This article was downloaded by: [Purdue University]

On: 03 April 2013, At: 04:03

Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK

Numerical Heat Transfer, Part B: Fundamentals: An

International Journal of Computation and Methodology
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/unhb20

HYBRID NUMERICAL/ANALYTICAL APPROACH TO

NONLINEAR DIFFUSION PROBLEMS
a
R. M. Cotta
a
Programa de Engenharia Mecānica, COPPE—Universidade Federal do Rio de Janeiro, Cidade
Universitária, C.P. 68503, Rio de Janeiro, RJ, 21945, Brasil
Version of record first published: 21 May 2007.

To cite this article: R. M. Cotta (1990): HYBRID NUMERICAL/ANALYTICAL APPROACH TO NONLINEAR DIFFUSION PROBLEMS,
Numerical Heat Transfer, Part B: Fundamentals: An International Journal of Computation and Methodology, 17:2, 217-226

To link to this article: http://dx.doi.org/10.1080/10407799008961740

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions

This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to
anyone is expressly forbidden.

The publisher does not give any warranty express or implied or make any representation that the contents
will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should
be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims,
proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in
connection with or arising out of the use of this material.
 Numerical Heat Transfer, Part B, vol. 17, pp. 217-226, 1990

HYBRID NUMERICAL/ANALYTICAL APPROACH TO

NONUNEAR DIFFUSION PROBLEMS

R. M. Cotta
Programa de Engenharia Mectinica, COPPE-Universidade Federal do
Rio de Janeiro, Cidade Universitaria, C.P. 68503, Rio de Janeiro,
RJ 21945, Brasil

A class of nonlinear diffusion-type problems is handled through a hybrid method. This

method incorporales the ideas in the generalized imegral transform technique to reduce
Downloaded by [Purdue University] at 04:03 03 April 2013

the original partial differenJial eqlUlJion into a iUnumerable system of coupled ordinary
differenlial equoiions. These eqlUlJions can then be solved through standard numerical
techniques, once the system is truru:aJed to a finite order. SuffJC~nt conditions for the
convergence ofthe truncaudfinite system are then examined. An appliJ:ation is considered
thai deals with a transient radiative fin problem, which is quite suitable for illustrating
the solution methodology and convergence behavior.

INTRODUCTION
Several numerical approaches have been proposed over the last few decades for the
solution of nonlinear heat and mass diffusion problems. Most of these approaches are
based on variations of the classical discrete approaches, the finite-difference and finite-
element methods. Only recently, and to a limited extent, have some hybrid numerical!
analytical schemes been proposed [1-3] that attempt to take advantage of well-known
analytical solution procedures for linear problems to achieve improved accuracy and
reliability, as well as decreased computing time, in the designed scheme for solving the
nonlinear versions of such problems.
The present contribution is one such attempt that uses the integral transform tech-
nique, recently presented within a unified frame for the exact solution of different classes
of linear diffusion problems [4] and generalized so that the solution of various other a
priori nontransformable classes of problems [5-15] are included. Here, the so-called
generalized integral transform technique is utilized to transform the original partial dif-
ferential system into an infinite system of coupled nonlinear ordinary differential equations
for the transformed potentials. The denumerable system is then truncated at the Nth row
and column and is numerically solved through well-established algorithms for systems
of ordinary differential equations [16-18], where N is of sufficiently large order for the
prescribed accuracy requirements. The explicit inversion formula is then recalled to
reconstruct the desired original potential. Therefore, it is also of interest to examine
sufficient conditions on the right-hand sides of the infinite system that warrant convergence
as N is increased [19].
The approach is illustrated through considering a transient heat conduction problem
of a radiating fin [20, 21], subjected to a nonlinear heat dissipation. Dimensionless heat

The author acknowledges [he financial support provided by CNPq from Brazil through grant no. 3023661
85-EM.

Copyright © 1990 by Hemisphere Publishing Corporation 217

 218 R. M. COTTA

NOMENCLATURE
d(x) coefficient in Eq. (I) 1 time or corresponding space variable
f(x) initial condition function T(x, I) potential (temperature. concentration,
H dimensionless surroundings velocity•... )
temperature in Eq. (29) To initial temperature
K(x) coefficient in Eq. (I) w(x) coefficient in Eq. (I)
N order of truncated system of ordinary x position vector
differential equations a(x), ~(x) boundary coefficients
radiation-conduction parameter in cl>(T) nonlinear boundary source function
Eq. (29) eigenvalues in Eq. (7)
Ni normalization integral from Eq. (7) eigenfunctions in Eq. (7)
pm nonlinear equation source function
Downloaded by [Purdue University] at 04:03 03 April 2013

fluxes and temperature profiles are computed for different situations and for various
truncation orders, so as to demonstrate the simplicity and effectiveness of the present
solution methodology.

ANALYSIS
Let us consider the following transient nonlinear diffusion problem, defined within
region V and boundary surface S

aT(x, t)
w(x) -'---'---'- V' K(x)VT(x, t) - d(x)T(x, t)
at
+ P(T(x, t» in x E V, t > 0 (1)

with boundary and initial conditions given, respectively, by

a(x) T(x, t) + f3(x)K(x) aT(x, t) q,(T(x, t» xES, t >0 (2)

an
T(x, t) = f(x) x E V, t = 0 (3)

Although the nonlinearities appear explicitly only in the source terms of both
equation (I) and boundary condition (2), such a formulation includes more general ones,
provided the linear operators

==
a
L
I
w(x)-
at (4)

L == V·K(x)V - d(x) (5)

B == a(x) + f3(x)K(x)-
a (6)
an
are taken as characteristic ones extracted from the original nonlinear versions, and the
remaining nonlinear terms are collapsed into the source functions P(T) and q,(T).
 NONLINEAR DlFFUSION PROBLEMS 219

By following the formalism in the generalized integral transform technique [5-15),

the auxiliary eigenvalue problem is taken as

V·K(x)VI\J(fL;, x) + [fLtW(X) - d(x»)I\J(fLi, x) =0 xEV (7)

ih\l;(x)
a(X)I\J(fLj, x) + ~(x)K(x)-- = 0 xES (8)
iJn

where solution is assumed to be known at this point, and allows the definition of the
following integral transform pair:

Transform: - f
Tj(t) =
t/ti(X)
v w(x) N 1/ 2 T(x, r) dv (9)
Downloaded by [Purdue University] at 04:03 03 April 2013

x 1 _
Inversion: T(x, t) = j~1 NF2 1\J;(X) T;(t) (10)

Equation (la) is now operated on with

f l\Ji(X) d
v NI12
I
V

to yield

(II)

where

-
g;(T) =
Jv
1\J;(x)
NI/2
;
P(T(x, r) dv
I
+ """112
N;
I.
s
K(x)
[t/t;(x) -----.::........:......:..
iJT(x, t)
iJn

- T(x, t) _iJl\J_;(_X)] ds (12)

. iJn

The surface integral above is now evaluated by making use of boundary conditions
(2) and (8), which after some manipulations yield

1 s
K(x) [I\J.(X) iJT(x, t) -
J an
T(x, r) iJt/t;(X)] ds
an
= r <l>(T(x,t» [I\J;(X) - K(x) [al\J;(X)/an1] ds (13)
Js u(x) + ~(x)
 220 R. M. COTTA

Also, the initial condition (3) is transformed by the operator

and the completely transformed problem is given by

j = 1,2, ... (14)

Ti(O) =I i = I, 2, ... ' (15)

Downloaded by [Purdue University] at 04:03 03 April 2013

where

-
8;(T) = NFI 2 {Jf(x, - T/t»l\Ij(x) dv

+ r<!>(x, T(t» {I\I;(X)' - K(X)[dl\l;(X)/c3n l } dS} (16)

Js} a(x) + l3(x)

and

I.- = ----vi
1
N;
f w(x)f(x) 1\I;(x) dv
v
(17)

Equations (14)-( 17) form an infinite system of nonlinear ordinary differential equa-
tions for the transformed potentials 'i; Once a solution is obtained, the inversion formula,
Eq. (10), can be recalled to provide the complete field T(x, t). For numerical purposes,
it would be highly desirable to approximate solutions of this now coupled system by
solutions of finite dimensional systems, since there are quite effective methods available
to compute such solutions. From an engineering point of view, it suffices to truncate the
infinite system at the Nth row and Nth column, solve this N X N system, and observe
the convergence behavior as N is increased. From a more formal point of view, it is
necessary to look for conditions on the right-hand sides of Eq. (14) ensuring that this
solution tends to the solution of the infinite system as N -+ 00.
Fortunately, a reasonably vast mathematical literature on infinite systems of ordinary
differential equations seems to be available. Here, we shall be particularly concerned
with the findings of Deimling [19], for both linear and nonlinear systems, summarized
in a couple of theorems now examined. We start with the denumerable system of linear
ordinary differential equations:

Y; 2:'" aij(t)Yj + bj(t)

j=1
(18)

yj(O) = c, i = 1,2, ... (19)

 NONLINEAR DIFFUSION PROBLEMS 221

Let aiit) and bi(t) be continuous functions defined in the t interval [0, o], and
define the matrix

where aij = max lai/t)1 (20)

(0, oJ

and

hi = max Ibi(t)1 (21)

[0. oJ

with the exponential matrix function exp(At) assumed to be defined in [0, (X]. Then if
c and b are such that for u ERN,
Downloaded by [Purdue University] at 04:03 03 April 2013

2: (eAO)ijlujl <
j~1
00 i ~ 1 (22)

and then, as shown in [19], the solution of the truncated system v".
N

(yfy = 2: ai/t) yf' + b;(t)

j~1
(23)

yf(O) = c, i = l , 2, ... , N (24)

is unique and tends to the infinite system solution Yi(t), uniformly on [0, (X] as N ~ 00.
Therefore, the procedure to investigate a priori the copvergence of the truncated
system is, basically, that of verifying whether vectors c and b satisfy condition (22), for
sufficiency, once the exponential function exp(At) has been somehow evaluated. Let us
now consider the nonlinear problem

Y; = hi(t, y) (25)

Yi(O) = c, i = 1,2, ... (26)

Let the functions hi be continuous for t E [0, (X] and be such that

i = 1,2, ... (27)

with constants aij ~ 0 such that exp(At) exists.

Also, we define

hi = max IhM, 0)1 (28)

[0.01
 222 R. M. COTI'A

Then, as shown in [19], if c and b above again satisfy Eq. (22), with A as obtained
from Eq. (27), the truncated nonlinear system has a unique solution x'', and yf(t) tends
to Ylt) as N ~ 00, uniformly in [0, a] for every i ~ 1.
The truncated finite system is now readily handled through standard numerical
techniques for ordinary differential equations [16, I7], widely available as well-established
scientific subroutine packages [18]. Eventually, systems such as Eqs. (14)-(17) above,
are likely to become "stiff," that is, systems with widely different t decays, especially
when the order N is increased significantly, due to possible quite different magnitudes.
of lower and higher order eigenvalues. However, specialized routines are also widely
available for such situations, such as Gear's method [16] implemented in the IMSL
subroutine DGEAR [18], posing no additional difficulty to the accurate and reliable
numerical solution of the present class of problems, which is now illustrated.
Downloaded by [Purdue University] at 04:03 03 April 2013

APPLICATION
We consider a transient analysis of a.radiating fin subjected to step changes in base
or ambient temperatures [20, 21], with a nonlinear surface dissipation term, written in
dimensionless form as

eru, t) a2T
- N (T 4 H4)
= - - in 0< x < I, t > 0 (29)
at ax 2 c

T(x,O) = To 0::5x::51 (30)

aT(x, t) I = 0 TO, r) (31)

ax x=o

The following eigenvalue problem is then taken

d~i;X) + J.LNi(X) = ° O<x<1 (32)

d~X) Ix=o = 0 l\JiO) = 0 (33)

which is readily solved as

1 (2i - 1)
l\Ji(X) = cos J.LiX N·I = -
2 J.Li =
2
'IT (34)

The equivalents to Eqs. (14)-(17) are then obtained as

-dO i
dt + J.L·6·
I I
2-
- g·(6)
I J
--
= 0 t> 0 (35)

i = 1, 2, ... , N (36)
 NONLINEAR DIFFUSION PROBLEMS 223

where

(37)

r. = ~ [( -1)j+1J (38)
f, NII2
j IJ.j

and for convenience we defined

8(x, t) = T(x, t) - (39)

Downloaded by [Purdue University] at 04:03 03 April 2013

and, consequently,

eo = To - 1 (40)

Once Eqs. (35)-(38) have been solved for the transformed potentials 8
j , the in-
version formula is recalled:

N
T(x, r) = 1 + v'2 LI cos. IJ.jX
j=
• 9;(t) (41)

Also of interest is the evaluation of the dimensionless heat transfer rates at the fin's
base, x = 1, given by [21]:

Qb(t)
_ ~ ett», t) I (42)
Nt ax x=1

where

N
er«. t) I = - v'2 L (- l)j+llJ.i 9;(t) (43)
ax x=1 ;=1

For best computational performance, however, Eq. (43) is not the most suitable
for evaluating the derivative at the boundary, since the problem is a nonhomogeneous
one. An alternative expression is readily obtained by integrating Eq. (29) and making
use of boundary condition (31) and the definition of average potential, yielding

(44)

where

1 + J: 8(x, t) dx (45)
 224 R. M. COTIA

or,

(46)

RESULTS AND DISCUSSION

To present some representative results, three cases were taken from the related
literature [21], and an analysis of convergence was performed for increasing values of
the order N and at different times t. All the numerical results reported were obtained by
making use of subroutine DGEAR from the lMSL package, with a relative error require-
ment of 10- 5 (five-digit accuracy).
Downloaded by [Purdue University] at 04:03 03 April 2013

Condition (22) was satisfied by taking

(47)

or

(48)

which provides a convenient diagonal matrix, and with

b. = h.(t 0) = N (H 4 -
I
1)-!.. (49)
I " c 60

The convergence rates can now be observed by considering different orders of

truncation, N, for both the dimensionless heat transfer rates and temperature profile. Table
1 illustrates the convergence of Qb(t) , computed both from Eqs. (43) and (46), with N =
5, 10, 15, 20, and 30. As expected, the inversion formula (41), when applied to the
Table I Convergence of Dimensionless Heal Fluxes at the Base (x = I)

N=5 N=IO N = 15 N = 30

Inversion Heal Inversion Heal Inversion Heal Inversion Heat

Case formula" balance" formula balance fonnula balance formula balance

I. To = 0.5 0.01 2.759 2.780 2.837 2.849 2.841 2.849 2.845 2.849
N, :; 0.6 0.05 1.300 1.323 1.311· 1.323 1.315 1.323 1.319 1.323
H:; 0.0 0.09 0.9986 1.022 1.010 1.022 1.014 1.022 1.018 \.022

2. To = 0.7 0.01 1.672 1.703 1.728 1.744 1.733 1.743 1.738 1.743
N, = 0.8 0.05 0.8369 0.8688 0.8524 0.8685 0.8578 0.8686 0.8632 0.8686
H = 0.0 0.09 0.6795 0.7115 0.6952 0.71\3 0.7006 0.71\4 0.7060 0.7114

3. To = 0.7 0.01 1.664 1.685 '1.7\5 1.727 1.719 1.727 1.723 1.727
Nc = 0.6 0.05 0.8096 0.8320 0.8206 0.8319 0.8243 0.8319 0.8281 0.8319
H:; 0.5 0.09 0.6410 0.6636 0.652\ 0.6635 0.6559 0.6635 0.6597 0.6635

"Equation (41).
bEqualion (44).
 NONLINEAR DIFFUSION PROBLEMS 225

Table 2 Convergence of Temperature Profile (Case I; I = 0.09)

x N=5 N = 10 N = 15 N = 20 or 30

0.0 0.5145 0.5146 0.5146 0.5146

0.1 0.5178 0.5178 0.5178 0.5178
0.2 0.5280 0.5279 0.5279 0.5279
0.3 0.5459 0.5459 0.5459 0.5459
0.4 0.5737 0.5739 0.5738 0.5738
0.5 0.6134 0.6133 0.6133 0.6133
0.6 0.666\ 0.6659 0.6659 0.6659
0.7 0.7319 0.7319 07320 0.7320
0.8 0.8107 0.8110 0.8110 0.8110
0.9 0.9016 0.9011 0.9012 0.9012
. 1.0 1.‫סס‬OO 1.‫סס‬OO i.oooo i.oooo
Downloaded by [Purdue University] at 04:03 03 April 2013

boundary quantity (43), converges much slower than the alternative expression (44)
obtained from the heat balance, which utilizes the inversion formula (41) integrated over
the whole domain, as shown in Eq. (45). As expected, for smaller values of t, a larger
number of equations is required to achieve the same degree of accuracy. The overall
behavior is excellent. however, since fully converged results are obtained for considerably
low truncation orders, with little computing effort. Table 2 shows the convergence of
the temperature profile at t = 0.09 and for case I (To = 0.5; N< = 0.6; H = 0.0).
Clearly, as the boundary is approached (x = 0.8 and 0.9), the convergence rate decreases,
though only slightly, as expected from utilizing the inversion formula. Again. the overall
performance is quite consistent, and an excellent agreement was found with the finite
difference results reported in [21].
The present approach can be readily extended to formulations involving nonlinear
convective terms. Its application to the solution of more involved problems shall follow.

REFERENCES
I. P. Li and C. J. Chen, Finite-Differential Method in Heat Conduction: Application of Analytic
Solution Technique, ASME Paper no. 79-WAlHT50, 1979. .
2. C. J. Chen and N. C. Chen, Development of Finite Analytic Numerical Method for Unsteady
Two-Dimensional Navier-Stokes Equations, J. Compo Physics, vol. 53, no. 2, pp. 209-226,
1984.
"3. P. V. Tsoi, Development of Method of Calculating Nonsteady Heat Transfer, High Temp.;
vol. 24. no. 3, pp. 388-395, 1986.
4. M. D. Mikhailov and M. N. Ozisik, Unified Analysis and Solutions of Heat and Mass Dif-
fusion, Wiley, New York, 1984.
5. M. N. Ozisik and R. L. Murray, On the Solution of Linear Diffusion Problems with Variable
Boundary Condition Parameters. J. Heat Transfer. vol. 96, pp. 48-51, 1974.
6. M. D. Mikhailov, On the Solution of the Heat Equation with Time Dependent Coefficient,
Int. J. Heat Mass Transfer. vol. 18, pp. 344-345, 1975.
7. R. M. Cotta and M. N. Ozisik, DiffusionProblemswith GeneralTime-DependentCoefficients,
Brazilian J. Mech. Sci. RBCM. vol. 9, no. 4. pp. 269-292, 1987.
8. R. M. Cotta and M. N. OZisik. Laminar Forced Convection in Ducts with Periodic Variation
of Inlet Temperature, Int. J. Heat Mass Transfer. vol. 29, no. 10, pp. 1495-1501. 1986.
 226 R. M. COTTA

9. R. M. Cotta, Steady-State Diffusion with Space-Dependent Boundary Condition Parameters,

Proc. Nat. Meeting of Thermal Sciences-ENCIT/86, pp. 163-166, Rio de Janeiro, Brazil,
December 1986.
10. J. B. Aparecido, R. M. Cotta, and M. N. Ozisik, Analytical Solutions to Two-Dimensional
Diffusion Type Problems in Irregular Geometries, J. Franklin lnst., vol. 326, no. 3, pp. 421-
434, 1989.
II. R. M. Cotta and C. A. C. Santos, Transient Diffusion Problems with Time-Dependent Bound-
ary Condition Coefficients, J. Eng. Phys., in press.
12. R. M. Cotta, On the Solution of Periodic Multidimensional Diffusion Problems, Int. Comm.
Heat Mass Transfer, vol. 16, no. 4, pp. 569-579, 1989.
13. R. M. Cotta and M. N. Ozisik, Transient Forced Convection in Laminar Channel Flow with
Stepwise Variations of Wall Temperature, Can. J. Chem. Eng., vol. 64, pp. 734-742, 1986.
14. J. B. Aparecido and R. M. Cotta, Steady-State Diffusion Problems with Variable Equation
Coefficients, Brazilian J. Mech. Sci. RBCM, in press.
Downloaded by [Purdue University] at 04:03 03 April 2013

15. J. B. Aparecido and R. M. Cotta, Laminar Flow Inside Hexagonal Ducts, Compo Mech., in
press.
16. C. W. Gear, Numerical Initial Value Problems in Ordinary Differential Equations, Prentice-
Hall, Englewood Cliffs, N.J., 1971.
17. G. Hall and J. M. Watt (eds.), Modern Numerical Methodsfor Ordinary Differential Equations,
Clarendon Press, Oxford, U.K., 1976.
18. IMSL Library, 7th ed., GNB Building, 7500 Ballaire Blvd., Houston, Tex. 77036, 1979.
19. K. Deimling, Ordinary Differential Equations in Banach Spaces, Lect. Notes Math., vol. 596,
1977.
20. J. G. Bartas and W. H. Sellers, Radiation Fin Effectiveness, J. Heal Transfer, pp. 73-75.
February 1960 ..
21. S. M. Saboya and A. C. Sanchez, Unsteady State of Finned Tube Thermal Radiators, Proc.
Latin American Congr. Heat and Mass Transfer, vol. I, pp. 232-243, Sao Paulo, Brasil,
May 1986.

Received 1 December 1988

Accepted 24 August /989