1

MODEL 6
Diffusion and Reaction in Catalyst Pellets

Mathematical Aspect: Perturbation methods and asymptotic analysis

Diffusion and reaction in a spherical particle

Many chemical reactions of practical relevance take place in the presence of a solid catalyst.
The reactants are present in a fluid phase and the reaction cannot occur until they contact the
catalyst that is deposited on the solid surface. Examples of applications are: catalytic cracking of
petroleum fractions, desulphurization of fuels and ammonia synthesis. In most applications, the
catalyst impregnates a porous particle and the fluid contacts the particles in a packed bed or a
fluidized bed. Modeling of this type of reactors requires a quantification of reaction rates in the
particles. This will be the main objective of the present model.
Consider a spherical, porous catalyst particle where a catalytic reaction takes place. The
particle is surrounded by a fluid with a reactant concentration c0, assumed to be uniform. The
reaction takes place in the pores of the particle, which lowers the reactant concentration in the
fluid filling the pores, and thus creates a concentration gradient that drives reactant into the
particle by diffusion. At steady state, the rate at which the reactant is transported by diffusion
will equal the rate of consumption by chemical reaction. The species continuity equation that
represents this balance in spherical coordinates, assuming that the concentration of reactant in the
pores is only a function of radial position, c(r), is

D e d  2 dc 
0= r  − rc (1)
r 2 dr  dr 

where De is the effective diffusivity of reactant in the catalyst particle and rc is the rate of
catalytic consumption of reactant, expressed as moles consumed per unit time and unit particle
volume. Appropriate boundary conditions are

c=c0, r=R (2)

dc
= 0 , r=0 (3)
dr

where c0 is the reactant concentration in the fluid surrounding the particle, and R is the radius of
the particle. The solution of this problem for a specific rc(c) gives the steady-state concentration
profile in the particle. From a practical standpoint, the most important parameter for the
quantification of the process is the total rate of consumption of reactant in the particle, W
(defined as moles of reactant consumed in the whole particle per unit time.) This parameter can
be quantified in two different ways:
(1) By direct integration of the reaction rate over the whole volume of the particle,

R
W= ∫ rc dVp = 4π ∫ rc r 2 dr (4)
Vp 0
 2

(2) From the diffusive flux going into the particle, as obtained from Fick's law,

dc
W = 4πR 2 D e (5)
dr r = R

This global consumption rate is usually represented by a dimensionless parameter called the
effectiveness factor, defined as the ratio between the actual consumption rate given by either of
the previous two equations and the consumption rate that would be obtained if the reaction
occurred at a uniform concentration equal to the external concentration c0; i.e.

W
η= (6)
4 3
πR rc (c 0 )
3

Let us consider the case for which the kinetics of the catalytic reaction can be expressed as

rc = kc n (7)

We will formulate the problem in dimensionless form by defining

r
x= (8)
R

c
y= (9)
c0

With these definitions, equations (1) to (3) become, after manipulations,

1 d  2 dy 
x  − φ2 y n = 0 (10)
x 2 dx  dx 

y=1, x=1 (11)

dy
= 0 , x=0 (12)
dx

where

kc 0n −1
φ=R (13)
De

is called the Thiele modulus. We can see that the dimensionless solution of this problem, y=y(x)
depends only on φ.
 3

Substituting equations (5) and (7) into equation (6), and expressing the result in terms of
dimensionless variables leads to

3 dy
η= (14)
φ 2 dx x =1

Analytical solutions for the problem given by equations (10) to (12) are not available for
arbitrary reaction order n (although the problem can be solved numerically), but an analytical
solution exists for n=1. This solution is found by performing the change of variables y=u/x and
then solving for u(x). The solution is

1 sinh φx
y= , n=1 (15)
x sinh φ

from which the effectiveness factor calculated from equation (14) is

3
η= (φ coth φ − 1) , n=1 (16)
φ2

Dimensionless concentration profiles generated by equation (15) are shown in Figure 1 for
various values of the Thiele modulus. The Thiele modulus (equation 13) represents a ratio
between the reaction rate and the diffusion rate. At very low Thiele modulus, diffusion is much
faster than reaction and the concentration profile tends to be flatter, which means that diffusion
can deliver the reactant to the inside of the particle faster that it is consumed. At very high Thiele
modulus, reaction is much faster than diffusion, which means that the reactant will be consumed
completely in a region close to the surface of the particle and so the profiles experience a rapid
drop for x<1 (Figure 1).
It is common in this type of problem to find that diffusion and reaction rates have magnitudes
that are very different, so that either φ«1, or φ»1. Hence, it is of practical significance to find how
the solution (concentration profile and effectiveness factor) behaves for either large or small
values of the Thiele modulus. For n=1, equations (15) and (16) serve this purpose, of course, but,
for n≠1, we will see that numerical solutions need not be obtained, but rather analytical equations
to represent the solution can be found using asymptotic analysis.

Introduction to perturbation theory and asymptotic analysis

Perturbation methods are used in the solution of mathematical models that contain a
parameter (or a group of parameters) that are either large or small (generally, in dimensionless
terms, a parameter that is large or small compared to unity,) but that still affects the solution.
This type of problem is very common in transport phenomena. Specific examples are:
(1) Boundary layer problems, in which changes in the dependent variable (velocity,
temperature, concentration) occur over a length scale that is appreciably smaller than the length
scale over which the solution is sought (in this case, the ratio between the characteristic lengths
of the two scales is a small parameter).
(2) Low Reynolds number fluid mechanics.
 4

(3) Mass transfer problems in which a transport mechanism dominates over another (e.g.
convection over diffusion).
(4) Reaction problems, such as the one considered in the present model.

1
y
0.9 φ =1
0.8
0.7
0.6 2
0.5
0.4
0.3
0.2
5
0.1 10 20 50
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x

Figure 1. Dimensionless reactant concentration profiles in a spherical catalyst pellet for a

first-order reaction (n=1), and various values of the Thiele modulus.

Our objective is the mathematical representation of models whose formulation contains a

small parameter ε, such that ε→0 (large parameters will be called 1/ε). The solution that we
would like to find changes with ε. If f is the function to be evaluated, there are three possible
cases in the limit ε→0:

0

lim f (ε) = α ≠ 0 (17)
ε→0 ± ∞


In some applications, the knowledge of this limit would be enough, but here we are concerned
with applications in which we would like to characterize the behavior of the function as the limit
is approached. One way of doing this would be to expand the function f(ε) in a Taylor series
centered at ε=0,

ε2
f ( ε ) = f ( 0) + f ' ( 0) ε + f " ( 0) +L (18)
2
 5

The value of f(0) would be the limit given by equation (17). Equation (18) is only useful when
the function has a finite limit.
The representation given by a Taylor series is very specific, in the sense that the function is
expressed in terms of a power series of ε. It is possible that there exists a known simple function
δ0(ε) (not necessarily a power series) that represents the behavior of f(ε) as ε→0. Formally, this
would mean that

f ( ε)
lim = a0, a0≠0, |a0|<∞ (19)
ε → 0 δ 0 (ε )

The function δ0(ε) that satisfies these conditions is called a gauge function, and we say that the
function f(ε) is of order δ0(ε),

f (ε) = O[δ 0 (ε)] (20)

where the symbol "O" does not represent numerical order of magnitude (as it is its common
usage). For example, we know that

sin ε
lim =1 (21)
ε→0 ε

so we can say that

sin ε = O(ε) (22)

When this notation is used, it is implicitly assumed that ε→0.

Equation (19) allows us to use the gauge function as an approximation for f(ε) as ε→0:

f (ε ) ~ a 0 δ 0 ( ε ) (23)

This equation is an example of what we will call an asymptotic representation of the function
f(ε). How good an approximation this representation is depends on the value of ε. A better
representation can be obtained if there is a gauge function available for f (ε) − a 0 δ 0 (ε) ; that is, if
there is a function δ1(ε) that satisfies

f (ε ) − a 0 δ 0 (ε )
lim = a1, a1≠0, |a1|<∞ (24)
ε→0 δ1 (ε)

Then, we can improve the degree of approximation of equation (23) by using

f (ε) ~ a 0 δ 0 (ε) + a1δ1 (ε) (25)

This process can be continued to obtain an approximation in the form of a series,

 6

f (ε) ~ a 0 δ 0 (ε) + a1δ1 (ε) + a 2 δ 2 (ε) + L (26)

Each coefficient of the series is given by

m −1
f (ε ) − ∑ a n δ n (ε )
am= lim n =0 , am≠0, |am|<∞ (27)
ε→0 δ m ( ε)

The representation of f(ε) in equation (26) is called an asymptotic expansion or asymptotic

series.
Taylor series can be used to generate asymptotic expansions, and yield a particular form of
equation (26), in which the gauge functions are power functions,

δ n (ε ) = ε n (28)

For instance, the Taylor series expansion of sin ε around ε=0 is an asymptotic expansion:

ε3 ε 5 ε 7
sin ε ~ ε − + − +L (29)
3! 5! 7!

The Taylor series method can be used in more complex situations to generate asymptotic
expansions. For example, let us try to find an asymptotic expansion for the error function, erf(x),
at low values of its argument. The error function is defined by

2 x −t 2
erf ( x ) = ∫ e dt
π0
(30)

Let x=ε in this equation. Since we are interested in low values of ε (and hence of t), we can start
by representing the integrand in an asymptotic series. For this purpose, we use the following
Taylor series for the exponential function

y 2 y3
e− y ~ 1 − y + − +L (31)
2! 3!

or, letting y=t2,

t4 t6
e− t 2 ~ 1 − t 2 + − +L (32)
2! 3!

Substituting this equation into equation (30) and integrating term by term leads to
 7

2  ε3 ε5 
erf (ε) ~  ε − + + L (33)
π 3 10 

where the nth coefficient has the form

2 (−1) n ε 2n +1
an = (34)
π (2n + 1)n!

Comparisons between the asymptotic expansion (33) and the exact values of the error
function are shown in Figure 2. Note that, in principle, the asymptotic expansion is supposed to
represent the function for ε→0. However, the figure shows comparisons up to relatively high
values of ε. At the scale of the plot, we can see that even the first term is enough to represent the
function adequately for ε less than about 0.3, while adding more terms improves the
representation for still higher values of ε. Calculations show that taking 8 terms in the evaluation
of the series gives a maximum absolute error of 10-5 in the calculation of the function for ε<1. To
keep the same maximum error, 16 terms are needed for ε up to 2, 31 terms for ε up to 3, and 75
terms for ε up to 5. However, when many terms are needed, the use of this asymptotic series
becomes impractical. An interesting aspect of the approximation provided by the truncated series
is that it can fail catastrophically at high values of ε. For example, some of the approximations in
Figure 2 diverge from the solution at large ε, giving even unrealistic values for the function (e.g.
negative values are obtained for the 2 and 4 term solutions for ε<2. This is characteristic of
asymptotic series, and it indicates that one must be careful trying to extend their application
beyond the asymptotic limit.

erf(ε) 1
1 term 4 terms Exact
0.9
0.8 3 terms
0.7
2 terms
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.5 1 1.5 2
ε

Figure 2. Comparison of the exact plot of the error function with the asymptotic expansion (33)
using the indicated number of terms.
 8

The Taylor series method, as applied before, cannot be used to find an asymptotic expansion
for the error function at large values of its argument; i.e. suppose that we are interested in finding
an asymptotic representation for erf(x) when x→∞. To treat this case, it is convenient to rewrite
the definition of the error function (equation 30) as follows,

2 x −t2 2  ∞ − t 2 ∞ 
− t 2 dt 
erf ( x ) = ∫
π0
e dt = ∫
π  0
e dt − ∫ e

(35)
x 

The first term on the right-hand side of this equation is erf(∞)=1, so that

2 ∞ −t2
π x∫
erf ( x ) = 1 − e dt (36)

The reason for making this transformation is to have an integral in which the independent
variable is always a large number (note that in the remaining integral x<t<∞). Now, we
recognize that

de − t 2
e − t 2 dt =− (37)
2t

and equation (36) becomes

1 t = ∞ de − t 2
erf ( x ) = 1 + ∫
π t =x t
(38)

We now perform integration by parts to get

 ∞ ∞ 
1  e− t 2 e − t 2 dt 
erf ( x ) = 1 +  +∫ (39)
π t t 2 
 x x 

e− t 2
Since lim = 0 , we get
t →∞ t

1  e − x 2 ∞ e − t 2 dt 
erf ( x ) = 1 + − +∫ (40)
π  x
x
t2 


Successive integrations by parts lead to

 9

e− x 2  1 1× 3 1× 3 × 5 
erf ( x ) = 1 − 1 − 2 x 2 + (2 x 2 ) 2 − (2 x 2 )3 + L (41)
x π  

Letting x=1/ε, we find an asymptotic series for the error function corresponding to large values
of its argument:

εe − (1 / ε) 2  1 2 1 × 3 4 1× 3 × 5 6 
erf (1 / ε) = 1 − 1 − 2 ε + 2 2 ε − 2 3 ε + L (42)
π

The asymptotic series for low (equation 33) and high (equation 41 or 42) values of the
independent variable generally lead to a representation of the function in a wide range of the
domain. For example, Figure 3 shows how the two series can be combined to represent the error
function. Note that, using three terms in each series still leaves a somewhat small range of the
independent variable for which neither series is adequate.

1
erf(x) Equation (33) (3 terms)
0.9
0.8 Exact
0.7
0.6
0.5 Equation (41) (3 terms)

0.4
0.3
0.2
0.1
0
0 0.5 1 1.5 2 2.5 3
x

Figure 3. Use of asymptotic expansions for the low (equation 33) and high (equation 41) limits
of the independent variable to represent the error function.

The method of integration by parts is a useful alternative to Taylor series to generate

asymptotic expansions. Both methods are very specific in the context of asymptotic analysis in
the sense that their objective is to represent a known function when the independent variable is
either large or small. In many applications, the function to be represented is unknown, and the
small parameter appears in the formulation (algebraic or differential equation) that leads to
determining the function. In the next two sections we will study how to solve this type of
problems.
 10

Regular perturbation analysis

Regular perturbation analysis refers to the solution of problems that contain either a small (ε)
or large (1/ε) parameter by using asymptotic expansions. This type of analysis can be applied to
certain types of algebraic and differential equations. In general, regular perturbation analysis
consists of the following steps:
(1) Propose a set of gauge functions δ0, δ1, δ2…
(2) Postulate that the solution of the problem is an asymptotic series based on the set of gauge
functions proposed (as in equation 26).
(3) Substitute the asymptotic series into the problem equations and use equation (27) or
equivalent to find the coefficients.
To illustrate the application of this procedure to algebraic equations, consider the following
example,

x 2 − (3 + ε) x + 2 = 0 (43)

In this case, the two roots of this equation can be easily obtained. However, the procedure that
we will apply can be followed for polynomial equations of degree greater than 4 and other
algebraic equations that have no analytical solution. We will use gauge functions that are powers
of ε. The proposed asymptotic expansion is

x ~ x 0 + ε x1 + ε 2 x 2 + L (44)

To find the coefficients, xi, we substitute equation (44) into equation (43),

( x 0 + εx1 + ε 2 x 2 + L) 2 − (3 + ε)( x 0 + εx1 + ε 2 x 2 + L) + 2 = 0 (45)

To find the coefficients, we now equate the coefficients of the same power of ε in this equation.
Equating coefficients of O(ε0) leads to

x 02 − 3x 0 + 2 = 0 (46)

whose roots are

x 0+ = 2 , x 0− = 1 (47)

where the superindices denote the sign used in the evaluation of the root.
Equating coefficients of O(ε1) in equation (45) leads to

x0
2x 0 x1 − x 0 − 3x1 = 0 ⇒ x1 = (48)
2x 0 − 3

There are two solutions for x1 only because there are two solutions for x0. The corresponding
values are
 11

x1+ = 2 , x1− = −1 (49)

Equating coefficients of O(ε2) in equation (45) leads to

x1 − x12
2 x 0 x 2 + x12 − x1 − 3x 2 = 0 ⇒ x 2 = (50)
2x 0 − 3

which leads to

x +2 = −2 , x −2 = 2 (51)

Two asymptotic series (equation 44) are found,

x + ~ 2 + 2ε − 2ε 2 + L (52)

x − ~ 1 − ε + 2ε 2 + L (53)

These solutions (including only the first three terms) yield an approximation to the roots of
equation (43) with an error ≤0.5% with respect to the exact solution for ε≤0.1 (Figure 4).
Furthermore, it can be shown that equations (52) and (53) are identical to the Taylor series
expansion around ε=0 of the exact solution

4
x
3.5

3 +
x
2.5

1.5
-
x
1

0.5

0
0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90
ε
Figure 4. Comparison between the exact roots of equation (43) as obtained from equation
(54), and the three-term asymptotic series (52) and (53). For ε≤0.2, the asymptotic and exact
solutions are indistinguishable at the scale of the plot.
 12

3 + ε ± (3 + ε) 2 − 8
x= (54)
2

The asymptotic solution of equation (43) illustrates how the method of regular perturbation
works in the solution of an algebraic equation. The usefulness of the technique would be better
appreciated in solving a more complex algebraic equation, especially one without an exact
analytical solution. The technique can be applied also to differential equations, as illustrated in
the application that follows.

Regular perturbation analysis of differential equations: solution of model 6

Here we will consider the solution of model 6, represented by equations (10) to (12), for small
values of the Thiele modulus, φ. For convenience, let ε=φ2. Equations (10) to (12) can be restated
as

1 d  2 dy 
x  − εy n = 0 (55)
x 2 dx  dx 

y=1, x=1 (56)

dy
= 0 , x=0 (57)
dx

The solution of this problem yields the dimensionless concentration profiles that depend on ε
as only parameter in the system. Hence, we want to find y=y(x;ε), where we use a semi-colon to
separate parameters in the equations (in this case only ε) from independent variables (x). Once
again, we will use powers of ε as gauge functions so that the proposed asymptotic expansion is

y( x; ε) ~ y 0 ( x ) + εy1 ( x ) + ε 2 y 2 ( x ) + L (58)

Substituting this asymptotic expansion into equations (55) to (57) yields

1 d  2 d ( y 0 + εy1 + ε 2 y 2 + L) 
 − ε( y 0 + εy1 + ε y 2 + L) = 0
x 2 n (59)
x 2 dx  dx 

y 0 + εy1 + ε 2 y 2 + L = 1 , x=1 (60)

d ( y 0 + εy1 + ε 2 y 2 + L)
= 0 , x=0 (61)
dx

The next step of the regular perturbation method would be to equate coefficients of the same
exponent in ε in these equations. However, the presence of the generic exponent n in equation
(59) makes this task difficult. Values of n≠1 make this term nonlinear so that it requires especial
treatment. Before we deal with this term in detail, we recognize that it should not have an impact
 13

on the terms of O(ε0), since it is at least first order on ε. We then equate coefficients of ε0 in
equations (59) to (61) to find

1 d  2 dy 0 
x =0 (62)
x 2 dx  dx 

y0=1, x=1 (63)

dy 0
= 0 , x=0 (64)
dx

Equation (62) can be integrated twice to find

A
y0 = +B (65)
x

Use of equations (63) and (64) to find the two integration constants leads to A=0, B=1, so that
the zeroth-order solution is

y0 ≡ 1 (66)

To treat the last term on the left-hand side of equation (59), we will expand yn in a Taylor
series with respect to ε around ε=0. By doing this, we can express this potentially inconvenient
function in terms of powers of ε. The Taylor series is

dy n 1 d2yn 1 d3y n
yn = yn + ε+ ε2 + ε3 + L (67)
ε =0 dε ε = 0 2 dε 2 ε = 0 3! dε 3 ε = 0

The derivatives required on the right-hand side of this equation can be evaluated by successive
differentiation as follows,

dy n dy
= ny n −1 (68)
dε dε

2
d2yn  dy  d2y
= n (n − 1) y n − 2   + ny n −1 (69)
dε 2  dε  dε 2

3
d3y n  dy  dy d 2 y d3y
= n (n − 1)(n − 2) y n − 3   + 3n (n − 1) y n − 2 + ny n −1 (70)
dε 3  dε  dε dε 2 dε 3

To evaluate the terms on the right-hand side of equation (67) at ε=0, we use equation (58) to
represent y as a function of ε, which yields, along with these three equations,
 14

yn = y 0n (71)
ε =0

dy n
= ny n0−1y1 (72)
dε ε = 0

d2yn
= n (n − 1) y n0− 2 y 21 + 2ny n0−1y 2 (73)
dε 2 ε = 0

d3y n
= n (n − 1)(n − 2) y n0− 3 y 31 + 6n (n − 1) y n0− 2 y1y 2 + 6ny n0−1y 3 (74)
dε 3 ε = 0

Using these derivatives, the asymptotic expansion (67) becomes

1
y n ~ y 0n + ny n0−1y1ε + [n (n − 1) y n0− 2 y 21 + 2ny n0−1y 2 ]ε 2 +
2
(75)
1
[n (n − 1)(n − 2) y n0− 3 y 31 + 6n (n − 1) y n0− 2 y1y 2 + 6ny n0−1y 3 ]ε 3 + L
3!

We can substitute this expansion in the place of yn in equation (59), and equate coefficients of ε1
in equations (59) to (61) to get

1 d  2 dy1  n
x −y0 =0 (76)
x 2 dx  dx 

y1=0, x=1 (77)

dy1
= 0 , x=0 (78)
dx

Using equation (66), the ODE (76) becomes

1 d  2 dy1 
x  =1 (79)
x 2 dx  dx 

Integrating twice leads to

x2 A
y1 = − +B (80)
6 x

Evaluation of the integration constants using the boundary conditions (77) and (78) yields A=0,
 15

B=−1/6, which leads to

1
y1 = ( x 2 − 1) (81)
6

Now we collect coefficients of ε2 in equations (59) to (61) to get

1 d  2 dy 2 
x  − ny n0−1y1 = 0 (82)
x 2 dx  dx 

y2=0, x=1 (83)

dy 2
= 0 , x=0 (84)
dx

Substituting equations (66) and (81) into equation (82), integrating twice and using the boundary
conditions (83) and (84) yields the following solution

 x4 x2 7 
y 2 = n  − +  (85)
 120 36 360 

Using only the first three terms, the asymptotic expansion (58) becomes

ε  x4 x2 7  2
y ~ 1 + ( x 2 − 1) + n  − + ε + L (86)
6  120 36 360 

In terms of the Thiele modulus, this equation becomes

φ2  x4 x2 7  4
y ~ 1 + ( x 2 − 1) + n  − + φ + L (87)
6  120 36 360 

The effectiveness factor can be evaluated from equation (14),

3 dy
η= (88)
ε dx x =1

which yields the asymptotic representation (using the first three terms of the expansion in
equation 86)

1
η ~ 1− nε + L (89)
15
 16

A second-order representation can be obtained by collecting coefficients of ε3 in equations (59)

to (61), which expands equation (86) to third-order terms, and then applying equation (88). The
result is

1 3n 2 − n 2
η ~ 1− nε + ε +L (90)
15 315

In terms of the Thiele modulus, this equation is

1 2 3n 2 − n 4
η ~ 1− nφ + φ +L (91)
15 315

For first-order reactions (n=1) and φ<0.5, the first two terms of equation (87) represent the exact
concentration profiles with errors below 0.1%. The asymptotic solution also works as a good
approximation of the profiles for even higher values of φ (Figure 5) if higher order terms are
included. Comparisons between effectiveness factors will be presented in the next section
(Figure 8).

1
y
0.98
0.96
0.94
0.92
0.9
0.88
0.86
Exact (equation 16)
0.84
Equation 87, two terms
0.82
Equation 87, three terms
0.8
0 0.2 0.4 0.6 0.8 1
x
Figure 5. Dimensionless concentration profile of a reactant in a spherical catalyst pellet for a
first-order reaction (n=1) for φ=1. Comparisons between exact and low Thiele modulus
asymptotic solutions are shown.

As seen in this section, application of regular perturbation analysis is generally straight

forward, and may lead to a good approximation of the solution of algebraic and differential
equations. However, this method breaks down for certain types of problems. In particular, when
the small parameter multiplies the term containing the derivative of the highest order in a
 17

differential equation, the solution in the limit as ε→0 involves a decrease in the order of the
differential equation, which generally means that one of the boundary conditions of the problem
will not be satisfied by the solution. In these cases, a different type of analysis is needed. The
next section explores an option for the asymptotic analysis of such problems.

Singular perturbation analysis: the method of stretched coordinates

In this section we will analyze Model 6 for the case of large Thiele modulus (φ→∞). Let

1
ε= (92)
φ2

Equations (10) to (12) can be written as

ε d  2 dy  n
x −y =0 (93)
x 2 dx  dx 

y=1, x=1 (94)

dy
= 0 , x=0 (95)
dx

To solve this problem using regular perturbation analysis, we would substitute the asymptotic
series (58) into equations (93) to (95). Collecting coefficients of ε0 in equation (93) would lead to
the trivial solution

y0 ≡ 0 (96)

This solution automatically satisfies boundary condition (95), which requires

dy 0
= 0 , x=0 (97)
dx

However, it is in contradiction with boundary condition (94), which would require

y0=1, x=1 (98)

Therefore, regular perturbation analysis cannot be applied to this problem. As ε→0, the
derivatives vanish from the ODE and the solution cannot be made to satisfy both boundary
conditions.
Before we decide how to approach this problem, let us explore the expected trend of the
solution. Figure 1 shows that the concentration profiles tend to become steeper close to the
external surface of the catalyst particle (x=1) as the Thiele modulus increases, due to the
predominance of the reaction rate with respect to diffusion. As φ→∞ (ε→0), we would expect
the profile to completely develop (i.e. drop to zero) in a very thin region adjacent to x=1. This
 18

means that ε→0 in this case leads to a significant shrinking of the part of the domain where the
solution is physically significant (i.e. the concentration will be different from zero only in a very
thin region). This relatively small portion of the domain is called a boundary layer. This
suggests that the problem would be more appropriately represented by a spatial variable that
effectively spans the region within the boundary layer. Consider the transformation

1− x
s= , α>0 (99)
εα

where we seek to use s as the new spatial variable, and α is an as yet undefined constant. This
transformation has two objectives: first, to shift the origin of coordinates to the surface of the
particle instead of the center, thereby defining the start of the boundary layer at the origin s=0;
and second, to stretch the spatial coordinate in such a way that the profile does not develop
completely through a small region spanned by the coordinate but rather it develops over "long"
distances in the new measure s (Figure 6). The transformation essentially maps the thin boundary
layer onto an infinite domain (in the limit as ε→0.)

Figure 6. The concept of the stretched coordinate defined by equation (99). As ε→0 the
boundary layer is mapped onto an (effectively) infinite domain.

Substituting the coordinate transformation (99) into equations (93) to (95) leads to

ε1− 2α d  dy 
 (1 − ε α s) 2  − y n = 0 (100)
(1 − ε α s) 2 ds  ds 

y=1, s=0 (101)

dy 1
= 0, s = →∞ (102)
ds εα
 19

The mathematical purpose of the stretched coordinate is to work in a context in which the
highest-order derivative does not vanish as ε→0. By inspection of the ODE (100), we can see
that this is achieved by making the choice α=1/2. This transforms equation (100) into

1 d dy 
 (1 − ε1 / 2s) 2  − y n = 0 (103)
(1 − ε1 / 2s) 2 ds  ds 

We will start the analysis by considering only terms of order ε0 in equations (103), (101) and
(102) or, alternatively, letting ε→0 in the same equations. The result is

d 2 y0
− y 0n = 0 (104)
ds 2

y0=1, s=0 (105)

dy 0
= 0, s → ∞ (106)
ds

This problem can be solved analytically. For this purpose, let

dy 0
u= (107)
ds

This implies, using the chain rule,

d 2 y 0 du du dy 0 du
= = =u (108)
ds 2 ds dy 0 ds dy 0

Substituting this result into equation (104) and separating yields

udu = y 0n dy 0 (109)

Integrating leads to

u 2 y 0n +1
= +K (110)
2 n +1

or

2 y n +1
1  dy 0 
  = 0 +K (111)
2  ds  n +1
 20

From equation (106), the first derivative goes asymptotically to zero as s→∞. This means that
the second derivative (and all higher order derivatives) should also vanish asymptotically for
large s. The ODE, which must be satisfied at any s, then implies that y0→0 as s→∞ (note that
n>0.) To find the integration constant in equation (111), we can evaluate it at s→∞, which
necessarily implies that K=0. Solving for the first derivative in equation (111) yields

dy 0 2 ( n +1) / 2
=± y (112)
ds n +1 0

Since y0 must decrease with s, the minus sign gives the only solution with physical significance.
Separating and integrating leads to

2 (1− n ) / 2 2
y0 = −s + A , n≠ 1 (113)
1− n n +1

and

ln y0=−s +B, n=1 (114)

From the boundary condition (105) we find A or B, and the solution is

2
 1− n 2  n −1
y 0 = 1 − s  , n≠1 (115)
 2 n +1

y 0 = e −s , n=1 (116)

Now, recalling that (equation 99)

1− x
s= = φ(1 − x ) (117)
ε1 / 2

Equations (115) and (116) lead to the following zeroth-order asymptotic approximations

2
 1− n 2  n −1
y ~ 1 − φ(1 − x )  , n≠1 (118)
 2 n +1

y ~ e −φ(1−x ) , n=1 (119)

Using these two equations along with equation (14), we find that the asymptotic representation
for the effectiveness factor is
 21

3 2
η~ (120)
φ n +1

These asymptotic representations are of zeroth order. Next, we will determine a higher order
representation for the dimensionless concentration profiles. Going back to equation (103), we see
that ε1/2 appears explicitly, which indicates that the asymptotic expansion used before (equation
58) might not work in this case. We will make the change of variable

δ=ε1/2 (121)

(note that δ=1/φ is still a small parameter) and seek an asymptotic representation for the
dimensionless concentration in terms of the power series

y ~ y 0 ( x ) + δy1 ( x ) + δ 2 y 2 ( x ) + L (122)

First we express equation (103) as follows

1 d  dy 
 (1 − δs) 2  − y n = 0 (123)
(1 − δs) ds 
2 ds 

This equation contains two terms that are not direct functions of powers of δ. These terms will be
represented by Taylor series in δ around δ=0. By analogy with equation (75) we can state

1
y n = y 0n + ny 0n −1y1δ + [n (n − 1) y 0n − 2 y12 + 2ny 0n −1y 2 ]δ 2 + K (124)
2

Also, using a Taylor series representation, it is possible to show that

1
= 1 + 2sδ + 3s 2 δ 2 + K (125)
(1 − δs) 2

Substituting equations (122), (124) and (125) into equation (123) and the boundary conditions
(101) and (102), and collecting terms of O(δ0), we recover equations (104) to (106). Collecting
terms of O(δ1) yields, after simplifications,

d 2 y1 dy
− 2 0 − ny 0n −1 y1 = 0 (126)
ds 2 ds

y1=0, s=0 (127)

y1=0, s→∞ (128)

For the general case n≠1, substitution of equation (115) into equation (126) leads to a
complex second order ODE that may not have an analytical solution. Nevertheless, a numerical
 22

solution could be found for a specific n. Here, we will limit our analysis to the case n=1 to
illustrate the potential advantages of a first-order solution. Letting n=1 in equation (126), and
substituting equation (116) yields

d 2 y1
− y1 = −2e −s (129)
ds 2

A particular solution of this equation is

y1, p = se −s (130)

whereas the homogeneous solution is

y1, h = Ae −s + Bes (131)

Letting y1=y1,h+y1,p and using the boundary conditions (127) and (128), the integration constants
are A=B=0, and the final solution is

y1 = se −s (132)

Combining equations (116), (122) and (132) leads to an improvement of equation (119) as an
asymptotic representation for the dimensionless concentration for n=1:

y ~ (2 − x )e −φ(1− x ) (133)

and the new effectiveness factor becomes

3 3
η~ − (134)
φ φ2

Figure 7 shows a comparison between the asymptotic and exact solutions for the
concentration profiles. For φ=10 and 20, the zeroth-order solution (equation 119) represents the
profiles fairly accurately, and the first-order solution (equation 133) is indistinguishable from it.
However, at φ=5, the first-order solution yields a considerable improvement.
The existence of a boundary layer becomes obvious as the Thiele modulus increases: note that
the thickness of the region in which the concentration is appreciable shrinks to about 20% of the
sphere radius at φ=20. The asymptotic solution can be used to calculate the boundary layer
thickness by establishing a criterion that defines the region of concentration change. For
example, if we say that the boundary layer is the region in which the dimensionless concentration
falls to 1% of the surface concentration, then the edge of the boundary layer (xbl) can be
calculated from equation (119) (here we are only concerned with really large values of the Thiele
modulus, say, φ≥20, so that the zeroth-order solution is adequate enough) by letting y=0.01. The
result is
 23

− ln 0.01
1 − x bl = (135)
φ

The most interesting aspect of this equation is that it shows that the boundary layer thickness is
inversely proportional to φ. It is worthwhile noticing that the reaction is occurring within the
boundary layer only, so that the rest of the catalyst particle (x<xbl) is not active.

1
y
0.9

0.8

0.7

0.6

0.5 φ =5

0.4

0.3
φ =10
0.2
φ =20
0.1

0
0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
x
Figure 7. Asymptotic (φ»1) dimensionless concentration profiles for a first order reaction
(n=1). Symbols: exact solution (equation 15); solid lines: first-order asymptotic expansion,
equation (133); dashed lines: zeroth-order asymptotic expansion, equation (119). For φ=10 and
20, both equations lead to curves that are indistinguishable in the scale of the plot.

A comparison between the various asymptotic solutions for effectiveness factors and the exact
solution for n=1 is shown in Figure 8. Note that the first-order solution yields a slight
improvement over the zeroth-order solution at low Thiele modulus (in the sense that it extends
somewhat the range in which it predicts η accurately), but it suddenly diverges from the
expected behavior as φ increases. The range of applicability of the first-order solution over the
zeroth-order solution is much wider for high Thiele modulus. It is interesting to see how two
relatively low order asymptotic solutions (equations 89 and 134) almost span the full range of
Thiele modulus in their approximation of the effectiveness factor.
 24

A second boundary layer problem: Michaelis-Menten kinetics

The use of stretched coordinates and the presence of boundary layers are not limited to
regions of relatively large spatial gradients. Rapid changes with time can lead to temporal
boundary layers. A practical case in which this is common is in the chemical kinetics of complex
reaction systems, in which reaction rates of various reactions may vary by orders of magnitude.
The existence of such disparities is usually treated by identifying the controlling (limiting)
reactions of a particular mechanism, which would set the kinetics of the system, or by assuming
that relatively fast reactions occur instantaneously. An example of this situation is the typical
reaction mechanism used to represent enzymatic reactions in which a substrate (A) complexes
with an enzyme (E) and then decomposes to form products (P), which release the enzyme back
into the reaction mixture:

A+E⇔C→P+E (136)

1
η
0.9
equation (91)
0.8 equation (89)

0.7
equation (120)
0.6
equation (134)
0.5
0.4
0.3
0.2
0.1
0
0.1 1 10 100
φ
Figure 8. Effectiveness factor for first order reaction (n=1) in a spherical catalyst. Symbols:
exact solution (equation 16); solid lines: asymptotic limits at high and low Thiele modulus.

where C is the enzyme/substrate complex. The complexation reaction is reversible, whereas the
decomposition reaction is irreversible. The enzyme is considered a catalyst, since it does not
suffer a permanent transformation in the reaction.
In a constant-volume batch system, the kinetics of the reaction system (136) can be expressed
as follows,

dc A
= − k 1c A c E + k 2 c C (137)
dt
 25

dc C
= k 1c A c E − (k 2 + k 3 )c C (138)
dt

where the reactions are considered to be elementary: k1 and k2 are the reaction rate constants of
the direct and inverse complexation reactions, respectively, and k3 is the constant associated with
the irreversible decomposition of the complex.
Note that, if the reactions are allowed to proceed for a long time, the steady state reached will
be a complete disappearance of both A and C, and the enzyme concentration will be equal to its
initial level. This is a consequence of the irreversibility of the decomposition reaction. Usually,
the reaction is started with initial substrate and enzyme concentrations cA0 and cET, respectively,
and no C present. It is common to find that cET«cA0 (note that relatively large amounts of enzyme
are not necessary, since it is regenerated as the reaction proceeds.) When the reaction starts, the
concentration of C rises rapidly since the complexation reaction proceeds faster than the
decomposition reaction. As the substrate is consumed, the rate of disappearance of C becomes
comparable to the rate of its production. It is usually assumed that in a relatively short time, these
rates will balance, leading to a pseudo-steady state at which

dc C
≈ 0 , pseudo-steady state hypothesis (139)
dt

In the analysis of this reaction system, it is usually assumed that condition (139) holds at all
times. If this is so, then, from equation (138) we obtain

k1
cC = cAcE (140)
k2 + k3

Also, the total enzyme concentration is unchanged, so that, at all times,

c ET = c E + c C (141)

Using equations (140) and (141) to eliminate cC and cE from equation (137) leads to

c ET c A
cC = (142)
K + cA

Kc ET
cE = (143)
K + cA

dc A kc A
=− (144)
dt K + cA

where

k=k3cET (145)
 26

k 2 + k3
K= (146)
k1

Equation (144) is known as the Michaelis-Menten kinetic expression, and it is generally

accepted as the equation governing enzymatic reaction kinetics. It can be solved by direct
integration, along with the initial condition

cA=cA0, t=0 (147)

The preceding analysis is the usual way in which this problem is treated in biochemistry
applications. Here, we will use asymptotic analysis to determine the applicability of the
approximation made, as well as the form of the solution at short times when the concentration of
complex builds up and the pseudo-steady state approximation is not valid. We start by making
the problem dimensionless to minimize the number of parameters in the equations. Consider the
following definitions of dimensionless variables,

cC
v= (148)
c ET

k 1c A
s= (149)
k 2 + k3

τ = k1c ET t (150)

In addition, from equations (141) and (148) we have

c E = c ET (1 − v) (151)

With these definitions, equations (137) and (138) become

ds
= −s(1 − v) + αv (152)
dτ

dv
ε = s(1 − v) − v (153)
dτ

where

k2
α= (154)
k2 + k3

k 1c ET
ε= (155)
k2 + k3
 27

Here, we will take ε→0 to be the small parameter in the formulation, since this will lead to the
pseudo-steady state approximation for v by making the transient term in equation (153) small.
The fact that ε multiplies the derivative in equation (153) will lead to the existence of a boundary
layer in dimensionless time as τ→0, since a transient for v will need a steep change of v with τ.
The initial conditions are

k 1c A 0
s = si = , τ=0 (156)
k2 + k3

v=0, τ=0 (157)

The first step in the solution is to apply regular perturbation analysis to the problem. We
postulate the existence of asymptotic expansions for the dependent variables:

s ~ s 0 + εs 1 + ε 2 s 2 + L (158)

v ~ v 0 + εv1 + ε 2 v 2 + L (159)

Next we substitute these series into the ODEs (152) and (153) and start collecting terms of the
same order in terms of the perturbation parameter. Terms of O(ε0) lead to

ds 0
= −s 0 (1 − v 0 ) + αv 0 (160)
dτ

0 = s 0 (1 − v 0 ) − v 0 (161)

This last equation implies that

s0
v0 = (162)
1 + s0

while substitution of this equation into equation (160) leads to

ds 0 (α − 1)s 0
= (163)
dτ 1 + s0

Notice that equations (162) and (163) are the dimensionless form of equations (142) and
(144), respectively. It would seem that the zeroth-order, regular perturbation analysis leads to the
Michaelis-Menten kinetic expression. However, it is clear that equation (162) does not satisfy the
initial condition that the expansion indicates (v=0 at τ=0), so that the initial conditions must be
left out of the formulation. Still, the partial solution found by regular perturbation (equations 162
and 163 with undefined initial conditions) should be valid outside the boundary layer, where
dv/dτ has moderate values. For this reason, this solution is called the "outer" solution. We will
change the notation of the results obtained to reflect this. The zeroth-order outer solution is then
 28

s (o)
v (o ) = 0
(164)
0
1 + s (o)
0

ds ( o) (α − 1)s (o )
0
= 0
(165)
dτ 1+ s (o)
0

with undefined initial conditions.

To find the solution inside the boundary layer (i.e., the "inner" solution), we apply the method
of stretched coordinates. First, we recognize that dv/dτ will have a large value inside the
boundary layer (rapid temporal change in the complex concentration). Therefore, a moderate
value for the derivative can only be obtained if the time scale is stretched. We define a stretched
time scale by

τ
θ= (166)
ε

Using this as a change of variable in equations (152) and (153) leads to

ds (i )
= −εs (i ) (1 − v (i ) ) + εαv (i ) (167)
dθ

dv (i )
= s (i ) (1 − v (i ) ) − v (i ) (168)
dθ

where the superscript (i) denotes inner solution. The inner solution must satisfy the initial
conditions:

s (i ) = s i , θ=0 (169)

v (i ) = 0 , θ=0 (170)

Now, we propose the asymptotic expansions

s (i ) ~ s (i ) + εs (i ) + ε 2 s (2i ) + K (171)
0 1

v (i ) ~ v (i ) + εv (i ) + ε 2 v (2i ) + K (172)
0 1

Next, we substitute equations (171) and (172) into equations (167) through (170) and start
collecting terms of O(εn). Consider terms of O(ε0):
 29

ds (0i )
=0 (173)
dθ

dv (i )
0
= s (i ) (1 − v (i ) ) − v (i ) (174)
dθ 0 0 0

s (0i ) = s i , θ=0 (175)

v (0i ) = 0 , θ=0 (176)

Equation (173) implies that s (0i ) is a constant inside the boundary layer and that constant must
be the initial value (equation 175):

s (0i ) ≡ s i (177)

Now, equation (174) becomes

dv (i )
0
= s i − (1 + s i ) v (i ) (178)
dθ 0

which can be easily integrated using the initial condition (176). The final result is

v (i ) =
si
[1 − e −(1+si )θ ] (179)
0 1 + si

In general, boundary layer type problems are similar to this one: one of the solutions is still
undefined after both outer and inner solutions are found. In this case, the outer solution is still
undefined. The final solution is found by forcing a match between the outer and the inner
solutions at the edge of the boundary layer (i.e. by imposing continuity between the solutions at
the different scales). A matching criterion for a function f is usually stated as follows:

(f inner ) →outer = (f outer ) →inner (180)

which means that the inner solution evaluated as the outer region is approached is equal to the
outer solution evaluated as the inner region is approached. This criterion works well for certain
classes of boundary layer problems. Translated into the problem at hand, equation (180) can be
expressed as (for the zeroth-order solutions):

v (i ) = v (o) (181)
0 θ →∞ 0 τ→0
 30

s (i ) = s (o) (182)
0 θ→ ∞ 0 τ→0

It is interesting to note that the matching criterion provides the initial conditions for the outer
solution in this case. The matching condition (182), along with equation (177), yields

s ( o) = s i , τ=0 (183)
0

This condition, along with equations (162) and (163), identifies the outer problem as the
dimensionless version of the Michaelis-Menten formulation. In other words, we have proven that
the Michaelis-Menten formulation is equivalent to the zeroth-order asymptotic solution for the
concentrations with respect to the dimensionless group defined by equation (155) at times long
enough to be represented by the scale τ. Note that the matching criterion (181), using equation
(179), leads to

si
v (o ) = , τ=0 (184)
0 1 + si

which is automatically satisfied by equations (164) and (183) (i.e. it does not yield additional
information).
The outer solution for s can now be found by integrating equation (165) subject to initial
condition (183). Separating and integrating yields

s (0o )
ln + s (0o) − s i = (α − 1)τ (185)
si

with v (o ) given by equation (164). Also, the inner solution can be written in terms of outer
0
variables by substituting (166) into equation (179):

 τ
s −(1+si ) 
v (i ) = i 1 − e ε (186)
0 1 + si  
 

with s (i ) given by equation (177).

0
In regular time scales, the outer solution is the desired outcome of the preceding analysis.
However, the inner solution provides the short-time behavior for the concentrations and it can be
used to calculate the boundary layer thickness, which in this case is interpreted as the time scale
over which the inner solution evolves. We can define the boundary layer thickness (τbl) as the
dimensionless time required for the inner solution to achieve 99% of its asymptotic value (which,
according to the matching criteria, is the initial condition of the outer solution). Mathematically,
τbl is obtained when
 31

si
v ( i ) = 0.99 , at edge of boundary layer (187)
0 1 + si

From equation (186) we obtain

− ε ln(0.01)
τ bl = (188)
1 + si

Figure 9 illustrates the behavior of the outer solutions for s and v in a regular time scale
(equations 164 and 185). These are the results that would be obtained in a simulation using
Michaelis-Menten kinetics. Figure 10 shows the inner solution for v (the inner solution for s is
uniform) in a short time scale, superimposed on the outer solution.

1
v, s
0.9
s
0.8

0.7

0.6

0.5 v

0.4

0.3

0.2

0.1

0
0 2 4 6 8 10 12 14
τ

Figure 9. Zeroth-order outer solutions (equations 164 and 185) for an enzymatic reaction
system with α=0.6, si=1 and ε=0.001. In this case, the boundary layer is confined to τ<0.0023, as
calculated by equation (188).

In this type of problems, sometimes it is inconvenient to have the solution partitioned into
separate equations for the inner and outer regions. To obtain one function that is valid in the
whole domain, a composite solution can be constructed, by adding the inner and outer solutions,
and then subtracting the value at their intersection (given by the matching condition 180). For a
general function f, the composite solution is
 32

f comp = f outer + f inner − (f outer ) → inner (189)

Applying this to our solution we obtain

τ
s ( o) s − (1+ s i )
v (c) = 0
− i e ε (190)
0 1 + s( o) 1 + si
0

s( c ) = s( o) (191)
0 0

0.6
v
outer solution
0.5

0.4

inner solution

0.3

0.2

0.1

0
0.0001 0.001 0.01 0.1 1
τ

Figure 10. Zeroth-order inner solution (equation 186) for the dimensionless concentration of
complex for the system in Figure 9.

where the superscript (c) denoted composite solution and s( o ) is a function of dimensionless time
0
given by equation (185) (note that an explicit representation cannot be found in this case.) A
comparison of these solutions with an exact solution of the problem (obtained by solving
numerically equations 152 and 153) is shown in Figure 11.
 33

0.6
v

0.5

0.4

0.3

0.2

0.1

0
0.0001 0.001 0.01 0.1 1
τ

Figure 11. Comparison between the exact solution (solid line) and the composite solution
(symbols) for the dimensionless concentration of complex.

Perturbation methods in stability analysis: bubble expansion and contraction

A common application of asymptotic analysis and perturbation methods is their use to
establish the stability of a physical system. In this section we illustrate this type of application
with an example: a gas bubble immersed in a quiescent liquid whose size varies due to changes
in external conditions. First, we will derive the governing equations to determine the temporal
change in bubble size. We will assume that the bubble is always a perfect sphere whose radius
varies with time, R=R(t). The only motion in the liquid surrounding the bubble will be caused by
the change in bubble radius (expansion or contraction). Using a spherical coordinate system with
origin at the center of the bubble, we will assume that the only nonzero component of the liquid
velocity vector is the radial component: vφ=vθ=0, vr≠0. We will also assume that the velocity is
symmetric with respect to both angular coordinates, so that vr=vr(t,r).
The continuity equation in spherical coordinates for incompressible flows is

1 ∂ (r 2 v r ) 1 ∂ ( v θ sin θ) 1 ∂v φ
+ + =0 (192)
r2 ∂r r sin θ ∂θ r sin θ ∂φ
 34

This simplifies to

∂ (r 2 v r )
=0 (193)
∂r

which yields, after integration,

F( t )
vr = (194)
r2

Here, we will consider the case in which changes in the bubble size are due to temporal
changes of the pressure in the liquid far from the bubble: a change in the liquid pressure will tend
to compress or expand the bubble, thereby inducing a temporal change in its size. We will
assume that the gas in the bubble does not dissolve in the liquid, and that it is at thermodynamic
equilibrium with the liquid at all times. This means that the partial pressure of the liquid species
in the bubble will be the vapor pressure of the liquid at the system temperature. Under these
conditions, changes in bubble size necessarily will imply that either liquid will evaporate into the
bubble or that vapor will condense to keep the partial pressure equal to the vapor pressure.
Nevertheless, if the density of the liquid phase is assumed to be much greater than the density of
the vapor (which is the common occurrence), this possible evaporation/condensation will have
negligible effect on the liquid velocity. Under these conditions, the velocity induced in the liquid
by the change in size of the bubble is

dR
vr r =R = (195)
dt

From equations (194) and (195), it follows that

dR
F( t ) = R 2 (196)
dt

which implies

R 2 dR
vr = (197)
r 2 dt

The equation that will eventually lead to determine R(t) is the radial component of the Navier-
Stokes equation in the liquid, which in this case simplifies to (neglecting gravitational effects)

 ∂v ∂v  ∂P  1 ∂  2 ∂v r  2 v r 
ρ r + v r r  = − + µ  r − (198)
 ∂t ∂r  ∂r  r 2 ∂r  ∂r  r 2 

Substitution of equation (194) into equation (198) leads to, after manipulations,
 35

∂P  1 dF 2F 2 
− = ρ −  (199)
∂r  r 2 dt r 5 

Note that the viscous terms cancel out. This equation can be integrated with respect to r to obtain

 1 dF F 2 
P = A + ρ −  (200)
 r dt 2r 4 

where A is an integration constant. We will assume that we know the liquid pressure far from the
bubble, and that this imposed pressure may be a function of time; i.e.

P = P∞ ( t ) , r→∞ (201)

This yields A=P∞ and the pressure profile is

 1 dF F 2 
P(r, t ) = P∞ ( t ) + ρ −  (202)
 r dt 2r 4 

This profile can be expressed in terms of the bubble radius by substituting equation (196) into
this equation to get

 R 2 d 2 R  dR  2  2R R 4 
P(r, t ) = P∞ ( t ) + ρ  +   −  (203)
 r dt 2  dt   r 2r 4 


The pressure of the liquid at the bubble surface can be related to the pressure inside the bubble
using a normal stress balance at the bubble surface. Here, we will leave out the details of the
derivation of this balance, but the final result is

∂v r 2γ
− P(R , t ) + 2µ = −PB ( t ) + (204)
∂r r = R R

The left-hand side of this equation represents the normal stress exerted by the liquid on the
bubble surface, which consists of pressure and viscous stresses, while the right-hand side reflects
the stress exerted by the gas on the bubble surface due to its pressure, PB, plus the surface tension
stress, where γ is the interfacial tension between the liquid and the gas, which can be considered
equal to the surface tension of the liquid, if the gas is ideal.
From equation (203) we have

 d 2 R 3  dR  2 
P(R , t ) = P∞ ( t ) + ρ R +    (205)
 dt 2 2  dt  
 36

Also, from equation (197) we get

∂v r 2 dR
=− (206)
∂r r = R R dt

Substituting equations (205) and (206) into (204) yields

2γ  d 2 R 3  dR  2  4µ dR
PB ( t ) − P∞ ( t ) = + ρ R +   + (207)
R  dt 2 2  dt   R dt

This equation is known as the Rayleigh-Plesset equation, and it is interpreted as an ODE whose
solution would yield R(t) for a specified pressure difference between the bubble and the liquid
away from the bubble. Under most practical applications, the viscous term in this equation can
be neglected. For example, for a bubble growing in water at ambient conditions, an order of
magnitude analysis shows that the surface tension term is several orders of magnitude greater
than the viscous term for reasonable bubble growth rates (i.e. dR/dt not surpassing a few cm/s).
We will neglect viscous terms and simplify equation (207) to

2γ  d 2 R 3  dR  2 
PB ( t ) − P∞ ( t ) = + ρR +    (208)
R  dt 2 2  dt  

The bubble pressure is the sum of the vapor pressure of the liquid (Pv) plus the pressure of the
inert gas in the bubble, which will be assumed ideal,

3nR̂T
PB = Pv + Pg = Pv + (209)
4πR 3

where n is the number of moles of inert gas in the bubble (constant), R̂ is the universal gas
constant, and T is bubble temperature (which will be assumed to be constant also). Let

3nR̂T
α= (210)
4π

then, using equation (209), equation (208) can be written as follows

2
d 2 R 3  dR  2γ α ∆P
R +   + − + =0 (211)
dt 2 2  dt  ρR ρR 3 ρ

where ∆P=P∞-Pv. Analytical solution of this ODE with two appropriate initial conditions and a
prescribed fluid pressure, P∞(t), is difficult because of the highly nonlinear nature of the equation
(all terms involving R(t) are nonlinear), but the equation can be solved numerically. Let us
consider the case in which the fluid pressure is constant. In that case, we see that a steady state
 37

solution of equation (211) would be

2γ α
− + ∆P = 0 (212)
R s R 3s

where Rs is the steady state or equilibrium bubble radius. This cubic equation can be rearranged
as follows

2γ 2 α
R 3s + Rs − =0 (213)
∆P ∆P

A convenient way of expressing this equation is to define a dimensionless equilibrium bubble

radius. To choose a characteristic length for the problem, we notice that, if the liquid pressure far
from the bubble equals the vapor pressure, ∆P=0 (incipient boiling or cavitation), the equilibrium
radius would be

α
Rs = , ∆P=0 (214)
2γ

This suggests the characteristic length

α
l= (215)
2γ

Now we define a dimensionless bubble radius by

R
x= (216)
l

Using this definition, equation (213) can be written as

1 2 1
x 3s + xs − = 0 (217)
β β

where

1/ 2
 α 
β = ∆P  (218)
 8γ 3 

Note that ∆P can be positive or negative: a negative pressure difference implies Pv>P∞, which
would indicate a supersaturated liquid. The solution to the cubic equation (217) has only one real
positive root for β>0, but two real positive roots for β<0. A plot with solutions of equation (217)
is presented in Figure 12. There is a minimum value of β below which there are no positive
 38

solutions of equation (217). This value of β can be obtained analytically by performing the
operation

dβ
=0 (219)
dx s β =β min

Solving equation (217) for β and applying this equation, we find: βmin=−2/31.5=−0.3849, at which
xs = 3 .

Figure 12. Dimensionless equilibrium bubble sizes, as obtained from equatiom (217). Two
solutions are possible for β<0.

The fact that two steady states are possible for β<0 raises the question if these steady states
are stable. One way to analyze the stability of a physical system is to use perturbation methods.
In what follows, we will use this case to illustrate how this is done. First, we will write equation
(211) in dimensionless form. Making use of equation (216) to make the radius dimensionless,
and using the following definition for dimensionless time

1/ 4
 (2γ ) 5 
θ = t  (220)
 ρ2α3 
 39

equation (211) becomes

2
x d 2 x 3  dx  1 1
+   + − +1 = 0 (221)
β dθ 2 2β  dθ  βx βx 3

Note that this equation simplifies to equation (217) at steady state.

Now we assume that the bubble is at its equilibrium size, x=xs, whose stability we wish to
analyze. At θ=0, we consider that the bubble size is modified by the presence of a small
perturbation to its radius; i.e.

x = x s (1 + ε) , θ=0 (222)

Since equation (221) is second order with respect to dimensionless time, we need a second
condition. We will assume that, since the bubble was initially at equilibrium, we can say that

dx
= 0 , θ=0 (223)
dθ

Note that, if ε=0, then x≡xs is, of course, a solution to the problem posed by equations (221) to
(223). The question now will be: if ε is small but nonzero, how will the bubble size evolve with
time? If the size returns to the initial steady state, then the steady state is stable. If the bubble size
evolves continuously away from the steady state, then the steady state is unstable and cannot be
achieved in practice.
The problem has now become a regular perturbation problem. To find the solution, we
postulate that the dimensionless bubble size can be represented by the following asymptotic
series

x ~ x s (1 + ψ1ε + ψ 2 ε 2 + L) (224)

Next, we substitute this equation into equations (221) and (223) and start collecting terms of the
same order in ε. Before we do this, however, we note that the third and fourth terms of equation
(221) need to be linearized. For this purpose, we expand them in Taylor series around ε=0 to get
the following asymptotic representations (for details, see treatment of term yn in the catalyst
particle problem, equations 67 to 75)

1 1  ψ2 2 
1 − ψ1ε + (ψ1 − 2 )ε + L
2
~ (225)
x xs

1 1  3 
1 − 3ψ1ε + 2 (4ψ1 − ψ 2 )ε + L
2 2
3
~ 3 (226)
x xs

It can be shown that substitution of equations (224) to (226) into equations (221) to (223) and
collection of terms of O(ε0) yields equation (217) (the steady state solution). Our main interest is
 40

in the first order solution, since it will reflect how the bubble size responds to the perturbation.
Collection of terms of O(ε1) from equations (221) to (223) leads to, after manipulations,

d 2 ψ1 3 − x s2
+ ψ1 = 0 (227)
dθ 2 x 5s

ψ1 = 1 , θ=0 (228)

dψ 1
= 0 , θ=0 (229)
dθ

The solution of this problem depends on the sign of the factor multiplying ψ1 in equation (227).
We will consider the three possible cases:

Case 1: 3 − x s2 > 0
The solution of equations (227) to (229) is

 3 − x2 
ψ1 = cos θ s  (230)
 x 5 
 s 

Case 2: 3 − x s2 < 0
The solution is

 x2 − 3  1   x2 − 3   
 + exp − θ x s − 3 
2
ψ1 = cosh  θ s  = exp θ s (231)
 x 5s  2   x 5s   x 5s 
   

Case 3: x s = 3
The solution is

ψ1 ≡ 1 (232)

Now, looking at equation (224), we see that the bubble size given by solutions (230) and
(232) will approach the steady state at all times if ε→0. On the other hand, the bubble size given
by solution (231) will grow indefinitely, regardless of how small ε is (i.e., ψ1→∞ as θ→∞). We
conclude that this solution is unstable and thus physically unattainable. Therefore, the bubble
will be unstable when x s > 3 . Going back to all possible steady state solutions (Figure 12), we
conclude that the upper branch of the curve where there are two possible solutions is always
unstable.