.

Analytical Solutions to Two-dimensional

D@usion Type Problems in h-regular
Geometries
by

J. B. APERECIDO

Department0 de Energia, Instituto Technolbgico de Aeron6utiq

IEME/IEM/ITA/CTA~S6o
JOG dos Campos, SP-12.200, Brazil
R. M. COTTA

Programa de Engenharia Mecanica, COPPEIUFRJ,

Cx Postal 68503, Rio de Janeiro, RJ-21945, Brazil
and

Cidude Universitaria-

M. N. 6ZIyK

Mechanical
University,

and Aerospace Engineering

Department,
Raleigh, NC 2769.5-7910, U.S.A.

North

Carolina

State

ABSTRACT:
A methodology,
bused on the ideas associated with the generalized integral
transform technique, is presented for developing analytical solutions to sufJiently
general
two-dimensional (i.e. x and t space coordinates) dtflision type problems in arbitrarily shaped
singl_v or doubly connected regions when the boundary contour in one of the space coordinates
can be e.xpressed in terms of the other. The model is also applicable for determining ,ftdly
developed velocity distribution in arbitrarily shaped straight ducts. To illustrate the application,
laminar,forcedJEow in a right angle triangle duct is considered and the results are compared
to the available exact solutions to assess the accuracy of the method.

Nomenclature
a, b
a(t), b(t)
A,,(% B,,(t)
A,*,(t),B,:(t)
n,,
.f
f;(x)
G(z)
dz

perpendicular
sizes of the right angle triangular duct
coefficients in operator Z,,
defined by Eqs (6d, e)
defined by Eqs (4d, e)
hydraulic diameter of right angle triangular duct
friction factor
prescribed boundary function in Eqs (Id, e)
imposed

pressure

Re

Reynolds

number

T(x> Y)

velocity field

,&The Franklinlnstitutc
0016-0032/89$3.00+0.00

gradient

in axial direction

,z

PT,,D,
= - ~~~
P

421

J. B. Aperecido

T,,
t, -x,Y
x(t)>x,(G
Greek synhols
X = h/u

et al.
average flow velocity
space coordinates
t-dependent boundary positions

P
/a(t)
VQi(0, x)

aspect ratio of right angle triangular duct

boundary condition coefficients
apex angle of right angle triangular duct
viscosity
t-dependent eigenvalues of Sturtn-Liouville
t-dependent eigenfunctions

Superscripts
_

integral transform

;.i*,l( = 0, 1

system

Subscripts

lowest order solution

iterated lowest order solution

I. Introduction
The analytical solution of steady-state heat and mass diffusion or laminar flow
in ducts having simple regular shapes has been studied extensively over the years,
and systematic presentation
of such results is available in Refs (l-3). An extensive
survey of various approaches with applications
to forced convection inside ducts
is given in Ref. (4).
In this work, we make use of the generalized integral transform technique (5-9)
to develop a methodology
for solving steady-state diffusion or laminar flow problems in a class of irregular domains in which the boundary contour defined by one
of the space variables can be expressed as a function of the other coordinate
variable.

ZZ.General Analysis
We consider sufficiently general two-dimensional
(i.e. t and x space coordinates)
steady-state diffusion type problem in an arbitrarily shaped region in which two of
the boundary contours in the variable x are expressed in terms of the space variable
t, that is, x = x(t) and x = x,(t) as illustrated in Fig. l(a).
The mathematical
formulation
of the problem is taken as
t)= P(x, t),

[~i(x)&+L]T(x,

T(x,t)=O

Bi,k%,
where the operators
422

at

0 = f&)

in

to < t < t,,

x=_~~(t),
at

t=tk,

for
for

x0(t) < x < x,(t),

k=O,l,
k=O,l,

(la)
(lb, c)
(ld,e)

L, and L are defined as

Journal of the Frankhn lnst~tute
Pergamon Prerr plc

2-D DifSusion Type Problems

Y
b

FIG. 1. Geometry

and coordinates
(i) general geometry and (ii) specific geometry
inside a right triangular straight duct.

and the boundary

condition

operator

Br,k =

a,-(-I)$
[

for flow

Br.k as

/ 1

for

k=1,2

(lh)

-where zk, yk, for k = I,2 are the coefficients.

If the above problem were defined in a regular domain, that is, the boundary
surfaces of the region coincide with the chosen coordinate
system, the problem
could readily be solved by the integral transform technique, as discussed in Refs
(2), (3). In the case of an irregular geometry such as specified above, the standard
integral transform techniques are not applicable. However, the basic ideas in the
generalized integral transform technique (5-9) can be extended to develop analytic
solutions to such class of problems as described below.
We assume that, in the problem defined by Eqs (I), the integral transform will
be applied only in the x variable but not in the t variable. With this consideration,
the eigenvalue problem appropriate
for the removal of the differential operator L
in the x variable is taken as
Vol 326. No. 3, pp. 421434.
Prm~d in Chat Britam

19X9

423

J. B. Aperecido et al.

in

WbitO, xl = d(O~WtiL4t0,xl,
with the boundary

x0(t) < x <

XI

(0

(24

conditions
~,&(t),x]

= 0,

Utilizing the eigenfunctions

transform pair is defined as

Inversion

x = xk(t),

k = 0, 1.

$[p,(t), x] of this eigenvalue

Transform

at

Tj(t) =

T(x,

t)

-v,(r)
w(x)Z,(x,
sIO(f)

Zj(X,

problem,

t)T(x, t) dx

t)TJt)

6% c)
the integral

(3a)

W)

,=O

where the kernel Z(x, t) is defined as

z,(x

t)

!@!!a?!

(3c)

N;"(t)

and the normalization

integral

N,(t) as
.x,(0

N,(t) =

wtx)[$h(t),
sx(o

41 dx.

(34

Method ofsolution. First we remove the differential operator L from the partial
differential equation (la) by the application
of the integral transform (3a) as now
described.
Equation (la) is operated on by the operator
x,(0
Z, (x, t) dx

(4)

s .~0(0
to obtain
-x,(1)
w(x)Zi(x,
s.r,(O

t)L,T(x,

t)dx+&(t)T~(t)

= g:(t)

(54

where

s
J,(I)

g*(t) =

Z,(x, t)P(x, t) dx.

.r,,(O

t5b)

The integral term on the left-hand side of Eq. (5a) is evaluated by replacing
the function
T(x, t) by its equivalent
inversion formula (3b) and performing
a
considerable
amount of manipulations.
In this process, the boundary conditions
(1 b, c) are utilized, hence, no longer needed.
The boundary conditions
(Id, e) are transformed
in a similar manner by operating them with the operator given by Eq. (4) and replacing the function T(x, t)
by its equivalent
inversion formula (3b). Finally, the integral transform
of the
problem given by Eqs (1) with respect to the x variable leads to the following

2-D DifSusion Type Problems

system of coupled ordinary
T,(t) :

differential

in the t variable

equations

for the transforms

d*T,(t)

b(t)7

in to < t < t,,

Subject to the boundary
cQ.Tj(&) - (- I)$

for

i= 1,2,3,

(64

conditions

d?dF)

+ f

i-

j=

A$(tJT,(tJ

fi.,,

t
for

k=O,l

and

i=1,2,3,.

(6b,cl

where
A,(f)

= Qt)&, +2b(t)A;(t),

B,j(t)

d(t)A$(t) +b(t)B$(t)

(74

d(t)

- ~
a(t)

0)

(7c)

s
X,(f)

dij =

w(x)ZI(x,
-ro(O

t)Z@,

t) dx =

for

i#j

for

i=j

(7d)

and dot denotes differentiation

with respect to t.
So far, the analysis is formal and exact, yielding an infinite system of coupled
second-order
ordinary differential equations.
Once T;(t) is determined
from the
solution of such a system, the function T(x, t) is obtained by the inversion formula
(3b). However, for computational
purposes we need to consider only a finite
number of such differential equations.
It is unlikely that closed-form solutions can be obtained for the above system of
ordinary
differential
equations ; however, sufficiently
accurate results can be
obtained by the application
of well established algorithms for systems of ordinary
differential
equations
with two point boundary
conditions,
such as a multiple
shooting scheme (10).
Here, we are also interested in developing straightforward,
explicit, approximate
analytical solutions as an alternative to the numerical evaluation of these equations,
hence we discuss below a method for obtaining approximate
analytic solutions to
the system of ordinary differential equations
(6). Once the transforms
Ti(t) are
known, the function T(x, t) is obtained by the inversion formula (3b).
Lowest Order Solution (LOS)
We assume that in the coefficients matrix for the equation formed from the
elements A,,(t) and B,(t), the diagonal elements are sufficiently dominant.
Then,
an approximate
system is obtained
by letting i =.j in the summations
of the
Vol 326. No. 3. pp. 421 434, 19x9
Prmtcd in Great Bntain

425

J. B. Aperecido et al.
complete
system

system given by Eqs (6), leading to the following

d2TL(t)+A,,(t)

b(t) dt2

-iL

.zta

lowest-order

+ Bii(t)F<,L(t) = h,*(t),
in

subject to the boundary

decoupled,

t,,<r<t,,

i=l,2,3

,...

@a)

conditions

[ui-(-l)y~A$(t~)]~i,~(tl)-(-l)iyx11_7;~q

= f;,k

for

k = 0, 1.
(gb, c)

This approximate
system is expected to produce
context of the assumptions
made.

a sufficiently

accurate

result in the

Iterated Lowest Order Solution (ILOS)

An analytical iteration of the lowest-order solution over the complete system (6)
can be employed to approximately
account for the effects of non-diagonal
elements
in matrices A,,(t) and Bij(t). Then, the resulting decoupled system becomes
d2Tih(t)
dT,,(t)
b(t)pd+
+A,;()-

+B,,(t)T,,,(t)

= Gi(t),

to < t < t,

i = 1,2,. . . ,

Pa>
and

dFi/z(t,)= pI,k

[X~-(-l)kY~A~(tk)l~i,,~(t~)-(-l)iY~~~

for

k = 0, 1
Pb, c)

where

ndt>= h:(t) -

lj;

d%(t)

A,,(t) dt

E,i,= .f,k$.1 (- v?kA;(tk)~,,L(tl,).

,= I

+Bi,(t)T,,,(t)

(94
(95 f)

i#r

Therefore,
once the lowest-order
solution
T,;,L(t), and/or the iterated version,
r,;.,,(t), have been obtained, the inversion formula (3b) can be utilized to yield an
approximation
for the function T(x, t) at any point inside the medium.

III. Application
In order to illustrate the application
of the analytical solutions presented here
we consider a fully developed laminar flow inside a right angle triangular duct, as

426

2-R D@sion

Type Problems

illustrated in Fig. lb, with the objective of developing explicit solutions for the
velocity field for arbitrary values of the aspect ratio. An exact solution is available
in Ref. [4] for 0 = 45, which corresponds
to an isosceles right angle triangular
duct ; however, analytical explicit solutions are not available for an arbitrary value
of the angle 8 (or the aspect ratio), although numerical results are available for
some cases (4). The mathematical
formulation
of the problem for the velocity
T(x, y) is given by
(?T(x, y) + a2 T(x, y)
------=-C
iYX2
dy2

in

O<y<b,

O<x<x,(y)

(lOa)

where

l w4
p dz

c = _

x,(y)

1-x

=a

(
and the no-slip

boundary

conditions

(lob)

for the velocity

T(x, y) are

T(O,.Y) = 0,

T[x,(y),yl = 0

(104 e)

T(x,O) = 0,

T(x, b) = 0.

(iOf, g)

From direct comparison of the systems (1) and (lo), the following
between them is obtained
Q(Y) = 1,
d(x) = 0,

W(X) = 1,

k(x) = 1

,fx(x, tr) = 0,

t+ y

b(y) = 1,
P(x, y) = c

and the solution of the corresponding

readily determined as

eigenvalue

problem

correspondence

(11)

in the x variable

(124

ti(k, x) = sin (w)

P,(Y) =

Zdx,y) =
The matrix coefficients

is

2L
xi (Y>

(12b)

&sin[$&I.

(12c)

A: and B$ become
i=j

0,

(13a, b)

AZ(Y) =
1

i#j

i=j,
C(Y)

(13c, d)
i#j,

Vol. 326. No. 3, pp. 421434.

Prmted in Great Bntain

,989

427

J. B. Aperecido et al.

Ai.,

= 24(v),

Bij(Y)= B$(~)-d(~)6ij,
/y(y)

= -

;
0

The complete

system of transformed

d2Ti
dy2 +2 5 A:(&!$)
,= 1

&,(y)[l+(-I)+].
-~
equations

+ f B,,(y)T,(y)
j= I

(1W

is given by
in

= &Q),

0 < y < 6,
i = 1,2,3,.

subject to the boundary

(144

conditions
FL(O) = 0,

Equations
determined.

(14) can be readily

Tl(b) = 0.

solved numerically

(1%
and the transforms

c)

T,(t) are

Lowest Order Solution (LOS)

As discussed previously, the lowest-order approximation
to the system (14) is
obtained by setting i = j in the summation terms and noting that A,T(y) = 0 by Eqs
(13a,b).
d$+

+ MJ)~.L(_v)

subject to the boundary

= WYY),

in

0 < y < b,

i = 1,2,3,.

..

(154

conditions
T,,L(0) = 0,

T!,L(b) = 0.

(15b, c)

Here, the subscript L is used to denote the lowest-order approximation.

Equation
(15) is solved and inverted by the inversion formula (3b) to obtain the lowest-order
solution for the velocity TL(x, y) in the form

I--l
i7cx

i= z,

i[: :I:l)i$C(il),]
. ~[.~,(y)]?-[a]5~~+(~,),.

[.~,(y)]~~~~(,)2}

(16a)

where
(A,), = -:+(b,,+d>,

(16b)

(2)) z = -4 - (b,, + $) I,

(16~)
(i7r)2.

The average flow velocity

428

(164

T,, is defined as
Journal oC the Frankhn lnsutute
Pergamon Press plc

2-D D@sion

Type Problems

ss
h

Ta, = ai

x,(v)

(17a)

T(x, Y) dx dy.

0 0

Introducing
Eq. (16a) into (17a) the lowest-order
velocity Tay,L becomes

solution

for the average

1
f
v~
~~
,=1,x,5,. .12[5+2(~1)11[3-2(j-i)21
The friction

factor can then be readily determined

j-Rex

flow

(17b)

from its definition

;$

(18a)
I

where the hydraulic

diameter

is given by
2cui
(18b)

and,
b
CI= ~ = aspect ratio.
a
Also of practical

interest

is the incremental

pressure

(18~)
drop number

K(W) defined

as
K(a)

(1W

-&(~)I

= 2[&@)

where

(19b)
is the kinetic energy correction

factor, and
(19c)

is the momentum
of the duct.

flux correction

factor, and A,. designates

Iteruted Lowest Order Solution (ILOS)

Once the lowest-order
solution has been
becomes

d*~,dy)
++Bi,(y)Tzi,h(y)

= I?;(y),

obtained,

0 -c y < b,

dy-

Ti,l,(0)
= 0;

T,,h(b) = 0

the cross-sectional

the system

area

of equations

i = 1,2,. . .

(204
(2Ob,c)

where
0,. 326. No. 3, pp 421434. 19x9
Printedin Great Bntam

429

J. B. Aperecido et al.

I?!(y) = h;(y) -2

5
;=
/f

The iterated lowest-order

the lowest-order solution

solution
as

(204
I
1

J+

can be expressed

in terms of a correction

Z./z(Y) = Ti,L(Y) + Ti,c(Y)

over

(21a)

where the correction term, Ti,C(y), is obtained in the same way as the lowest-order
solution obtained from the Euler equation, to yield
_ [5(alb!a~>l

j=I ~5/2+(~j~11[5/2+(~j)*1

if

[5/2+(/li),1[5/2+(1,,),1

where,
hf=

Finally,
solution

after inversion
becomes

_($)2(~Y$+(_l)Ji-il

(21c)

a$ = A$(Y)-~I(Y>

(214

b; = B;(y)x:(y).

(21e)

by the inversion

formula

(3b), the iterated

lowest-order

T,, 6, Y) = r,_ (x, Y) + T, k J)

(22a)

where

(22b)
All the other related quantities,
such as average flow velocity. friction factor,
and incremental
pressure drop number, can be similarly obtained by the direct
application of Eqs (17a, 18a, 19ac).

IV. Results and Discussion

The velocity distribution
in laminar flow inside right angle triangular ducts is
reviewed in Ref. (4) and the exact solution is presented for the case 0 = 45 (i.e.
right angle triangular
isosceles duct). In addition, values off Re, K,(W), &(cc)
and K(co), obtained from the numerical solutions available in the literature, are

430

Journal

olthc Frenkhn
Pcrgamon

Instilule
Press plc

2-D Dt~ision

Type Problems

TABLE I

A comparison of the product f Rejor various apex angles l3 obtaixedfrom the present analysis
with the practically exact numerical solution jiom ReJ (4)
.f Re
0
Apex

5
10
15
20
25
30
35
40
45

Present analysis
_

Ref. (4)
Exact numerical
solution

Complete solution
with N = 25

12.000
12.27
12.49
12.68
12.83
12.94
13.034
13.09
13.13
13.154

LOS

12.0000
12.2556
12.4734
12.6587
12.8127
12.9366
13.0319
13.0992
i3.1394
13.1528

12.000
12.252
12.466
12.647
12.801
12.933
13.049
13.157
13.264
13.381

% error

ILOS

% error

0.000
-0.147
-0.192
-0.260
-0.226
-0.054
0.115
0.512
1.021
1.726

12.000
12.249
12.454
12.621
12.754
12.860
12.942
13.009
13.068
13.126

0.000
-0.171
-0.288
-0.465
-0.592
-0.618
-0.706
-0.619
- 0.472
-0.213

presented for various apex angles. Here we examine the accuracy of the present
solutions by comparing with the exact results presented in Ref. (4).
Table I lists the practically exact values of f Re obtained either by a purely
numerical approach (4) or by the complete solution given here, together with the
results obtained from our lowest-order and iteratedlowest-order
solution for various
apex angles from Q = 0 (i.e. parallel-plates)
to 0 = 45 (right angle-isosceles
triangle duct). The eigenvalue problem for the present analysis leads to exact results
for 0 = 0. As f3is increased, the regular parallel-plate
geometry is distorted and
the problem becomes one of irregular geometry. Apparently,
as the most critical,
---

--w_
l5-

I
5

FIG. 2. K,(m), K,(a)

Vol.326,No. 3.pp.421434, 1989
Printed,"Great Bntaln

K,(m)
--k-u

---_._
I
IO

and K(a)

I
15

.-.-

ILOS
LOS

EXACT

--LIyii~L-.-.
I
20

I
25

I
30

I
35

----_.=A
I
40

for LOS, ILOS and NUMERICAL

I
45

solutions.
431

J. B. Aperecido

et al.

Y
FIG. 3.

Comparison of LOS, ILOS and EXACT solutions (X = 0.10).

distorted geometry of right isosceles triangle is approached,

the relative error for
the lowest-order
solution increases to a value of about 1.7% at 8 = 45; but the
variation of the error is not monotonical.
The iterated lowest-order
solution, to
some extent, corrects for the relative importance
of non-diagonal
elements. The
complete solution obtained with the subroutine
DUCPR of the IMSL subroutine
package (lo), by taking 25 terms in the related summation,
practically coincides
with the purely numerical solutions reported previously.
In Fig. 2, K,,(a) and K,(m) obtained from the lowest-order and iterated lowestorder solutions plotted against the apex angle, 8. Included on this figure are the
results obtained from numerical solutions. For 0 = 0 (i.e. parallel-plate
geometry),
the agreement between the numerical and the present results are excellent; some
noticeable accuracy loss occurs as 0 -+ 45 or a + 1. The iterated lowest-order
solution underestimates
the incremental pressure drop number ; especially for 0 -+

005

FIG. 4. Comparison

432

oi

015

02

025

03

035

04

of LOS, ILOS and EXACT solutions

045

05

(Y = 0.50).
Journal

of the Franklin
Pergamon

Institute
Press plc

2-D Diffusion

.,

.-.-

LOS

EXACT

Type Problems

05-

005

I
01

I
015

I
02

I
025

I
03

I
035

I
04

I
045

0.5

FIG. 5. Comparison

of LOS, ILOS and EXACT solutions (Y = X).

45, it is less accurate than the lowest-order

solution for K?(a). Such behavior is
due to the fact that the normalized
velocity profile obtained with the iterated
solution is flattened up with respect to the lowest-order
solution. This effect is
magnified when the profile is raised to the third power and integrated.
Finally, Figs 3-5 show the locally fully developed velocity profiles along selected
lines for the most critical situation 0 = 45. The lines selected include x = 0.10 (the
bottom of the triangle), y = 0.50 (at the midplane
of the triangle), and y = x
(passing through the centroid).
Clearly, the agreement
is reasonable,
and the
iterated solution corrects the lowest-order solution to some extent.
In conclusion,
the lowest-order
solution presented here is accurate enough for
most practical purposes and provides a simple, straightforward
approximate
analytic solution for steady-state, diffusion type problems in a class of irregular geometries. The iterated version, although it provides some correction over certain range
of the parameters,
might not always be worthwhile
to apply due to the extra
computations
required. The complete solution, on the other hand, provides a fast
and reliable alternative to time consuming purely numerical approaches.

V. Conclusions
A methodology
is presented for the analytic solution of a sufficiently general,
two-dimensional,
steady-state, diffusion type equation in arbitrarily shaped singly
or doubly connected regions when the boundary contour in one of the coordinate
systems can be expressed in terms of the other. The lowest-order
solution can
provide sufficiently accurate results when the diagonal elements of the coefficient
matrix are dominant compared to the nondiagonal
elements.
Acknowledgement
One of the authors (J.B.A.) wishes to acknowledge
the financial support
CAPES/Brazil
and UNESP-Ilha
Solteira-Brazil
during his stay at ITA.
Vol. 326, No 3, pp. 421434,
Prmted in Great Britain

provided

by

1989

433

J. B. Aperecido

et al.

References
(1) M. N. &isik, Heat Conduction,
Wiley, New York, 1980.
(2) M. D. Mikhailov and M. N. ijzigik, Unified Analysis and Solutions of Heat and
Mass Diffusion, Wiley, New York, 1984.
(3) M. D. Mikhailov and M. N. ijzigik, An alternative general solution of the steadystate heat diffusion equation,
ht. J. Heat Mass Transfer, Vol. 23, pp. 6099612,
1980.
(4) R. K. Shah and A. L. London, Laminar flow forced convection in ducts, Adt. Heat
Tran@r, Suppl. 1, 1978.
(5) M. N. Ozigik and R. L. Murray, On the solution of linear diffusion problems with
variable boundary condition parameters,
ASME Paper No. 74-HT-I, 1974.
(6) R. M. Cotta and M. N. ijziaik, Transient forced convection in laminar channel flow
with stepwise variations of wall temperature,
Gun. J. Chem. Engng, Vol. 64, pp.
734742, 1986.
(7) R. M. Cotta and M. N. oziaik, Laminar forced convection in ducts with periodic
variation of inlet temperature,
ht. J. Heat Mass Transfer, Vol. 29, no. 10, pp.
1495-1501, 1986.
(8) R. M. Cotta, Steady-state
diffusion with space-dependent
boundary condition parameters, Proc. First Nutional Meeting Thermal Sciences-ENCIT-86,
pp. 163-166,
Rio de Janeiro, Brazil, December 1986.
(9) R. M. Cotta and M. N. ijzigik, Diffusion problems with general time-dependent
coefficients, Braz. J. Mech. Sci.-RBCM,
submitted for publication.
(10) ZMSL Library, Edition 7, GNB Building, 7500 Ballaire Blvd, Houston, Texas, 77036,
1979.

Journal

434

of the Franklin
Pergamon

lnst~tute
Press plc