INTRODUCTION
Natural convection is a mechanism of heat transport in which no external source influences the fluid motion. In fact, density difference due to the gravitational field is the major causality of this type of heat transport. Due to its potential applications in industry and engineering, this mechanism has attracted a great deal of attentions among researchers. In many cases, accurate anticipation of the behavior of fluid flows can remove harmful effects such as mechanical noises and vibrations from industrial systems (Jianu and Rosen, 2017). A very conventional industrial application of natural convection is free cooling of electronic devices. This heat transfer mechanism can exist in smallscale devices such as computer chips to largescale physical systems such as geothermal systems, heat exchangers, chemical catalytic reactors, fibre and granular insulation, packed beds, and petroleum reservoirs. In the last couple of decades, natural convection heat transfer for Newtonian and nonNewtonian fluids streaming between two infinite parallel vertical plates has been studied by different researchers. In these systems, heat is transferred by natural convection mechanism from vertical plates to fluids. Analysis of this mechanism is significant especially when the moving fluid is minimally affected by forced convection (McCabe et al., 1993). Natural convection heat transfer between vertical flat plates was analysed for several different nonNewtonian fluids by Bruce and Na (Bruce and Na, 1967). In another investigation, they considered and analysed several natural convection problems in laminar regimes (Shenoy and Mashelkar, 1982). Rajagopal (Rajagopal and Na, 1985) conducted a thermodynamic analysis on different constitutive functions in the area of fluid mechanics. It should be pointed out that in fluid mechanics problems, to yield reliable results, inherent nonlinear behavior of systems must be considered. However, solving nonlinear equations is often very difficult. To overcome the complexities, various methods such as instance Homotopy Perturbation Method (HPM) (Nayfeh, 1973; Ganji et al., 2010; Rafei and Ganji, 2006; He, 2006; Soltanian et al., 2010; Esmaeilpour and Ganji, 2007; Ganji et al., 2007; Ganji et al., 2006), Variational Iteration Method (VIM) (Wang and He, 2007), Homotopy Analysis Method (HAM) (Hassan and Rashidi, 2013; Shahzad and Ramzan, 2012), Adomian Decomposition Method (ADM) (Abbasbandy, 2007), Least Square Method (Shi and Ren, 2010), Galerkin Method (Hessari, 2014) and Collocation Method (CM) (Petroudi et al., 2014) have been introduced by different researchers. It is worth mentioning that the aforementioned methods offer accurate successive approximations for solution of ODEs. Etbaeitabari and Domairy (Etbaeitabari et al., 2013) employed VIM to evaluate the natural convection of a nonNewtonian fluid streaming between two infinite vertical plates (Etbaeitabari et al., 2013). Ziabakhsh and Domairy (Ziabakhsh and Domairry, 2009) analytically investigated the natural convection of a nonNewtonian fluid between two infinite parallel vertical flat plates considering the nature of the fluid. Hatami et al. (Hatami et al., 2014) numerically and analytically performed a heat transfer and flow analysis on a nonNewtonian third grade nanofluid flow in porous medium considering magnetic field. Sheikholeslami and Ganji (Sheikholeslami and Ganji, 2015) analytically solve the problem of nanofluid hydrothermal behavior in presence of magnetic field and Brownian motion of nanoparticles using Differential Transformation Method. Sheikholeslami et al. (Sheikholeslami et al., 2015) used differential transformation method to develop a theoretical solution for the case of laminar steady magnetohydrodymanics nanofluid streaming in a semipermeable channel. Azimi et al. (Azimi et al., 2014) analytically studied the heat transfer of an unsteady nanofluid flow between two moving parallel plates applying Galerkin Optimal Homotopy Asymptotic Method. In another paper, Azimi et al. (Azimi and Riazi, 2016) proposed an analytical solution for the problem of a steady nanofluid streaming through a convergentdivergent channel utilizing Reconstruction of Variational Iteration method. Shahriari et al. (Shahriari et al., 2018) analytically investigated the effect of viscous dissipation on temperature distribution of a twodimensional unsteady nanofluid streaming between two moving parallel plates utilizing AkbariGanji Method. The main purpose of the present paper, for the first time, is to carry out a momentum and energy analysis on nonNewtonian fluid flow between two infinite vertical flat plates in a changeable semipermeable medium considering variable heat source and magnetic field terms. Initially, partial differential equations of momentum and energy are derived and turned into ordinary differential equations using a similarity solution. Afterwards, a system of differential equations is solved by LSM and subsequently reliable functions are presented for temperature and velocity distributions. In order to evaluate the reliability of this method, the governing equations are also solved and compared by a numerical method and an analytical approach known as GM. Subsequently, the influences of Hartman number, porosity and heat source parameters on the velocity and temperature distributions are presented. Finally, variations of velocity with change types of porosity and magnetic field are studied.
GOVERNING EQUATIONS
Figure 1 illustrates the schematic of the problem analysed in the current paper. As can be seen in this figure, the system includes two vertical flat plates (the distance between the plates is considered to be 2b) and a nonNewtonian fluid streaming between the plates. In this investigation, a steady, incompressible, nonNewtonian fluid flow between two parallel vertical plates in a semipermeable medium with variable heat source and magnetic field terms is considered. The schematic of the problem is shown in Figure 1. The temperatures of the walls located at \(x = + b\) and \(x =  b\) are constant and assumed to be \(T_{2}\) and \(T_{1}\) (\(T_{1}\)>\(T_{2}\)), respectively. This temperature difference results in upward motion of fluid near the wall at \(x =  b\) and downward motion of fluid near the wall at \(x = + b\).
Generally, a viscous fluid is governed by continuity and Navier–Stokes equations. Since the fluid is considered to be incompressible, the conservation equations of momentum and energy are stated as follows (Rajagopal and Na, 1985; Hatami et al., 2014; Sheikholeslami and Ganji, 2015):
\[\rho_{f}\left( \frac{\partial u}{\partial t} + u.\nabla u \right) = divT^{*}  \frac{\text{μφ}}{k^{*}}\left( 1 + \lambda_{r}\frac{\partial}{\partial t} \right)u + \rho_{f}\left\lbrack 1  \beta_{T}\left( T  T_{m} \right) \right\rbrack g + J \times B\] 
(1) 
\[{(\rho c)}_{f}\left( \frac{\partial T}{\partial t} + u.\nabla T \right) =  k\nabla^{2}T + T^{*}.L + Q_{0}(T  T_{m})\] 
(2) 
where \(\varphi\) and \(Q_{0}\) are porosity of medium and heat source parameter, respectively. Stress in a thirdgrade nonNewtonian fluid (\(T^{*}\)) is presented in the following (Hatami et al., 2014):
\[T^{*} =  pI + \left\lbrack \mu + \beta_{3}\left( \text{tr}A_{1}^{2} \right) \right\rbrack A_{1} + \alpha_{1}A_{2} + \alpha_{2}A_{1}^{2}\] 
(3) 
where \(\beta_{3}\), \(\alpha_{1}\) and \(a_{2}\) are temperaturedependent material modules which. In Eq. 3, \( pI\) represents the spherical stress related to the restraint of incompressibility and \(L\) is velocity gradient. The definitions of the kinematical tensors, \(A_{1}\) and \(A_{2}\), can be found in (Hatami et al., 2014).
Variable magnetic field, porosity of medium and heat source are defined as follows:
\[\begin{matrix} J \times B =  \sigma B_{0}^{2}u\left( 1  \frac{x}{b} \right), & \frac{\text{μφ}}{k^{*}}u = \frac{\text{μφ}}{k^{*}}\left( 1  \frac{x}{b} \right)u, & Q = Q_{0}(T  T_{m}) \\ \end{matrix}\] 
(4) 
By using the following similarity variables, the governing momentum and energy equations can be defined in simpler forms (Rajagopal and Na, 1985):
\[\begin{matrix} U = \frac{u}{u_{0}}, & \eta = \frac{x}{b}, & \theta = \frac{T  T_{m}}{T_{1}  T_{2}} \\ \end{matrix}\] 
(5) 
By substituting the above parameters into the Navier–Stokes and energy equations, following equation are obtained:
\[\frac{d^{2}U}{d\eta^{2}} + 6\delta\left( \frac{\text{dU}}{\text{dη}} \right)^{2}\frac{d^{2}U}{d\eta^{2}} + \theta  PU\left( 1  \eta \right)  \text{Ha}^{2}U(1  \eta) = 0\] 
(6) 
\[\frac{d^{2}\theta}{d\eta^{2}} + Ec.Pr\left( \frac{\text{dU}}{\text{dη}} \right)^{2}\frac{d^{2}U}{d\eta^{2}} + 2\delta E.Pr\left( \frac{\text{dU}}{\text{d\ η}} \right)^{4} + M\theta = 0\] 
(7) 
where \(\Pr\), \(\text{Ec}\), \(\delta\), \(P\), \(\text{Ha}\) and \(M\) are Prandtl number, Eckert number, dimensionless nonNewtonian viscosity, dimensionless porosity parameter, Hartman number and dimensionless heat source parameter, respectively. These parameters are defined in the following:
\[\begin{matrix} Ec = \frac{{U_{0}}^{2}}{c(T_{1}  T_{2})}, & Pr = \frac{\text{μc}}{k}, & \delta = \frac{6\beta_{3}{U_{0}}^{2}}{\mu b^{2}}, & P = \frac{\varphi b^{2}}{k^{*}}, & M = \frac{Q_{0}b^{2}}{k}, & Ha = B_{0}b\sqrt{\frac{\sigma_{f}}{\mu}} \\ \end{matrix}\] 
(8) 
Accurate boundary conditions for this problem are presented in the following (Rajagopal and Na, 1985):
\[U = 0,\theta = + \frac{1}{2}\ \text{at}\ \eta =  1\] 
(9) 
\[U = 0,\ \theta =  \frac{1}{2}\ \text{at}\ \eta = + 1\] 
(10) 
In this study, for solving the coupled and nonlinear Eqs. (6) and (7), least square and Galerkin methods are used.
THEORETICAL METHODS
Least square and Galerkin methods are analytical methods operating based on Weighted Residuals Methods. In these methods, a differential operator represented by \(D\) is applied on the function of \(u\) to generate function \(p\).
\[D\left( u\left( x \right) \right) = p(x)\] 
(11) 
where function \(u\) is approximated by \(\widetilde{u}\) and is a linear combination of a set of functions.
\[u \cong \widetilde{u} = \sum_{i = 1}^{n}{c_{i}\varphi_{i}}\] 
(12) 
Since \(p\) cannot be calculated by applying operator \(D\) on the approximated function \(u\), an error will be appeared in the calculations which can be defined as follows:
\[E\left( x \right) = R\left( x \right) = D\left( \widetilde{u}\left( x \right) \right)  p(x) \neq 0\] 
(13) 
Minimizing the average residual of Eq. 13 is the main purpose of these three procedures.
\[\begin{matrix} \int_{}^{}{R\left( x \right)W_{i}\left( x \right) = 0}, & i = 1,2,\ldots,n \\ \end{matrix}\] 
(14) 
In LSM, \(W_{i}\) can be calculated by the following correlation:
\[\begin{matrix} W_{i} = \frac{\partial R}{\partial c_{i}}, & i = 1,2,\ldots,n \\ \end{matrix}\] 
(15) 
In GM, \(W_{i}\) is defined as follows:
\[\begin{matrix} W_{i} = \frac{\partial\widetilde{u}}{\partial c_{i}}, & i = 1,2,\ldots,n \\ \end{matrix}\] 
(16) 
It is notable that the number of weight functions are equal to the number of unknown parameters of the approximated function which must satisfy the considered initial and boundary conditions.
RESULTS AND DISCUSSION
Comparison of Analytical and Numerical Results
As previously expressed, LSM is used for solving the nonlinear governing equations. Analysis and solution of the governing equations are implemented in Matlab. By applying the boundary conditions, following approximated functions are considered for velocity and temperature distributions, respectively. It is worth mentioning that these functions are calculated by using tryanderror technique.
\[U = 0.5\eta + c_{1}\left( \eta^{2}  1 \right) + c_{2}\left( {\eta  \eta}^{3} \right)\] 
(17) 
\[\theta = c_{3}\left( \eta^{2}  1 \right) + c_{4}\left( {\eta  \eta}^{3} \right)\] 
(18) 
By using the velocity and temperature approximated functions and applying the LSM, unknown coefficients in Eqs. 17 and 18 will be obtained. For validation purpose, the values of nondimensionalized numbers including \(\delta\), \(\Pr\), \(\text{Ha}\), \(P\), \(M\) and \(\text{Ec}\) are taken to be unity and constant. By doing some algebraic computations, following equations are achieved for velocity and temperature distributions:
\[U =  1.565 \times 10^{ 5}\left( \eta^{2}  1 \right)  0.0758\left( {\eta  \eta}^{3} \right)\] 
(19) 
\[\theta = 0.5\eta  8.3463 \times 10^{ 5}\left( \eta^{2}  1 \right)  0.0885\left( {\eta  \eta}^{3} \right)\] 
(20) 
Results calculated by the current numerical and analytical solutions for flow velocity and temperature are indicated in Figures 2 and 3, respectively. As can be observed in these figures, there is an appropriate agreement between the numerical and analytical results. Hence, it can be concluded that for this problem LSM is an efficient technique for solving the momentum and energy equations.
Tables 1 and 2 list the values of velocity and temperature calculated by using the numerical and analytical approaches which includes LSM and GM for two different cases. As presented in Table 1, for the first case, \(\delta = Pr = Ec = M = P = Ha = 1\). In this case, porosity and magnetic field parameters are constant along the channel width. The average error related to LSM for nondimensionalized functions of velocity and temperature are found to be 0.034% and 0.16% in this case, respectively. The average error related to GM for nondimensionalized functions of velocity and temperature are found to be 0.035% and 0.17%, respectively. By comparing the values of errors, it is implied that precision and efficiency of both of the methods are considerable high.
Table 1. Results obtained by LSM, GM and numerical (Num) methods for U(𝞰) and θ(𝞰) when \(\mathbf{\delta = Pr = Ec = M = P = Ha = 1}\) and are constant along the channel width
\[\mathbf{\delta = Pr = Ec = M = P = Ha = 1}\]

𝞰

U

θ


LSM

GM

Num

LSM

GM

Num

1

0

0

0

0.5

0.5

0.5

0.9

0.01296

0.013011

0.013146719

0.465147

0.465708161

0.466013

0.8

0.021828

0.021913

0.022204199

0.425513

0.426457054

0.427241

0.7

0.027059

0.027163

0.027552464

0.381631

0.382797641

0.384146

0.6

0.029106

0.029218

0.029639184

0.334031

0.335280886

0.337197

0.5

0.028426

0.028534

0.028947952

0.283244

0.28445775

0.286877

0.4

0.025473

0.025568

0.025971096

0.229801

0.230879199

0.233686

0.3

0.0207

0.020775

0.021192268

0.174232

0.175096193

0.178145

0.2

0.014563

0.014613

0.015078778

0.117069

0.117659697

0.120793

0.1

0.007517

0.007538

0.008081113

0.058842

0.059120673

0.062191

0

1.57E05

6.53E06

0.000636827

8.35E05

3.01E05

0.002913

0.1

0.00749

0.00753

0.006823278

0.05868

0.059061105

0.05645

0.2

0.01453

0.0146

0.013868076

0.11691

0.117601934

0.1153

0.3

0.02067

0.02076

0.020058962

0.17408

0.175041439

0.17305

0.4

0.02545

0.02556

0.024943712

0.22966

0.230828656

0.2291

0.5

0.0284

0.02852

0.028052555

0.28312

0.284412623

0.28287

0.6

0.02909

0.02921

0.028898561

0.33392

0.335242377

0.33383

0.7

0.02704

0.02716

0.026985106

0.38155

0.382766955

0.38145

0.8

0.02182

0.02191

0.021823057

0.42545

0.426435393

0.42527

0.9

0.01295

0.01301

0.012957597

0.46511

0.465696729

0.4649

1

0

0

0

0.5

0.5

0.5


Table 2. Results obtained by LSM, GM and numerical (Num) methods for U(𝞰) and θ(𝞰) when \(\mathbf{\delta = Pr = Ec = M = P = Ha = 1}\) and vary along the channel width
\[\mathbf{\delta = Pr = Ec = M = P = Ha = 1}\]

𝞰

U

θ


LSM

GM

Num

LSM

GM

Num

1

0

0

0

0.5

0.5

0.5

0.9

0.016768

0.015655

0.014942

0.495076

0.485015

0.48933

0.8

0.029167

0.026988

0.025954

0.481861

0.462687

0.469198

0.7

0.037629

0.034445

0.03339

0.460947

0.433625

0.440541

0.6

0.04259

0.038472

0.03762

0.432925

0.398439

0.404365

0.5

0.044483

0.039514

0.039024

0.398385

0.357739

0.361712

0.4

0.043743

0.038017

0.037981

0.357919

0.312135

0.313631

0.3

0.040803

0.034426

0.034871

0.312117

0.262234

0.261151

0.2

3.61E02

0.029188

0.030073

0.26157

0.208648

0.205254

0.1

0.030063

0.022747

0.023968

0.206869

0.151986

0.146857

0

0.023131

0.01555

0.016948

0.148605

0.092857

0.086792

0.1

0.015736

0.008042

0.009417

0.087369

0.031871

0.025795

0.2

0.008313

0.000668

0.001803

0.023752

0.03036

0.0355

0.3

0.001295

0.00613

0.00544

0.04166

0.09323

0.09658

0.4

0.00488

0.01189

0.01181

0.10826

0.15613

0.15704

0.5

0.00979

0.01619

0.01677

0.17548

0.21845

0.21661

0.6

0.01298

0.01857

0.01976

0.24271

0.27958

0.27511

0.7

0.01404

0.01858

0.02016

0.30937

0.33891

0.33254

0.8

0.01251

0.01579

0.01735

0.37487

0.39583

0.38897

0.9

0.00798

0.00975

0.01079

0.43861

0.44973

0.44466

1

0

0

0

0.5

0.5

0.5


As listed in Table 2, calculations are conducted for the case in which\(\ \delta = Pr = Ec = M = P = Ha = 1\). For this case, porosity and magnetic field parameters are proportional to \(1  \eta\) and linearly vary along the channel width. In this case the average error related to LSM for velocity and temperature functions are found to be 0.5% and 3.36%, and for GM the values of average error are found to be 0.088% and 0.42%, respectively. Results show that when porosity, heat source and magnetic field parameters are constant along the channel width, both of the method are efficient. Nevertheless, for the case in which the parameters vary linearly, Galerkin method possesses better performance.
Effects of Porosity on Velocity and Temperature Distributions
Figures 4 and 5 describe the influences of dimensionless porosity parameter on the velocity and temperature when the values of the other nondimensionalized parameters including \(\delta\), Pr, Ec, Ha and M are unity. According to these figures, it can be seen that porosity parameter which appears in the form of \(PU(1  \eta)\) linearly varies. With regard to Figure 4, it can be observed that with an increase in the value of porosity parameter, velocity reduces. In other words, rising this parameter results in an increment in porosity and a reduction in permeability leading to a greater resistance against the fluid flow and subsequently a decline in the value of velocity. The effect of channel width on the temperature distribution for several values of porosity parameter is delineated in Figure 5. As can be seen in Figure 5, variation of temperature with channel width for different values of porosity parameters are very close to each other. Therefore, it can be concluded that porosity has negligible impact on the flow temperature.
Figure 6 depicts the variation of velocity with position for several change types of porosity. In descending order, constant value and ascending order of porosity, porosity parameters are appeared in forms of\(\ PU(1  \eta)\), \(\text{PU}\) and \(PU(\eta + 1)\) in the momentum equation, respectively. As can be seen in Figure 6, when porosity parameter is constant, velocity distribution is symmetric along the channel width. In descending order of porosity, velocity of the flow is zero before reaching the centerline of the channel and also on the lefthand side of the centerline. In this case, the value of resistance force due to the flow impermeability on the lefthand side of the centerline is greater on the lefthand side of the centerline. In this regard, velocity of the flow will be zero in this position. In ascending order of porosity, velocity of the flow is zero on the righthand side of the centerline.
Effects of Heat Source Term on Velocity and Temperature Distributions
In Figures 7 and 8, the influences of nondimensionalized heat source parameter on velocity and temperature distributions when the other parameters including\(\ \delta\), Ha, P, Pr and Ec are unity are indicated. Regarding Figure 7, with an increment in the value of heat source parameter, velocity of flow reduces along the channel width. The reason is that any change in temperature due to existence of heat source will lead to remarkable variations in buoyancy force and subsequently the flow velocity. With respect to Figure 8, it can be seen that an enhancement in the value of heat source parameter results in a reduction in temperature along the channel width. According to Figure 8, it must be pointed out that by adding heat source term in energy equation, temperature varies nonlinearly. With increasing the value of heat source parameter, curvature of temperature distribution increases.
Effects of Magnetic Field on Velocity and Temperature Distributions
Figures 9 and 10 indicate the variations of velocity and temperature with position for different values of Hartman number. In these figures, all the considered dimensionless parameters except the Hartman number are assumed to be unity. In this case, magnetic field parameter is appeared in the form of \(\text{Ha}^{2}U(1  \eta)\) in momentum equation. As shown in Figure 9, with an increment in the value of Hartman number, velocity in each point of the channel width declines. In fact, increasing the Hartman number rises the Lorentz force which is related to the magnetic field leading to an increment in resistance against the flow and subsequently reduction in the value of velocity. In Figure 10, the effect of Hartman number on temperature distribution is shown. According to this figure, Hartman number has insignificance impact on temperature. In other words, the effect of magnetic field on temperature is negligible. Figure 11 illustrates the variation of velocity with position for several change types of magnetic field. It should be noted that in this figure all the considered nondimensionalized parameters are taken to be unity. In descending order, constant value and ascending order of magnetic field, this parameter is appeared in forms of\(\ \text{Ha}^{2}U(1  \eta)\), \(\text{Ha}^{2}U\) and \(\text{Ha}^{2}U(\eta + 1)\), respectively, in the momentum equation. As can be seen in Figure 11, when magnetic field parameter is constant, velocity distribution is symmetric along the channel width. In descending order of magnetic field, velocity of the flow is zero before reaching the centerline of the channel as well as the lefthand side of the centerline. In this case, the value of resistance force due to the magnetic field on the lefthand side of the centerline is further. In this regard, velocity of the flow will be zero in this position. In ascending order of magnetic field, velocity of the flow is zero on the righthand side of the centerline because of lower resistance in this region.
Nomenclature
\[u\] 
velocity in y direction 
U 
dimensionless velocity 
\[P\] 
porosity dimensionless parameter 
\[M\] 
heat source dimensionless parameter 
\[\Pr\] 
Prandtl number 
\[b\] 
half of plates distance 
\[c\] 
specific heat of the fluid 
\[k\] 
coefficient of conductivity 
\[\text{Ec}\] 
Eckert Number 
\[\text{Ha}\] 
Hartman Number 
\[\Pr\] 
Prandtl number 
T 
temperature field 
\[T_{m}\] 
mean temperature 
\[u_{0}\] 
inlet velocity 
\[Q_{0}\] 
heat source parameter 
\[k^{*}\] 
permeability 
\[B_{0}\] 
magnetic field 
Greek symbols

\[\delta\] 
dimensionless nonNewtonian viscosity 
\[\eta\] 
dimensionless variable of distance 
\[\mu\] 
dynamic viscosity 
\[\theta\] 
dimensionless temperature 
\[\rho_{f}\] 
fluid density 
\[\beta_{T}\] 
coefficient of thermal expansion 
\[\varphi\] 
porosity 
\[\sigma\] 
electrical conductivity 
CONCLUSION
This study tried to perform a reliable analysis on solution of momentum and energy equations for nonNewtonian fluid flow between two infinite vertical flat plates taking into account the effects of the porosity of medium, heat source and magnetic field. In this paper, porosity, heat source and magnetic field terms are considered to be variable. In the first phase of this investigation, governing PDEs were converted into ODEs using a similarity solution. Afterwards, LSM was employed to solve a system of ODEs. To evaluate the accuracy of this method, the equations were also solved by an analytical approach called Galerkin method and a numerical technique. By comparing the results of these methods, it was revealed that for the case in which magnetic field, heat source and porosity terms are constant along the channel width, both of the analytical methods reached promising results. However, when the aforementioned terms linearly vary along the channel width, results obtained by Galerkin method were more accurate. Based on the results, with an increase in the value of porosity parameter, the velocity declined while the temperature profile varied constantly. Furthermore, enhancing the value of heat source parameter led to reductions in the values of velocity and temperature. It should be pointed out that by adding heat source term, a curvature appeared in temperature profile. In other words, the temperature distribution varied nonlinearly considering this term and the profile curvature grew with rising the value of heat source parameter. With an increase in the value of Hartman number, velocity would reduce while temperature was constant. By enforcing variable porosity and magnetic field terms to momentum equation, velocity distribution would be nonsymmetric and stagnation point shifted to the right or lefthand sides of the channel centerline. The position of stagnation point relative to the channel centerline was specified by ascending, descending and constant lines.