series solutions for unsteady
laminar MHD flow near forward stagnation point of an impulsively rotating and translating
sphere in presence of buoyancy forces, Nonlinear Analysis: Real World Applications, 11:1159-
1169, 2009.

385 [37] H. Takhar, A. Chamkha and G. Nath, Unsteady laminar MHD flow and heat transfer in the
stagnation region of an impulsively spinning and translating sphere in the presence of buoyancy
forces, Heat Mass Transfer, (37):397-402, 2001.

[38] E. M. Sparrow, E. R. Eckert and W. J. Minkowycz, Transpiration cooling in a magnetohydro-


dynamic stagnation point flow, Advances in Applied Science Research, 11:125147, 1962.

40
390 [39] P. D. Ariel, Hiemenz flow in hydromagnetics, Acta Mechanica, 103:3143, 1994.

[40] T. Watanabe and I. Pop, Thermal Boundary Layers in Magnetohydrodynamic flow over a Flat
Plate in the Presence of a Transverse Magnetic Field, Acta Mechanica, 105:233238, 1994.

41