In this work, the kinetically reduced local Navier-Stokes equations are applied to the simulation of two- and three-dimensional unsteady viscous incompressible flow problems. The reduced differential transform method is used to find the new approximate analytical solutions of these flow problems. The new technique has been tested by using four selected multidimensional unsteady flow problems: two- and three-dimensional Taylor decaying vortices flow, Kovasznay flow, and three-dimensional Beltrami flow. The convergence analysis was discussed for this approach. The numerical results obtained by this approach are compared with other results that are available in previous works. Our results show that this method is efficient to provide new approximate analytic solutions. Moreover, we found that it has highly precise solutions with good convergence, less time consuming, being easily implemented for high Reynolds numbers, and low Mach numbers.
1. Introduction
Many of the physical phenomena in fluid mechanics are formulated according to the unsteady viscous incompressible Navier-Stokes (INS) equations, which has the non-dimensional formula consisting of the momentum equations and the continuity equation [1–8](1)∂tu+u·∇u+∇p=1Re∇2u,(2)∇·u=0,where t is the physical time, u is the velocity field, p is the pressure, and Re is the Reynolds number (Re=UL/ν, where U is the scale velocity field, L is the characteristic length, and ν is the kinematic viscosity of the fluid).
Analytical and numerical solutions of INS equations are known difficulty because they are non-linear equations, and they do not find the time evolution equation for the pressure that must be determined by solving the Passion equation at each time step, which requires effort and time. Therefore, there are a lot of studies that have developed an alternative formula description of incompressible fluid flows. One of these alternative formulas is the kinetically reduced local Navier-Stokes (KRLNS) equations which was suggested in [1] for the thermodynamic description of incompressible fluid flows at low Mach numbers. The system of KRLNS equations is(3)∂tu+u·∇u+∇p=1Re∇2u,(4)∂tg=-1Ma2∇·u+1Re∇2g,such that(5)p=g+u22,where Ma is the Mach number ( Ma=U/Cs is the ratio of the characteristic flow speed U to the isentropic sound speed Cs ), and g is the grand potential. The time scale in INS equations related to that of KRLNS equations; tKRLNS(τ)=Ma×tNS.
All studies which have presented the KRLNS equations for simulation of unsteady incompressible viscous flow problems, used the numerical schemes for solving these equations. The KRLNS equations are proposed for the simulation of low Mach number flows in [2], and used the spectral element method to find the numerical solution of the three-dimensional Taylor Green vortex flow. In [3], two-dimensional KRLNS system is simplified and compared with a Chorin’s artificial compressibility method for steady state computation of the flow in a lid-driven cavity at various Reynolds numbers, the Taylor Green vortex flow is demonstrated that the KRLNS equations correctly describe the time evolution of the velocity and of the pressure, for this purpose, the explicit Mac Cormack scheme is used. In [5] the KRLNS equations are applied to two-dimensional simulation of doubly periodic shear layers and decaying homogeneous isotropic turbulence, to solve these equations have been used the central difference scheme for the spatial discretization in both advection and diffusion terms and four stages Runge-Kutta method for the time integration, the numerical results are compared with those obtained by the artificial compressibility method, the lattice Boltzmann method, and the pseudospectral method. Higher order difference approximations are used in [6] to find the solutions of the KRLNS equations which are applied for two-dimensional simulations of Womersley problem and doubly periodic shear layers.
The main purpose of this paper is to find new approximate analytical solutions for two- and three-dimensional unsteady viscous incompressible flow problems. To achieve this objective, the flow problems that are described by alternative formulas of Navier-stokes equations, which are named KRLNS (3) and (4), the reduced differential transform method (RDTM) is proposed. The reasons that encourage us to propose RDTM to solve the present problems are being an effective and efficient method to find approximate analytical solutions for nonlinear equations and we believe that it has been achieved for the first time in its study. Moreover, we extend the application of RDTM and compare its reliability and efficiency with other methods. New approximate analytical solutions for two- and three-dimensional unsteady viscous incompressible flows were found using RDTM. The results that we obtained are better than others, refer to the results in [4, 8] in accuracy, convergence, and CPU time.
The structure of this paper is organized as follows: In Section 2, we begin with some basic definitions and the use of the RDTM on the KRLNS equations. Section 3 explains the manner we adopted to discuss the convergence of the solutions. In Section 4, we apply this method to solve four flow problems of different dimensions in order to show its ability and efficiency in finding new approximate solutions. Section 5 introduces conclusions of the present work.
2. Reduced Differential Transform Method
The RDTM is an iterative procedure for obtaining a Taylor series solution of differential equations. This method is similar to the differential transform method which was first introduced by Zhou [9]. RDTM has been successfully used to many nonlinear problems [10–19] since it does not require any parameter, discretization, linearization, or small perturbations; thus it reduces the size of computational work and is easily applicable.
The main idea of this method depends on the representation the function of two variables u(x,t) as a product of single-variable function, i.e., u(x,t)=f(x)g(t), then the function u(x,t) can be represented as(6)ux,t=∑i=0∞Fixi∑j=0∞Gjtj=∑k=0∞Ukxtk,
Definition 1.
If function u(x,t) is analytic and differentiated continuously with respect to time t and space x, then(7)Ukx=1k!∂k∂tkux,tt=0,is called t-dimensional spectrum function of u(x,t), and is the transformed this function.
Definition 2.
The reduced differential inverse transform of Uk(x) is defined as(8)ux,t=∑k=0∞Ukxtk.
Then, the inverse transformation of the set of Uk(x) gives the n-terms approximation solution as follows:(9)unx,t=∑k=0nUkxtk,and the exact solution is(10)ux,t=limn→∞unx,t.
To show some basic properties of (n+1) dimensional RDTM [18], we have to consider X=(x1,x2,…,xn) to be a vector of n variables, and function u(X,t) is analytic and continuously differentiable with respect to time t and space in the domain of interest, then the fundamental mathematical operations performed by RDTM are readily obtained and listed in Table 1.
Reduced differential transformation.
Functional form
Transformed form
w(X,t)=u(X,t)±v(X,t)
Wk(X)=Uk(X)±Vk(X)
w(X,t)=αu(X,t)˙
Wk(X)=αUk(X)˙, α is constant
w(X,t)=u(X,t)v(X,t)˙
Wk(X)=∑i=0kUi(X)Vk-i˙(X)
w(X,t)=∂r∂tru(X,t)
Wk(X)=(k+1)…(k+r)Uk+r(X)˙=(k+r)!k!Uk+r(X)˙
w(X,t)=∂r1+r2⋯+rn∂x1r1∂x2r2…∂xnrnu(X,t)
Wk(X)=∂r1+r2⋯+rn∂x1r1∂x2r2…∂xnrnUk(X)
For the application of this method with the KRLNS equations to find the approximate analytical solutions for INS equations, we referred to as (KRDTM) in this paper. In order to do that we suppose that X=(x,y,z), u=(u,v,w), and Uk=(Uk,Vk,Wk), where u(X,t), v(X,t) and w(X,t) are the fluid velocity components in the x, y, and z directions and Uk(X), Vk(X), Wk(X), Gk(X), and Pk(X) are t-dimensional spectrum functions of u(X,t), v(X,t), w(X,t), g(X,t), and p(X,t), respectively, we get(11)k+1Uk+1X=-Ak+Bk+Ck+∇PkX+1Re∇2UkX,(12)k+1Gk+1X=-1Ma2∇·UkX+1Re∇2GkX,such that(13)PkX=GkX+Uk2X2,Ak=∑i=0kUiX∂Uk-iX∂x,Bk=∑i=0kViX∂Uk-iX∂y,Ck=∑i=0kWiX∂Uk-iX∂z,where k=0,1,2,3,…, U0(X)=u(X,0), V0(X)=v(X,0), W0(X)=w(X,0), and G0(X)=g(X,0). Then the exact solution is obtained as follows:(14)uX,τ=limn→∞unX,τ,where(15)unX,τ=∑k=0nUkXτk.
3. Analysis of Convergence
The convergence of the approximate analytical solutions that are resulted from the application of RDTM to INS equations is discussed by relying on the approach followed in [20, 21].
Let us consider the Hilbert space H=L2((a,b)3×[0,T]) define by(16)u:H→Rwith∫a,b3×0,Tu2X,tdXdt<∞,and the norm u2=∫(a,b)3×[0,T]u2(X,t)dXdt. Define(17)u=u,v,w:H3→R3with∫a,b3×0,Tu2X,t+v2X,t+w2X,tdXdt<∞,such that u2=u2+v2+w2.
We consider the INS equation in the following form:(18)LuX,t=NuX,t+RuX,t,which is equivalent to the following form:(19)uX,t=FukX,t,where L=∂t is the linear partial derivative with respect to t, N is a nonlinear operator, R is a linear operator, and F is a general nonlinear operator involving both linear and nonlinear terms. According to RDTM(20)k+1Uk+1X=NUkX+RUkX,and the solutions(21)uX,t=∑k=0∞UkXtk=∑k=0∞Bk,where Bk=(B1k,B2k,B3k). It is noted that the solutions by RDTM are equivalent to determining the sequence(22)S0=U0X=B0,S1=U0X+U1Xt=B0+B1,S2=U0X+U1Xt+U2Xt2=B0+B1+B2,⋮Sn=∑k=0nUkXtk=∑k=0nBk,such that Sn+1=F(Sn).
The sufficient condition for convergence of the series solution {Sn}0∞ is presented in the following theorems
Theorem 3.
The series solution {Sn=(Rn,Sn,Tn)}0∞ converges whenever there is γ such that 0<γ<1, γ=γ1+γ2+γ3, and Bi(k+1)⩽γiBik
Proof.
We show that {Sn=(Rn,Sn,Tn)}0∞ is a Cauchy sequence in the Hilbert space H3. For this reason, consider(23)Rn+1-Rn=B1n+1⩽γ1B1n⩽γ12B1n-1⩽⋯⩽γ1n+1B10,Sn+1-Sn=B2n+1⩽γ2B2n⩽γ22B2n-1⩽⋯⩽γ2n+1B20,Tn+1-Tn=B3n+1⩽γ3B3n⩽γ32B3n-1⩽⋯⩽γ3n+1B30,Using triangle inequality(24)Sn-Sm=Rn,Sn,Tn-Rm,Sm,Tm=Rn-Rm,Sn-Sm,Tn-Tm⩽Rn-Rm+Sn-Sm+Tn-Tm⩽Rn-Rn-1+Rn-1-Rn-2+⋯+Rm+1-Rm+Sn-Sn-1+Sn-1-Sn-2+⋯+Sm+1-Sm+Tn-Tn-1+Tn-1-Tn-2+⋯+Tm+1-Tm⩽γ1n+γ1n-1+⋯+γ1m+1B10+γ2n+γ2n-1+⋯+γ2m+1B20+γ3n+γ3n-1+⋯+γ3m+1B30⩽γn+γn-1+⋯+γm+1B10+B20+B30=γm+1γn-m-1+γn-m-2+⋯+1B10+B20+B30⩽γm+11-γB0since B0<∞ and 0<γ<1, we get limn,m→∞Sn-Sm=0; thus, we conclude that {Sn}0∞ is a Cauchy sequence in the Hilbert space H3, then the series solution {Sn}0∞ converges to some {S}∈H3.
Theorem 4.
Let F=(F1,F2,F3) be a nonlinear operator satisfies Lipschitz condition from a Hilbert space H3 into H3 and u(X,t) be the exact solution of INS equations. If the series solution {Sn}0∞ converges, then it is converged to u(X,t).
Proof.
Let u1(X,t),u2(X,t), and we have(25)Fu1-Fu2=F1u1,F2u1,F3u1-F1u2,F2u2,F3u2=F1u1-F1u2,F2u1-F2u2,F3u1-F3u2⩽F1u1-F1u2+F2u1-F2u2+F3u1-F3u2⩽γ1u1-u2+γ2u1-u2+γ3u1-u2=γ1+γ2+γ3u1-u2=γu1-u2.Therefore, there is a unique solution of the problem (18) by the Banach fixed-point theorem. Now we should prove that {Sn}0∞ converges to u(X,t)(26)uX,t=FuX,t=F∑k=0∞Bk=Flimn→∞∑k=0nBk=limn→∞F∑k=0nBk=limn→∞FSn=limn→∞Sn+1=S.
Definition 5.
For i=1,2,3 and k∈N⋃{0}, we define(27)γik=Bik+1Bik,Bik≠0,0,Bik=0.then we can say that ∑k=0∞Uk(X)tk converges to the exact solution u(X,t) when γk=γ1k+γ2k+γ3k and 0<γk<1 for all k∈N⋃{0}.
4. Test Problems
In this section, the KRDTM is applied to find approximate analytical solutions of four unsteady viscous incompressible flow problems, two of these problems have exact solutions and the others do not have the exact solutions. We applied KRDTM for each problem to get some approximate analytical solutions. Then the convergence of these solutions has been discussed theoretically and numerically. Finally, the results have been reviewed through some figures, which represent the velocity components and the vorticity functions, which satisfy(28)Ω=∇×u,and explain the time development with the enstrophy, which is defined as(29)ε=12V∫VΩ2dV,where V is volume for three-dimension flow problems. Our results are computed by using various value of Reynolds numbers at some time levels. All our calculations are run by Maple 18 software.
Firstproblem (P1) is two-dimensional Taylor decaying vortices flow [4, 7, 8], which describes an initially periodical vortex structure convected by the flow field and exponentially decaying due to the viscous decaying. The exact solution of this problem that achieves (1) and (2) is(30)ux,y,t=-cosxsinye-2t/Re,vx,y,t=sinxcosye-2t/Re,px,y,t=-14cos2x+cos2ye-4t/Re,when we used KRDTM for solving two-dimensional (1) and (2) equations, where X=(x,y), u=(u,v), Uk=(Uk,Vk), and tKRLNS:=τ; we get(31)k+1Uk+1x,y=-A1k+B1k+∂Pkx,y∂x+1Re∂2Ukx,y∂x2+∂2Ukx,y∂y2,(32)k+1Vk+1x,y=-A2k+B2k+∂Pkx,y∂y+1Re∂2Vkx,y∂x2+∂2Vkx,y∂y2,(33)k+1Gk+1x,y=-1Ma2∂Ukx,y∂x+∂Vkx,y∂y+1Re∂2Gkx,y∂x2+∂2Gkx,y∂y2,such that(34)Pkx,y=Gkx,y+Uk2x,y+Vk2x,y2,A1k=∑i=0kUix,y∂Uk-ix,y∂x,B1k=∑i=0kVix,y∂Uk-ix,y∂y,A2k=∑i=0kUix,y∂Vk-ix,y∂x,B2k=∑i=0kVix,y∂Vk-ix,y∂y,where k=0,1,2,3,…, U0(x,y)=u(x,y,0), V0(x,y)=v(x,y,0), and G0(x,y)=g(x,y,0); the solutions are produced as follows:(35)ux,y,τ=limn→∞unx,y,τ,(36)vx,y,τ=limn→∞vnx,y,τ,(37)gx,y,τ=limn→∞gnx,y,τ,where(38)unx,y,τ=∑k=0nUkx,yτk,vnx,y,τ=∑k=0nVkx,yτk,gnx,y,τ=∑k=0nGkx,yτk,such that(39)U1x,y=2Recosxsiny,U2x,y=-2Re2cosxsiny-2Resin2xcos2y,⋮,V1x,y=-2Resinxcosy,V2x,y=2Re2sinxcosy-2Recos2xsin2y,⋮,and(40)G1x,y=1Recos2x+cos2y+2Recos2xsin2y-sin2xsin2y-cos2ycos2x+cos2ysin2x,G2x,y=-2Re2cos2x+cos2y-8Re2cos2xsin2y-sin2xsin2y-cos2ycos2x+cos2ysin2x,⋮To prove that the condition of the convergence of these solutions verified on the domain [0,2π]2 apply Definition 5(41)γ10=U1x,yτU0x,y=2τRe,γ20=V1x,yτV0x,y=2τRe,γ30=G1x,yτG0x,y=3.771236166τRe,γ11=U2x,yτ2U1x,yτ=Re2+4τ2Re,γ21=V2x,yτ2V1x,yτ=Re2+4τ2Re,γ31=G2x,yτ2G1x,yτ=3.162277660τ2Re,⋮,such that γ0=γ10+γ20++γ30,γ1=γ11+γ21+γ31,…. For example, if t=0.5, Ma=0.001, and Re=100 such that τ=Ma×t, for all x and y in this domain, then(42)γ0=0.3885618083×10-4<1,γ1=0.5159113783×10-3<1,…and if Re=1000 then(43)γ0=3.885618083×10-6<1,γ1=0.5015821387×10-3<1,…if t=2 and Re=100 then(44)γ0=0.1554247233×10-3<1,γ1=0.2063645515×10-2<1,…and if Re=1000 then(45)γ0=0.1554247233×10-4<1,γ1=0.2006328556×10-2<1,…
The errors measurements L1, L2, and L∞-norm resulting from the application of KRDTM and the implicit central compact method (ICCM) in [8] for the computed u velocity component with CPU time for various grids at time level t=0.5, Ma=0.001, and Re=100 are tabulated in Table 2. In Table 3, L2-norm errors for u are calculated at time level t=5 and Ma=0.001 for various Reynolds numbers. The contours of the vorticity and pressure are explained in Figure 1 at t=2, Ma=0.01, and Re=100. The comparisons of the computed u and v velocity components with the exact solution along the vertical and horizontal center lines at time levels t=2, Ma=0.01, and Re=100 are shown in Figure 2. In Table 2, it can be noticed that the accuracy of new solutions and the size of the calculated errors of KRDTM are not often affected by the grid size which has been used in comparison with numerical results. Moreover, the results of KRDTM are better than ICCM [8]. Also, from this table, it is clearly shown that the KRDTM results in less computation time (CPU) than ICCM [8]. In Table 3, the same facts have been shown with various Reynolds numbers and grid spacing for L2-norm at t=5, where in some cases the CPU time reached zero. The results are given in Figure 1 show that the profiles of vorticity and pressure as contour plot are equivalent and identical with other results in [4, 7, 8]. Moreover Figure 2 compares between the exact and new approximate analytical solutions, and the identical is confirmation of the efficiency of KRDTM in solving INS equations with good convergence for different time.
The L1, L2, and L∞-norm errors for u of P1 at t=0.5 and Ma=0.001.
Grid size
method
L1-norm
L2-norm
L∞-norm
CPUs
11×11
KRDTM
3.34×10-8
5.82×10-9
1.19×10-9
0.062
ICCM [8]
3.56×10-2
4.24×10-2
7.37×10-2
6.83
21×21
KRDTM
3.19×10-8
5.69×10-9
1.19×10-9
0.172
ICCM [8]
4.71×10-4
5.73×10-4
1.09×10-3
27.81
41×41
KRDTM
3.15×10-8
5.62×10-9
1.91×10-9
0.967
ICCM [8]
7.15×10-6
8.72×10-6
1.69×10-5
107.45
81×81
KRDTM
3.15×10-8
5.59×10-9
1.91×10-9
2.57
ICCM [8]
1.67×10-7
2.05×10-7
4.01×10-7
432.14
161×161
KRDTM
3.15×10-8
5.57×10-9
1.91×10-9
10.8
ICCM [8]
3.64×10-8
4.45×10-8
8.71×10-8
1770.30
The L2-norm errors for u of P1 at t=5 and Ma=0.001.
Grid size
Re=40
Re=100
Re=500
Re=1000
Max CPUs
11×11
1.46×10-6
5.82×10-7
1.16×10-7
5.82×10-8
0
21×21
1.42×10-6
5.69×10-7
1.14×10-7
5.69×10-8
0.016
41×41
1.40×10-6
5.62×10-7
1.12×10-7
5.62×10-8
0.047
81×81
1.40×10-6
5.59×10-7
1.12×10-7
5.59×10-8
0.172
161×161
1.39×10-6
5.57×10-7
1.11×10-7
5.57×10-8
0.655
321×321
1.39×10-6
5.56×10-7
1.11×10-7
5.56×10-8
2.53
Contours plots of vorticity and pressure of P1.
The comparison of the computed u(π,y) and v(x,π) with the exact solutions of P1.
Secondproblem (P2) is Kovasznay flow [4, 7, 8], which is the laminar flow of viscous fluid behind a two-dimensional grid, with x-axis normal to the grid and the velocity field is assumed to be such that u:=U+u and v:=v, where u(x,y,t) and v(x,y,t) are the components of velocity; U is the average velocity in the x-direction. Thus, the two-dimensional INS equations with a periodicity in one direction may represent the wake of a two-dimensional grid the same as (1) with replacing the convective terms by ((U+u,v)·∇)(u,v), where U refers to one in this test. When we solved two-dimensional equations (1) and (2) by using KRDTM for this test problem, we get the same equations (31), (32), and (33) with(46)Ak=∂Ukx,y∂x+∑i=0kUix,y∂Uk-ix,y∂x,Bk=∑i=0kVix,y∂Uk-ix,y∂y,Dk=∂Vkx,y∂x+∑i=0kUix,y∂Vk-ix,y∂x,Ek=∑i=0kVix,y∂Vk-ix,y∂y.
The exact solution of the steady state of this problem [7] considers the initial conditions in this test(47)ux,y,0=1-eλxcos2πy,vx,y,0=λ2πeλxsin2πy,px,y,0=p0-12e2λx,where λ=Re/2-(Re)2+16π2/2, p0 is a reference pressure (an arbitrary constant), and p0=0 in the test. The solutions produce similar solutions in (35), (36), and (37) with(48)U1x,y=exλcos2πy4π2-λ2Re+2λ,U2x,y=-12exλcos2πy4π2-λ2Re+2λ2+2λe2xλ-12exλcos2πy-λ4π2-λ24π2e2xλsin22πy-λ2Re82π2-λ2e2xλcos22πy-4π2-λ2exλcos2πy-24π2+λ2e2xλ-λ4π2exλsin22πy,⋮,V1x,y=-λ2πexλsin2πy4π2-λ2Re+2λ,V2x,y=λ2πexλsin2πy4π2-λ2Re+2λ2+exλ4πsin2πy4π2+λ2-2cos2πy4π2-λ24π2-λ2Re+2λ-4π2-λ22πReexλsin2πyπ2-exλcos2πy4π2-λ2,⋮,and(49)G1x,y=exλRe42π2-λ2cos22πy-4π2+λ2-λ42π2sin22πyexλ-4π2-λ2cos2πy,G2x,y=exλπ2Re2λ6-12λ4π2+48λ2π4-64π6exλcos2πy+0.5λ4π2-4λ2π4+8π6cos22πy-λ6+24λ2π4-32π6exλ,⋮These solutions satisfy the conditions of convergence in the domain [-0.5,1.5]2,(50)γ10=U1x,yτU0x,y=λ2-39.47841760-2λRe2e3λ-e-λτ22Re28λ+e3λ-e-λ,γ20=V1x,yτV0x,y=λ2-39.47841760-2λRe2τ2Re2,γ30=G1x,yτG0x,y=2τ2Re2e8λ-10.5625λ8+59.21762641λ6+7013.454554λ4-46146.68129λ2+455449.4888+e4λ-1eλ584.4545462λ4-46146.68129λ2+910898.97751/2÷e8λ-10.140625λ4+18.50550825λ2+1388.079547+eλe4λ-129.60881320λ2+5844.545462+4675.636370λe2λ1/2,⋮For example, if Ma=0.01, t=0.1 such that τ=Ma×t, and Re=20, for all x and y, then(51)γ0=0.417945328×10-2<1,γ1=0.6772122966×10-1<1,…if Re=40 then(52)γ0=0.2106111743×10-2<1,γ1=0.8546834826×10-1<1,…and if Re=100 then(53)γ0=0.8607760699×10-3<1,γ1=0.1755994678<1,…
In Tables 4 and 5, the L2-norm error for the computed velocity component at Ma=0.01 for some values of Reynolds numbers is compared with the numerical results of the upwind compact finite difference method (UCFDM) in [4]. We noticed that the accuracy of the results obtained from KRDTM is higher and better than the results of UCFDM for different values of Reynolds numbers. It is clear that the maximum CPU time for all cases is not more than 30.0s, and iterations number (3) of KRDTM is less than iterations number of UCFDM (number of iterations ⩽3000). This shows that KRDTM is faster convergence and more accurate than UCFDM. The influence of the Reynolds number value on the computed vorticity and stream function ψ(x,y,t) which is satisfied(54)∂yψx,y,t=ux,y,t,∂xψx,y,t=-vx,y,t
The L2-norm error for u of P2 at t=0.1 and Ma=0.01.
Grid size
method
Re=20
Re=40
Re=100
Re=500
Max CPUs
11×11
KRDTM
4.34×10-7
5.24×10-8
1.12×10-8
2.30×10-9
0.063
UCFDM [4]
6.06×10-2
3.26×10-2
2.75×10-2
1.33×10-2
—
21×21
KRDTM
3.42×10-7
4.30×10-8
9.21×10-9
1.90×10-9
0.218
UCFDM [4]
1.40×10-2
7.37×10-3
3.85×10-7
2.33×10-3
—
41×41
KRDTM
3.02×10-7
4.03×10-8
8.98×10-9
1.87×10-9
0.624
UCFDM [4]
1.73×10-3
9.48×10-4
5.06×10-4
3.08×10-4
—
81×81
KRDTM
2.83×10-7
3.90×10-8
8.86×10-9
1.86×10-9
2.57
UCFDM [4]
1.62×10-4
9.56×10-5
5.18×10-5
3.21×10-5
—
161×161
KRDTM
2.74×10-7
3.83×10-8
8.80×10-9
1.85×10-9
10.2
UCFDM [4]
1.48×10-5
9.60×10-6
5.74×10-6
3.38×10-6
—
321×321
KRDTM
2.69×10-7
3.80×10-8
8.77×10-9
1.85×10-9
41.8
UCFDM [4]
1.40×10-6
9.59×10-7
6.64×10-7
3.94×10-7
—
The L2-norm error for v of P2 at t=0.1 and Ma=0.01.
Grid size
Re=20
Re=40
Re=100
Re=500
Max CPUs
11×11
2.54×10-7
2.73×10-8
3.02×10-9
1.30×10-10
0.047
21×21
2.11×10-7
2.46×10-8
2.90×10-9
1.27×10-10
0.172
41×41
1.88×10-7
2.32×10-8
2.83×10-9
1.26×10-10
0.671
81×81
1.77×10-7
2.25×10-8
2.79×10-9
1.25×10-10
2.50
161×161
1.72×10-7
2.22×10-8
2.77×10-9
1.24×10-10
9.80
321×321
1.69×10-7
2.20×10-8
2.76×10-9
1.24×10-10
39.2
is explained in Figure 3, so that the pairs of bound eddies produced behind the single elements of the grids and at large distance downstream. However, the streamlines are parallel and equidistant as shown by the short lines on the right side of the figure for all values of Reynolds numbers. We also note that when the value of Reynolds number increases, the whole flow pattern is expanded uniformly in the direction of main flow. Moreover, it can be observed that the rate of change of the flow is very great, and the length of vortices increases towards the downstream flow with the increase in the Reynolds number.
The vorticity and streamlines contour plots for P2 at t=0.1 and Ma=0.01.
The vorticity Ω
The streamlines ψ
Thirdproblem (P3) is three-dimensional Taylor decaying vortices flow, whose initial conditions [22–26] are given by(55)ux,y,z,0=UsinxLcosyLcoszL,vx,y,z,0=-UcosxLsinyLcoszL,wx,y,z,0=0,px,y,z,0=p0+ρ0U216cos2xL+cos2yLcos2zL+2,
with periodic boundary conditions in all directions, where p0 is a reference pressure (an arbitrary constant), U is characteristic velocity, ρ0 is the density, and L is the inverse of the wave number of the minimum frequencies (the largest length scale of flow). We used in this test p0=0, U=1, ρ0=1, and L=1 [23] and applied the KRDTM for solving three-dimensional equations (1) and (2), where X=(x,y,z), u=(u,v,w), and Uk=(Uk,Vk,Wk), and we get(56)k+1Uk+1x,y,z=-A1k+B1k+C1k+∂Pkx,y,z∂x+1Re∂2Ukx,y,z∂x2+∂2Ukx,y,z∂y2+∂2Ukx,y,z∂z2,(57)k+1Vk+1x,y,z=-A2k+B2k+C2k+∂Pkx,y,z∂y+1Re∂2Vkx,y,z∂x2+∂2Vkx,y,z∂y2+∂2Vkx,y,z∂z2,(58)k+1Wk+1x,y,z=-A3k+B3k+C3k+∂Pkx,y,z∂z+1Re∂2Wkx,y,z∂x2+∂2Wkx,y,z∂y2+∂2Wkx,y,z∂z2,(59)k+1Gk+1x,y,z=-1Ma2∂Ukx,y∂x+∂Vkx,y∂y+∂Wkx,y∂z+1Re∂2Gkx,y∂x2+∂2Gkx,y∂y2+∂2Gkx,y∂z2,
such that(60)Pkx,y,z=Gkx,y,z+Uk2x,y,z+Vk2x,y,z+Wk2x,y,z2,A1k=∑i=0kUix,y,z∂Uk-ix,y,z∂x,B1k=∑i=0kVix,y,z∂Uk-ix,y,z∂y,C1k=∑i=0kWix,y,z∂Uk-ix,y,z∂z,A2k=∑i=0kUix,y,z∂Vk-ix,y,z∂x,B2k=∑i=0kVix,y,z∂Vk-ix,y,z∂y,C2k=∑i=0kWix,y,z∂Vk-ix,y,z∂z,A3k=∑i=0kUix,y,z∂Wk-ix,y,z∂x,B3k=∑i=0kVix,y,z∂Wk-ix,y,z∂y,C3k=∑i=0kWix,y,z∂Wk-ix,y,z∂z,
where =0,1,2,3,…, U0(x,y,z)=u(x,y,z,0), V0(x,y,z)=v(x,y,z,0), W0(x,y,z)=w(x,y,z,0), and G0(x,y,z)=g(x,y,z,0).
Then the exact solution is obtained as follows:(61)ux,y,z,τ=limn→∞unx,y,z,τ,(62)vx,y,z,τ=limn→∞vnx,y,z,τ,(63)wx,y,z,τ=limn→∞wnx,y,z,τ,(64)gx,y,z,τ=limn→∞gnx,y,z,τ,
such that(66)U1x,y,z=-3Resinxcosycosz-18sin2xcos2z,U2x,y,z=92Re2sinxcosycosz-316Re1-cos2y1+cos2zsin2x+sinxsinycosz32Recosxsinycosz-116sin2ycos2z-sinxcosycosz-3Recosxcosycosz-14cos2xcos2z-cosxcosycosz-3Resinxcosycosz-18sin2xcos2z+116cos2x+cos2ysin2zsinxcosysinz-12Resin2x1+2cos2y+3cos2ycos2z,⋮,V1x,y,z=3Recosxsinycosz-18sin2ycos2z,V2x,y,z=-92Re2cosxsinycosz-316Re1-cos2x1+cos2zsin2y-sinxsinycosz-32Resinxcosycosz-116sin2xcos2z+cosxsinycosz3Recosxcosycosz-14cos2ycos2z+cosxcosycosz3Recosxsinycosz-18sin2ycos2z-116cos2x+cos2ysin2zcosxsinysinz-12Resin2y1+2cos2x+3cos2xcos2z,⋮,W1x,y,z=18cos2x+cos2ysin2z,W2x,y,z=-1Recos2x+cos2ysin2z+12Re1-3cos2xcos2ysin2z+18coszsin2zsinxcosysin2x-cosxsinysin2y-sinxcosycosz3Resinxcosysinz+14sin2xsin2z+cosxsinycosz-3Recosxsinysinz+14sin2ysin2z,⋮,
The flow is computed with in a periodic square box defined as -π<x,y,z<π. The condition of the convergence of these solutions is verified by applying Definition 5(68)γ10=U1x,y,zτU0x,y,z=Re2+288τ42Re,γ20=V1x,y,zτV0x,y,z=Re2+288τ42Re,γ30=W1x,y,zτW0x,y,z=0,γ40=G1x,y,zτG0x,y,z=22τRe,γ11=U2x,y,zτ2U1x,y,zτ=3Re4+70Re2+864τ2Re2+288Re,γ21=V2x,y,zτ2V1x,y,zτ=3Re4+70Re2+864τ2Re2+288Re,γ31=W2x,y,zτ2W1x,y,zτ=Re2+1232τRe,γ41=G2x,y,zτ2G1x,y,zτ=247τ11Re,⋮,
such that γ0=γ10+γ20+γ30+γ40, γ1=γ11+γ21+γ31+γ41,…. For example, if Ma=0.01, Re=100, and t=0.5, then(69)γ0=0.0020275629<1,γ1=0.0100995152<1,…,
if t=2 then(70)γ0=0.0081754148<1,γ1=0.0435106065<1,…,
and if t=5 then(71)γ0=0.0204385369<1,γ1=0.1087765162<1,…
Tables 6 and 7 show L2-norm error for u and w with CPU time at time levels t=0.5,2,5, Re=100, Ma=0.005, and Ma=0.01. It is clear that the value of the calculated error is acceptable with different time levels; in addition to that the longest period of CPU time for all cases is 1140s. So, we can say that these solutions have a good accuracy and convergence low Mach numbers. The relationship of the change of time with the enstrophy is shown in Figure 4 at Re=20,40,100,500 for Ma=0.1,0.01. In Figure 5, we explained the change in the contours of the z-component of the vorticity and the velocities with time on the surface z=0 at Re=100 and Ma=0.05.
The L2-norm errors for u and w of P3 at Re=100 and Ma=0.005.
Grid size
t=0.5
t=2
t=5
Max CPUs
uvelocity
33×33×33
1.82×10-8
1.17×10-6
1.82×10-5
2.12
65×65×65
1.73×10-8
1.11×10-6
1.73×10-5
18.4
129×129×129
1.69×10-8
1.08×10-6
1.69×10-5
126
257×257×257
1.67×10-8
1.07×10-6
1.67×10-5
995
wvelocity
33×33×33
1.21×10-8
7.74×10-7
1.21×10-5
2.42
65×65×65
1.19×10-8
7.60×10-7
1.19×10-5
18.4
129×129×129
1.18×10-8
7.52×10-7
1.18×10-5
143
257×257×257
1.17×10-8
7.49×10-7
1.17×10-5
1120
The L2-norm errors for u and w of P3 at Re=100 and Ma=0.01.
Grid size
t=0.5
t=2
t=5
Max CPUs
uvelocity
33×33×33
1.46×10-7
9.32×10-6
1.46×10-4
2.22
65×65×65
1.39×10-7
8.87×10-6
1.39×10-4
17.7
129×129×129
1.35×10-7
8.65×10-6
1.35×10-4
128
257×257×257
1.33×10-7
8.53×10-6
1.33×10-4
1000
wvelocity
33×33×33
9.68×10-8
6.19×10-6
9.68×10-5
2.47
65×65×65
9.50×10-8
6.08×10-6
9.50×10-5
18.5
129×129×129
9.40×10-8
6.02×10-6
9.40×10-5
144
257×257×257
9.36×10-8
5.99×10-6
9.36×10-5
1140
The enstrophy for P3.
The Contours plots of z-component of vorticity, u and v velocity components for P3 on z=0 at Re=100 and Ma=0.05.
z-component of vorticity Ω
u velocity component
v velocity component
Fourthproblem (P4) is one type of three-dimension Beltrami flow [25, 27, 28], which yield a family of velocity and pressure fields depending on the selection of a and d. In this test, we selected a=π/4 and d=π/2. This problem has the exact solution satisfying (1) and (2), which is given by(72)ux,y,z,t=-asinay+dzeax+cosax+dyeaze-d2t/Re,vx,y,z,t=-asinaz+dxeay+cosay+dzeaxe-d2t/Re,wx,y,z,t=-asinax+dyeaz+cosaz+dxeaye-d2t/Re,px,y,z,t=-a22e2ax+e2ay+e2az+2sinax+dycosaz+dxeay+z+2sinay+dzcosax+dyeaz+x+2sinaz+dxcosay+dzeax+ye-2d2t/Re,
and when we applied KRDTM for solving (1) and (2), we get the same solutions in (61), (62), (63), and (64) with(73)U1x,y,z=ad2Resinay+dzeax+cosax+dyeaz,U2x,y,z=-ad42Re2sinay+dzeax+cosax+dyeaz-2aReeax+ya2a2-d2sinay+dz+d2cosay+dzsinaz+dx+a3d2sinay+dz-cosay+dzcosaz+dx+eay+za2+d2cosax+dy-adsinax+dya2sinaz+dx+a2d2cosax+dy+2a3dsinax+dycosaz+dx+eaz+xa4cosax+dy+a4sinax+dysinay+dz+2a3dsinax+dycosay+dz-2a2e2axa2+d22,⋮,V1x,y,z=ad2Resinaz+dxeay+cosay+dzeax,V2x,y,z=-ad42Re2sinaz+dxeay+cosay+dzeax-2aReeay+za2a2-d2sinaz+dx+d2cosaz+dxsinax+dy+a3d2sinaz+dx-cosaz+dxcosax+dy+eaz+xa2a2+d2cosay+dz-adsinay+dzsinax+dy+d2a2cosay+dz+2a3dsinay+dzcosax+dy+eax+ya4cosay+dz+sinay+dzsinaz+dx+2a3dsinay+dzcosaz+dx-2a2e2aya2+d22,W1x,y,z=ad2Resinax+dyeaz+cosaz+dxeay,W2x,y,z=-ad42Re2sinax+dyeaz+cosaz+dxeay-2aReeay+za4sinaz+dx+cosaz+dxsinax+dy+2a3dcosax+dysinaz+dx+eaz+xa2a2-d2sinay+dz+2a3dcosay+dzsinax+dy+a2d2sinay+dz-adcosay+dzcosax+dy+eax+ya2a2+d2cosaz+dx-adsinaz+dxsinay+dz+a2d2cosaz+dx+2adsinaz+dxcosay+dz-2a2e2aza2+d22,⋮,
These solutions satisfy the conditions of convergence at the domain [-1,1]3,(75)γ10=U1x,y,zτU0x,y,z=2.4674011τRe,γ20=V1x,y,zτV0x,y,z=2.4674011τRe,γ30=W1x,y,zτW0x,y,z=2.4674011τRe,γ40=G1x,y,zτG0x,y,z=2.917691460τRe,γ11=U2x,y,zτ2U1x,y,zτ=3.0495132689Re2-0.020148681Re+0.1636659976τRe,γ21=V2x,y,zτ2V1x,y,zτ=3.0495132689Re2-0.020148681Re+0.1636659976τRe,γ31=W2x,y,zτ2W1x,y,zτ=3.0495132689Re2-0.020148681Re+0.1636659976τRe,γ41=G2x,y,zτ2G1x,y,zτ=4.3109179800τRe,⋮,
such that γ0=γ10+γ20+γ30+γ40,γ1=γ11+γ21+γ31+γ41,….
For example, if Ma=0.01, Re=100, and t=2, then(76)γ0=0.2063978953×10-2<1,γ1=0.1838160431<1,…,
if t=5 then(77)γ0=0.5159947383×10-2<1,γ1=0.4595401079<1,…,
if Re=1600 and t=2 then(78)γ0=0.1289986845×10-3<1,γ1=0.1830235364<1,…,
and if t=5 then(79)γ0=0.3224967113×10-3<1,γ1=0.4575588408<1,…
The L2-norm error for the u velocity component with CPU time is calculated in Table 8 to study the accuracy of these approximate solutions; the results show an excellent accuracy of our method for all values of Reynolds number at t=0.5,2,5 and Ma=0.01, with good implementation period ranging between 2.09s,1080s. The computed enstrophy is compared with their exact values in the same period of time in Figure 6 at Re=100,1600 for two Mach numbers. In Figure 7, we explained the z-component of the computed vorticity on surface z=0 at Re=100 and t=5 in two domains [-1,1]2 and [-5,5]2. Through these figures, we could notice the relationship between the accuracy of these approximate solutions and Mach numbers which is with decreasing Mach number.
The L2-norm errors for u of P4 at Ma=0.01.
Grid size
t=0.5
t=2
t=5
Max CPUs
Re=100
33×33×33
6.18×10-6
1.00×10-4
6.44×10-4
2.44
65×65×65
5.89×10-6
9.54×10-5
6.15×10-4
20.4
129×129×129
5.74×10-6
9.31×10-5
6.00×10-4
151
257×257×257
5.67×10-6
9.19×10-5
5.93×10-4
1080
Re=500
33×33×33
1.24×10-6
2.00×10-5
1.29×10-4
2.9
65×65×65
1.18×10-6
1.91×10-5
1.23×10-4
15.9
129×129×129
1.15×10-6
1.86×10-5
1.20×10-4
127
257×257×257
1.13×10-6
1.84×10-5
1.19×10-4
996
Re=1600
33×33×33
3.86×10-7
6.26×10-6
4.03×10-5
2.20
65×65×65
3.68×10-7
5.96×10-6
3.84×10-5
16.8
129×129×129
3.59×10-7
5.82×10-6
3.75×10-5
130
257×257×257
3.54×10-7
5.75×10-6
3.71×10-5
1020
The enstrophy for P4.
Re=100
Re=1600
The surface plots of the z-component of the computed vorticity for P4 on z=0 at Re=100 and t=5.
-1<x,y<1
-5<x,y<5
5. Conclusions
In this paper, the simulations of two- and three-dimensional unsteady viscous incompressible flow problems are presented by using the kinetically reduced local Navier-Stokes equations with the reduced differential transform method. New approximate analytical solutions obtained by KRDTM are tested in terms of accuracy and convergence. The results show that the new solutions have good accuracy and convergence, especially with high Reynolds numbers and low Mach numbers. The comparison explained that the computational time of these solutions is faster than that of other numerical solutions. Therefore, KRDTM is an effective and accurate method for solving the unsteady viscous incompressible flow problems.
Data Availability
The data used to support the findings of this study are included within the article.
Conflicts of Interest
The authors declare that they have no conflicts of interest.
Acknowledgments
We would like to thank the English proofreader Assistant Professor Mahdi Mohsin Mohammed for his careful reading.
AnsumaliS.KarlinI. V.ÖttingerH. C.Thermodynamic theory of incompressible hydrodynamics20059482-s2.0-18144365509KarlinI. V.TomboulidesA. G.FrouzakisC. E.AnsumaliS.Kinetically reduced local Navier-Stokes equations: An alternative approach to hydrodynamics20067432-s2.0-33749380281BorokS.AnsumaliS.KarlinI. V.Kinetically reduced local Navier-Stokes equations for simulation of incompressible viscous flows20077662-s2.0-37049002912ShahA.YuanL.KhanA.Upwind compact finite difference scheme for time-accurate solution of the incompressible Navier-Stokes equations201021593201321310.1016/j.amc.2009.10.001MR2576808Zbl1352.760842-s2.0-71749084413HashimotoT.TannoI.TanakaY.MorinishiK.SatofukaN.Simulation of decaying two-dimensional turbulence using kinetically reduced local Navier-Stokes equations88Proceedings of the Seventh International Conference on Computational Fluid Dynamics2012715718Zbl1391.76101HashimotoT.TannoI.TanakaY.MorinishiK.SatofukaN.Higher order numerical simulation of unsteady viscous incompressible flows using kinetically reduced local Navier-Stokes equations on a GPU2015110108113MR3131226Al-SaifA. S. J.Analytical approximate solutions for two-dimensional incompressible Navier-Stokes equations2015496986ShahA.FayyazH.RizwanM.Fourth-order central compact scheme for the numerical solution of incompressible Navier-Stokes equations201794122492250710.1080/00207160.2017.1284315MR3740662Zbl1394.353722-s2.0-85012299185ZhouJ. K.1986Wuhan, ChinaHuazhong University PressKeskinY.OturancG.Reduced differential transform method for partial differential equations200910674147749MR2654185KeskinY.OturancG.Reduced differential transform method for generalized KdV equations2010153382393MR2654185Zbl1198.352232-s2.0-7814924221010.3390/mca15030382GuptaP. K.Approximate analytical solutions of fractional Benney-Lin equation by reduced differential transform method and the homotopy perturbation method20116192829284210.1016/j.camwa.2011.03.057MR2795407Zbl1221.652762-s2.0-79955466924AbazariR.AbazariM.Numerical simulation of generalized Hirota-Satsuma coupled KdV equation by RDTM and comparison with DTM201217261962910.1016/j.cnsns.2011.05.022MR2834421RawashdehM.Using the Reduced Differential Transform Method to solve nonlinear PDEs Arises in biology and physics2013238103710432-s2.0-84880581541BenhammoudaB.Vazquez-LealH.Sarmiento-ReyesA.Modified Reduced Differential Transform Method for Partial Differential-Algebraic Equations20142014927948110.1155/2014/279481MR3278350NeogB. C.Solutions of some system of non-linear PDEs using reduced differential transform method2015113744WangK.LiuS.Analytical study of time-fractional Navier-Stokes equation by using transform methods2016Paper No. 61, 1210.1186/s13662-016-0783-9MR3465513Zbl06957472YuJ.JingJ.SunY.WuS.(n+1) dimensional reduced differential transform method for solving partial differential equations201627369770510.1016/j.amc.2015.10.016MR34277882-s2.0-84946426878MohamedM. S.GepreelK. A.Reduced differential transform method for nonlinear integral member of Kadomtsev-Petviashvili hierarchy differential equations20172511710.1016/j.joems.2016.04.007MR3591245Zbl06675928Az-ZobiE. A.Al-KhaledK.A new convergence proof of the Adomian decomposition method for a mixed hyperbolic elliptic system of conservation laws201021784248425610.1016/j.amc.2010.10.040MR2739669Saberi NikH.SoleymaniF.A taylor-type numerical method for solving nonlinear ordinary differential equations20135235435502-s2.0-8488344671210.1016/j.aej.2013.02.006TaylorG. I.GreenA. E.Mechanism of the production of small eddies from large ones193715889549952110.1098/rspa.1937.0036BrachetM. E.MeironD. I.NickelB. G.MorfR. H.FrischU.OrszagS. A.Small-scale structure of the taylor-green vortex19831304114522-s2.0-002061885210.1017/S0022112083001159van ReesW. M.LeonardA.PullinD. I.KoumoutsakosP.A comparison of vortex and pseudo-spectral methods for the simulation of periodic vortical flows at high Reynolds numbers201123082794280510.1016/j.jcp.2010.11.031MR27743172-s2.0-79951772562PiatkowskiM.MüthingS.BastianP.A stable and high-order accurate discontinuous Galerkin based splitting method for the incompressible Navier-Stokes equations201835622023910.1016/j.jcp.2017.11.035MR3743637BoY.WangP.GuoZ.WangL.-P.DUGKS simulations of three-dimensional Taylor-Green vortex flow and turbulent channel flow2017113MR3688493EthierC. R.SteinmanD. A.Exact fully 3D Navier–Stokes solutions for benchmarking19941953693752-s2.0-002849909210.1002/fld.1650190502Zbl0814.76031BottiL.Di PietroD. A.A pressure-correction scheme for convection-dominated incompressible flows with discontinuous velocity and continuous pressure2011230357258510.1016/j.jcp.2010.10.004MR27454442-s2.0-78649328662