Case Studies in Thermal Engineering 3 (2014) 4350

Contents lists available at ScienceDirect

Case Studies in Thermal Engineering

journal homepage: www.elsevier.com/locate/csite

A finite-difference method of high-order accuracy for the

solution of transient nonlinear diffusiveconvective problem
in three dimensions
Marco Donisete de Campos a,b,n, Estaner Claro Romo c,
Luiz Felipe Mendes de Moura b
a
Federal University of Mato Grosso, Institute of Exact and Earth Sciences, Av. Governador Jaime Campos, 6390, 78600-000 Barra do
Garas, Mato Grosso, Brazil
b
Mechanical Engineering Faculty, State University of Campinas, PO Box 6122, 13083-970 Campinas, So Paulo, Brazil
c
Department of Basic and Environmental Sciences, EEL-USP, University of So Paulo, 12602-810 Lorena, So Paulo, Brazil

a r t i c l e i n f o abstract

Article history: This paper presents an efficient technique of linearization of the nonlinear convective
Received 18 November 2013 terms present in engineering problems involving heat and mass transfer and fluid
Received in revised form mechanics. From two numerical applications, this technique with a method of high-
13 February 2014
order finite differences is validated by numerical solution of transient nonlinear diffusive
Accepted 7 March 2014
convective problem in three dimensions.
Available online 3 April 2014
& 2014 The Authors. Published by Elsevier Ltd. This is an open access article under the
Keywords: CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/).
Linearization technique
Nonlinear problems
Finite differences

1. Introduction

Numerical simulation of engineering problems has been successfully performed since decades ago. Problems governed
by linear differential equations have been solved easily and successfully by most numerical methods proposed in the
literature, among them we can mention the works involving, e.g., finite difference method, approximate methods for
fractional ordinary differential equations, finite element method, spectral method and matrix approach.
Based on the high-order finite difference method, that is the focus of the work, several authors have proposed various
methodologies for numerically solving the transient diffusiveconvective problems. Among these papers [1], established an
exponential high-order compact alternating direction implicit method for the numerical solution of unsteady 2D
convectiondiffusion problems using the CrankNicolson scheme for the time discretization and an exponential fourth-
order compact difference formula for the spatial discretization. The unconditionally stable character of the scheme has been
verified by a discrete Fourier analysis and the computational results showed the accuracy, efficiency and robustness of the
method [2] developed a high order compact alternating direction implicit method for solving two-dimensional unsteady
convectiondiffusion problems. The method is unconditionally stable and second-order accurate in time and fourth-order
accurate in space. Three numerical studies are carried out to demonstrate its high accuracy and efficiency and, although it

n
Corresponding author at: Federal University of Mato Grosso, Institute of Exact and Earth Sciences, Av. Governador Jaime Campos, 6390, 78600-000
Barra do Garas, Mato Grosso, Brazil. Tel.: 55 663 405 317.
E-mail address: marco_donisete@yahoo.com.br (M.D. de Campos).

http://dx.doi.org/10.1016/j.csite.2014.03.001
2214-157X/& 2014 The Authors. Published by Elsevier Ltd. This is an open access article under the
CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/).
44 M.D. de Campos et al. / Case Studies in Thermal Engineering 3 (2014) 4350

has been proposed for convectiondiffusion problems, can also be used to solve pure diffusion or pure convection problems
[3] using the compact difference scheme with the operator-splitting technique solved the multidimensional time fractional
diffusion problem, investigating the stability and accuracy of the scheme [4] developed an exponential high order compact
alternating direction implicit method for solving three dimensional unsteady convectiondiffusion equations. The method is
fourth-order in space, second order in time, and proved to be unconditionally stable. Three numerical experiments are
performed to demonstrate its high accuracy and efficiency and to show its superiority over the classical DouglasGunn
alternating direction implicit scheme and the Karaa's high-order alternating direction implicit scheme [5].
In this work, the aim is to present a formulation that uses an accurate linearization technique with easy to implement
and that uses only one iteration at each time step, optimizing the calculation of the velocity profile.

2. Formulation

In this work, we propose a solution, by the high-order finite difference method for the transient nonlinear convection
diffusion equation in three dimensions which is given by

2 
u u u u u 2 u 2 u
u u u 2 2 1
t x y z x 2 y z
where u(x,y,z,t) is the velocity field in the x,y,z-directions and is the kinematic viscosity.
For the temporal discretization, considering the equation given by Eq. (1), we use the method called family of
approximation [6], in which a weighted average of the time derivative of a dependent variable is approximated on two
consecutive time steps by linear interpolation of the values of the variable at two steps
 n  n 1
u fug~ n 1 fug~ n u u
 1  2
t t n 1 t t
where 0 r r1; tA [tn, tn 1], n 0,1,2,,mt, where mt is the number of steps in time, and { }n refers to the value of the
enclosed quantity at time n and t n 1 t n 1  t n is the (n 1)th time step. Adopting 0:5; we obtain the well-known
CrankNicolson method to carry out the time discretization as follows:

n1 
2 n1 
u un u 2 un 1 2 un 1 un 1 un 1 un 1
0:5  un 1  un 1  un 1
t x 2 y 2 z 2 x y z

2 n 
u 2 un 2 un n u
n
n u
n
n u
n
0:5 u u u 3
x2 y2 z2 x y z
In order to linearize the convective terms of the previous equation will be used the technique proposed by [7] according
to which for a sufficiently small time step, these terms can be expanded into a Taylor series after the first-derivative terms.
The result is as follows:
un 1 un 1 un un
un 1  un un 1  un : 4a
x x x x

un 1 un 1 un un
un 1  un un 1  un 4b
y y y y

un 1 un 1 un un
un 1  un un 1  un 4c
z z z z

This method is commonly known as Newton's method since it provides a quadratic convergence [8]. Note that this
technique does not require an iterative linearization in each time step, making quicker the calculation of u.
And, substituting Eqs. (4a)(4c) into the convective terms of Eq. (2), it yields

n1 
2 n1
u un u 2 un 1 2 un 1 un 1 un un un 1
0:5  un  un 1 un  un
t x2 y2 z2 x x x y
n1

u n
u n
u u n
u n
 un 1 un  un  un 1 un
y y z z z

2 n 
u 2 un 2 un u n
u n
un
0:5  un  un un
x2 y 2 z2 x y z
n1

n 1 u
n
n 1 u
n
n u n 1 u
n
un 1
u u u u F 5
x y z z t
where

2 n 
un u 2 un 2 un
F 0:5 :
t x 2 y 2 z 2
 M.D. de Campos et al. / Case Studies in Thermal Engineering 3 (2014) 4350 45

Now, discretizing in the space by Central Difference Method of O(x2) for nodes with dx, dy or dz distance from the
boundary, we have
! !
uni 1;j;k
1
2uni;j;k 1
uni 1;j;k
1
uni;j11;k  2uni;j;k 1
uni;j11;k
 0:5 0:5
x2 y2
1 1 1
! ! !
uni;j;k 1
2uni;j;k uni;j;k 1 n
uni 1;j;k
1
 uni 1;j;k
1
n
uni;j11;k uni;j11;k
 0:5 0:5u i;j;k 0:5u i;j;k
z2 2x 2y
n1 n1
! !
u i;j;k 1
 u i;j;k  1 1 u n
i;j;k u n
i;j;k u n
i;j;k 1
0:5uni;j;k 0:5 0:5 0:5 uni;j;k F1
2z t x y z
n
! n
! n
!
0:5 0:25ui;j;k n 1 0:5 0:25ui;j;k n 1 0:5 0:25ui;j;k n 1
)  2 ui;j;k  1  2  ui;j  1  2  ui  1;j;k
z z y y x x
! !
1 uni;j;k uni;j;k uni;j;k 1
0:5 uni;j;k
x2 y2 z2 t x y z
n
! n
! n
!
0:5 0:25ui;j;k n 1 0:5 0:25ui;j;k n 1 0:5 0:25ui;j;k n 1
 2 ui 1;j;k  2 ui;j 1;k  2 ui;j;k 1 F 1 : 6
x x y y z z

where
!
uni;j;k uni 1;j;k  2uni;j;k uni 1;j;k uni;j 1;k 2uni;j;k uni;j  1;k uni;j;k 1  2uni;j;k uni;j;k  1
F1 0:5 :
t x2 y2 z2

Now, considering the internal nodes using the Central Difference Method with O(x4), we have
 uni 2;j;k
1
16uni 1;j;k
1 1
30uni;j;k 16uni 1;j;k
1
 uni 2;j;k
1
uni;j12;k 16uni;j11;k  30uni;j;k
1
16uni;j11;k uni;j12;k
 0:5
12x2 12y2
1 1 1 1 1
!
 uni;j;k 2
16uni;j;k 1
 30uni;j;k 16uni;j;k 1
uni;j;k 2
 uni 2;j;k
1
8uni 1;j;k
1
 8uni 1;j;k
1
uni 2;j;k
1
0:5uni;j;k
12z2 12x
!
 uni;j12;k 8uni;j11;k  8uni;j11;k uni;j12;k 1
 uni;j;k 2
1
8uni;j;k 1
1
 8uni;j;k 1
1
uni;j;k 2

12y 12z
!!
1 uni;j;k uni;j;k uni;j;k 1
0; 5 uni;j;k F2
t x y z
! ! !
uni;j;k n1 2 uni;j;k n 1 uni;j;k
) u   u un 1
24z2 24z i;j;k  2 3z2 3z i;j;k  1 24y2 24y i;j  2;k
! ! !
2 uni;j;k n1 uni;j;k n1 2 uni;j;k n 1
  u u   u
3y2 3y i;j  1;k 24x2 24x i  2;j;k 3x2 3x i  1;j;k
! ! !
uni;j;k n1 2 uni;j;k n 1 uni;j;k
) u   u un 1
24z2 24z i;j;k  2 3z2 3z i;j;k  1 24y2 24y i;j  2;k
! ! !
2 uni;j;k n1 uni;j;k n1 2 uni;j;k n 1
  u u   u
3y2 3y i;j  1;k 24x2 24x i  2;j;k 3x2 3x i  1;j;k
!!
1:25 1:25 1:25 1 uni;j;k uni;j;k uni;j;k 1
0:5 uni;j;k
x2 y2 z2 t x y z
! ! !
2 uni;j;k n 1 uni;j;k n1 2 uni;j;k
 u  u  un 1
3x2 3x i 1;j;k 24x2 24x i 2;j;k 3y2 3y i;j 1;k
! ! !
uni;j;k n1 2 uni;j;k n 1 uni;j;k
 u  u  un 1 F 2 7
24y2 24y i;j 2;k 3z2 3z i;j;k 1 24z2 24z i;j;k 2

where
uni;j;k uni 2;j;k 16uni 1;j;k  30uni;j;k 16uni 1;j;k  uni 2;j;k uni;j  2;k 16uni;j  1;k  30uni;j;k 16uni;j 1;k  uni;j 2;k
F2 0:5
t 12x2 12y2
!
 uni;j;k  2 16uni;j;k  1 30uni;j;k 16uni;j;k 1 uni;j;k 2
:
12z2
46 M.D. de Campos et al. / Case Studies in Thermal Engineering 3 (2014) 4350

3. Numerical applications

Here, we present two numerical applications: first, from the proposal to an exact solution to Eq. (1), will be considered
the greatest error (L1 error) for different spatial and temporal refinements, besides the computational time and the
convergence achieved for meshes adopted. All computations were run on a Intel Core i7/2.4G private computer using double
precision arithmetic. In the second application, using the sinusoidal velocity profiles, it was demonstrated situations in
which softens the mesh refinement of the solution numerical oscillations of the velocity field.

3.1. Application 1

Considering the exact solution of Eq. (1) given by ux; y; z; t expx y 2z 6t=, was done the comparison with the
numerical solution, with 1, t 0.1, 0 rx,y,zr0.1 and using a regular mesh with hx yz 0.1.
Table 1 shows the errors associated with the numerical solution compared with the analytical solution according to the
L1 norm, considering the maximum error for stopping criterion for the GaussSeidel on the order of 10  10, the CPU time
and the rate of convergence (CR) defined by CR logE1 =E2 = logt 1 =t 2 , where E1 and E2 are the L1 norm errors with
the temporal grid sizes t 1 and t 2 , respectively.
The boundary and initial conditions are directly taken from this analytical solution. At first we set h0.005, t0.1 and
change t from 0.005 to 0.001. The L1 norm errors and the CPU time used for the present method are depicted in Table 1.
We notice that the present method achieves the accuracy of 2.03E 08 with a temporal refinement of t0.001 whereas for
h 0.0025 achieves the accuracy of 9.03E  09.
Fig. 1 shows the standard error L1 and velocity profile u in the xy-plane, considering x yzLx/20 and
Lx Ly Lz 0.1. It can be noted that the greatest error occurs in the center of the plane analyzed.

Table 1
L1 error, CPU time and the rate of with t 0.1.

t h 0.005 h 0.0025

L1 norm CPU time CR L1 norm CPU time CR

0.00500 4.69E  07 60.781 4.58E  07 1744.875

0.00250 1.16E  07 103.172 2.02 1.05E  07 2873.437 2.12
0.00167 5.29E  08 137.375 1.93 4.14E  08 3708.547 2.29
0.00125 3.06E  08 165.328 1.91 1.92E  08 4842.578 2.67
0.00100 2.03E  08 189.110 1.84 9.03E  09 5391.891 3.38

Fig. 1. L1 error (left) and profile u (right) both in the xy-plane, for z 0.05 and t 0.1.
 M.D. de Campos et al. / Case Studies in Thermal Engineering 3 (2014) 4350 47

Fig. 2. Two dimensional velocity profile in the xy-plane for z 0.5 (left), and in the xz-plane for y0.5 (right) considering 1.

Fig. 3. Two dimensional velocity profile in the xy-plane for z 0.5 (left), and in the xz-plane for y0.5 (right), considering 0.1.

3.2. Application 2

Now, considering the computational domain as being an unit cube and t 0.1, with the initial condition
ux; y; z; 0 sin 2x cos 2yz and the boundary condition ux; y; z; t sin 2x cos 2yzet . Fig. 2 shows the bidimen-
sional profile, respectively, of u-velocity in the xy-plane with z0.5 and in the xz-plane with y 0.5 considering 1.
Figs. 3 and 4 show the two dimensional velocity profile u in the xy-plane with z0.5 and in the xz-plane with y0.5 are
given, respectively, 0.1 and 0.01, with xyz Lx/20 and t Lt/20.
Aiming to decrease the numerical oscillations in the previous cases, the mesh was refined, considering now
x yzLx/40 and t Lt/20. Thus, in the conditions of the previous case, was obtained the velocity profile shown
in Fig. 5, in which we note that the refinement softened the numerical oscillations.
48 M.D. de Campos et al. / Case Studies in Thermal Engineering 3 (2014) 4350

Fig. 4. Two dimensional velocity profile in the xy-plane for z 0.5 (left), and in the xz-plane for y 0.5 (right) considering 0.01.

Fig. 5. Two dimensional velocity profile in the xy-plane for z 0.5 (left), and in the xz-plane for y 0.5 (right), considering 0.01.

Figs. 6 and 7 illustrates the numerical results for two dimensional velocity profile u in the xy-plane with z 0.5 and in the
xz-plane with y0.5 are given, respectively, 0.001 with x y zLx/190 and tLt/100 and 0.0001, with
x yz Lx/20 and tLt/1000.

4. Conclusions

In this work, an efficient technique of linearization of the nonlinear convective terms present in engineering problems
involving heat and mass transfer and fluid mechanics is presented.
This technique is very simple, stable and accurate and technique does not require any type of iterative process within
each time step, which represents a major saving on the computational time. It used acopled with the method with a method
 M.D. de Campos et al. / Case Studies in Thermal Engineering 3 (2014) 4350 49

Fig. 6. Two dimensional velocity profile in the xy-plane for z 0.5 (left), and in the xz-plane for y 0.5 (right), considering 0.001.

Fig. 7. Two dimensional velocity profile in the xy-plane for z 0.5 (left), and in the xz-plane for y 0.5 (right), considering 0.0001.

of high-order finite differences is validated by two numerical applications of tridimensional transient nonlinear diffusion-
convection equation in three dimensions.

Acknowledgements

We acknowledged the support from the National Council of Scientific Development and Technology, CNPq, Brazil (Proc.
500382/2011-5) and Mato Grosso Research Foundation, Fapemat, Brazil (Proc. 292470/2010).

References

[1] Tian ZF, Ge YB. A fourth-order compact ADI method for solving two-dimensional unsteady convectiondiffusion problems. J Comput Appl Math
2007;198:26886.
[2] Tian ZF. A rational high-order compact ADI method for unsteady convectiondiffusion equations. Comput Phys Commun 2011;182:64962.
50 M.D. de Campos et al. / Case Studies in Thermal Engineering 3 (2014) 4350

[3] Cui M. Convergence analysis of high-order compact alternating direction implicit schemes for the two-dimensional time fractional diffusion equation.
Numer Algorithms 2013;62:383409.
[4] Ge YB, Tian ZF, Zhang J. An exponential high order compact ADI (EHOC ADI) method for 3D unsteady convectiondiffusion problems. Numer Methods
Partial Differ Equ 2013;29:186205.
[5] Karaa S. A high-order compact ADI method for solving three-dimensional unsteady convectiondiffusion problems. Numer Methods Partial Differ Equ
2006;22:98393.
[6] Reddy JN. An introduction to the finite element method. 3rd ed. McGraw-Hill; 2013.
[7] Dennis Jr JE, Schmabel RB. Numerical methods for unconstrained optimization and nonlinear equations. Prentice-Hall; 1983.
[8] Jiang BN. The least-squares finite element method: theory and applications in computational fluid dynamics and electromagnetics. Springer; 1998.