Вы находитесь на странице: 1из 13

ENGR 9420 - Engineering Analysis: Scaling and Asymptotic Analysis

2
2.1

Scaling and Asymptotic Analysis


Introduction

We begin with an introduction to scaling methods and asymptotic analysis. Scaling provides
a quick route to understanding the physical behaviour of an ordinary or partial differential
equation. In some cases the scaling analysis provides asymptotically exact behaviour, while
in most cases it provides the asymptotic behaviour correct within an order of magnitude. In
the latter case this information is still useful, particularly for putting numerical simulations
or experimental measurements into context.
We will review the basic rules for conducting a scaling analysis, review some simple
examples which illustrate the concepts, and then conclude with a discussion of asymptotic
model development using scaled solutions.

2.2

Scaling Methods

Scaling analysis is the art of examining the order of magnitude of terms appearing in the
governing equations describing a physical problem. In some texts, it is referred to as order
of magnitude analysis or just scaling. Scaling analysis may be used to simplify governing
equations through the determination and elimination of terms which have little effect on
the balance that constitutes the governing equation or it may also be used to obtain crude
solutions which are correct within an order of magnitude sense.
Scaling analysis should not be confused with the method of dimensional analysis, i.e.,
the procedure of non-dimensionalization using the well known Buckingham method. The
Buckingham method is a variable reduction technique, whereby a group of dimensional
variables is reduced in number through combination into dimensionless groups. Its greatest
shortcoming is that it often produces multiple solutions for dimensionless groups, many of
which are often not physically meaningful. On the other hand, since scale analysis begins
with the equations describing the physical process, it will produce physically meaningful
solutions for the dimensionless groups. These issues will be dealt with in more detail later.
2.2.1

Basic Rules for a Scaling Analysis

Bejan (1995) recommends scale analysis as the premier method for determining the most
knowledge for the least effort. For this reason, it should be ones first approach to understanding a problem. The method of scale analysis has five basic rules, Bejan (1995). These
are:
Rule 1 - Always define the spatial extent of the problem under consideration if relevant. This
could be a physical dimension associated with the system geometry if is finite, or a boundary
layer thickness (penetration depth) when the system or region of interest is semi-infinite in
character.
Two important concepts in problems which characterize the nature of a field are: penetration depth and volumetric rise, Arpaci (1966) and Arpaci and Larsen (1984). These
concepts differentiate the nature of the transport process.

ENGR 9420 - Engineering Analysis: Scaling and Asymptotic Analysis

10

Penetration depth (or boundary layer) problems are the result of step changes imposed at
a boundary which result in a penetrating effect of the species of interest, into the domain or
geometry. In these types of problems, only part of the geometry or system is affected by the
change imposed at the boundary or boundaries. These types of problems are characteristic of
short time diffusion problems. In volumetric rise problems, an external change in the system
is felt everywhere within the domain or geometry of interest, i.e distributed sources/sinks
such as a pressure gradient, internal heat generation, or chemical reaction. We will address
both of these concepts later when formulating solutions to several problems.
Rule 2 - A differential equation constitutes an equivalence between two dominant terms
appearing in the equation. Other terms in the equation of interest may dominate under
certain conditions, i.e. short time versus long time, upstream versus downstream, etc. The
art of selecting the dominant scales is the basis of Rules 3-5.
Rule 3 - If in the sum of two terms in which the order of magnitude of one term is greater
than the order of magnitude of the other term, then the order of magnitude of the sum is
dictated by the dominant term. i.e. given c = a + b, then if
O(a) > O(b) then O(c) O(a)

(1)

Rule 4 - If in the sum of two terms, the order of magnitude is the same for each term, then
the sum is of the same order of magnitude. i.e. given c = a + b, then if
O(a) O(b) then O(c) O(a) O(b)

(2)

Rule 5 - In any product or quotient of two terms, the order of magnitude of the product
or quotient is given by the product or quotient of the order of magnitudes of each term. i.e.
given c = a b, then if
O(c) O(a)O(b)

(3)

O(c) O(a)/O(b)

(4)

or if given c = a/b, then

We will illustrate these concepts by considering a few simple problems from heat transfer,
fluid dynamics, and mechanics, to illustrate the method.
2.2.2

Applications of Scaling Analysis

We begin by demonstrating the method for two simple problems. One is from fluid mechanics,
the other from heat transfer. In the first example we consider a finite domain, while in the
second we consider a semi-infinite domain. These lead to two important concepts in transport
phenomena which characterize the nature of the flow or transport field: penetration depth
and volumetric rise, Arpaci (1966) and Arpaci and Larsen (1984).

ENGR 9420 - Engineering Analysis: Scaling and Asymptotic Analysis

11

Penetration depth (or boundary layer) problems are the result of step changes imposed at
a boundary which result in a diffusion of the species of interest, into the solution domain or
geometry. In these types of problems, only part of the geometry or system is affected by the
change imposed at the boundary or boundaries. These types of problems are characteristic
of short time diffusion processes.
In volumetric rise problems, an external change in the system is felt everywhere within
the domain or geometry of interest. These types of problems are characteristic of long time
diffusion processes, or systems with distributed sources/sinks such as a pressure gradient,
internal heat generation, or chemical reaction.
We will address both of these concepts when formulating solutions to several problems.
Example 1 - Scaling in a Finite Domain: Fully Developed Laminar Fluid Flow
Consider one dimensional fully developed fluid flow in a simple channel of width 2b is
governed by the following momentum equation:
dp
d2 u
2 =
dx }
dz
| {z
|{z}

F riction

(5)

P ressure

The momentum equation is subject to the no slip condition, u = 0, at the channel


walls (x = b), and a boundedness condition (or plane of symmetry), u 6= , along the
centerline of the channel (x = 0). The above equation represents a balance between friction
and pressure forces. The equation also may be written to represent a balance between two
approximate scales using the geometry and flow characteristics:
p
U

(6)
2
b
L
where we have chosen b to be the representative length scale of the channel, dp/dz p/L
and u U , the mean flow velocity. In this problem, we only consider one characteristic
length scale related to the cross-section, b, but in many problems more than one length scale
may be present, i.e. a rectangular or elliptical duct. We now obtain the pressure drop as:

LU
b2

(7)

or
LU
b2
We can easily obtain the shear stress at the wall from Newtons law of viscosity:

du
U
=


dn wall
b
p = C1

or

(8)

(9)

U
(10)
b
Finally, we may also define a friction factor as is frequently done in fluid mechanics:
= C2

ENGR 9420 - Engineering Analysis: Scaling and Asymptotic Analysis

f=

C2 U/b
C2
=
=
2
2
U
U
Reb

12

(11)

Thus, this simple analysis has shown that the friction factor is:
O(1)
(12)
ReL
The exact solution to this problem for fully developed flow in non-circular ducts is found
in all basic fluid texts. The constant C2 = 12 when L = Dh = 4A/P = 4b, the hydraulic
diameter of the duct. This was a relatively simple analysis. However, for multi-dimensional
problems, we find that the same characteristic results are obtained.
f

Example 2 - Scaling in a Semi-Infinite Domain: Unsteady Heat Conduction


In the second example, we consider transient conduction or diffusion into a semi-infinite
domain initially at some uniform temperature Ti at t = 0, and whose surface is raised to a
temperature Ts > Ti , when t > 0. The governing differential equation is:
1 T
=
|{zt}

Storage

2T
x2
|{z}

(13)

Conduction

This equation may be written more simply to represent the following balance
Ts Ti
Ts Ti

(14)
t
2
In this problem, there is no finite dimension associated with the geometry, thus x is
the instantaneous penetration depth. We may now solve the above equation to obtain the
intrinsic or natural diffusion length scale:

(15)

Next, we desire the heat flux at the surface. This may be obtained from Fouriers law:

T
q = k
(16)
x s
The heat flux scales according to
k(Ts Ti )
k(Ts Ti )

t
The exact solution for the heat flux q found in most heat transfer texts is
q

(17)

k(Ts Ti )

(18)
t
In
the above expression, the denominator is referred to as the conduction thickness
= t. We see from this example, that although Eq. (13) is a partial differential
q=

ENGR 9420 - Engineering Analysis: Scaling and Asymptotic Analysis

13

equation, the two important characteristics of this diffusion problem are easily found in an
approximate sense. We also see, that the conduction penetration depth is proportional to
the intrinsic penetration depth .
This simple approach predicted the correct form of the solutions to within an order of
magnitude. This was a rather simple exercise in both problems as there were only two scales
to consider in the balance. More complex transport problems represent balances between
more than two scales. In these cases, we must examine the asymptotic behaviour of these
problems by considering pairings of the dominant scales over a range of a single characteristic
dependent variable.
Example 3 - Scaling with Multiple Scales: Falling Body Dynamics
In our final example, we consider the problem of a body in free fall. The force balance
on the body dictates that three forces dominate. These are the acceleration or inertial force,
the gravitational force, and viscous drag forces. These forces yield the following equation of
motion when Newtons law is applied:
 2
dx
d2 x
(19)
m 2 = mg k
dt
dt
or in terms of velocity:
dv
= mg kv 2
(20)
dt
where g is the gravitational acceleration, and k is the damping or drag coefficient. The
above equation is a non-linear ordinary differential equation which is also subject to the
initial condition v(0) = 0.
The differential equation is a balance of three forces:
m

dv
m
| {zdt}

= mg |{z}
kv 2
|{z}

Acceleration

Gravity

(21)

Drag

We can characterize the short time and the long time solution behaviour by considering
the two fundamental balances: acceleration gravity and gravity drag.
At very small times after the initial release of the object, when velocities are expected to
be very small and drag minimal, the balance is between acceleration and gravity:
v
m mg
t

(22)

which yields
v gt

t0

(23)

At larger times the balance is between gravity and drag forces as the body reaches
terminal velocity:
mg kv 2
(24)
which yields
r
v

mg
k

(25)

ENGR 9420 - Engineering Analysis: Scaling and Asymptotic Analysis

14

We may now introduce dimensionless variables and present the scales more conveniently:
v
gt
t
p mg q t?
m
v
k

t? 0

(26)

gk

and

v
1
t?
(27)
v
The above scaled asymptotes actually represent the exact asymptotes as we shall see
shortly. The long time limit should be evident, since there is no acceleration in this state
if a terminal velocity is achieved, and therefore the scaled force balance also represents the
exact limit. The exact solution to the above problem can be found by separating variables
and integrating:
Z
Z
mdv
(28)
dt =
mg kv 2
Integration using simple trigonometric substitutions and application of the initial condition v(0) = 0, gives the following result:
r !
r
mg
gk
v(t) =
tanh t
(29)
k
m
or in dimensionless form:

v(t)
p mg = tanh(t? )

(30)

It can be shown that the above exact solution has the same limits as the scaled solution,
since tanh() = 1 and tanh(x) x for small values of x, i.e. the leading term in a Taylor
series expansion is x.
Finally, we can estimate the value of t? where all three forces interact by equating the
two limiting solutions. This gives:
t? 1

(31)

Thus we may summarize the asymptotes as follows:


v(t)
t? , t? << 1
v
(32)
v(t)
1, t? >> 1
v
The exact solution and asymptotes are plotted in the figure below. It is easily seen that
the transition or non-linear behaviour is in the region 0.5 < t? < 2.5 which is clearly of O(1).
It should be noted that much of the important information about this problem has been
found in less time than it will take you to integrate the exact solution.

O restart;
O ode:=diff(v(t),t)=g-k/m*v(t)^2;
d
kv t 2
ode :=
v t =gK
dt
m
O dsolve({ode,v(0)=0},v(t));
t gmk
tanh
gmk
m
v t =
k
O Vstar:=tanh(tstar);
Vstar := tanh tstar

(1)

(2)

ENGR 9420 - Engineering Analysis: Scaling and Asymptotic Analysis

15

(3)

O plot({Vstar,tstar,1},tstar=0..4,view=[0..4,0..1.5],labels=[`t*`,
`v*`],labelfont=[Helvetica,16],color=black);

1.5

1.0

v*
0.5

0
0

t*

O series(tanh(t),t);

1
2
Fig. 1 - Plot of the exact solution
for the(4)falling body.
tK
t C and
t C Oasymptotes
t
3
15
3

O tanh(infinity);

(5)

O tanh(0.01);

0.009999666680
(6)
Example 4 - Scaling with Multiple Equations:
Boundary Layer
Drag
In the final example we will examine a more complex system of partial differential equations from fluid dynamics. The classic problem of laminar boundary layer drag on a semiinfinite plate is defined as follows:

and
u

u v
+
=0
x y

(33)

u
u
2u
+v
= 2
x
y
y

(34)

where = / and are subject to the following boundary and initial conditions:

u(0, y) = U
v(0, y) = 0
u(x, 0) = v(x, 0) = 0
u(x, y ) = U

(35)
(36)
(37)
(38)

This problem is already drastically simplified from the full equations of motion. Despite
its significant simplification, it is still a complex problem to solve. The coupled equations
were originally solved using a reduction of variable technique using a similarity variable and
an asymptotic series. A topic we may address later.
Since this problem has no explicitly defined length scale, it will be a penetration type
problem requiring knowledge of the fluid boundary layer thickness which we will define as
= (x). The u velocity scale will be assumed to scale as u U and the dimensional scales
y and x x, which is the controlling variable. In fluid dynamics, the boundary layer
thickness and local shear stress (drag) on the surface will be functions of x along the surface.
In other words, their magnitude changes with local position downstream from the leading
edge of the plate.

ENGR 9420 - Engineering Analysis: Scaling and Asymptotic Analysis

16

We begin the scaling analysis by rewriting the the differential equations using the approximate scales:
U v
, 0
x

U
x

(39)

and

U U
U
U
U2 U2
,v 2
,
2

(40)
x

x x

Thus, since the left hand side of Eq. (40) has the same order of magnitude, we may
conclude that:
r
x
(41)

U
U

The local shear stress is defined from the constuitive relationship (or for this type of flow
from Newtons law of viscosity) which gives:

u
U
U
=
p x

(42)
y y=0

U
The above result is traditionally presented as a friction coefficient Cf defined similar to
Eq. (11):
Cf =

2
U
Rex

(43)

where Rex = U x/ is the dimensionless Reynolds number.


The analytical approach to solving the problem requires reducing the two partial differential equations into one non-linear ordinary differential equation for a new solution variable:
f f 00 + 2f 000 = 0

(44)

subject to appropriate boundary conditions (Schlichting, 1978). The above requires a full
numerical solution if all information about the velocity field is required, however an analytic
solution for the value of the shear stress at the wall was obtained using an asymptotic series
expansion which yields the solution for the above drag coefficient of:
0.664
(45)
Cf =
Rex
One can easily see that the effort to obtain the correct physical behaviour was quite
minimal, and if you were analyzing the problem experimentally or numerically, the trend
would be correct.

2.3

General Asymptotic Characteristics

In many problems found in the fields of transport phenomena, solid mechanics, and fluid
dynamics, there exists an asymptotic characteristic of a general solution for small and large

ENGR 9420 - Engineering Analysis: Scaling and Asymptotic Analysis

17

values of a particular parameter. For example, in the case of fluid flow, the first type of
behaviour which comes to mind is the flow of fluid at small and large Reynolds numbers.
However, in this type of situation, there often exists a transition region which may not be
smooth. However, many complex physical systems exhibit a smooth transition between two
asymptotic solutions. The smooth transition means that there are no discontinuities and no
sudden changes in slope within the transition region.
If we confine ourselves to a particular flow regime such as laminar flow, there exist
many fundamental problems which display smooth transition for flow from small to large
Reynolds number, from short to long tubes, for small to large Prandtl numbers, for small to
large Knudsen numbers, and in the case of mass transfer with reaction we often encounter
diffusion dominated versus reaction dominated flows.
Other asymptotes appear in steady and transient internal and external conduction, steady
conduction across a gas layer between two large parallel isothermal plates, radiation through
a highly porous substance between two large isothermal walls, contact between a rigid
sphere and a softer substrate, natural and forced internal and external convection. There
are many other examples from conduction, fluid flow, and mass transfer, which will be
considered later.
Since the various problems have different dependent and independent parameters and
different symbols are used, it is necessary to present the asymptotes and the method of
combining them in a general form based on dimensionless dependent and independent parameters.
The dimensionless dependent parameter is called and the independent parameter is
called . The functional dependence () can be relatively simple or quite complex depending
on the physical problem.
For now we will assume that the dimensionless parameter has two asymptotes corresponding to very small and very large values of the dimensionless independent parameter
:
(
0 = C0 m as 0

(46)
= C n as or 1
The asymptotes 0 and are based on analytical solutions and they consist of a constant
which is a positive, real number. The two constants are denoted as C0 as 0 and as C
for or 1. The exponents m and n can have the values such as 0, 1, 1/2, 1/4, 1/3
for example.
It is known from analysis, experiments and numerical work that frequently transitions
in a smooth manner between the two asymptotes.
The plots in Figures 2 (a-d) illustrate four basic, but different, trends. Figure 2a shows
both asymptotes 0 and increasing with increasing values of , and the solution is
concave upwards. For example, external forced and natural convection from single isothermal
convex bodies exhibit this trend. Figure 2b shows both asymptotes 0 and decreasing
with increasing values of , and the solution is concave upwards. For example, steady
conduction in spherical wall and external transient conduction from isothermal convex bodies
exhibit this trend. Figure 2c shows both asymptotes 0 and increasing with increasing
values of , and the solution is concave downwards. For example, internal forced and natural

ENGR 9420 - Engineering Analysis: Scaling and Asymptotic Analysis

18

convection within ducts and channels exhibit this trend. Figure 2d shows the asymptote 0
to be constant independent of and the asymptote decreases with increasing values of
, and the solution is concave downwards.

2.4

Composite or Blended Models

Many complex problems lend themselves to asymptotic analysis. For these problems, we may
develop a composite solution by considering only the asymptotic or limiting characteristics.
One may ask why we need only consider the limiting characteristics of a problem. It will
become more apparent as we proceed, but simply put, it is often easier to obtain exact
solutions for limiting conditions of a particular variable of interest, than it is to find a
solution which is exact over a wide range of that particular variable.
We will see from scale analysis, that we may easily determine the mathematical character
of asymptotes, and from the various solution methodologies in applied mathematics, we can
easily obtain exact or approximate solutions to these asymptotes. Once these asymptotes
are determined we may develop a composite model by combining them in one of two ways.
The asymptotes 0 and can be combined in a series or parallel manner depending on the
trends of 0 and with respect to .
If 0 > as 0, the solution is concave upwards, and the asymptotes are combined
in the following manner:
= [p0 + p ]1/p
(47)
If 0 < as 0, the solution is concave downwards, and the asymptotes are
combined in the following manner:
1
=



1
0

p


+

p 1/p
(48)

The parameter p is a fitting parameter whose value can be found in a simple manner.
To find a value of p we can select an intermediate value of = i for which (i ) is known
through analytical, experimental or numerical methods. For the intermediate value of we
can write
1/p
(i ) = [(C0 im )p + (C in )p ]
(49)
or
1
=
(i )



1
C0 im

p


+

1
C in

p 1/p
(50)

In both relationships the fitting parameter p is unknown. It can, however, be calculated by


numerical methods or by means of computer algebra systems such as Maple, Mathematica,
and Mathcad. In the event that a data set contains multiple values or contains scatter, we
may obtain p by minimizing the root mean square error (rms) between the data and model
predictions. In this case the composite model is a best fit of all available data.

ENGR 9420 - Engineering Analysis: Scaling and Asymptotic Analysis

(a)

(b)

(c)

(d)

19

Fig. 2 - Four Common Model Forms, Yovanovich (2003).


Often the approximate solution is presented in a form which is based on one of the two
asymptotes. If this is done, then the two models can expressed as:

0 [1 + ( /0 )p ]1/p
(51)
Series
=
[1 + ( / )p ]1/p

0 / [1 + (0 / )p ]1/p
Parallel
=
(52)
/ [1 + ( / )p ]1/p

0
Finally, it is often of interest to know the point of transition from one asymptote to the
next. This is is easily done using the asymptotes themselves. Given,
0 = C0 m

and

= C n

(53)

ENGR 9420 - Engineering Analysis: Scaling and Asymptotic Analysis

20

where C0 , C and m, n are constants, we define c as the transition point and solve:
0 =

or

C0 m = C n

(54)

Solving for c gives:



c =

C0
C

1/(nm)
(55)

This criterion can be used as an approximate measure of the transition point. For example, if we consider developing laminar flow in a tube, the criterion will be an estimate of the
dimensionless entrance length or the transition from developing flow to fully developed flow.
Example 5 - Asymptotic Model Construction
Returning to our third scaling example, we see that this problem is characteristic of Fig.
2c. The exact solution gives = v/v = 0.7616 when t? = 1. We may use this to find that
p = 2.545 in Eq. (39). Or in an approximate sense:
1
=

1
" 
# 2.545
2.545
1
+1
t?

(56)

Thus if we were limited in some sense in our efforts to obtain more than one point, we
can easily approximate the solution. This will become more apparent for those examining
physical behaviour using experiments or numerical solutions, where effort is more significant.

References
Arpaci, V., Conduction Heat Transfer, Addison-Wesley, 1966.
Arpaci, V. and Larsen, P., Convection Heat Transfer, Prentice-Hall, 1984.
Bejan, A., Convection Heat Transfer, Wiley, 2005.
Bejan, A.,Simple Methods for Convection in Porous Media: Scale Analysis and the Intersection of Asymptotes, International Journal of Energy Research, Vol. 27, pp. 859-874,
2003.
Bird, R., Stewart, Lightfoot, E., Transport Phenomena, Wiley Inc., New York, NY, 1960.
Churchill, S.W., Viscous Flows: The Use of Practical Theory, Butterworth-Heinemann,
Boston, MA, 1987.
Churchill, S.W. and Usagi, R., A General Expression for the Correlation of Rates of Transfer
and Other Phenomena, American Institute of Chemical Engineers, Vol. 18, pp. 1121-1128,
1972.

ENGR 9420 - Engineering Analysis: Scaling and Asymptotic Analysis

21

Leal, L.G., Laminar Flow and Convective Transport Processes: Scaling Principles and Asymptotic Analysis, Butterworth-Heinemann, Boston, MA, 1992.
Schlichting, H.V., Boundary Layer Theory, McGraw-Hill, New York, NY, 1979.
Yovanovich, M.M., Asymptotes and Asymptotic Analysis for Development of Compact
Models for Microelectronics Cooling, Keynote Paper, Semi-Therm, 2003.

Вам также может понравиться