CHAPTER

ase
•
on ca 1n once

4.1 INTRODUCTION
As discussed in Chapter 3, order-of-magnitude analysis is crucial in identifying valid approx-
imations. Using it to find length and time scales and to predict temperature, concentration,
or velocity scales occasionally provides all one needs to know. Usually, however, more pre-
cise results are desired. Because problems cannot always be reduced to one-dimensional
forms that are steady or pseudosteady, techniques are needed for solving partial differen-
tial equations. Even ordinary differential equations sometimes require methods beyond
what sufficed in Chapter 3 or is reviewed in Appendix B. Accordingly, this chapter and the
next focus on methods for finding analytical solutions. The methods to be presented were
selected both for their wide applicability and for their conceptual value, but by no means
are they a comprehensive set. Familiarity with such methods has at least two benefits, even
with the excellent computational resources that are widely available. First, an analytical
result is almost always preferable to a numerical one, in that it displays the relationships
in a more general and explicit manner. Second, a good analytical background leads to the
more effective use of numerical methods (which are beyond the scope of this book). Even
if the "real" problem must be solved numerically, an analytical solution to a related one
can help test a new code. Moreover, experience in solving partial differential equations
analytically can warn of pitfalls that sometimes arise when using commercial software.
The topics in this chapter are the similarity method (Section 4.2) and regular and sin-
gular perturbation analysis (Sections 4.3 and 4.4, respectively). They are grouped together
because all are related to the scaling concepts discussed in Chapter 3. The similarity method
typically requires that a dynamic length scale be much smaller than the system dimensions,
and perturbation methods are applicable when a problem with scaled variables contains a
parameter that is either small or large compared with unity. Similarity and perturbation also
share the feature of not being restricted to linear boundary value problems, unlike the

113
114 SOLUTION METHODS BASED ON SCALING CONCEPTS

eigenfunction expansion and singularity methods presented in Chapter 5. Indeed, pertur-

bation analysis is a widely used tool for making nonlinear problems tractable.

4.2 SIMILARITY METHOD

All methods for solving partial differential equations accomplish their aim by somehow reduc-
ing the task to that of solving an ordinary differential equation. The similarity method does
this by combining two original, independent variables into one composite variable. Accord-
ingly, it is sometimes called combination of variables. Similarity analysis is applicable to
certain problems in which the length scales are determined only by rate processes and not by
geometric dimensions. Such problems involve domains that may be viewed as infinite or semi-
infinite. Examples include transient diffusion or conduction with small penetration depths and
steady, two-dimensional flows involving velocity, temperature, or concentration boundary lay-
ers. It should be mentioned that, even when geometric length scales are absent, the similarity
method will not always work. However, testing its applicability usually requires only a mod-
est effort, and when it does work the solution is often very elegant.
In the simplest applications of the similarity method, the dependent variable (veloc-
ity, temperature, or concentration) has profiles that have the same shape for all times or
positions, differing only by the length scale over which the variations occur. In other words,
the profiles are sufficiently self-similar to be superimposed graphically by introducing a
time-dependent or position-dependent scale factor. This is illustrated for transient diffusion
in Fig. 4-1, which shows concentration profiles for a system in which the concentration at
x = 0 is suddenly increased from 0 to C0 at t = 0. As shown on the left, when the abscissa
is x the profiles at various times differ. However, as shown on the right, when the time-
dependent penetration depth is taken into account by changing the abscissa to xj 8(t), a sin-
gle curve is obtained. This ability to superimpose all the curves shows that the concentration
is expressible as a function of just one variable, a particular combination of x and t. Such
a function will be governed by an ordinary differential equation instead of a partial

c c

0 L-L....::::::::....._l...._---=::t=--=:::::::::::...._ 0
0 81 X 0 xl~t)

Figure 4-1. The effect of a similarity transformation in a transient diffusion problem. On the left, the solute
concentration is shown as a function of position at three times (t3 > t2 > t 1) following a sudden increase at x = 0.
In this particular illustration the penetration depth 8(t) is defined (arbitrarily) as the distance corresponding to
7/8 of the concentration change, and 8i = 8(ti)· On the right, the abscissa has been changed to xj8(t), yielding
a single curve that is valid for all x and t.
 4.2 Similarity Method 115

differential equation. If there were a length scale other than the penetration depth, such as
a domain thickness (L), this strategy would not work. That is, to satisfy a boundary condi-
tion at x == L, x would have to be kept separate from t.
In Fig. 4-1 it was sufficient to adjust the coordinate using the time-dependent length
scale, while leaving the concentration alone. In certain other problems the dependent vari-
able must be transformed as well, using a scale factor of its own. In that case, both axes of
a plot are "stretched" in a way that causes a family of curves to superimpose. For steady
boundary-layer flows, the length scale is the boundary layer thickness, which typically
depends on position along a surface. In such problems the scale factors are functions of
position instead of time.

EXAMPLE 4.2-1 Diffusion from a Boundary The similarity solution developed here applies to the
earliest times in the membrane diffusion problem in Example 3.5-1. Whereas the previous pseu-
dosteady analysis required that t > L 2 jD, the similarity approach is applicable when t << L 2jD. At
such early times the boundary at x =Lis not "felt," and the membrane acts as if it were semi-infinite.
What makes this type of problem a good candidate for similarity analysis was discussed already in con-
nection with Fig. 4-1.
Using the dimensionless concentration e = Cj(KC0 ), the governing equations for @(x,t) at
early times are
ae a2e
--=D-- (4.2-1)
at ax2
@(x,O)=O (4.2-2)
@(O,t) = 1 (4.2-3)
@(oo,t) = 0. (4.2-4)
Equation (4.2-3) assumes that, with little time for changes to occur in the external compartments, the
concentration at x = 0 remains nearly constant. Equation (4.2-4) expresses the fact that the membrane
behaves as if it were semi-infinite.
It is assumed that e = @(TJ) only, where

X
(4.2-5)
TJ = g(t)

and g(t) is a scale factor that is to be determined. This scale factor is proportional to the penetration
depth. The idea that e is expressible as a function of the composite variable TJ is only a hypothesis,
and it must be verified by showing that the partial differential equation and auxiliary conditions can
all be satisfied. The differential equation will be transformed first, and then the initial condition and
boundary conditions will be considered. The underlying philosophy is to proceed until one of two
outcomes occurs: (1) all equations are satisfied, showing that the problem has been solved; or (2) an
irresolvable contradiction occurs, indicating that some other approach is needed.
Using subscripts as reminders of which variable is being held constant, the partial derivatives
in Eq. (4.2-1) are transformed as

ae aTJ de -71g' de
(4.2-6)
at x at x dTJ g dTJ
ae aTJ de 1 de
(4.2-7)
ax t ax t dTJ g dTJ
a2e 1 aTJ d de
(4.2-8)
ax2 t g ax t dTJ dTJ
116 SOLUTION METHODS BASED ON SCALING CONCEPTS

where g' = dgjdt. After substitution and rearrangement, Eq. (4.2-1) becomes

d2e gg' de
- -2 + TJ =0. (4.2-9)
dTJ D dTJ
If the similarity hypothesis is correct, then neither t nor x can appear separately in the equa-
tions that govern e. Because g = g(t), time will be eliminated as a separate variable in Eq. (4.2-9)
only if the product gg' is a constant. Accordingly, we assume that
gg'
=c (4.2-10)
D
where c is a dimensionless constant. Because g corresponds to the penetration depth, which grows
with time, we need g > 0, g' > 0, and c > 0. Otherwise, the value chosen for c is unimportant, as
will be shown. Recognizing that 2gg' = (g 2)' and that the initial penetration depth is zero,
Eq. (4.2-10) becomes

g(O) = 0. (4.2-11)

This has the solution

g(t) = (2cDt) 112 (4.2-12)

which confirms that there is indeed a function g(t) that permits e

to depend on TJ alone, at least in
the differential equation for e. This form for g(t) could have been anticipated from the discussion
of penetration depths in Section 3.5. That is, it could have been reasoned at the outset that g(t) must
equal a positive constant times t 112 . Expressing the constant as in Eq. (4.2-12), the desired compos-
ite variable is
X
TJ= (4.2-13)
v2clli"
How best to choose the value of c is discussed later.
Returning now to the differential equation for e(TJ), Eq. (4.2-9) becomes

(4.2-14)

Integrating once gives

de= ae-cr//2 (4.2-15)
dTJ
where a is a constant that will be found from the boundary conditions. A second integration results in

e(TJ) =a e-c17212dTJ +b (4.2-16)

where b is another constant. It is convenient to rewrite this general solution as

1J

e(TJ) = e(TJo) + a e-csz/2ds (4.2-17)

1Jo

where s is a dummy variable that will disappear after the integration. Using a definite integral replaces
b with another unknown constant, e(TJo). Equation (4.2-17) is as general as Eq. (4.2-16) and is valid
for any reference position such that TJo > 0.
It remains to be shown that the initial condition and boundary conditions can be satisfied.
Because g(O) = 0, Eqs. (4.2-2) and (4.2-4) are equivalent to one another. That is, t = 0 and x = oo
both correspond to TJ = oo. Accordingly, Eqs. (4.2-2) through (4.2-4) reduce to

e(O) = 1, e(oo) = 0. (4.2-18)

 4.2 Similarity Method 117

This collapse of three auxiliary conditions into two is crucial, in that the second-order differential
equation for @('Y}) can satisfy only two conditions. This reduction would not have occurred if the
membrane had not been modeled as semi-infinite; this is where the absence of a geometric length
scale is necessary. In applying Eq. (4.2-18) to Eq. (4.2-17), either 'YJo = 0 or 'YJo = oo would be a good
choice. Using 'YJo = 0 and evaluating a, the concentration is found to be
,---TJ
~
a('YJ) = 1 - 2c e -cs2/2d
s. (4.2-19)
7T
0
This confirms that the similarity method works for this problem. However, a few more steps are
needed to put the solution in the best form, including choosing the optimal value for c.
An integral like that in Eq. (4.2-19) appears in both the error function (erf) and the comple-
mentary error function (erfc), which are defined as
X
2 2
erf(x) = y:;;. e-u du, erfc(x) =1 - erf(x). (4.2-20)
0
Plots of these functions are given in Appendix B. Using the variable change u = sv-;j2, Eq. (4.2-19)
becomes
(4.2-21)

Recalling Eq. (4.2-13), note that 'Y}v-;j2 = x j (2Viit). That is, upon returning to the variables x and
t, the constant c cancels out. Thus, c can be assigned any convenient positive value. The arguments
of erf and erfc are simplest if c = 2. Accordingly, the final result is written as
X
@('Y}) = 1 - erf('Y}) = erfc('Y}), 'Y}= (4.2-22)
2Viit"
The desirability of choosing c = 2 could have been anticipated when first integrating Eq. (4.2-14),
in that it leads to the simplest argument for the exponential in Eq. (4.2-15).
Concentration profiles corresponding to three times are shown in Fig. 4-2. To construct this
plot, an arbitrary time t0 was chosen as a reference, and dimensionless time and position were defined

CD
..._....
c: 0.8 -

0
·-••co•
'
........
c:
CD 0 .6
0 1" = 10
c
0
0
fJ)
U) 0.4 -

-0c::
([)

·- r =1
U)
c:: 0 .2
CD
.E -
r= 0 .1
0
0
0 0.5 1 1.S 2 2.5 3
Dim1ensionless Position (x)

Figure 4-2. Concentration as a function of position (X) at three times (T) following a step change at a bound-
ary. The variation of erfc(x/ T112) with xis shown forT = 0.1, 1, and 10. The concentration is scaled using the
value at the boundary, and dimensionless position and time are as defined in Example 4.2-1.
118 SOLUTION METHODS BASED ON SCALING CONCEPTS

as T = tjt0 and x = xj g(t0), respectively. Thus, 1J = x/T 112 and what is plotted is erfc(x/T 112) as a
function of x for T = 0.1,1, and 10. The actual concentration profiles in Fig. 4-2 resemble those
sketched in Fig. 4-1. Given that erfc(1) = 0.16, most of the concentration decrease occurs by
1J = 1 or x = g(t). Thus, the scale factor g(t) is indeed a representative penetration depth.
The solute flux at x = 0 is

Nx
ac
=-D-
DKC0 d@ D 1/2
KC0 . (4.2-23)
x=o ax x=O g d1J 7]=0 7Tl

The concentration gradient is inversely proportional to the penetration depth. Thus, the flux varies
as t- 112 because the penetration depth varies as t 112 .

EXAMPLE 4.2-2 Diffusion from an Instantaneous Point Source Suppose that at t = 0 a certain
amount of an inert solute is released at the plane x = 0 (e.g., injected carefully into a column of liq-
uid). The number of moles introduced per unit area is m 0 . Diffusion will cause the initially sharp
concentration peak to spread, and the maximum concentration will decrease over time, as illustrated
in Fig. 4-3. (This is actually a plot of the final solution, although its qualitative features can be antic-
ipated.) Because the total amount of solute is constant, the area under each curve must be the same;
hence, the decrease in the peak concentration. It is desired to predict C(x, t) for t > 0.
It is apparent from Fig. 4-3 that the concentration profiles at various times cannot be super-
imposed simply by dividing x by a time-dependent scale factor, as in Example 4.2-1. That is, chang-
ing the abscissa will not suffice. Because the maximum concentration decreases with time, a second
scale factor must be sought. Accordingly, it is assumed that

X C(x, t)
fJ(TJ) = f(t) (4.2-24)
1J = g(t) '

where f(t) is the scale factor for the concentration. Although 1J is dimensionless, @ is not. Thus,
although f(t) is called a scale factor, it is not actually a concentration.

t"=1
1

0.8
T=2
/ - ........
0.6 /

-c:: 0.4 ~
·-0en
c:
Q)
E 0.2 ,' ...
' ......
....

0 1 2 3
Dimensionless Position Ct)
Figure 4-3. Concentration as a function of position (X) at three times (T) following a release of solute at the
origin. The variation of exp(-x2/T)jT 112 with xis shown forT= 1, 2, and 3. The concentration is scaled using
the peak value at T = 1, and dimensionless position and time are as defined in Example 4.2-1.
 4.2 Similarity Method 119

Choosing the spatial domain as x > 0 (because of the symmetry), the equations that govern
C(x, t) are

ac a2
-=D--
c (4.2-25)
at ax2
()()

mo
- = C(x, t)dx (4.2-26)
2
0

-
ac (O,t) = 0, C(oo,t) = 0. (4.2-27)
ax
Equation (4.2-26) states that the amount of solute in the region being modeled is always half of
that added initially. Thus, it is assumed that the concentration distribution is symmetric about
x = 0. Aside from this implied symmetry, the details of the initial concentration distribution
prove to be unimportant.
Before transforming the governing equations to introduce the similarity variables, it is help-
ful to use order-of-magnitude analysis to anticipate the functional forms of the scale factors.
Equation (4.2-25) indicates that g(t) rv (Dt) 112 , as in Example 4.2-1. The required form for f(t) is
revealed by the integral constraint. Denoting the maximum concentration (that at x = 0) as Cm(t),
Eq. (4.2-26) indicates that m 0 j2 rv Cm(t)g(t), or

Cm(t) rv 1/2. (4.2-28)

2(Dt)

Because f(t) must mimic the time dependence of Cm, it is apparent now that f ex t- 112.
The derivatives in Eq. (4.2-25) are transformed as

ac = ,e _ g'f de
at 1 g TJ dTJ
(4.2-29)

ac f de
(4.2-30)
ax g dTJ

a2c f d2e
(4.2-31)
ax2 g2 dTJ2

where f' = df/dt and g' = dgjdt. The resulting differential equation for @(TJ) is

(4.2-32)

If the similarity hypothesis is correct, t cannot appear independently in Eq. (4.2-32). Thus, the
coefficients involving f(t) and g(t) each must be constant. Because gg' / D was constant also in
Example 4.2-1, it is convenient to use the same function g(t) as before,

g(t) = 2(Dt) 112 . (4.2-33)

With g specified in this manner, the remaining coefficient becomes

2
f' g f'
- - =4t =K (4.2-34)
Df f
120 SOLUTION METHODS BASED ON SCALING CONCEPTS

where K is a constant that must be determined. Equation (4.2-34) is a first-order, linear differential
equation for f(t), which has the general solution
f(t) = atK/ 4 ( 4.2-35)

where a is another constant. The order-of-magnitude analysis indicated that f ex t-112 , so that K = -2.
As will be shown shortly, the value of a is immaterial (provided that a i:- 0), so that a= 1 is chosen for
convenience. Thus, the scale factor for the concentration is
f(t) = t- 112. (4.2-36)

The differential equation for l3(TJ) is now

(4.2-37)

One integration yields a first-order linear equation. The solution that is consistent with both bound-
ary conditions in Eq. (4.2-27) is
2
l3(TJ) = be-11 • (4.2-38)

The only equation that has not yet been satisfied is the integral constraint. Substituting Eq. (4.2-38)
into Eq. (4.2-26) and performing the integration gives b = (m 0 j 2)(1TD)- 112. If a= 1 had not been
specified, this would simply be the value of the product ab; thus, as stated above, a has no independ-
ent significance. The final result for l3(TJ) is
mo 2
l3(TJ) = 1/ 2 e-11 • (4.2-39)
2(1TD)
Returning to dimensional variables, the concentration is given by

C(x,t) = mo e-x2/ 4Dt. (4.2-40)

2(1TDt) 112
This resembles a normal or Gaussian probability distribution, with a peak centered at x = 0 and a
variance of 2Dt. The maximum concentration is
mo
Cm(t)=--- (4.2-41)
2(1TDt)112
which differs from the order-of-magnitude estimate in Eq. (4.2-28) only by a factor of 1/ 1r 1/ 2.
In Fig. 4-3 the variables are each made dimensionless using values at an arbitrary time t0 , as in Exam-
ple 4.2-1. What is plotted is T- 112 exp (-x 2/ T) as a function of x forT= 1, 2, and 3.

More extensive discussions of similarity analysis are provided in the monographs by

Barenblatt (1996), Dresner (1983), and Hansen (1964). Many examples of its application to
problems in fluid mechanics and convective heat and mass transfer appear later in this book,
beginning in Chapter 7.

4.3 REGULAR PERTURBATION ANALYSIS

Perturbation methods are applicable to problems that contain a small dimensionless param-
eter, usually denoted as e. If scaled variables are used, as in this book, then "small" means
e << 1. Because a large parameter can always be inverted and renamed, perturbation meth-
ods are applicable then as well. The basic strategy is to express the solution as a series
involving e. In a regular perturbation problem the result is a power series. The leading term
2
corresponds to the solution for e = 0, and succeeding ones involving e, e , etc., provide
 4.3 Regular Perturbation Analysis 121

corrections to the initial approximation. Ideally, the procedure for generating the coeffi-
cients in the series replaces an original problem (which may be intractable) with a sequence
of much simpler ones. This section focuses on regular perturbation analysis. Singular per-
turbation problems, which are distinguished by the fact that setting e == 0 does not provide
a good first approximation and which do not necessarily involve power series in e, are dis-
cussed in Section 4.4.
The power series in a regular perturbation solution is equivalent to a Taylor expan-
sion about e == 0, with e being treated as a variable. For the smooth functions encountered
in transport problems, such a series will be convergent. That is, for a fixed value of e,
adding more terms will improve the accuracy of the result. However, in singular problems
the series may be nonconvergent. Although not necessarily improved by adding more terms
withe fixed, it becomes exact as e ) 0. Such expressions, termed asymptotic series, can
be remarkably accurate with just a few terms, their lack of convergence notwithstanding.
Introductions to asymptotic series and to perturbation methods applied to both algebraic
and differential equations are provided in Bender and Orszag (1978), Hinch (1991), and
Holmes (1995).
In constructing perturbation solutions we need a notation to indicate which terms are
negligible at a given level of approximation. The statement "f(e) == O(en) as e ) 0," read
as "f(e) is of order en as e ) 0," means that lf(e )I/ en < m as e ) 0, where m is any pos-
itive constant. In other words, f(e) vanishes fast enough to keep the ratio f(e)jen bounded.
Any sum of tem1s involving en, en+l, en+2 and so forth is O(en), so that for a polynomial
in e the order in the "0" sense is determined by the smallest exponent. When using the
"0" notation the statement "as e ) 0" is usually omitted, the limit already being clear from
the context. There is a distinction between order as expressed by "0" and order of magni-
tude, as denoted in this book by"~". This is because the former is based on boundedness
and the latter on magnitude. For example, iff== ke, where k is a constant, f == O(e)
whether k == 1 or 100. Nonetheless, when the variables are scaled, the order of magnitude
of a term and its "0" order are often the same. That is, scaling tends to keep proportional-
ity constants such as k from being large or small. 1
Following are three examples of regular perturbation solutions of differential equations.
The frrst two are variations on the type of steady, one-dimensional problem in Section 3.2. In
both cases the ordinary differential equation encountered initially is nonlinear and has no ana-
lytical solution, but can be solved readily when a parameter is small. The third example
involves two dependent variables, each functions of time and position, which are governed by
coupled, nonlinear, partial differential equations. Remarkably, the perturbation methodology
leads to a straightforward solution, by creating a sequence of pseudosteady, linear problems.

EXAMPLE 4.3-1 Temperature-Dependent Thermal Conductivity The heated-wire problem in

Example 3.2-5 is modified here by assuming that the thermal conductivity and heating rate are both
functions of temperature. The energy conservation equation for the wire is then

1 d dT
r dr rk(T) dr + Hv (T) = 0. (4.3-1)

1
Another standard symbol for mathematical order is "o." If f(e) fen~ 0 as e ~ 0, then "f(e) = o(en)." The use
of "0" is preferred because it provides a closer bound on the limiting behavior. Whether one writes f = O[g( e)]
or f = o[g(e)], g(e) is termed the gauge function. Gauge functions are typically powers of e, as above, but in sin-
gular perturbation problems other functions sometimes arise; see Van Dyke (1964) for examples.
122 SOLUTION METHODS BASED ON SCALING CONCEPTS

Assuming that the temperature range is small enough to make the thermal properties linear functions
ofT, the conductivity and heating rate are expressed as
k(T) = k 00 [1+ a(T - T00 )] (4.3-2)
Hv(T) = Hoo[1 + a(T- Too)]. (4.3-3)

The dimensional parameters are Too, the ambient temperature; koo, the thermal conductivity at the
ambient temperature; Hoo, the volumetric heating rate at the ambient temperature; and a, a constant
that expresses the sensitivity of k and Hv to T. The local rate of heat generation is Hv = i 2 I K, where
i is the current density and K is the electrical conductivity of the wire. From Ohm's law, i = KVI L,
where Vis the voltage drop and Lis the length of the wire. Thus, Hv = KV 2I L 2. Because conduction
of heat and electricity in metals is by the same mechanism, it may be assumed that K ex k, which is
what leads to the same coefficient a in Eqs. (4.3-2) and (4.3-3).
To simplify the analysis, it is assumed that Bi = hRI koo >> 1, so that the surface temperature
2
is Too· Following Example 3.3-1, the scale for the temperature change is HooR Ikoo. Accordingly,
the dimensionless quantities are defined as

e= T-Too (4.3-4)
HooR 2lkoo '
The differential equations and boundary conditions become

1 d de
- 7](1 +ee) + 1 +ee=O (4.3-5)
1J d1J d1J
de
d1J (0) = 0, 8(1) = 0. (4.3-6)

Equation (4.3-5) is nonlinear because the coefficient of deld1J depends on e. It is this nonlinearity
that precludes solving the problem for unrestricted values of e. However, if the dependence of k and
Hv on Tis relatively weak, such that e << 1, the problem can be solved using regular perturbation.
The objective is to evaluate e(rJ) under those conditions.
Although e would be a constant for any given material, in a perturbation analysis it is advan-
tageous to view it as a variable and examine what happens as e ) 0. Accordingly, the temperature
is expressed as
00

e(7]) = L en(rJ)en (4.3-7)

n=O
as for a Taylor expansion about e = 0. The radial variations in temperature are described by the coef-
ficient functions, en(1J), which are independent of e. Only e 0 ( 17) and e 1(17) will be computed here,
but that will show how additional terms could be obtained. Substituting Eq. (4.3-7) into Eq. (4.3-5)
yields

1 d 2 deo del
- 7](1 + eeo + O(e )) +e + O(e2) + 1 + eeo + O(e2) = 0. (4.3-8)
1J d1J d1J d1J
Anticipating that only e0 and e 1 will be evaluated, all terms involving e 2 or higher powers are
recorded simply as "O(e2)." Expanding the boundary conditions in this manner gives

d8o (0) + e d8t(O) + O(e2) =0 (4.3-9)

d1J d1J
eo(1) + ee1(1) + O(e2) = 0. (4.3-10)

Equations (4.3-8) through (4.3-10) lead to a sequence of problems, the first of which is the
0(1) problem that governs e0 . The 0(1) problem, which may be thought of also as the O(e0 ) prob-
 4.3 Regular Perturbation Analysis 123

lem, is identified by ignoring all terms that involve positive powers of e. Given that e ) 0, such
terms are too small to matter at this frrst level of approximation. The 0( 1) problem is
1 d deo
- TJ +1=0 (4.3-11)
TJ dYJ dYJ
deo
dYJ (0) = 0, 8 0(1) = 0. (4.3-12)

Alternatively, Eqs. (4.3-11) and (4.3-12) could have been obtained just by setting e = 0 in the orig-
inal problem. These are the equations for constant k and Hv, and their solution is

(4.3-13)

This is the same as Eq. (3.2-49) with Bi >> 1, except with a factor 1/4 caused by a different choice
of temperature scale in defining e.
The O(e) problem that governs e 1(TJ) is obtained by selecting only those terms in Eqs. (4.3-8)
through (4.3-10) that are multiplied by e. The idea is that the 0(1) terms have already been satisfied,
whereas the O(e2) ones are still too small to matter. Note that in a homogeneous equation such as
Eq. (4.3-10), the zero on the right-hand side is interpreted as 0 + (O)e + (O)e2 .... Thus, the O(e) prob-
lem is

1 d deo del
- TJ eo + + eo = 0 (4.3-14)
TJ dYJ dYJ dYJ
del
dYJ (0) = 0, 8t(l) = 0. (4.3-15)

Equation (4.3-14) seems unpromising until it is recalled that e 0 ( 71) is known. That is, it is simply a
linear differential equation for e 1(TJ), with nonhomogeneous terms arising from eo(TJ). Using
Eq. (4.3-13) to evaluate the nonhomogeneous terms and then integrating, the solution is found to be

(4.3-16)

This function provides a frrst approximation to the effects of having a variable k and Hv.
Assembling the results from the 0( 1) and 0(e) problems, the first two terms of the expansion
fore are

(4.3-17)

The procedure used above could be extended indefinitely to find successively better approximations
to e. If this were done, it would be found that the governing equations at O(en) are linear for all
n > 1. This occurs in all regular perturbation analyses, whether or not the 0( 1) problem is linear, and
the automatic linearization greatly facilitates finding the higher-order solutions. In this example, not
only was the 0(1) problem linear also, but the simple form of its solution (a polynomial) helped
make the one at 0(e) simple as well. Thus, the success of a regular perturbation analysis hinges
mainly on whether the 0(1) problem can be solved, at least in a convenient form.
To summarize the methodology, once a scaled problem with a small parameter has been for-
mulated, four steps are needed in a regular perturbation analysis: (1) the power series involving e is
stated, thereby defining the coefficients; (2) the governing equations are expanded to the extent
needed to evaluate the desired number of coefficients; (3) the problems that govern the individual
coefficients are identified and solved, until enough coefficients are known; and (4) the coefficients
are substituted into the series in e to get the final result.
124 SOLUTION METHODS BASED ON SCALING CONCEPTS

EXAMPLE 4.3-2 Second-Order Homogeneous Reaction Suppose that the irreversible reaction
2A ) B takes place at steady state in a liquid film, at a rate given by RvA = -kC] = -2RvB· Species
A enters the liquid at x = 0, where its concentration is C0 , and there is an inert and impermeable sur-
face at x = L. It is desired to determine the rate of consumption of A, assuming that the reaction is
slow relative to diffusion (although not so slow as to yield CA = C0 throughout the film). Although
the differential equation for CA (x) is made nonlinear by the reaction rate law, for slow reactions this
can be solved as a perturbation problem.
With dimensionless quantities defined as 1J = xl L, e = CAl Co, and 8 = Da = kCoL 2IDA' the
reactant concentration is governed by

d@
@(0) = 1, dYJ (1) = 0 (4.3-18)

where 8 << 1. Following the standard procedure for a regular perturbation problem, the concentra-
tion is expanded as

(4.3-19)

To evaluate the first three coefficients, the differential equation is written as

(4.3-20)

and the expanded boundary conditions are

€Jo(O) + 8€J1(0) + 8 2 €J2(0) + 0(83 ) = 1 (4.3-21)

d8o (1) + e d8t (1) + e2d82 (1) + O(e3) = 0. (4.3-22)

d1J d1J d1J

Selecting the terms in Eqs. (4.3-20) through (4.3-22) that do not involve 8, the 0(1) problem
is found to be
d€Jo
@ 0(0) = 1, dYJ (1) = 0. (4.3-23)

This has the fairly trivial solution

(4.3-24)
which corresponds to an unreactive system. Picking out the terms multiplied by 8 and using @0 from
Eq. (4.3-24), the 0(8) problem is

@1(0) = 0, (4.3-25)

which has the solution

(4.3-26)
2
Finally, choosing the terms that involve only 8 and using Eqs. (4.3-24) and (4.3-26), the problem
for e2 is

(4.3-27)
 4.3 Regular Perturbation Analysis 125

The solution to this is

(4.3-28)

Assembling the information on the coefficients, the final result for the reactant concentration is
1]2 2 1]4
fJ(TJ) = 1+e - TJ +~ - 173 + 21] + O(s3). (4.3-29)
2 3 4
The rate of consumption of species A per unit area must equal its flux at x = 0, which is propor-
tional to

(4.3-30)

Returning to dimensional quantities, the flux of A at the open boundary is

DACo d@ 2
NAx(O) = - d (0) = kC~L 1 - -s + O(s2) . (4.3-31)
L TJ 3
The leading term is just the reaction rate (per unit area) based on the concentration Co at the bound-
ary, and the next one is a correction for the spatial variation in the reactant concentration. Because
the concentration of A declines with increasing distance into the liquid, the reaction rate is reduced,
as indicated by the smaller flux. For fast reactions, the flux is given by Eq. (3.2-76).

EXAMPLE 4.3-3 Reaction with an Immobile Substrate In this problem it is assumed that species
A enters a porous or fibrous material at x = 0, beginning at t = 0, and reacts there with B according
to A+ B ) C. The rate of the bimolecular reaction is given by RvA = RvB = -kCACB. Of impor-
tance, species B is immobile. For example, the medium might be an array of fibers with fluid-filled
interstices through which A can diffuse, and B might be a reactive group on the fibers. The product
C, which will not be considered, might be either a chemically altered fiber or a soluble species. It is
assumed that there is an impermeable and inert surface at x = L and that the initial concentration of
B in the porous material and the concentration of A at the open boundary are both C0 . It is desired
to predict CA(x,t) and CB(x,t) for a reaction that is relatively slow.
In terms of dimensional quantities, the problem formulation is

acA a2cA
=DA -kCACB (4.3-32)
at ax 2

CA (x,O) = 0, _a_c _A (L, t) = o (4.3-33)

at
acB
- - =-kCACB (4.3-34)
at
CB(x, 0) = C0 . (4.3-35)

The differential equations are coupled via the reaction rate term, the form of which makes the prob-
lem nonlinear. There is no a2CBfax 2 term in Eq. (4.3-34) because B does not diffuse, and thus there
are no boundary conditions for CB. Nonetheless, CB will depend on x as well as t.
2
There are two time scales embedded in this set of equations, a diffusion time (td = L / DA) and
a reaction time [tr = (kC0)- 1]. Their ratio is the Damkohler number, s = td/tr = kC0 L 2/DA. By
assumption, s << 1 or td << tr. There are three time regimes that might be analyzed. In the earliest,
where t << td, species A has just begun to diffuse into the porous medium, and there has not yet been
enough time for the reaction to become important. The solution to this penetration problem without
reaction is given in Example 4.2-1. Fort rv td, the boundary at x = L begins to have an effect, but
the reaction is still negligible. This transient diffusion problem in a finite domain, without reaction,
126 SOLUTION METHODS BASED ON SCALING CONCEPTS

can be solved using the methods in Chapter 5. In the third time regime, where t rv tn the reaction
finally becomes important and there will be noticeable consumption of B. Because the supply of A
is unlimited (its concentration at x = 0 remaining always at C0), all of the B eventually will disap-
pear. It is this last, long period that is to be analyzed.
Fort rv tr the scaled time variable is T = t/tr. With the other dimensionless variables defined
as TJ = xj L, (} = C A!C 0 , and cp = CB! C0 , the problem is reformulated as
ao a 2o
e - e(}cp (4.3-36)
aT aTJ2
ao
(}(TJ, 0) = 0, 0(0,T) = 1, -(1,T)=0 (4.3-37)
aTJ

acp = -Oc/J (4.3-38)

aT
c/J(TJ,O) = 1. (4.3-39)

The small parameter (e) does not appear in Eq. (4.3-38) because the only time scale that affects
CB is tr-
To solve this as a perturbation problem, both dependent variables must be expanded in e. The
series are of the form

O(TJ, T) = Oo(TJ, T) + e0 1(TJ, T) + O(e2) (4.3-40)

c/J(TJ,T) = c/Jo(TJ,T) + ecfJI(TJ,T) + O(e2). (4.3-41)

A sequence of relatively simple problems is obtained by alternating between (} and c/J. Beginning
with (}, the expanded differential equation and boundary conditions are
aoo 2
a 0o 2
a o1
e + O(e2 ) = + e - eOoc/Jo + O(e2) (4.3-42)
aT aTJ 2 aTJ2
0 0 (0, T) + e0 1(0, T) + 0(e2) = 1 (4.3-43)
a00 a01
- (1,T)+e (1,T)+O(e2)=0. (4.3-44)
aTJ aTJ
The 0(1) problem is therefore
a2oo ao0
2 = 0, 00(0, T) = 1, - (1,T) = 0. (4.3-45)
aTJ aTJ
The solution to this pseudosteady problem is simply

Oo(TJ, T) =1 (4.3-46)

which corresponds to the uniform concentration that would exist for steady diffusion without reac-
tion. Because the analysis focuses on times such that t rv tr >> td, diffusion across the entire layer of
medium is effectively instantaneous. That is why the initial condition, 0(TJ, 0) = 0, will not enter into
the analysis.
Turning to cp, the expanded differential equation and initial condition are

acfJo acfJI 2 2
- + e- + O(e ) = -Ooc/Jo- e(Ooc/Jl + 01c/Jo) + O(e ) (4.3-47)
aT aT
2
c/Jo(TJ,O) + ec/Jl(TJ,O) + O(e ) = 1. (4.3-48)

The resulting 0(1) problem is

c/Jo(TJ, 0) = 1 (4.3-49)
 4.4 Singular Perturbation Analysis 127

which has the solution

c/Jo(TJ, T) = e-r. (4.3-50)
Withe a constant at this level of approximation, cp decays as if B were consumed by a first-order
reaction.
Returning to e, the differential equation for the 0(s) problem is
2
a eo a e1
- eocfJo. (4.3-51)
aT aTJ 2
Using e0 = 1 and c/Jo = e- 7
, the differential equation and boundary conditions are

(4.3-52)

Again, the problem is pseudosteady; this is true of the governing equations for all en· The solution
to Eq. (4.3-52) is

-r
e . (4.3-53)

Back again to cp, the 0(s) problem and its solution are
2
acpl +A.. - - TJ -2r
(4.3-54)
aT o/l - TJ 2 e '
2
c/Jl(TJ,T) = TJ- ~ (e-r- e-2r). (4.3-55)

If more coefficients were desired, this alternation could be continued indefinitely.

Assembling the results, the final expressions for the concentrations are
TJ2
e(TJ, T) = 1 - e TJ- e-7 + O(s2) (4.3-56)
2
2
c/J(TJ, T) = e-7 + e TJ- ~ (e- 7 - e- 27) + O(s2). (4.3-57)

Note from Eq. (4.3-56) that, although A is consumed by the reaction, e actually increases with time
during the period being modeled. The reason is that the reaction slows as B is consumed, allowing
diffusion to maintain concentrations of A in the interior that are ever closer to the fixed value at TJ = 0.
As indicated by Eq. (4.3-57), the concentration of B declines monotonically with time, there being
no mechanism to replenish that reactant.

4.4 SINGULAR PERTURBATION ANALYSIS

Regular perturbation analyses describe relatively small effects, in that a solution obtained
by setting 8 == 0 provides a good first approximation to one with 8 << 1. Singular pertur-
bation analyses can yield more surprising results. Indeed, the defining characteristic of a
singular problem is that the solution for 8 == 0 differs markedly from that for 8 ) 0. The
counterintuitive nature of such problems is at the root of certain paradoxes in fluid mechan-
ics that were encountered in the 19th century, as discussed later in this book. Boundary lay-
ers, thin regions in which there are large gradients in field variables such as velocity or
temperature, are ubiquitous in nature and in engineered systems. When a boundary layer
exists, the governing equations inevitably define a kind of singular perturbation problem.
Although a good physical understanding of boundary layers began to emerge in the early
1900s, largely explaining what had been so puzzling, methods for solving singular pertur-
bation problems were not well developed until the 1950s.
128 SOLUTION METHODS BASED ON SCALING CONCEPTS

As has been stressed, scaling of variables is essential in identifying valid approxima-

tions. Scaling has added importance in singular problems, in that they have at least one
independent variable with multiple scales. For example, in a boundary-layer problem a sys-
tem dimension and one or more boundary-layer thicknesses might each be important length
scales. Serious errors can stem from a failure to recognize this and properly scale the equa-
tions. With at least two scales being important, the approximations that apply for one range
of the independent variable (e.g., for certain positions) generally differ from those for
another range. This multiregion character causes the solutions to be localized. Whereas
regular perturbation solutions are uniformly valid everywhere, singular ones are not, and
local results generally must be woven together to describe the overall behavior. A princi-
pal method for doing this involves matched asymptotic expansions. Singular perturbation
with matched asymptotic expansions has been described in many texts on applied mathe-
matics. For information beyond the introduction given here, see Bender and Orszag (1978),
Hinch (1991), Holmes (1995), Kevorkian and Cole (1981), Lin and Segel (1974), Nayfeh
(1973), or Van Dyke (1964).
In the first example that follows, a singular perturbation analysis is applied to a steady
reaction-diffusion problem with concentration boundary layers. The second example is a
problem in chemical kinetics in which a small time interval takes the place of a thin region.
The final example, which revisits steady heat transfer in a fin, illustrates how singular per-
turbation may be used to combine a one-dimensional solution that is valid in one region
with a two-dimensional solution that applies in another.

EXAMPLE 4.4-1 Consecutive Reactions As shown in Fig. 4-4, suppose that species B is gener-
ated from a substrate S within a liquid film and then reacts with species A. The rate of formation of
B in the first reaction is constant at R 0 , as might occur if the reaction is enzymatic and S is present
in great excess. Species A is supplied at a boundary, thereby maintaining a steady state. The rate of
consumption of B in the second reaction is kCACB· For simplicity, the concentration of A is assumed
to vary linearly from 0 at x = 0 to C0 at x = L. This would occur if A were abundant enough to make
the effect of reaction on CA (x) negligible. The objective is to determine CB(x) when the second reac-
tion is arbitrarily fast.
With RvB = Ro- kCACB, the governing equations for CB(x) are
d2CB
DB + Ro- kCACB = 0 (4.4-1)
dx 2
dC dC
_ B (0) = 0 = B (L). (4.4-2)
dx dx

Liquid
ICA=O
s~ a

dCtfdX = 0 dCtfdx = 0
A+B -+ C

X=O X= L
Figure 4-4. Consecutive, irreversible reactions in a liquid film. Species B is formed continuously from a sub-
strateS and reacts with A to form the product C. Species A is supplied at x =Land leaves at x = 0, whereas B
is confined to the film.
 4.4 Singular Perturbation Analysis 129

Initially, TJ = xj L will be chosen as the dimensionless coordinate. The dimensionless concentration

is cp = CB/C;, where c; is to be determined. That concentration scale must be inferred from the dif-
ferential equation because no information on it is gained from the no-flux boundary conditions. With
c; not yet known, Eq. (4.4-1) is rewritten as
d 2cp R L2
-- + 0 (4.4.3)
dYJ 2 DBc;
where CA(TJ) = CoTJ has been used. Because the overall consumption of Bat steady state must bal-
ance its formation, both reactions are important. This suggests that c;
be chosen to make the terms
in parentheses in Eq. (4.4-3) equal, or c;
= R0 jkC0 . The Damkohler number for the bimolecular
reaction (the second term in parentheses) is Da = kC0 L 2/ DB· With Da >> 1 by assumption, the small
parameter for the perturbation analysis is chosen as 8 = Da- 1. The dimensionless problem is then
d2cp
8 - TJcfJ = -1 (4.4-4)
dYJ2

dc/J (0) = 0 = dc/J (1). (4.4-5)

dYJ dYJ
A solution of Eq. (4.4-4) that satisfies Eq. (4.4-5) is desired for arbitrarily fast reactions, or
8 ) 0. It is tempting to seek a first approximation by setting 8 = 0 in Eq. (4.4-4). If this is done, the
differential equation becomes an algebraic equation with the simple solution cp = 17- 1. However,
dc/J/ dYJ = -17-2 does not vanish at TJ = 0 or TJ = 1, so that neither boundary condition is satisfied. The
underlying problem is that setting 8 = 0 reduces Eq. (4.4-4) from a second-order differential equa-
tion to a "zeroth-order" differential equation. Any loss of order in a differential equation removes a
degree of freedom needed to satisfy a boundary condition. Having the highest order term in the dif-
ferential equation be multiplied by the small parameter is sufficient (although not necessary) to make
a problem singular.
From a scaling perspective, the mistake in the simple solution was the implicit assumption that
8d2cpj d17 2 = 0(8) for all TJ or that d 2cpj d17 2 = 0(1) everywhere. That is what would allow the diffu-
sion term to be neglected in comparison with the reaction terms, both of which are 0(1). The failure
of the proposed solution is evidence that this assumption is erroneous, at least for certain ranges of
TJ. The inability of the simple solution to satisfy either boundary condition suggests that where the
assumed scaling breaks down is near the boundaries. In particular, it indicates that 8d 2cpj d17 2 = 0(1)
near the boundaries, making the diffusion term comparable to the reaction terms. In other words, it
suggests that d 2cpfdTJ 2 = 0(8- 1) near TJ = 0 and TJ = 1. If ilcp rv 1 and if mathematical order and
order of magnitude are similar, then il TJ rv 8 112 . In other words, there are thin, dynamically distinct
regions near TJ = 0 and TJ = 1, which are concentration boundary layers. More generally, a concen-
tration boundary layer is any region in which concentration variations are rapid enough to make the
second derivative (the diffusion term) important, although it is negligible elsewhere.
The remainder of this analysis will focus on the "interior region" and the boundary layer at
TJ = 1; the boundary layer at TJ = 0 is the subject of Problem 4-16. Separate series solutions will be
developed for the two regions, and then the solutions will be combined via asymptotic matching.
Because the boundary layers are thin, L must be the correct length scale for the interior region
and...., there is no apparent need to modify the interior variables. The interior concentration is denoted
as c/J(TJ). As with a regular perturbation problem, it is expanded in 8 as
(4.4-6)

Equation (4.4-4) then becomes

2-
d c/Jo 2 - - 2
8 2
+ 0(8 ) - YJ[c/Jo + 8cp1 + 0(8 )] = -1. (4.4-7)
dYJ
130 SOLUTION METHODS BASED ON SCALING CONCEPTS

The boundary conditions are not considered yet, because they are not applicable to the interior region.
The 0(1) solution to Eq. (4.4-7) is what was obtained already,

- 1
c/Jo(TJ) = - . (4.4-8)
TJ
-
This indicates that cp rv
-
1 in most of the interior and thus confirms that the scaling of cp is satisfac-
tory (except near TJ = 0). The O(e) solution is found to be

(4.4-9)

so that the interior solution through 0(e) is

- 1 2e
c/J(TJ) = - + 4 + O(e2). (4.4-10)
TJ TJ
Additional terms can be obtained easily if desired, because
2-
- 1 d c/Jn
c/Jn+1(TJ) = - d 2 (n > 0). (4.4-11)
TJ TJ
However, the two terms in Eq. (4.4-10) will suffice.
For the boundary at TJ = 1, the distance into the medium is dTJ = 1 - TJ. From the order-of-
magnitude-analysis, the thickness of the boundary layer is dTJ rv e 112. Accordingly, following the
examples in Section 3.3, a suitably scaled coordinate is

y = (1 - TJ)e-1/2. (4.4-12)

That is, the small distance 1 - TJ is stretched by the large factor e- 112 to make y rv 1 in the bound-
ary layer. Changing coordinates in Eq. (4.4-4) and denoting the concentration in the boundary layer
as c/J(Y), the differential equation for that region is
2-
d cp - (1 - e1/2Y)c/J = -1. (4.4-13)
dY 2
Because the small parameter appears now only as e 112 , the boundary layer concentration is expanded
in half-integer powers, such that

(4.4-14)

With this assumed form for the expansion, the differential equation and boundary condition become

2-
d c/J1 - -
2 - c/J1 + Yc/Jo +e (4.4-15)
dY

(4.4-16)

The only boundary condition applicable here is that at Y = 0, corresponding to TJ = 1.

To obtain a boundary-layer solution that is correct through O(e), three coefficients in the series
must be evaluated. Each is governed by a second-order differential equation that has a solution involv-
ing two constants. One constant is found as usual from the boundary condition, and the other is deter-
mined by asymptotic matching of the boundary layer and interior solutions, as will be explained. The
 4.4 Singular Perturbation Analysis 131

0(1) problem obtained from Eqs. (4.4-15) and (4.4-16) is

2-
d 4>o - dc/Jo
-ct>o=-1, dY (0)=0. (4.4-17)
dY 2
The solution that satisfies the boundary condition is

cp0(Y) = 1 + A cosh Y (4.4-18)

where A is a constant that must be determined by the matching procedure.

In a boundary-layer problem, there is not a specific location at which the concentration and
flux in a given region can be equated with those in an adjacent region. Rather, the regions overlap,
in the sense that either differential equation will be reasonably accurate over some common range of
positions. Although in the present problem the interior solution worsens as the boundary at Y = 0 or
TJ = 1 is approached, and the boundary layer approximation deteriorates as Y becomes large, the
changes are gradual and there is a "gray area" in which either might be applied. In asymptotic match-
ing the overlap is expressed as
.....,
lim c/J(TJ) = lim c/J(Y), (4.4-19)
77~1 y~oo

The idea that 'rJ can be nearly 1 (or 1 - 'rJ nearly 0) at the same time that Y is nearly infinite may seem
contradictory. Recall, however, that in defining Y the small quantity 1 - TJ was multiplied by the fac-
tor s- 112 , which becomes arbitrarily large ass ) 0. Thus, if sis small enough, there will indeed be
a range of positions for which 1 - TJ is very small and Y very large simultaneously. The existence of
dual length scales is what permits this, with "small" according to one metric being "large" accord-
ing to another.
In implementing the matching requirement in Eq. (4.4-19), the interior and boundary layer
solutions must be expressed in terms of a common coordinate, so that they can be compared. In what
follows, both are expressed in terms of Y. Converting Eq. (4.4-10) to Yusing 'rJ = 1 - s 112Y, the inte-
rior solution is rewritten as

(4.4-20)

There are two terms here of the form ( 1 - x )-n, with x << 1 and n a positive integer. To eliminate s
from the denominators, they are expanded using
n(n + 1)
(1 - x)-n = 1 + nx + x 2 + O(x 3). (4.4-21)
2
Retaining terms only through O(s), the result is

(4.4-22)

Note that changing the coordinate from TJ toY converted the original series with integer powers of s
into one involving half-integer powers. Equation (4.4-22) is the form of the interior solution that will
be used in the asymptotic matching.
From Eq. (4.4-18), for large Y the 0(1) boundary-layer solution has the form

lim c{J(f) = 1 + A ey + O(e 112 ). (4.4-23)

y~oo 2

Because the ey term has no counterpart in Eq. (4.4-22), it is concluded that A= 0 and

c/Jo(Y) = 1. (4.4-24)

Thus, the boundary-layer concentration at 0(1) is simply a constant, which matches what is found
by evaluating the 0( 1) interior solution at TJ = 1.
132 SOLUTION METHODS BASED ON SCALING CONCEPTS

Moving now to the O(e 112) contribution to the boundary layer solution, the differential equa-
tion and boundary condition are
2-
d c/Jl - - dcpl
2
- c/11 = -Yc/Jo = -Y, dY (0) = 0. (4.4-25)
dY
The solution that satisfies the boundary condition is
c/J1(Y) = (1 + B)e-Y +Bey+ Y (4.4-26)
where B is the next constant to be determined from the matching requirement. Adding this term to
the series, the limiting form of the boundary-layer solution is now
lim cp (Y) = 1 + e 112 (Bey+ Y) + O(e). (4.4-27)
')'~()()

The e-Y term is absent because it vanishes for large Y. Comparing this result with Eq. (4.4-22), the
ey term again is seen to have no counterpart in the interior solution. Accordingly, B = 0 and
c/Jl(Y) = y + e-Y. (4.4-28)

The comparison of Eqs. (4.4-27) and (4.4-22) also shows that theY terms match exactly, suggesting
that the boundary-layer solution is correct through O(e 112).
One more coefficient in the boundary-layer series will be evaluated. The O(e) problem for the
boundary layer is
- -Y
-Ycpl= -Y(Y+e ), : ; (0) = 0. (4.4-29)

Once the boundary condition is satisfied the solution is

c/J2(f) = : +C e-Y + Cey + Y 2 + 2 + : (f 2 + Y)e-Y (4.4-30)

where Cis another constant. The limiting form of the boundary-layer solution is now
lim cp(Y) = 1 + e 112 y + e(Cey + Y 2 + 2) + O(e 112). (4.4-31)
y~oo

Comparing once again with Eq. (4.4-22), it is seen that the interior and boundary-layer solutions will
match exactly through O(e) if C = 0. Thus, the O(e) coefficient for the boundary layer is

c/J2(f) = Y 2 + 2 + _!_ (f 2 + Y + l)e-Y (4.4-32)

4
and the final boundary-layer solution is

cp(Y) = 1 + e 112 (Y + e-Y) + e Y 2 + 2 + _!_

4 (Y 2 + Y + l)e-r] + O(e312 )
. (4.4-33)

Solutions that are valid through O(e) have been obtained now for the interior region and the
boundary layer. To plot the concentration profile or calculate quantities that depend on the concen-
tration (e.g., the local reaction rate), what is needed is an expression that spans both regions. A com-
posite solution that fills that need is constructed by adding the interior and boundary-layer results
and (to avoid double counting) subtracting the overlap solution from that sum. Thus,

(4.4-34)

Either expression may be used, because their equality is ensured by the asymptotic matching. Choos-
ing Y as the coordinate and using Eqs. (4.4-20) and (4.4-33), the composite expression is
1 1 1 8
cp(Y) = (1 - e 1 Y)- + 2e(1 - e 1 Y)- + e 1 e-Y + 4 (Y 2 + Y + 1)e-Y + O(e312) .
2 2 4 1 2 (4.4-35)

The interior, boundary-layer, and composite solutions are compared in Fig. 4-5 for e = 0.01. The
composite solution is nearly identical to the boundary-layer result for Y < 1 and mimics the interior
 4.4 Singular Perturbation Analysis 133

Figure 4-5. Concentration

profiles for a reactive inter-
2.2 mediate in a system with
consecutive, irreversible
2 - -- - - ~nterior reactions, as described in
- - - Boundary Layer Example 4.4-1. The dimen-
1.8 - - ComposUe
cp(Y) sionless concentration (4J) is
shown as a function of

1.4 - - dimensionless position (Y)

near one boundary, based on
the solution for the interior
1.:2 ,. region [Eq. (4.4-22)], the
----
~~

- - -
...
.-.
.

boundary-layer solution
1 -- I I
[Eq. (4.4-33)], or the com-
0 1 2 3 4 posite solution [Eq. (4.4-35)].
y The results are for s = 0.01.

result for y > 2. This confrrms that y rv 1 in the boundary layer or that the boundary layer thickness
(relative to L) is rvs 112 . This is consistent with the order-of-magnitude reasoning that led to
Eq. (4.4-12). It also indicates that Y = 2 is large enough to resemble Y = oo. The actual concentra-
tion at Y = 0 is 12% greater than that predicted by the initial approximation, namely cp = TJ- 1.

EXAMPLE 4.4-2 Quasi-Steady-State Approximation in Chemical Kinetics Suppose that the

homogeneous reactions
kl k2
A ;:::::=~ B -~ C
k_l

take place in a well-stirred batch reactor, with the intermediate (species B) being highly reactive. If
only species A is present initially (at concentration C0), the time-dependent concentrations will be
governed by
dCA
dt = -k1CA + k_ 1CB, CA(O) = c0 (4.4-36)

dCB
dt = k1CA- (k-1 + k2)CB, CB(O) = 0 (4.4-37)

dCc
dt = k2CB, Cc(O) = 0. (4.4-38)

However, if B is reactive enough that its concentration remains low, it is common in kinetic analysis
to assume that dCB! dt = 0. Neglecting the time derivative for a highly reactive intermediate is termed
the quasi-steady-state approximation, or QSSA. (It could also be called a pseudosteady approxima-
tion, but "quasi" is common usage in kinetics.) The QSSA reduces Eq. (4.4-37) to an algebraic equa-
tion, with the solution

(4.4-39)

After using this to eliminate CB' Eq. (4.4-36) is readily integrated to obtain C A(t). Evaluating CB(t).
using that result, it is straightforward to compute the rate of formation of the final product from
Eq. (4.4-38). The objective is to examine the limitations of this very useful approximation. That will
be done by comparing the QSSA result with a more exact solution, obtained from a singular pertur-
bation analysis. It will be shown that there are two very different time scales and that the QSSA is
valid only for the longer one.
134 SOLUTION METHODS BASED ON SCALING CONCEPTS

The dimensionless concentrations of A and B and dimensionless time are defined as

CA
O(T) = - (4.4-40)
Co
CB k-1 + k2
c/J(T) = C k (4.4-41)
0 1

(4.4-42)
The scale for CA is clearly C0 , the assumed scale for C B is based on the QSSA, and the assumed time
scale is based on the forward step of the first reaction. Equation (4.4-41) presumes that the QSSA is
not greatly in error, and the time scale in Eq. (4.4-42) reflects the rate of change in CA' but not nec-
essarily the rates for CB and Cc. The dimensionless parameters that arise are

k1
e= - - - (4.4-43)
k_1 + k2
k2
A= k . (4.4-44)
-1

The intermediate will be "highly reactive" if e is small, so that the objective is to find a solution for
e << 1 that can be compared with that found using the QSSA. Withe small, it is expected from Eq.
(4.4-41) that CB <<CA. The parameter A is not necessarily large or small.
With these definitions, Eqs. (4.4-36) and (4.4-37) become

dO
- =-(}+ 0(0) = 1 (4.4-45)
dT

dcp
edT= 0 - cp, c/J(O) =0 (4.4-46)

Equation (4.4-38) is not considered further, because Cc does not influence C A or CB and because cal-
culating Cc requires only a straightforward integration once CB (or cp) is known.
Adopting the QSSA would mean neglecting the time derivative in Eq. (4.4-46), with the result
that
(4.4-47)

Replacing the differential equation by this algebraic equation precludes satisfying the initial condi-
tion for B. This indicates that the QSSA fails for small T, when cp increases rapidly from 0 to approx-
imately 0. Because the right-hand side of Eq. (4.4-46) is ""'1 during this period, it appears that
dcpj dT'"'"' 1/e or that T '"'"'e. This motivates defining a new time variable for the initial period as
w = Tje. (4.4-48)
With time rescaled in this manner, the governing equations for small times are

dO
- =-eO+ 0(0) = 1 (4.4-49)
dw

c/J(O) = 0. (4.4-50)

A A

The...., concentrations
....,
for small times are denoted hereafter as O(w) and c/J(w), and those for large times
as O(T) and c/J(T).
 4.4 Singular Perturbation Analysis 135

For large times, the concentrations are expanded in integral powers of e as

O(T) = Oo(T) + 01(T)e + O(e2) (4.4-51)

$(T) = ~o(T) + ~1(T)e + O(e2 ) (4.4-52)

Only the first two terms in each series will be evaluated. Using these expansions, Eqs. (4.4-45) and
(4.4-46) become

(4.4-53)

(4.4-54)

The 0(1) problem at long times is therefore

,...,
d 80 -
-d-T = -Oo+ (4.4-55)
,..., ,...,
4>o = Oo. (4.4-56)
Comparing this with Eq. (4.4-47), it is evident that the QSSA is simply a first approximation to the
long-time behavior. The solution to Eqs. (4.4-55) and (4.4-56) is
80(T) = a 0e-Arl 0 + A)= ~o(T) (4.4-57)

where a0 is a constant to be determined by matching. The initial conditions are not applicable yet
because Eq. (4.4-57) is not valid for small times.
Inspection ofEqs. (4.4-53) and (4.4-54) reveals that all higher-order (n > 1) corrections in the
long-time solution are governed by

(4.4-58)

(4.4-59)

Thus, each pair of terms in the long-time expansions is governed

,...,
by one differential equation and
one algebraic equation. Using Eq. (4.4-57) to evaluate 4>0 and solving the problem for n = 1,
the long-time expansion through O(e) is found to be

,. ., =a e-ArI (1 + A)+ e a e-ArI (1 + A)+

8(T) aoA Te-ArI (1 + A) + O(e2) (4.4-60)
o 1 (1 + A)2

2(T) =a e-Arl (l + A) + e a e-Arl (l + A) + aoA 1+ T e-Arl (l +A) + O(e2) (4.4-61)

0 1
'fJ (1 + A) (1 + A)
where a 1 is another constant to be determined.
Considering now short times, the expansions are

O(w) = Oo(w) + 01(w)e + O(e2) (4.4-62)

~(w) = ~o(w) + ~1(w)e + O(e2) (4.4-63)

where again only two terms in each series will be evaluated. Expanding Eqs. (4.4-49) and (4.4-50),
the coefficient functions are governed by
A

d8 0 A

dw = O, 80 (0) = 1 (4.4-64)
136 SOLUTION METHODS BASED ON SCALING CONCEPTS

(4.4-65)

dcp 0 A A A

dw = Oo - c/Jo, c/Jo(O) =0 (4.4-66)

~n(O) = 0 (4.4-67)

where n > 1. In this case there are two differential equations at each level of approximation. Because
this set of equations is valid for small times, the initial conditions are relevant now. Solving the 0(1)
and O(e) problems gives
A e
O(w) = 1 - (1 + Aw- e-w) + O(e2) (4.4-68)
(1 +A)
A e
c/J(w) = 1- e-w- ( +A) [(1- A)(1- e-w) + w(A- e-w)] + O(e2). (4.4-69)
1
In this case there are no undetermined constants.
To complete the solution through O(e), it is necessary to evaluate the constants a0 and a 1 that
appear in the long-time solution. Because the same two constants are involved in both concentration
variables, the asymptotic matching may be done using either (} or cp. Choosing c/J, the matching
requirement is expressed as
A A

limc/J(T) = lim c/J(w). (4.4-70)

7~0 w~oo

Because the limitations of the QSSA will be apparent only at short times, w will be used as the com-
mon independent variable. The limiting behavior of cp is found by expanding the exponentials in -
Eq. (4.4-61) for small T. The result is

. . ., AT 2 AT 2
lim cp = a 0 1 - + 0(T ) + ea 1 1 - + \ + 0(T )
7~o 1 +A 1 ll

A T AT + O(T2) + O(e2).
+ea0 1 +-- 1- (4.4-71)
1+A 1+A 1+A
Changing tow using T =we, Eq. (4.4-71) reduces to
. . . ., aoA 2
hm cp = ao + e a1 - (w - 1) + O(e ). (4.4-72)
7~o 1 +A
A

Neglecting only the decaying exponentials in Eq. (4.4-69), the limiting behavior of cp is
A e
lim cp = 1- [1 + A(w - 1)] + O(e2). (4.4-73)
w~oo 1 +A
For these last two expressions to match, it is necessary that
1
a 0 = 1, a =- (4.4-74)
1 1 +A.
Using Eq. (4.4-73) for the overlap solution, the composite solution for cp is
(4.4-75)
....., A

where cp and cp are evaluated from Eqs. (4.4-61) and (4.4-69), respectively. This completes the solu-
tion through O(e).
 4.4 Singular Perturbation Analysis 137

1 ---- ·-
.------..,

-- ---- - -------
-.. ~

--. ------ ~ .......

..-- ~ -. ~
0.8
!±S4ooo,o, -- ._,
=

0.6
~ OS SA
0 .4 - --Long time
- - - -- Short time
0.2 - - Composite

0 ~~~~~~~~~~~~~~~~~~~--
0 2 4 5
(JJ

Figure 4-6. Time-dependent concentration of a reactive intermediate in a well-stirred batch reactor, as described
in Example 4.4-2. The dimensionless concentration(¢) is shown as a function of dimensionless time (w), based
on the quasi-steady-state approximation [QSSA, Eq. (4.4-57)], the solution for long times [Eq. (4.4-72)], the solu-
tion for short times [Eq. (4.4-69)], or the composite solution [Eq. (4.4-75)]. The results are fore= 0.1 and A= 2.

The various results for c/J(w) are compared in Fig. 4-6 for 8 = 0.1 and A= 2. The composite
solution interpolates smoothly between those for specific ranges of time. The solution obtained using
the QSSA is inaccurate for short times, as anticipated, but is very close to the composite result for
w > 3. In summary, this example demonstrates that the QSSA provides an approximation to the long-
time solution that is accurate at 0(1). The 0(8) correction that was derived gives the error involved
for small but nonzero 8. This singular perturbation methodology has been applied to several other
reaction schemes; for examples see Bowen et al. (1963) and Segel and Slemrod (1989).

EXAMPLE 4.4-3 Limitations of the Fin Approximation The objective is to analyze a somewhat
more general version of the heat-transfer problem in Example 3.4-2 (see Fig. 3-12). The temperature
at the base of the fin (z = 0) is assumed now to depend on the transverse coordinate, such that
To= T0 (x). In Example 3.4-2 it was mentioned that, when Bi << 1 and 'Y >> 1, the approximate tem-
perature profile will be accurate for most of the fin, but not necessarily near the base. The difficulty
is that a given function T0(x) may be inconsistent with the neglect of temperature variations in the x
direction. In other words, the fin approximation might not satisfy the boundary condition at z = 0.
As seen in the preceding examples, the failure of an approximate solution to satisfy a boundary con-
dition or initial condition is a hallmark of singular perturbation problems. By treating the part of the
fin near the base as a boundary layer, more rigorous bounds can be placed on the accuracy of the fin
approximation. The objective is to determine those bounds and provide a procedure for improving
the temperature profile obtained previously.
The dimensionless temperature and coordinates that will be used are

@= T-Too X
t = z.
TJ = W' (4.4-76)
To- Too ' w
This notation is the same as that in Example 3.4-2, except that e is now the local temperature, rather
than the cross-sectional average. The cross-sectional average temperature at the base is
w
1
T0 = W T0 (x)dx. (4.4-77)
0

For simplicity, it is assumed that T0 (x) is symmetric about x = 0.

138 SOLUTION METHODS BASED ON SCALING CONCEPTS

The small parameter here is the Biot number, so that e = Bi = hW/ k. The analysis is simpli-
fied somewhat by assuming that the fin is dynamically long (corresponding to A >> 1 in Example
3.4-2), in which case the aspect ratio ('Y = L/W) will not appear in the model. The two-dimensional
problem for e(TJ, ~) is
a2 e a2 e
-- + =0 (4.4-78)
aTJ2 a~2

To- Too
e(TJ,O) = f(TJ) ===---
To- Too
(4.4-79)

e(TJ, 00) = 0 (4.4-80)

ae
- (0,~) =0 (4.4-81)
aTJ

(4.4-82)

Outer solution. In singular perturbation parlance the region where the field variable changes
slowly is the outer region; a boundary layer is an inner region. The z coordinate appropriate for the
outer region (most of the fin) was identified previously as
Z = e 1 12~. (4.4-83)
....,
Denoting the outer solution as e(TJ, Z), the governing equations for the outer problem are
a2€J a2€J
- 2 + 2 =0 (4.4-84)
aTJ az
....,
e(TJ, oo) = 0 (4.4-85)

(4.4-86)
....,
ae ..,
- (1,Z) = -ee(1,Z). (4.4-87)
aTJ
The boundary condition at the base has been omitted because it does not apply to the outer problem.
The outer expansion is assumed to be of the form
(4.4-88)
....,
The governing equations for the coefficients en are then
2
ae 2
a en 2
a en-1
- - = o· n> 1 (4.4-89)
aTJ2 ' aTJ2 az 2 '
en(TJ, oo) = 0, n> 0 (4.4-90)
....,
aen
- (0, Z) = 0, n> 0 (4.4-91)
aTJ

n > 1. (4.4-92)
....,
The differential equation and boundary conditions for the 0(1) problem indicate that e0 is
independent of TJ. Accordingly, the solution is written as
(4.4-93)
 4.4 Singular Perturbation Analysis 139

where A 0(Z) must be determined. Equation (4.4-90) indicates that A 0 (oo) = 0, but that is all that is
known yet about this function. Proceeding to the O(e) problem, it is found that

(4.4-94)

-
where A 1(Z) is another unknown function. However, if this result for @ 1 is differentiated with respect
to 'YJ and that derivative used in Eq. (4.4-92) with n = 1, a differential equation for A0(Z) is obtained.
That equation is
d2Ao
2 = Ao. (4.4-95)
dZ
The solution that satisfies A0( oo) = 0 is
A 0(Z) = C0 e-z (4.4-96)
where C0 is a constant that will be determined by matching with the inner solution. Equations
(4.4-90), (4.4-94), and (4.4-96) indicate now that A 1(oo) = 0. Proceeding to O(e), Eq. (4.4-89) with
n = 2 gives

(4.4-97)

where A 2(Z) is another unknown function. Equation (4.4-92) with n = 2 now provides the differen-
tial equation for A 1, which is

d2A1 -Co -z
dz2 - A1- 3 e . (4.4-98)

The solution that satisfies A 1( oo) = 0 is

(4.4-99)

where C 1 is another constant to be determined from matching. The outer solution through O(e) is

(4.4-100)

The outer solution, together with the matching requirement, can be used to anticipate the form
A

of the inner (boundary-layer) solution, which is denoted as @('YJ, ~).Noting from Eq. (4.4-83) that,
in the overlap region, large values of ~ will correspond to small values of Z, the asymptotic match-
ing is expressed as
(4.4-101)

Evaluating the limit of the outer solution by expanding the exponential for small Z gives

- z2 'YJ2 z
lim e = 1- Z + + O(Z3) C0 + e -Co + C1 + Co- (4.4-102)
z~o 2 2 6

Upon changing variables from Z to~ (i.e., adopting~ as the common coordinate), this becomes

(4.4-103)

which indicates that the inner expansion must contain half-integer powers of e.
140 SOLUTION METHODS BASED ON SCALING CONCEPTS

Inner solution and matching. The scaled variables for the inner problem are those in the orig-
inal problem statement. Based on Eq. (4.4-103), the inner expansion is assumed to be of the form

e(TJ,~) = eo(TJ,~) + e 1 /2 @ 1 (TJ,~) + ee2(TJ,~) + O(e312). (4.4-104)

Substituting this into the original problem statement gives

n>O (4.4-105)

A A

fJo(TJ, 0) = f(TJ); en (TJ, 0) = 0, n>1 (4.4-106)

(4.4-107)
A

aeo
aTJ ( 1, () = o; n > 2. (4.4-108)

The solution to the 0(1) inner problem, obtained using the methods of Chapter 5, is
00

eo (TJ, ~) = 1 + bo~ + V2 L [(!n- bn)e-n7T( + bnen7T(] cos n7TTJ (4.4-109)

n=l
1

fn = V2 f(TJ)COS n7TYJdTJ, n>1 (4.4-110)

0

where the bn are constants that remain to be determined. The O(e 112) inner solution is simply

(4.4-111)
where a is another constant to be found from matching. To obtain the correct behavior for ~ ) oo, the
en1r( terms in Eq. (4.4-109) must be absent; compare with Eq. (4.4-103). Accordingly, bn = 0 for all
n > 0, and
(4.4-112)

A comparison of Eqs. (4.4-112) and (4.4-103) indicates now that C0 = 1, b0 = 0, and a= -1. An
additional term in the inner expansion would be needed to find the remaining constant in the outer
solution, C1.
With the constants evaluated as just described, the inner and outer solutions are

(4.4-113)

00

&J(TJ,~) = 1+ V2 L fne-n7T( cos n7TTJ- e 1 12~ + O(e). (4.4-114)

n=l

As expected, the leading term in the outer solution (e- 2 ) is identical to the result from the fin approx-
imation; see the discussion following Eq. (3.4-27). Moreover, Eq. (4.4-113) shows that the error in
the fin approximation is O(e) in the outer region. The error in the inner region depends on the tem-
perature profile that is specified at the base. If To is constant, corresponding to f(TJ) = 1, it may be
confirmed from Eq. (4.4-110) thatfn = 0 for all n > 1. In that case, the infinite series in Eq. (4.4-114)
vanishes and the fin approximation is as accurate in the inner region as in the outer one; that is, the
inner temperature equals e-z through O(e 112). However, if To is not constant, the infinite series does
not vanish and the errors in the fin approximation in the inner region are 0(1).
 References 141

The scaling used for the inner region assumes that~"" 1 there, or that the inner region has a dimen-
sional thickness of ""W. This may be checked by examining the infinite series in Eq. (4.4-114)
because that is the largest term that distinguishes the inner and outer solutions. The most slowly
decaying term in the sum, e-'TTC, is already quite small for ~ = 1, which is consistent with the scal-
ing assumption. This illustrates a point mentioned in Section 3.4, which is that edge effects in con-
duction and diffusion are important over only about one thickness of the region being modeled. That
is, deviations from the fin approximation near the base are a type of edge effect.

References
Abramowitz, M. and I. A. Stegun. Handbook of Mathematical Functions. U.S. Department of Com-
merce, National Bureau of Standards, Washington, DC, 1970.
Arnold, J. H. Studies in diffusion: III. Unsteady-state vaporization and absorption. Trans. AIChE 40:
361-378, 1944.
Barenblatt, G. I. Scaling, Self-Similarity, and Intermediate Asymptotics. Cambridge University Press,
Cambridge, UK, 1996.
Bender, C. M. and S. A. Orszag. Advanced Mathematical Methods for Scientists and Engineers.
McGraw-Hill, New York, 1978.
Bowen, J. R., A. Acrivos, and A. K. Oppenheim. Singular perturbation refinement to quasi-steady
state approximation in chemical kinetics. Chern. Eng. Sci. 18: 177-188, 1963.
Dresner, L. Similarity Solutions of Nonlinear Partial Differential Equations. Pitman, Boston, 1983.
Hansen, A. G. Similarity Analyses of Boundary Value Problems in Engineering. Prentice Hall, Engle-
wood Cliffs, NJ, 1964.
Hinch, E. J. Perturbation Methods. Cambridge University Press, Cambridge, UK, 1991.
Holmes, M. H. Introduction to Perturbation Methods. Springer-Verlag, New York, 1995.
Kamke, E. Differentialgleichungen, Vol. 1. Akademische Verlagsgesellschaft Becker & Erler Kom.-
Ges., Leipzig, Germany, 1943, p. 475.
Kevorkian, J. and J.D. Cole. Perturbation Methods in Applied Mathematics. Springer-Verlag, New
York, 1981.
Lin, C. C. and L. A. Segel. Mathematics Applied to Deterministic Problems in the Natural Sciences.
Macmillan, New York, 1974.
Nayfeh, A. H. Perturbation Methods. Wiley-Interscience, New York, 1973.
Segel, L.A. and M. Slemrod. The quasi-steady-state assumption: A case study in perturbation. SIAM
Rev. 31: 446-477, 1989.
Van Dyke, M. Perturbation Methods in Fluid Mechanics. Academic Press, New York, 1964.

Problems
4-1. Transient Conduction for Two Plates in Contact
Suppose that two plates of indefinite thickness, initially at different temperatures, are brought
into contact at t = 0. Plate 1, with thermal properties a 1 and kb is at x < 0 and has initial tempera-
ture T1; plate 2, with properties a 2 and k 2 , is at x > 0 and has initial temperature T2 .
Determine the temperatures in both plates, expressed as @ = (T- T2)j(T1 - T2). Is the tem-
perature at x = 0 constant or does it vary with time? (Hint: It will be helpful to use similarity vari-
ables of the form TJi = xj[2(ait) 112 ] .)
142 SOLUTION METHODS BASED ON SCALING CONCEPTS

4-2. Velocity of a Solidification Front

Suppose that a pure liquid, initially at temperature Too, occupies the space x > 0. At t = 0 the
temperature at x = 0 is lowered to T0 , which is below the freezing temperature TF. Solidification will
begin, with the solid-liquid interface advancing in the x direction. As shown in Fig. P4-2, the inter-
face is at x = B(t). Assume that the liquid and solid have the same density, but different thermal con-
A

ductivities (kL and ks, respectively). The latent heat of fusion is A. In this situation B(t) may be
determined exactly, without invoking a pseudosteady approximation.

(a) State the differential equations and boundary conditions governing the solid and liquid tem-
peratures, Ts(x, t) and TL(x, t), respectively.
(b) Taking into account the fixed temperatures at x = 0 and x ) oo, but ignoring for the
moment the interfacial conditions at x = B(t), show that the solid and liquid temperatures
may each be expressed in terms of error functions. (Hint: Two similarity variables are use-
ful, namely TJi = xj[2(ait) 112], where i = S or L.)
(c) Requiring that T = TF at the solid-liquid interface, show that B(t) ex t 112. Let
12
B(t) = 2'Y(ast) 1
where 'Y is a constant that remains to be determined.
(d) Use the interfacial energy balance to show that 'Y is given implicitly by

e/'2 kLalf2(Too- TF)el'2asfaL

erf 'Y ksa}!2(TF- T0)erfc['Y(asfaL) 112 ]

Complete the solution forTs (x, t) and TL(x, t) in terms of 'Y·

(e) Consider now the special case where Too = TF. Show that a pseudosteady analysis involv-
ing Ts(x, t) gives
2ks(TF- T0)t 112
D(t) = A •

Ap
Comparing this with the exact results of parts (c) and (d), show that the pseudosteady and
exact expressions for B(t) coincide if Cps(TF- To)/A << 1. (Hint: erf(x) ) 2x/~ as
x ) 0.) Use an order-of-magnitude analysis of the energy equation and boundary conditions
to explain why this condition makes the problem pseudosteady.

Solid Liquid

T5 (x,t) Too
•

X= 0 X= 8(t)

Figure P4-2. Moving boundary in a solidification process, in which 8 increases with time. The temperatures at
x = 0 and x = oo are To and Too, respectively.
 Problems 143

4-3. Transient Conduction with a Constant-Flux Boundary Condition

Suppose that a solid of indefinite thickness is subjected to a constant heat flux q0 at x = 0,
beginning at t = 0. It occupies the space x > 0 and its initial temperature is uniform at Too·
(a) State the equations that govern T(x, t), and use order-of-magnitude reasoning to infer how
the temperature at x = 0 varies with time. What variable transformations does this suggest
when seeking a similarity solution?
(b) Use the similarity method to determine T(x, t). [Hint: The solution involves an integral of
the complementary error function, denoted as ierfc and defined as
00

ierfc(x) = erfc(s)ds.
X

The properties of this and other integrals of erfc are described in Abramowitz and Stegun
(1970, p. 299). This integral error function can be evaluated from erfc(x) as
-x2
e
ierfc(x) = -xerfc(x) + .y.;·l
(c) How does T(O, t) from the similarity solution in part (b) compare with the order-of-
magnitude estimate in part (a)?

4-4. Similarity Solutions for Time-Dependent Boundary Concentrations

Consider a transient diffusion problem like that in Example 4.2-1, but with the concentration
at x = 0 a given function of time, C0(t). Using C0(t) as a time-dependent concentration scale, setting
'YJ = xj g(t) as before, and supposing that K = 1, a solution could be sought of the form
C(x, t)
8('ry) = Co(t) .

(a) Show that this will work only if C0(t) = a{Y, where a and 'Yare constants. With C0(t) in this
form and 'YJ = xj[2(Dt) 112 ], show that the governing equations become
d 2() d()
2
+ 2'Y] d - 4'Y() = 0, 0(0) = 1, ()(oo) = 0.
d'Y] 'YJ
(b) If the dependent variable is changed, the problem in part (a) can be solved in terms of known
2
functions. As given in Kamke (1943), show that the transformation () = 'YJ-ll e-71 12w('Y]2)
2
yields
2
d 2w 1 K ((1/4)- J.L )
--+ --+-+ w=O
dz2 4 z z2
2
where z = 'YJ , K = -['Y + (1/4)], and J.L = 1/4. The differential equation for w(z) is called
Whittaker's equation, and its solutions (Whittaker functions) are expressible in terms of the
confluent hypergeometricfunctions M and U (Abramowitz and Stegun, 1970, p. 505). Using
this information, and assuming that 'Y > 0, the solute concentration and the flux at the
boundary are found to be
F(1 + 'Y) 2
8(71) = \;"; e-71 U[(1/2) + 'Y, (1/2), 'YJ 2 ]

F(1 + 'Y) D 1/2

N = - C0(t)
x x=O T[(1/2) + 'Y] t
where F(x) is the gamma function (Appendix B). Note that for 'Y = 1/2 the flux is inde-
pendent of time. For 'Y = 0 the concentration at the boundary is constant, and the results of
144 SOLUTION METHODS BASED ON SCALING CONCEPTS

Example 4.2-1 are recovered. This follows from the properties U(1/2, 1/2, 71 2) =
V7T exp(TJ2) erfc(TJ), T(1) = 1, and T(1/2) = V7T.

4-5. Transient Diffusion in a Permeable Thbe with Open Ends

The objective is to examine in more detail the effects of the capillary ends in the permeabil-
ity experiment described in Example 3.5-2. In that experiment (Fig. 3-16) an open-ended cylindrical
tube (isolated kidney capillary) of radius a and length 2L was immersed initially in a fluid with solute
concentration C0 . At t = 0 the concentration in the surrounding fluid was changed rapidly to a new
constant value, C 1 < C0 , causing outward diffusion both across the wall and through the open ends.
The membrane permeability is km.
(a) In Example 3.5-2 it was found that radial concentration variations within the tube can be
neglected. Show that the concentration in the lumen, C(z, t), is governed then by

ac
- =D
a2
c -
2km
(C- C1).
at az 2 R

(b) It was also pointed out in Example 3.5-2 that, fort<< L 2j D, the concentration in the cen-
tral part of the tube (near the observation windows) is independent of z. Show that the dimen-
sionless concentration there is given by

Sketch qualitatively the complete concentration profiles for several values of t, including
times both less than and greater than L 2/ D.
(c) For the early times discussed in part (b), use similarity to obtain a solution that includes posi-
tions near the ends. Choose z = 0 at one end and assume that C(O, t) = C 1 for all t > 0. For
the similarity method to work, how must the concentration variable be modified?

4-6. Diffusion from a Continuously Stretched Polymer Sheet*

After a certain elastic polymer is synthesized in sheet form, a small amount of volatile, unre-
acted monomer remains. It has been proposed to speed the removal of the monomer by stretching
the sheet at a constant velocity for a certain period of time. This would continuously reduce the thick-
ness (diffusion distance) and increase the area for mass transfer. The type of deformation to be
employed is a planar extension, as described in Problem 2-6. The coordinates, thickness, and stretch-
ing velocity are as defined there.

(a) State the partial differential equation and auxiliary conditions that govern the two-
dimensional monomer concentration, C(x,y,t). Assume that the initial concentration is uni-
form at C0 and that rapid mass transfer in the air makes C negligible at all exposed surfaces.
Interpret the position x = 0 in Fig. P2-6 as a symmetry plane, rather than an exposed sur-
face. Be sure to include the convective terms in the conservation equation, with Vx and vy
related to the rate of strain as in Problem 2-6(a).
(b) If the sheet length is sufficient to make the edge effect at x = L(t) unimportant (i.e., to per-
mit the boundary condition there to be ignored), show that a one-dimensional problem
results in which C = C(y,t) only. The diffusion problem can be simplified further, to one
with fixed boundaries, if the remaining independent variables are transformed as

*This problem was suggested by A. Chakraborty.

 Problems 145

y t
T= -
TJ = B(t)' f(t)

where B(t) is the half-thickness given in Problem 2-6(b) and f(t)is a suitable time scale. Find
the function f(t) that gives
ae a2 e
aT aTJ2

as the differential equation for the one-dimensional problem, where e = cI Co.

4-7. Time-Dependent Evaporation
Suppose that a pure liquid composed of species B is brought into contact with a stagnant gas
that initially contains only species A. Evaporation of the liquid begins at t = 0, leading to a mixture
of A and B in the gas. Assume that A does not enter the liquid. The gas-liquid interface is at z = 0,
with the gas at z > 0. The mole fraction of B in the gas at z = 0 is x 0 . The system is at constant tem-
perature and pressure and the gas mixture is ideal. This problem is identical to that in Example 3.5-3,
except that here xB reaches zero at z = oo rather than at a finite height. Also, it is assumed here that
there is an indefinitely large amount of liquid, so that the evaporation is never complete. The similar-
ity solution outlined below is from Arnold (1944).

(a) Show that the mole fraction of B in the gas, x B(Z, t), is governed by

XB(Z, 0) = 0, X B(O, t) = x 0,
(b) Using a similarity variable (TJ) analogous to that in Example 4.2-1, show that
d 2@ d@
+2(TJ-cp)d =0
2
dTJ TJ
@(0) = 1, @(oo) = 0
1 x0 d@
cp = - -2 1- x 0
-
dTJ
(0)

where @(TJ) = xBfx0. The constant cp cannot be determined until later [see part (d)];
although x 0 is given, the concentration gradient at the interface is not yet known.
(c) Solve the problem in part (b) to determine @(TJ).
(d) Show how to compute cp for a specified value of x 0 . [Hint: An implicit expression is
obtained, of the form cp = f( cp).]

4-8. Slow Bimolecular Reaction

This problem concerns the reaction A + B ) C in a liquid film at steady state. The system is
the same as in Problem 3-6, except that the reaction is assumed now to be relatively slow. As shown
in Fig. P3-6, species A is introduced at x = 0 and species B at x = L. The bimolecular rate constant
is k. For simplicity, assume that the diffusivities and boundary concentrations are equal, such that
DA = DB = D and CAO = CBL = Co.
(a) Show that the concentration of A is governed by
d28
= s8(8 + 2TJ - 1), 8(0) = 1, 8(1) = 0
2
dTJ
146 SOLUTION METHODS BASED ON SCALING CONCEPTS

where(}= C AfC0, TJ = xj L, and e = Da = kC0L 2/ D. [Hint: First combine the conservation

equations for A and Band solve for CB(x) in terms of CA(x).]
(b) Use a perturbation analysis to determine l3(TJ) for slow reaction kinetics, for which e << 1.
Include terms through O(e2); that is, find the first three terms in the expansion.

4-9. Slow Reaction with Michaelis-Menten Kinetics

Suppose that an enzyme that catalyzes the reaction A ) B is immobilized in a gel or porous
support that is shaped as a long cylinder of radius R. The reaction follows the rate expression

where k0 and Km are constants. The reactor is small enough that e = Da = k0 R 2 fDACo << 1, where
C0 is the concentration of A at the surface. The other dimensionless parameter, 'Y = Km/C0 , is not
necessarily small or large.
Use a regular perturbation scheme to determine the steady concentration profile for A and also
its flux at the surface. Evaluate three terms in the series for the concentration [i.e., ones through
O(e2)].

4-10. Autoxidation of Nitric Oxide in Cell Culture Medium

In the scaling analysis for macrophage cultures (Example 3.3-4) it was concluded that the NO
concentration is governed by Eq. (3.3-25). Although the differential equation is nonlinear, an analyt-
ical solution can be obtained if the depth of culture medium (L) is small enough to make Da2 << 1
and Da4 << 1, thereby permitting a perturbation analysis. To avoid having to expand the NO con-
centration in two separate small parameters, Da2 and Da4 can be replaced by two other dimension-
less groups, one (e) that approaches zero as L is reduced and another (a) that is independent of L.
Recalling that Da2 ex L and Da4 ex L 2 , a suitable depth-independent parameter is a = Da4 112/Da2 .
(a) Three possible choices for the small parameter are (i) e = Da4, (ii) e = Da2, and (iii)
e = aDa 2 . Compare the differential equations that result from each definition and discuss
their relative merits, taking into consideration the actual magnitude of a.
(b) Choosing definition (iii) from part (a), evaluate the first three nonzero terms in a regular per-
turbation expansion for(} A(TJ) fore<< 1. For simplicity, assume that 'Y = 0 (i.e., neglect the
effect of o2- synthesis).

4-11. Transport in a Membrane with Oscillatory Flow

Suppose that a membrane of thickness L that is in contact with aqueous solutions is subjected
to a pulsatile pressure on one side, leading to time-periodic flow across the membrane. The trans-
membrane velocity is given by

U(t) = Uo + U1 sin 1rwt

where U0 , Ub and w are constants. Consider convective and diffusive transport of a solute whose flux
is given by

N(x,t) = -D
ac + U(t)C
ax
where C(x, t) is the solute concentration in the membrane. The concentrations at the boundaries are
constant, with C(O, t) = C0 and C(L, t) = 0. Assume that the pressure pulses are relatively slow, such
that e = wL2/ D << 1. It is desired to investigate solute transport after many pressure pulses have
occurred (i.e., ignoring initial transients).
 Problems 14 7

(a) List all the potential time scales and identify an appropriately scaled time, T. Letting
@ = C/Co and X= xjL, state the equations governing @(X,T).
(b) Show that a regular perturbation problem is obtained for s << 1. Write the differential equa-
tions and boundary conditions governing the first two coefficients in the expansion.
(c) Solve for the first coefficient, @0 (X, T).

4-12. Absorption with a Nearly Irreversible Reaction

Assume that gaseous species A is absorbed in a stagnant liquid, resulting in the formation of
species B in the liquid via the reversible reaction A 2B. The reaction rate is given by
RvA = -kv(CA- KCg). (Note that the equilibrium constant K has units of inverse concentration.)
The reaction equilibrium strongly favors B, but not quite to the extent that the reaction can be con-
sidered irreversible. The system is at steady state and the liquid layer is indefinitely thick. The liq-
uid concentrations at the gas-liquid interface (x = 0) are CA(O) = C 0 and CB(O) = 0. For simplicity,
assume equal diffusivities, DA =DB= D.

(a) State the differential equations and boundary conditions governing the concentrations
@(TJ) =CA./Co and P(TJ) = CB/C0 . What is the length scale needed for converting x to the
dimensionless coordinate TJ?
(b) Assuming that s = KC0 << 1, use a perturbation analysis to calculate @(TJ) and P(TJ) and
find the flux of A into the film, NA(O). Determine the frrst two terms in the expansion for
the flux and discuss the effect of reaction reversibility. (Hint: Expand both concentration
variables as regular perturbation series.)

4-13. Deviation from Pseudosteady Diffusion in a Membrane

The system to be considered is the same as in Example 3.5-1, except that it is supposed now
that the volume of compartment 2 is so large that C 2 ---- 0 at all times. It is desired to predict C 1(t)
following the sudden addition of solute to compartment 1. In particular, the objective is to identify
the correction to a pseudosteady result that would arise if s = td/tp were not vanishingly small.

(a) For the dimensionless equations governing the solution concentration (c/J) and membrane
concentration (0) to be written as

dc/J ao
= (0, T), c/J(O) = 1
dT aTJ
ao a2o
e ()(TJ, 0) = 0, ()(0, T) = cp(T), ()(1, T) = 0
aT aTJ2'

how must TJ, T, cp, (), and s be defined? In particular, justify the definition ofT.

(b) Assume that for smalls the concentrations may be expanded as

2
c/J(T)= c/Jo(T) + scp1(T) + O(s )
2
O(TJ,T) = Oo(TJ,T) + s0 1(TJ,T) + O(s ).

Evaluate the frrst two terms in each series. Will the actual decay of the solution concentration C 1 be
slower or faster than what is implied by a pseudosteady analysis?

4-14. Heat Transfer in a Packed Bed Reactor

Suppose that a long, cylindrical reactor of radius R is packed with catalyst pellets. Gas flow
through the packed bed supplies a component that reacts exothermically at the surface of the pellets,
releasing heat at a constant volumetric rate Hv. The pellet surface area per unit volume is a and heat
transfer from the pellets to the gas is characterized by a heat transfer coefficient h. For simplicity, the
bulk gas temperature TG and reactor wall temperature Tw are each assumed to be constant. The pack-
ing has an effective thermal conductivity k.
148 SOLUTION METHODS BASED ON SCALING CONCEPTS

(a) Show that if axial conduction is neglected, a steady-state energy balance for the packing
leads to

where T(r) is the catalyst temperature. (Temperature variations at the level of a catalyst pel-
let have been neglected; the coordinate r refers to radial position in the reactor.)
(b) Convert the energy equation of part (a) to dimensionless form, using
ha(T- Tw)
()= .
H
Under what conditions will there be a thermal boundary layer, and where is it located? What
are the physical reasons for the boundary layer?
(c) For the boundary layer conditions of part (b), use a singular perturbation analysis to obtain
the first two terms of a series expansion for the temperature in each region, and construct a
composite solution.

4-15. Nonuniform Catalyst Distribution*

Suppose that a certain catalyst is prepared by immersing a sheet of porous material in a solu-
tion that contains the catalytic agent, thereby depositing it in the pores. The resulting catalyst con-
centration Cc(x) is found to be nonuniform. Assume that it is given by

Cc(x) = Cc0 [1 - (xj L)2 ]

where L is the sheet thickness and Ceo is a constant. The supported catalyst is to be used in a labo-
ratory experiment involving the irreversible reaction A ) B, which follows the rate expression

RvA = -kCcCA·
Entry of species A will occur only at one surface of the sheet, where the concentration will be main-
tained at CAo; the opposite surface will be placed next to an impermeable barrier. The reaction is fast
enough that Da = kCcoL 2/ D A >> 1.
(a) It is intended that the side with depleted catalyst (x = L) be placed next to the impermeable
barrier. For this arrangement, use a singular perturbation analysis to determine the steady-
state flux of A entering the sheet. Evaluate the first two terms in the expansion for the flux.
(b) Suppose that the sheet is accidentally reversed, so that the side with high catalyst concen-
tration is placed next to the barrier. Show that the first term in a perturbation expansion for
CA is governed by the Airy equation, which is
d 2W
2
= zW.
dz
The solutions to this equation, called Airy functions, are given in Abramowitz and Stegun
(1970, pp. 446-450).

4-16. Consecutive Reactions with One Unknown Concentration

The analysis of diffusion with consecutive reactions in Example 4.4-1 focused on the interior
region and one of the boundary layers. The objective is to complete the analysis by considering the
other boundary layer (that at x = 0 or TJ = 0). The starting point is the dimensionless formulation
given by Eqs. (4.4-4) and (4.4-5), in which e << 1.

* This problem was suggested by R. A. Brown.

 Problems 14 9

(a) Within this boundary layer TJ << 1 and cp >> 1. Thus, neither the coordinate nor the con-
centration variable is properly scaled here. If new variables are defined as

determine the values of m and n that will correctly scale X and l/J. (Hint: Based on the first
approximation to cp in the interior region, namely cp = 1/TJ, how must m and n be related?)
(b) Determine the 0( 1) term in a perturbation expansion for l/J(X), including asymptotic match-
ing with the solution for the interior region. [Hint: The solution involves the Airy functions
denoted as Ai(x), Bi(x), and Gi(x). The properties of these functions are summarized in
Abramowitz and Stegun (1970, pp. 446-450).]

4-17. Consecutive Reactions with Two Unknown Concentrations

This is a variation on the steady reaction-diffusion system analyzed in Example 4.4-1 and
Problem 4-16, in which the concentrations of both participants in the fast bimolecular reaction now
are unknown. As shown in Fig. P4-17, it is assumed that species B is generated at a constant rate R 0
from a substrate S in a liquid film. Species A enters the film continuously and reacts with B at a rate
given by RvA = -kCACB. Aside from the entry of A at x = 0 (where its concentration is maintained
at C 0), there are no fluxes at the boundaries. A convenient set of dimensionless variables and param-
eters is

RoL2 DB
a= e=
DACo ' kC0 L 2 .

The concentration scale for species B is explained in Example 4.4-1, and e once again is the inverse
of the Damkohler number for the consumption of B. The bimolecular reaction is assumed to be fast,
such that e << 1. The other Damkohler number (a) is neither large nor small.

(a) Show that the governing equations are

d2()
- a()cp = 0, ()(0) = 1,
dTJ 2
d2cp
e - ()cp = -1, dc/J (0) = 0 = dc/J (1).
dTJ2 dTJ dTJ

(b) Confirm that this is a singular problem by obtaining a solution for e = 0 and showing that
one of the four boundary conditions cannot be satisfied. Where is the boundary layer?

Liquid
dCA!dx= 0

dCafdx= 0
• dCafdx ·= 0

X=O x=L
Figure P4-17. A system with consecutive reactions, in which species B is generated continuously in a liquid
film from a substrate S and reacts there with A to form the product C. Species B is confined to the film and A
enters only at x = 0.
150 SOLUTION METHODS BASED ON SCALING CONCEPTS

Porous Catalyst

C=O
·--
u

X=O X= L
Figure P4-18. A porous catalyst with a second-order reaction and a uniform velocity U across the slab.

- -
(c) Let ()(TJ) and c/J(TJ) represent the concentrations in the interior region. Solve for the 0(1)
terms in both expansions. (Hint: Expand both concentrations as power series in e. There will
be one unknown constant to be determined later by matching.)
- -
(d) Let ()(Y) and cp(Y) represent the concentrations in the boundary layer. How should Y (the
scaled coordinate for the boundary layer) be defined? Determine the first two terms in each
boundary layer expansion, using asymptotic matching with the interior solution as neces-
sary. (Hint: Expand the concentrations using half-integer powers of e.)

4-18. Diffusion and Convection in a Porous Catalyst*

It is proposed to increase the rate of reaction in a porous catalyst by using flow through the
porous material to augment delivery of the reactant. As shown in Fig. P4-18, assume that the reac-
tion occurs in a slab of thickness L and that there is a uniform velocity U across the slab. The irre-
versible reaction follows second-order kinetics, such that Rv = -kC 2 for the reactant. The reactant
concentrations at the upstream (x = L) and downstream (x = 0) sides are Co and 0, respectively, the
flow being in the -x direction.
(a) Show that the dimensionless reactant concentration is governed by

1 d 2() Pe d()
--+
2
-()2 =0 ()(0) = 0, ()(1) = 1.
Da dTJ Da dTJ '
How must the Damkohler number (Da) and Peclet number (Pe) be defined?

(b) Suppose that the reaction is fast relative to diffusion, but not fast relative to convection. That
is, assume that Da >> 1 but Da/Pe rv 1. For e = Da- 1 ) 0 and a= Da/Pe = 0(1), this
becomes a singular perturbation problem. There is just one boundary layer. Where is it and
how should the coordinate by rescaled there?
(c) Find the first term in the expansion for() in each region. Apply asymptotic matching and
construct a composite solution.

*This problem was suggested by M. Z. Bazant.