Heat Transfer: Fundamentals of

Fundamentals Of
Heat Transfer
Theory and Applications
(Class Notes for ME 371)
Michael C. Wendl
Department of Mechanical Engineering
and School of Medicine
Washington University
Saint Louis, U.S.A.

(2005)
Fundamentals of Heat Transfer Theory and Applications
c
1999, 2003, 2005 by Michael C. Wendl
Permission is granted to copy, distribute and/or display this document under

the terms of the Attribution–NoDerivs–NonCommercial License, Version 1.0
or any later version published by Creative Commons. A copy of the license
is included in the section entitled “Creative Commons Public License”. For
more information, contact Creative Commons, 559 Nathan Abbott Way,
Stanford, California 94305, USA.
Non–lawyer summary: this license allows you to freely reproduce, distribute, and
display verbatim copies of this document, in electronic or print form, for non–
commercial purposes.
This document was implemented entirely with open–source, a.k.a. “free” software:
Typeset using LATEX 2 with graphics by Xfig and plots by Xmgrace, all running
on the Linux operating system (Debian distribution)
Document Version 2.1 (August 2005), see Appendix D for revision history
This document contains no color figures
Wustl–E67–371
Contents
Preface vi
Chapter 1. Introduction 1
1.1. Physical Mechanisms 2
1.2. Concept of Conservation of Energy 3
1.3. The Continuum Assumption 4
1.4. Absolute Versus Relative Temperature 6
Chapter 2. Elementary Heat Conduction 7

2.1. Fourier’s Law 7
2.2. Thermal Properties of Matter 8
2.3. The Conduction Equation 9
2.4. Boundary and Initial Conditions 13
Chapter 3. One–Dimensional Steady Conduction 15

3.1. General Solution 16
3.2. Circuit Analogy 17
3.3. Cylindrical Configurations 21
3.4. Heat Generation 23
3.5. Fin Analysis 27
3.6. Fin Performance Metrics 34
Chapter 4. Transient Conduction 37

4.1. Generalities 37
4.2. Lumped Capacitance Analysis 37
4.3. Applicability of Lumped Capacitance 39
4.4. Casting the General One–Dimensional Problem 42
4.5. Transient Analysis in One Dimension 45
4.6. The Similarity Technique 48
4.7. Multi–dimensional Transient Conduction 50
Chapter 5. Introduction to Convection 53

5.1. Boundary Layer Introduction 54
5.2. Governing Equations 57
5.3. Dimensionless Parameters 60
5.4. Reynolds–Colburn Analogy 63
Chapter 6. External Convection 66

iii
CONTENTS iv
6.1. Laminar Flow Over A Flat Plate 67

6.2. Karman–Pohlhausen Integral Solution 69
6.3. Empirical Correlations 74
Chapter 7. Internal Convection 77

7.1. Laminar Pipe Flow 77
7.2. The Case of Constant Heat Flux 80
7.3. The Couette Problem 85
Chapter 8. Natural Convection 90

8.1. Wall Bounded Convection on a Vertical Flat Surface 91
8.2. Dimensionless Formulation 93
8.3. Similarity Solution for the Vertical Plate 94
Chapter 9. Heat Exchangers 99

9.1. The Thumbnail Diagram 100
9.2. Overall Heat Transfer Coefficient 102
9.3. LMTD Analysis 104
9.4. Correction Factors for Complex Configurations 109
9.5. –NTU Analysis 110
Chapter 10. Introduction to Radiation 116

10.1. Solid Angle and Radiative Quantities 117
10.2. Blackbody Radiation 122
10.3. Radiation Characteristics of Real Surfaces 126
10.4. Kirchhoff’s Law and Gray Surfaces 130
Chapter 11. Radiation Exchange 134

11.1. The View Factor 134
11.2. View Factor Algebra 136
11.3. Blackbody Radiation Exchange 141
11.4. Exchange Among Diffuse Gray Surfaces 141
11.5. The Graybody Matrix Problem 145
11.6. Additional configurations 147
Appendix A. Transient Conduction Example: The Homogeneous

Cooling Problem 149
A.1. Separation of Variables Method 149
A.2. Solution Procedure 151
A.3. Determining Mode Coefficients 152
A.4. Example: The Unit Initial Condition 154
Appendix B. Laminar Forced Convection in a Pipe 156

B.1. Volume Flow Rate 156
B.2. Integration of Mean Temperature Equation 156
CONTENTS v
Appendix C. Blackbody Radiation 158

C.1. Wien’s Displacement Law 158
C.2. Radiation Functions 159
Appendix D. Document History 161
About the Author 162
Creative Commons Public License 163
Bibliography 166
Index 170
Preface
Heat transfer is fundamentally important in almost all aspects of en-

gineering, physics, and the life sciences. Issues related to this branch of
thermodynamics arise in such varying circumstances as aerospace, biology,
consumer products, machinery, medical devices, geo–environmental applica-
tions, chemical processes, and nano–technology. There are many excellent
texts that introduce this subject and I do not consider this assembly of
notes to be any sort of replacement. This merely represents a collection of
concepts I consider to be the most valuable in a one–semester introductory
course. The material is therefore intended to amplify and augment that
which appears in your primary course textbook. In large part, we follow
Incropera and Dewitt (2002) in terms of the way material is arranged. A
sufficient familiarity with mathematics is presumed, especially calculus and
its theorems (e.g. the Chain Rule), linear differential equations, linear al-
gebra, and simple combinatorics. Moreover, much of this material depends
upon the understanding of a basic level of fluid mechanics (for example as
covered by an introductory course), especially the concepts of conservation
of mass and momentum. Problems and examples are minimal, since this
volume serves as a collection of notes rather than a complete text. Sup-
plementary margin notes appear at appropriate places suggesting example
problems from Incropera and Dewitt (I & D) that could be discussed in
class.
The style of presentation leans toward the theoretical and mathematical
side of the spectrum rather than the empirical side. My experience is that
instilling good mathematical habits early in an undergraduate career is of
significant benefit. Moreover, this allows us not to ask readers to take too
much on faith. For the most part, we shall prove everything from first
principles so that the student can derive a truly fundamental understanding
of the material. Where formal proofs are beyond the scope of what might
be presented in an introductory format, we will cite appropriate sources.
Phenomena which cannot be presented according to first principles, e.g.
some aspects of convection, turbulence, etc. will be described in terms of
a few empiricisms, but will largely be left for in–class discussion by the
instructor. There is no mention of English units of measurement. Where
units are necessary, we work exclusively in the International System (SI).
Michael C. Wendl
December 2003
vi
CHAPTER 1
Introduction
Heat Transfer can be described as the process of energy transmission due

to a gradient in temperature T , which is a measurable quantity 1.1. Thus,
determining the temperature distribution for a given problem is the desired
solution since other quantities of interest, such as heat flux can be derived
from it. The subject of heat transfer, which deals with non–equilibrium
processes, is essentially an extension of an introductory course in thermo-
dynamics, which is usually limited to equilibrium states. For example, a
typical problem in thermodynamics might be to determine the final equilib-
rium T when an annealed steel machine part is quenched in a vat of water.
The extension of this problem in heat transfer might be to find the rate of
cooling of the steel (T as a function of time and location), which is required
to predict the resulting hardness in various regions of the part. What are a
few common examples and associated problems and/or analyses?
• Automotive cooling system: implement a design such that power-
plant and working fluid temperatures remain in a specified operat-
ing range while increasing overall engine efficiency (Taymaz et al.,
2003)
• Solar power generation: design and optimize hardware for maximal

efficiency in heating a working fluid (Odeh et al., 2000)
• HVAC1.2: size required systems for ultra–high–rise office towers and

predict resulting temperature distribution in offices and common
spaces (Sinha et al., 1998)
• Biomechanics: analyze body temperature increases as a function of

frictional heating by physical exertion (Bergmann et al., 2001)
• Electronics: increase effectiveness of laptop computer cooling sys-

tems in response to increasing CPU power dissipation (Pastukhov
et al., 2003)
1.1Temperature is a surprisingly difficult concept to define in terms of first principles.

This is usually deferred to graduate courses in thermodynamics where it arises from con-
siderations of energy and entropy. Here, we will take the simplistic, but typical approach
of understanding temperature merely as a quantity that indicates thermal energy.
1.2Heating, Ventilation, and Air Conditioning
1
1.1. PHYSICAL MECHANISMS 2
• Medicine: analyze effectiveness of hyperthermia treatments for can-

cer (Alexander, 2003)
• Aerodynamics: design fuselage to limit effects of aerodynamic heat-

ing (Lee et al., 2003)
In summary, it is difficult to find a problem totally devoid of heat transfer
considerations.
1.1. Physical Mechanisms

There are basically 3 main modes of heat transfer that we’ll study: con-
duction, convection, and radiation.
• Conduction occurs through microscopic mechanisms, such as lattice
vibrations and electron movement. There is no bulk motion of the
medium — it is strictly a diffusion process. Example: the heat felt
when holding one end of a long copper bar in a open fire.
• Convection relies on bulk motion of the medium. Example: cooling

effect realized by standing in front of a fan after sprinting the 400m.
• Radiation occurs by way of electromagnetic phenomena as described

by Maxwell’s equations and does not necessarily depend upon a
medium (energy transfer can occur across a vacuum). Example:
heating sensation felt by bare skin in the noon sun.
In the “real world”, a problem often relies on all three modes. For exam-
ple consider the M–16 rifle in automatic fire mode, so that the temperature
in the barrel is approximately constant (Fig. 1.1). Heat from expanding
barrel
conduction
gas tube
handguard
radiation
heat shield
convection
Figure 1.1. M–16 cross section showing barrel, gas tube, heat
shield, and outer handguard.
combustion gas in the bore and friction between the bullet and the bore is
conducted through the barrel to its outer surface. The barrel is cooled by
1.2. CONCEPT OF CONSERVATION OF ENERGY 3
convected air as heat is absorbed, causing a change in density and subse-

quent air movement. This heat is carried away as the air leaves through
holes in the top of the handguard. The barrel is further cooled by radiating
energy to inner cooling surface inside the handguard.
Analyses of such real–world problems are usually quite difficult. For
example, geometry is complicated and, as we shall see, the full governing
equations are non–linear partial differential equations 1.3. Thus, we invari-
ably require theoretical, experimental, and computational procedures. How-
ever, for this course we’ll look primarily at idealized “model” systems that
can be solved in closed form.
1.2. Concept of Conservation of Energy

The cornerstone of heat transfer is the law of conservation of energy,
which is described in terms of a specific volumetric space called the con-
trol volume (CV ) and a bounding surface called the control surface (CS)
that encloses the CV (Fig. 1.2). In later sections, we will also see that
E gen E stored
CV
E in CS
E out
Figure 1.2. Control volume and control surface schematic.

the conservation laws for mass and momentum arise in certain situations.
Conservation of energy can be stated symbolically as
dEstored
(1.1) = Ėstored = Ėin + Ėgen − Ėout ,
dt
where E represents energy and the dot notation connotes a rate process 1.4.
In other words, the rate of energy increase in the control volume is equal to
the rate at which it is generated internally 1.5 plus the rate at which it comes
into the control volume minus the rate at which it leaves. This energy can
be of various types — usually we mean thermal energy, but it can also be
mechanical work. The terms Ėin and Ėout describe phenomena occurring at
(across) the CS, while Ėgen is a volumetric term associated with the CV . I&D Ex. 1.3
pp 17
1.3Recall that these same complexities were present in analyzing fluid mechanics
problems.
1.4We have not said anything about describing these Ė quantities yet.
1.5
Energy can be generated by a variety of means, for example nuclear decay. One
of the most common situations is electrical dissipation via current flowing through a wire
having non–zero electrical resistance.
1.3. THE CONTINUUM ASSUMPTION 4
Note that the equation integrated over any time period ∆t must also hold
true:
(1.2) ∆Estored = Ein + Egen − Eout .
There will be occasions where the conservation of energy principle will
be required at a surface, for example as a boundary condition at a solid–gas,
liquid–gas, or solid–liquid interface (Fig. 1.3). The thickness of this interface
fluid
energy convected away from surface
energy conducted to surface

t solid
Figure 1.3. Conservation of energy at an interface. Dotted

lines indicate control surfaces.
t is taken to be vanishingly small, so that the control surface encloses no

mass or volume. The storage and generation terms are therefore irrelevant
and the conservation of mass equation reduces to
(1.3) Ėin = Ėout .
Configurations will typically involve conduction on the solid side and con-
vection on the liquid or gas side, so that Eq. (1.3) reduces to the statement
that conduction heat transfer equals convection heat transfer at a surface.
1.3. The Continuum Assumption

We will be concerned with heat transfer in both solids and fluids. In
fluid mechanics, the so–called continuum assumption played an important
role in how theory was constructed, especially for gases. For liquids and
solids, the continuum assumptions is usually taken for granted. Let us have
a brief review of this concept.
If we were to look at a mass of fluid at the microscopic level, what we
would see are individual molecules interacting with each other. We are not
actually interested in the behavior of individual molecules for our engineer-
ing applications. Rather, we want to understand the overall (or macroscopic)
behavior of the system as a whole. That is, it is the macroscopic proper-
ties such as density, temperature, or pressure drop that are of engineering
interest. What we are doing from the mathematical perspective is taking
averages over small elemental volumes. These volumes must be large enough
such that they contain enough molecules at any instant in time to yield a
statistically significant average. Yet they must also be small enough so that
the statistical average does not vary over the volume — it should be a con-
stant. If these conditions are met, the properties will have definite point
1.3. THE CONTINUUM ASSUMPTION 5
values. In other words, they will be continuous functions of space and time. the continuum
This is the so–called continuum assumption. assumption
To illustrate this concept, consider the density ρ of a fluid. A region
of fluid is shown in Fig. 1.4(a). We desire to find the density at a point C
C (x ,o y ,oz ) o
δm ρ
δV defined
here
x δV
(a) (b)
z
Figure 1.4. (a) Defining density as a point function according

to the continuum assumption. Large region has volume V and mass
m, while the point region C has volume δV and mass δm. (b) Value
of density according to size of δV .
located at coordinates (x0 , y0 , z0 ), where density is defined simply as mass

per unit volume. For the larger region, the average density within its volume
V would simply be ρ = m/V . Since this is an average, we would expect ρ
to vary within V since V is large. Specifically, the average ρ would not be
expected to be equal to the density at point C. To determine the density at
C, we compute δm/δV . The question governing the continuum assumption
is how small or large is δV ?
Let us plot δm/δV as a function of δV as in Fig. 1.4(b) starting from
large values of δV . The average density tends to approach an asymptotic
value as the volume is lowered to the point such as to enclose only homoge-
neous fluid in the immediate neighborhood of point C. Below this point, the
volume is small enough such that the number of molecules contained at a
given instant of time is not constant in the average sense. Rather, the num-
ber varies erratically, causing a corresponding variation in mass δm, which
in turn leads to an erratic value for density δm/δV . There is clearly a lower
limit for δV which restricts the continuum assumption. Luckily, this limit
is very small compared to the scales we are interested in for engineering
calculations1.6. Therefore, we are permitted mathematically to approximate
δV → 0 relative to the size of our engineering length scales. By this, we
mean that the volume approaches but does not reach 0. Therefore, the con-
tinuum assumption will be valid for our problems of interest. Panton (1984)
contains a more detailed discussion.
1.6For example, a cubic meter of air at standard temperature and pressure
(15 C, 101.3 kP a) contains about 2.5 × 1025 molecules. Therefore, the number of molecules
in a volume about the size of a grain of sand, about 10−12 cubic meters, would be 2.5×1013 ,
which is large enough to ensure that the average mass would be constant.
1.4. ABSOLUTE VERSUS RELATIVE TEMPERATURE 6
Since the (x0 , y0 , z0 ) location of point C is arbitrary, ρ at any point in

the region could likewise be determined. If ρ were computed simultaneously
for all δV in the fluid, we could formulate an expression as a function of
the location and time of the measurements, i.e. ρ = ρ(x, y, z, t). Thus,
the continuum assumption leads naturally to functional definitions for the
properties of interest.
1.4. Absolute Versus Relative Temperature

We will strictly use the International System (SI), with the most common
quantities and units being given in Table 1.1. It is important to remember
Table 1.1. Some common SI units
Quantity Unit
length meter (m)
mass kilogram (kg)
time second (s)
temperature kelvin (K)
that temperature can be interpreted in an absolute sense, that is, relative

to absolute zero, or in a relative sense. The unit of kelvin (K) refers to the
absolute scale, while celsius1.7 (o C) is a relative quantity. Their relationship
is given by
(1.4) Tkelvin = Tcelsius + 273.15 .
Some problems are intrinsically formulated upon the premise of absolute
temperature, for example the Ideal Gas Law given by
P = ρRT ,
where P is pressure, ρ is fluid density, and R is the ideal gas constant. Sim-
ilarly, we will see the same requirement for radiation problems, where the
absolute temperature raised to the fourth power becomes important. Us-
ing relative temperature is incorrect for these problems! Conversely, other
problems, for example in convection, depend only on temperature differ-
ences, for which it is easy to verify that both relative or absolute units are
appropriate1.8.
1.7This is also referred to as centigrade in some countries.

1.8For example, we can express an absolute temperature difference ∆T as
∆T = Tk,2 − Tk,1 = (Tc,2 + 273.15) − (Tc,1 + 273.15) = Tc,2 − Tc,1 ,

which is clearly equivalent to the corresponding relative temperature difference.
CHAPTER 2
Elementary Heat Conduction
Here we will introduce the basic concepts of conduction heat transfer.
2.1. Fourier’s Law

In chapter 1, we stated that heat conduction is a diffusion process
that occurs via molecular mechanisms. The fundamental governing law is
Fourier’s Law of Heat Conduction. In one dimension this is 2.1
dT
(2.1) q 00 = −k ,
dx
where q 00 is heat flux, k is thermal conductivity, and dT /dx is tempera-
ture gradient along an independent coordinate x. Thus, temperature is a
function of position, and is written in the 1–D case as T (x). Eq. (2.1) is
a phenomenological law derived from numerous experimental observations
rather than first principles. It holds for all the physical configurations we
shall study here, but additional terms are required for more complicated ma-
terials, for example some animal tissues 2.2. Heat flux is clearly a directional
quantity. The general three–dimensional form written in vector notation is

00 ∂T ∂T ∂T
(2.2) q~ = −k ∇T = −k î + ĵ + k̂ .
∂x ∂y ∂z
Although the negative sign in Eqs. (2.1) and (2.2) looks rather strange,
it is required for consistency of the physics. Heat energy is conducted along
a temperature gradient. That is, energy flows from a high temperature
region to one of lower temperature. For example, in Fig. 2.1 energy flows
in the positive x direction, therefore q 00 must be a positive quantity. The
temperature gradient is
dT T2 − T 1
= ,
dx x2 − x 1
which is clearly a negative quantity, since x 2 − x1 is positive, but T2 − T1 is
negative. In fact, T2 − T1 must be negative, otherwise the resulting gradient
2.1
We follow the notation of Incropera and Dewitt (2002) where double–prime in-
dicates a flux, whereas lack thereof indicates the quantity has been integrated over an
area.
2.2This situation is quite analogous to Newtonian versus Non–Newtonian fluids. Re-
call that for a Newtonian fluid, shear stress and rate of strain are linearly related, where
the constant of proportionality is the viscosity. For Non–Newtonian fluids, additional
terms result in much more complicated relationships.
7
2.2. THERMAL PROPERTIES OF MATTER 8
T1
T2
x
x x2
1
heat flow
Figure 2.1. Heat is conducted along a negative gradient from

high temperature toward low temperature, shown in the positive
direction here.
would not allow energy to flow as shown. Since k is defined as positive, the
leading negative sign is clearly required for consistency. In this sense, the
temperature gradient is very similar to the pressure gradient in pipe flow:
movement proceeds along a negative gradient.
2.2. Thermal Properties of Matter

We mentioned the thermal conductivity k as a material parameter in
Fourier’s Law. Two other quantities are important to the problems we wish
to study: the volumetric heat capacity, ρc p , and the thermal diffusivity, α.
Unites for these quantities are list in Table 2.1. Other properties will arise
Table 2.1. Thermal properties
Property Notation Units

thermal conductivity k W/(m K)
volumetric heat capacity ρ cp J/(m3 K)
thermal diffusivity α = k/(ρ cp ) m2 /s
as well, including many of those associated with fluid mechanics problems

and surface properties for radiation. These will be introduced as needed.
Recalling that the physical basis for conductivity is at the molecular
and atomic levels2.3, we would suspect that solids would generally have the
highest conductivities, followed by liquids, and finally gases. This is in fact
the case and is largely due to differences in the molecular spacings (Fig. 2.2).
As mentioned, the desire for heat transfer problems is to determine the
temperature distribution for a given problem. The subject of materials
2.3At the atomic levels, this is primarily associated with the concept of free, or mobile
electrons, especially for metals.
2.3. THE CONDUCTION EQUATION 9
pure metals
alloys
non−metallic solids
liquids
gases
0.01 0.1 1 10 100 1000

thermal conductivity (W / m K)
Figure 2.2. Approximate ranges of thermal conductivity for

various classifications of matter.
science deals much more comprehensively with such issues. Actual values are
available in many references (e.g. Incropera and Dewitt, 2002, Appendix A).
2.3. The Conduction Equation

It was mentioned in the opening sentences of Chapter 1 that the goal of
heat transfer is to determine the temperature distribution. In the general
case, this will depend on all three spatial dimensions and on time, so that
T = T (x, y, z, t). We introduced Fourier’s Law in Eq. (2.1), but this alone
does not provide a foundation for calculating T . Instead, we must base
our theoretical framework on the conservation law for energy, which was
introduced only at a conceptual level in Chapter 1.
We extend the conceptual treatment using the classic differential ap-
proach; The resulting equation will be valid for every differential point in
a problem domain. This is characteristic of differential formulations: the
conservation law must be satisfied simultaneously for all (x, y, z, t). In this
sense, the differential formulation is exact, however, it typically presents
a more challenging mathematical situation than the integral approach 2.4.
We define an infinitely small control volume 2.5 with a properties of inter-
est defined in the center. Any properties depending upon a flux can be
extrapolated to the boundaries using truncated 1–term Taylor series 2.6.
Define the differential element according to the volume δx × δy × δz
(Fig. 2.3). According to the concept of conduction, the material is either a
solid, or a non–moving fluid. We formulate all quantities at a given instant
2.4The integral approach is not typically as important in heat transfer as in fluid
mechanics, but will be introduced later for certain configurations.
2.5Once again, the concept of “infinitely small” remains restricted to the continuum
assumption introduced in Chapter 1. A term more suitable is then perhaps “differential”,
which connotes, small enough, but not too small.
2.6Terms of second–order and higher can all be neglected since they involve products
of the differential quantities, e.g. (δx)2 , which are exceedingly small.
y
qt
q
f
δy
.
q q, E st q
l r
.
E gen
x
q
n
q δz
b
δx
z
Figure 2.3. Differential volume showing energy generation

Ėgen and energy storage Ėst terms in the center and heat fluxes at
the boundaries.
of time: the energy generation and storage terms, Ėgen and Ėst , represented
as volumetric entities in the element, and all heat conduction terms at the
boundaries. Expressing these terms as 1–term Taylor series expansions with
respect to their associated values defined in the center, we find
∂qx δx
(2.3) q l = qx −
∂x 2
∂qx δx
(2.4) q r = qx +
∂x 2
∂qy δy
(2.5) q b = qy −
∂y 2
∂qy δy
(2.6) q t = qy +
∂y 2
∂qz δz
(2.7) q f = qz −
∂z 2
∂qz δz
(2.8) q n = qz +
∂z 2
Energy generation is simply
(2.9) Ėgen = q̇ δx δy δz ,
where q̇ is the per unit volume rate of generation 2.7. Finally, the rate of
change of thermal energy stored in the element can be expressed as
∂T
(2.10) Ėst = ρ cp δx δy δz .
∂t
These terms can be used directly in the conceptual conservation law
in Eq. (1.1) on pp. 3, where the rate of energy entering the element is
Ėin = qb + qf + ql and the rate of energy leaving is Ėout = qn + qr + qt . This
operation yields
∂T
(2.11) ρ cp δx δy δz = q̇ δx δy δz + qb + qf + ql − (qn + qr + qt ) ,
∂t
which becomes

∂T ∂qy δy ∂qz δz
(2.12) ρ cp δx δy δz = q̇ δx δy δz + qy − + qz −
∂t ∂y 2 ∂z 2

∂qx δx ∂qz δz ∂qx δx ∂qy δy
+ qx − − qz + − qx + − qy + ,
∂x 2 ∂z 2 ∂x 2 ∂y 2
after substituting Eqs. (2.3) through (2.6) for the heat conduction terms.
We then cancel and combine terms appropriately to get

∂T ∂qx ∂qy ∂qz
(2.13) ρ cp δx δy δz = q̇ δx δy δz − δx + δy + δz .
∂t ∂x ∂y ∂z
Eq. (2.13) represents the limit in terms of what can be derived strictly
from theory. It cannot be solved, because T and q are both unknowns 2.8.
However, we have additional relationships between T and q in the form of
Fourier’s Law! In light of Eq. (2.2), we can express the heat terms as
∂T
(2.14) qx = −k δy δz
∂x
∂T
(2.15) qy = −k δx δz
∂y
∂T
(2.16) qz = −k δx δy
∂z
which can be substituted into Eq. (2.13) to obtain

∂T ∂ ∂T
(2.17) ρ cp δx δy δz = q̇ δx δy δz − −k δy δz δx
∂t ∂x ∂x

∂ ∂T ∂ ∂T
+ −k δx δz δy + −k δx δy δz .
∂y ∂y ∂z ∂z
2.7This is the unfortunate notation used in Incropera and Dewitt (2002), which can
be easily confused with heat conduction q and heat flux q 00 .
2.8We assume that heat generation q̇ would be prescribed, or would be measurable
for a problem.
All volume terms δx×δy ×δz cancel and all double–negatives cancel, leaving

∂T ∂ ∂T ∂ ∂T ∂ ∂T
(2.18) ρ cp = q̇ + k + k + k
∂t ∂x ∂x ∂y ∂y ∂z ∂z
as the general three–dimensional time–dependent form of the heat conduc-
tion equation. We have explicitly derived this equation in the Cartesian
(rectangular) coordinate system, which will be sufficient for many of the
problems we will study. However, in some instances we will require the
cylindrical and spherical forms of the conduction equation. Most reference
texts address these cases (e.g. Incropera and Dewitt, 2002).
Eq. (2.18) expresses conservation of energy in terms of the single de-
pendent variable of temperature T . We can solve this in principle because
there is one equation and exactly one unknown. In mathematical terms,
this is a linear partial differential equation, since T does not appear in the
form of any products of itself, or its derivatives 2.9. Thus, these problems
are amenable to theoretical treatments. Non–linear systems (which do exist
in many problems) typically require advanced theory and/or computational
treatment.
You should have a conceptual understanding of what the terms in this
equation mean physically. For example, if we multiply ∂/∂x (k ∂T /∂x) in
Eq. (2.18) by dx and compare to terms in Eq. (2.11), we see

∂ ∂T
(2.19) k dx = ql00 − qr00
∂x ∂x
is essentially the net heat flux into the control volume in the x direction.
Similar interpretations hold for the corresponding terms in the y and z
directions. Thus, Eq. (2.18) specifies that the net rate of conduction into
the control volume plus the rate of generation is equal to the rate of change
of energy stored at any point in the domain.
Notice that we have not made any assumptions regarding the conduc-
tivity k in Eq. (2.18). However, if k is a constant, it can be moved outside
of the derivatives, so that
1 ∂T q̇ ∂2T ∂2T ∂2T
(2.20) = + + + ,
α ∂t k ∂x2 ∂y 2 ∂z 2
where q̇ is again the energy generation rate per unit volume (W/m 3 ) and α
is the thermal diffusivity. Mathematically, this is more straightforward to
work with than Eq. (2.18). Further simplifications are possible, for example
2.9Actually, the assumption of linearity of the overall equation also depends on the
heat generation term q̇ as being itself linear. However, if q̇ depended upon T in some
non–linear fashion, e.g. as a power of T , then the equation would no longer be linear.
We shall not discuss such complicated cases here, as the theoretical treatment is quite
difficult.
2.4. BOUNDARY AND INITIAL CONDITIONS 13
if conduction is also steady (stationary), Eq. (2.20) reduces to

q̇ ∂2T ∂2T ∂2T
(2.21) + + + = 0.
k ∂x2 ∂y 2 ∂z 2
Moreover, if energy generation is absent, we obtain
∂2T ∂2T ∂2T
(2.22) + + = 0,
∂x2 ∂y 2 ∂z 2
which can also be written ∇2 T = 0 using the Laplacian operator
∂2 ∂2 ∂2
(2.23) ∇2 = + + .
∂x2 ∂y 2 ∂z 2
Finally, if temperature varies only in one coordinate direction, we obtain
d2 T
(2.24) = 0,
dx2
where the regular derivative d has replaced the partial derivative ∂ since
temperature is a now a function of only a single variable, i.e. T = T (x).
2.4. Boundary and Initial Conditions

The various forms of the conduction equation we have just discussed
govern the physics of how heat is conducted in the interior of some pre–
defined domain. However, this still does not completely define the problem.
We must also have some specification of temperature (or its derivatives)
on the boundaries of the domain — these are called boundary conditions.
Moreover, if the problem is also unsteady, we must also know the initial
condition for the problem. That is, we must have the value of T (or its
derivative) at some specific time.
As you may remember from differential equations, there are two required
boundary conditions for each spatial coordinate in the problem since these
terms are in the form of second derivatives. Conversely, there is only one
initial condition needed for any unsteady problem because the temporal
term appears as a first derivative. Almost always, the latter takes the form
of prescribing T itself at some instant of time, usually by convention at
t = 0, which can be written in a few different ways
(2.25) T (~x, t) |t=0 = T (~x, 0) = T0 (~x) ,
where ~x is one, two, or three spatial coordinates, depending upon the prob-
lem, and T0 is a function of T that gives the temperature distribution
throughout the domain at the particular instant of time.
Boundary conditions, on the other hand, can take a number of forms,
depending upon the physics of the problem. The most obvious case is similar
to the initial condition, where a specific value of T is specified at a coordinate
location. For example we could write
(2.26) T (~x, t) |x=0 = Ts ,
2.4. BOUNDARY AND INITIAL CONDITIONS 14
where Ts is the temperature at the surface coinciding with the coordinate

location x = 0. Eq. (2.26) is commonly known as a boundary condition of
the first kind, or a Dirichlet boundary condition, and is common to heat
transfer problems. For example, if the boundary is in contact with a medium
undergoing a phase change, the constant temperature boundary condition
is a good model.
A boundary condition of the second kind, or Neumann boundary con-
dition, involves the derivative of T . For instance, Fourier’s Law in Eq. (2.1)
on pp. 7 is a ready–made Neumann boundary condition when applied at a
physical boundary, e.g.

∂T
(2.27) −k = qs00 .
∂x x=0
In other words, the heat flux qs00 at the x = 0 boundary is related to the
temperature gradient ∂T /∂x at x = 0 as prescribed by this equation 2.10.
The special case of the adiabatic (perfectly insulated) boundary is given by

∂T
(2.28) = 0.
∂x x=0
Boundary conditions of the third type, also called Robbins boundary
conditions, specify a balance between energy conduction at the boundary
and the rate at which this energy is convected away. This case was described
conceptually in Fig. 1.3 and Eq. (1.3) on pp. 4 in Chapter 1. The form is
something of a combination of the first two types of boundary condition in
that it contains both temperature itself and its derivative

∂T
(2.29) −k = h (T∞ − T |x=0 ) ,
∂x x=0
where T∞ is a temperature characterizing the fluid which convects heat
away and h is a convection coefficient which we discuss at length in later
chapters2.11.
2.10Note that we have replaced the regular derivative d with the partial derivative
∂ to indicate that the temperature may be multi–dimensional, whereas we had not yet
introduced this concept in Eq. (2.1).
2.11For the moment, we will assume that h is known for problems involving Eq. (2.29).
Later, we will see that convection heat transfer is largely the study of determining h.
CHAPTER 3
One–Dimensional Steady Conduction
Here, we will examine problems where conduction occurs along only one
primary dimension, say the x direction, and where the problem is steady.
Recall, that Eq. (2.24) on pp. 13 governs this situation, i.e.
d2 T
(3.1) = 0.
dx2
This equation is useful for conceptual understanding of conduction, but is
also a good engineering approximation for cases where there is dominance in
one dimension. For example, consider the two–dimensional domain shown
in Fig. 3.1 and assume that the temperature on the boundaries is given by
the equations
y1
x
(0, 0) x1
Figure 3.1. Two–dimensional domain with one dimension

much longer than the other, i.e. x1 y1 .
T |x=0 = T |y=0 = 0 and T |x=x1 = T |y=y1 = T1 .

According to discussions in Chapter 2, particularly Eq. (2.22) on pp. 13,
we could deduce that this problem is governed formally by the conduction
equation
∂2T ∂2T
(3.2) + = 0.
∂x2 ∂y 2
However, if we look at the relative sizes of temperature gradients using a
simple finite–difference approximation,
∂T T1 − 0 ∂T T1 − 0
≈ and ≈ ,
∂x x1 − 0 ∂y y1 − 0
it is clear that gradients in the x direction will be small compared to those in
the y direction. Thus, such a problem could be satisfactorily approximated
as one–dimensional.
15
3.1. GENERAL SOLUTION 16
3.1. General Solution

The differential equation for the steady one–dimensional heat conduction
problem is trivial to solve by direct integration. Integrating Eq. (3.1) once
gives
dT
(3.3) = C1
dx
and integrating once again yields
(3.4) T = C 1 x + C2 ,
where C1 and C2 are constants to be determined using the boundary con-
ditions. An interesting physical observation can be gleaned for all one–
dimensional problems, regardless of their boundary conditions. Specifically,
Eq. (3.3) shows that the temperature gradient in the domain is a constant,
so that the temperature variation given by Eq. (3.4) is linear. Thus, the
temperature distribution is a line whose endpoints are the boundary temper-
atures and, according to Fourier’s Law, the heat flux is constant throughout
the domain.
Example 3.1:
The heads of the Titanic’s 29 coal–fired steam boilers can be modeled as
simple circular plates having a radius of r = 2.4m. Assuming a b = 0.05m
plate thickness, a thermal conductivity of k = 60W/mK, a hot–side steam
temperature of Th = 400K, and a stoker room temperature of T c = 310K,
determine the heat transfer through one of the heads.
x
r = 2.4
400 K 310 K
b = 0.05
Figure 3.2. Frontal and planar view of Titanic steam boiler

head.
Gradients in the radial direction will be small compared to gradients

through the head since the dimensions differ by about a factor of fifty, there-
fore, we can treat this problem one–dimensionally. According to the figure,
the boundary conditions are T (0) = T h and T (b) = Tc . Using the gen-
eral solution in Eq. (3.4), we can solve for the constants as C 2 = Th and
3.2. CIRCUIT ANALOGY 17
C1 = (Tc − Th )/b. This gives a temperature distribution of

Tc − T h
T (x) = x + Th .
b
The total heat transfer is simply the heat flux times the area of the boiler
plate. From Fourier’s Law, we obtain
dT 310 − 400
q = −k A = −kπr 2 C1 = −60 · π · 2.42 = 1.95 M W .
dx 0.05
Thus, the heat transfer is almost 2 million Watts. ♦♦♦
3.2. Circuit Analogy

The linear nature of the solution in Eq. (3.4) suggests an analogy between
heat conduction and electrical conduction. Recall that in a linear circuit,
current I flows along gradients in voltage ∆V = V 2 −V1 through a resistance
R (Fig. 3.3). These entities are related through the simple linear circuit
R
V1 V2
Figure 3.3. Linear circuit showing current flowing through a

resistor along a voltage gradient.
equation
(3.5) ∆V = I R
Looking more closely at Eqs. (3.3) and (3.4), we see that the linear drop
in temperature gradient coupled with Fourier’s Law allows us to write the
conduction heat transfer in the exact form as
dT T2 − T 1 T1 − T 2 kA
(3.6) q = −k A = −k A = kA = ∆T ,
dx L L L
where the generic problem is defined according to boundary conditions
T (0) = T1 and T (L) = T2 . A little algebra shows the equivalent relationship
L
(3.7) ∆T = q = q Rt ,
kA
whose form is identical to Eq. (3.5) where ∆T is analogous to voltage (driv-
ing potential), q is analogous to current (what “flows” in the circuit), and
Rt is a thermal resistance. More specifically, R t in Eq. (3.7) is the thermal
resistance of conduction, for which we can use the slightly more descriptive
notation
L
(3.8) Rt,cond = .
kA
In Eq. (2.29) on pp. 14, we introduced the idea of convection heat trans-
fer as quantified by a convection coefficient h. This equation is a form of
Newton’s Law of Cooling
(3.9) q = h A ( T |x=0 − T∞ ) ,
which shows that this form of convection is also linear. Again, some simple
algebra allows us to arrange this equation as
1
(3.10) T |x=0 − T∞ = q .
hA
Or, put more directly
(3.11) ∆T = q Rt,conv ,
where
1
(3.12) Rt,conv =
hA
is the thermal resistance for convection.
Are such analogs useful for practical problem solving? To answer this
question, refer back to the idea at the beginning of this chapter where we
discussed the pure one–dimensional problem versus the multi–dimensional
problem for which the size of the terms allowed us to use a one–dimensional
approximation. The former case is easily analyzed by solving Eq. (3.1) and
applying boundary conditions. But how can we treat the latter situation,
for example with respect to a problem having composite layers all of differ-
ent thermal conductivities as shown in Fig. 3.4? Here, the top and bottom
R2
2 R1 R4
T1 1 T2
4
T1 R3 T2
3
L1 L 2,3 L4
Figure 3.4. One–dimensional problem in a composite domain

along with corresponding thermal circuit model.
boundaries are insulated, and temperature variation is only important in the

x direction. So the problem is one–dimensional, but cannot conveniently be
treated using Eq. (3.1). However, it readily yields to the circuit analogy.
Specifically, we need only reduce the analogous circuit diagram to the sim-
plest possible arrangement using the standard rules of circuit analysis:
• Resistors in series have additive resistance, e.g. for N resistors in

series
Rt,total = Rt,1 + Rt,1 + · · · + Rt,N
• Resistors in parallel are governed by the reciprocal relationship,

e.g. for N resistors in parallel
1 1 1 1
= + + ··· +
Rt,total Rt,1 Rt,1 Rt,N
These resistances are of the appropriate variety: conductive as given by
Eq. (3.8) or convective, as in Eq. (3.12). Nodes function exactly as in cir-
cuit analysis as well. For example, the temperature at the interface between
two layers can be computed from the corresponding nodal value in the cir-
cuit diagram. Thus, the entire conduction problem can be solved using the
standard tools from circuit analysis. Moreover, the circuit model allows us
to neatly account for other factors, such as the large drop in temperature
across the interface of between two materials. Microscopically, an interface
contains small gas–filled voids that may present a significant thermal “con-
tact resistance” Rt,c , which we would expect to be able to determine based
on the two materials (Incropera and Dewitt, 2002). Lets look at an example
that combines all of these phenomena.
Example 3.2:
A manufacturer of pre–fabricated components for high–rise buildings is
determining the thermal rating of a new panel design. The system consists
of a header made up of an exterior cladding material having thermal con-
ductivity kc , and an inner insulator of conductivity k i , window glass with
conductivity kg , and a footer of the same construction as the header. It can
be assumed that heat transfer occurs one–dimensionally through the panel.
What would the thermal circuit for this system be assuming that there is
a thermal contact resistance of Rc between the cladding and insulator com-
ponents?
The panel height and width are w × L and the header, pane, and footer
each have a height of w/3. The total thickness is t, half of which is cladding
and half of which is insulating material. The following factors must be
accounted for
• convection boundary condition that relates the temperature of the
outside air T∞ to the temperature of the panel’s outer surface T s,o
through a known convection coefficient h — based on a total surface
area of wL, Eq. (3.12) indicates this resistance would be
1
Rconv =
wLh
• heat transfer through the header and footer, which is resisted by the
cladding, the insulator, and the contact resistance between them,
which we assume is known to be Rc — these three resistances are
w/3
cladding
insulation
w/3
glass
w/3
t/2
L
t/2
heat transfer
Figure 3.5. Pre–fabricated window panel for a high–rise build-

ing.
in series, so they are additive, with the conductive resistances for

cladding and insulators given by Eq. (3.8). The total for each of
the header and footer is

t/2 t/2
Rh,f = + Rc +
kc L w/3 cladding ki L w/3 insulation
• heat transfer through the glass pane, whose resistor is also given
by Eq. (3.8) as

t
Rg =
kg L w/3 cladding
where we note the the full thickness t appears in the numerator

since the glass is twice as thick as the cladding and insulation.
• convection boundary condition that relates the temperature of the

inside office environment Tof f ice to the temperature of the panel’s
inner surface Ts,i through a known convection coefficient h — since
the area and the coefficient are the same as for the external side,
this resistor is given by Rconv above
We can assemble these into the circuit diagram shown in Fig. 3.6, where the
nodes give the appropriate temperatures. This can easily be reduced to a
single circuit via the rules for resistors in parallel and in series. ♦♦♦
3.3. CYLINDRICAL CONFIGURATIONS 21
R h,f
T oo Rg
T office
Ts,o T s,i
R conv R conv
R h,f
Figure 3.6. Circuit model for pre–fabricated window panel.
3.3. Cylindrical Configurations

As with the Cartesian system, the circuit analogy is applicable to other
coordinate systems if a one–dimensional model of conduction is justified. If
we were to go through the same derivation of the conduction equation in
cylindrical coordinates (r, θ, z) as shown back in Chapter 2, we would find

∂T 1 ∂ ∂T 1 ∂ ∂T ∂ ∂T
(3.13) ρ cp = q̇ + kr + 2 k + k
∂t r ∂r ∂r r ∂θ ∂θ ∂z ∂z
(see e.g. Incropera and Dewitt, 2002, for derivation). If we assume one–
dimensional steady conductivity in the radial direction, a constant thermal
conductivity k, and no heat generation, Eq. (3.13) simplifies to

k d dT
(3.14) r = 0.
r dr dr
When integrated twice, this yields
(3.15) T (r) = C1 ln r + C2 ,
where C1 and C2 are constants to be determined using the boundary con-
ditions.
Assume we have an annular section where T = T h at r = rh (an inner
“hot” side) and T = Tc at r = rc (an outer “cool” side) where rc > rh > 0.
Substituting, we get Th = C1 ln rh + C2 and Tc = C1 ln rc + C2 . Therefore,
Th − Tc = C1 (ln rh − ln rc ) = C1 ln(rh /rc ), which can be solved for the
constant as
Th − T c
(3.16) C1 = .
ln (rh /rc )
Backsubstituting, the value of the other constant is found to be
Th − T c
(3.17) C 2 = Tc − ln rc ,
ln (rh /rc )
so that the resulting temperature distribution is
(Th − Tc ) ln (r/rc )
(3.18) T (r) = + Tc .
ln (rh /rc )
3.3. CYLINDRICAL CONFIGURATIONS 22
Application of the Fourier Law yields

dT Th − T c rc 1
(3.19) q = −k A = −k × (2 π r L) × ,
dr ln (rh /rc ) r rc
where L is the length along the z coordinate direction of the domain. With
a little algebra, this can be simplified to
2 π L k (Th − Tc )
(3.20) q = ,
ln (rc /rh )
where we have absorbed the negative sign according to ln (r c /rh ) = − ln (rh /rc ).
Recalling from Eq. (3.7) that the thermal resistance R t = ∆T /q, we see that
the thermal circuit representation for conduction in a cylindrical domain is
ln (rc /rh )
(3.21) Rt = .
2πLk
Recall, we have assumed rc > rh > 0 thus far, so we could write Eq. (3.21)
in a slightly more generic form as
ln (ro /ri )
(3.22) Rt = ,
2πLk
where subscripts “o” and “i” denote the outer and inner boundaries, respec-
tively.
Let’s go back for a minute to the M–16 cooling problem. The bores of
GI barrels are chrome–plated for increased wear and anti–fouling properties,
however, this adds another “layer” of thermal resistance (Fig. 3.7). Let r i
ri To
Ti
ro convection
T oo
rc
Figure 3.7. Chrome–plated GI M–16 barrel.
be the radius of the bore, rc the radius of the chrome plating, and r o be
the outer radius of the barrel. We assume that the plating is “perfect”, i.e.
there are no interstitial voids, so there is no thermal contact resistance. Also
let’s assume that conditions in the bore are such that the bore temperature
equals the temperature at the inner surface of the barrel. How can this
system be represented by a thermal circuit? We simply construct a series
circuit representing the thermal resistance of the chrome plating, the barrel
itself, and the convection occurring at the outer surface of the barrel, as
depicted in Fig. 3.8. Applying the circuit analogy, we find
3.4. HEAT GENERATION 23
Ti To T oo
R plating R barrel R convection
Figure 3.8. Circuit for M–16 barrel problem.
Ti − T ∞
(3.23) q = ,
Rplating + Rbarrel + Rconvection
where
ln (rc /ri )
(3.24) Rplating = ,
2 π L kchrome
ln (ro /rc )
(3.25) Rbarrel = ,
2 π L kbarrel
1
(3.26) Rconvection = .
2 π ro L h
Here, L is the length of the barrel and h is the convection coefficient at its
outer surface. We notice that for this problem, the thickness of plating is
very thin, therefore rc /ri is just slightly more than unity . The natural log I&D Ex. 3.4
of this is close to zero, therefore, we conclude that chrome–plating the barrel pp 107
does not measurably affect the heat transfer situation.
3.4. Heat Generation

We now want to look at the important case of heat generation occurring
in the medium itself, e.g. conversion of electrical energy to heat energy
through the action of Ohmic resistance. Simplifying Eq. (2.21) for one–
dimensional conduction in a Cartesian domain, we obtain
d2 T q̇
(3.27) + = 0,
dx2 k
where for example q̇ = I 2 Relec per unit volume for the electric problem.
This equation is straightforward to solve by direct integration, which yields
q̇x2
(3.28) T (x) = − + C1 x + C 2
2k
as the general solution. Again, we must use the boundary conditions to
evaluate the constants of integration.
Let us consider an application of Eq. (3.28) to the problem of DNA
fragment separation using Polyacrylimide Gel Electrophoresis 3.1. This tech-
nique, used to generate the raw data of the Human Genome Project (Lander
et al., 2001), separates fragments by size by exploiting the difference in net
charge (Fig. 3.9). The axial voltage difference leads to heat generation within
3.1Fred Sanger won the 1980 Nobel Prize in Chemistry for inventing the sequencing
reaction that supplies the input for this process.
x
Ts
lower voltage
x=+L
high voltage
Ts x=−L
migration of DNA fragments
Figure 3.9. Polyacrylimide Gel Electrophoresis.
the gel, which is a significant design consideration in sequencing lab instru-

mentation. If there is a sufficient temperature rise, DNA will de–nature in
the gel. Let the gel thickness be 2L and the ohmic dissipation per unit
volume be q̇ = I 2 Rgel . Also, let the temperature at both surface of the gel
be Ts . We can solve for the constants as C1 = 0 and C2 = 0.5 q̇ L2 /k + Ts ,
so that the exact solution is
q̇L2 x2

(3.29) T (x) = 1 − 2 + Ts .
2k L
This is a parabolic profile with a maximum at x = 0 (the center of the gel)
of
q̇L2
(3.30) Tmax = + Ts .
2k
Eq. (3.30) describes the influences that design parameters have on the pro-
cess.
• the surface temperature of the gel T s results from the surrounding
temperature in the laboratory environment, which we assume is
not a systematically controlled variable
• heat generation q̇ results from the voltage gradient and is therefore

a function of the experimental requirements — this is maximized
to the greatest degree possible to achieve reasonable results
• the gel conductivity k is essentially a fixed parameter
• the gel thickness 2L can be directly controlled — gels are therefore

made as thin as possible
Since it appears as a square, L is a strong variable, for example halving it
reduces the temperature rise by a factor of four. I&D Ex. 3.6
Solutions are developed in a similar fashion for the cylindrical coordinate pp 117
system. The governing equation is

1 d dT q̇
(3.31) r + = 0.
r dr dr k
Once again, integrating twice, we get the general solution

q̇ r 2
(3.32) T (r) = − + C1 ln r + C2 .
4k
Now let’s look at an interesting combination problem for which the circuit
analogy can be applied for part of the problem, but for the other part having
heat generation it can’t.
Example 3.3:
Find the current–carrying capacity of a wire (Fig. 3.10) assuming the
failure mode is the onset of insulation melting and find the temperature
in the wire at this condition. Assume failure occurs at T = 330K. The
following parameters are given, where “w” denotes a wire property and “i”
denotes an insulation property:
• radii: rw = 0.002m and ri = 0.003m
• conductivities: kw = 80W/(mK) and ki = 10W/(mK)
• convection parameters: T∞ = 300K and hi = 20W/(m2 K)
• electrical resistance: ρ0w = 0.008Ω/m
T oo
ri
rw
w
ire
in
su
la
tio
n
Tw
Ti
Figure 3.10. Electrical power transmission wire.

Current capacity: Intuitively, we know that the maximum insulation
temperature will occur at the wire surface r w because if heat is flowing
radially outward, the temperature in the insulation at larger radii must be
lower. Since there’s no energy generation in the insulation, we can use
the thermal circuit analogy to solve for the limiting current I. The circuit
equation is q = (Tw − T∞ )/Rtotal , where the resistors are added in series as
ln (ri /rw ) 1
Rtotal = Rinsul + Rconvec = +
2 π L ki 2 π r i L hi
and L is the length of the wire. This simplifies to
0.00645 + 2.6526 K 2.659 K
Rtotal = = .
L W L W
Tw Ti T oo
R insul R convec
Figure 3.11. Circuit representing the wire insulation.
Because the system is steady–state, the heat generated by the dissipation

all goes through the insulation, therefore q is simply the dissipation rate,
which is I 2 Relectric = 0.008LI 2 W . Therefore, from the circuit equation, we
have
2.659
0.008LI 2 = Tw − T∞ ,
L
where Tw − T∞ = 30K. This results in
30
I2 = ,
0.008 · 2.659
which yields 37.6 Amps3.2.
Max wire temperature: For the wire, no circuit analogy can be used
because of the heat generation. Therefore, we have to refer back to the
general solution in Eq. (3.32) for temperature with heat generation:
q̇ r 2
T (r) = − + C1 ln r + C2 .
4k
We see that because the wire is a solid cross–section that C 1 = 0, otherwise
the term would blow up because ln 0 → −∞, so that
q̇ r 2
T (r) = − + C2 .
4k
We have to know the volumetric heat generation rate, q̇, which can be
computed as
Ėgen I 2 Relectric 0.008 L 37.62 W
q̇ = = 2 L
= = 897, 828 3 .
V ol π rw π 0.0022 L m
Now we can evaluate C2 using the fact that the temperature at the wire—
insulation boundary is known to be 330 K at failure, therefore
q̇r 2 897, 828 · 0.0022
C2 = T (0.002) + = 330 + = 330.01 K .
4k 4 · 80
The exact solution is therefore
q̇ r 2
T (r) = − + 330.01 ,
4k
3.2This is about enough to carry the load of forty 100 W light bulbs. Of course,
with an appropriate safety factor defined by a municipal electrical code, the maximum
prescribed load would be somewhat less.
3.5. FIN ANALYSIS 27
which allows us to determine the temperature at the center of the wire in a

straightforward fashion. Evaluating T (0), we find
T (0) = C2 = 330.01 K .
We see that there’s hardly any difference in temperature between the center
of the wire and the edge of the wire. It’s essentially at a constant tempera-
ture. ♦♦♦
3.5. Fin Analysis

Extended surfaces, or “fins”, are often used to increase heat transfer, for
example as applied to air conditioner coils, I/C chips, heat exchangers, car
radiators, lawn mower engines, pipes, etc. Why is this the case? If we recall
the concept of Newton’s Law of Cooling introduced in Eq. (3.9) on pp. 18,
we can write a generic form of this convection equation as

(3.33) q = h A T |surf ace − Tf luid ,
where h is once again the convective heat transfer coefficient. How can the
various components of this equation be modified to increase q?
• Tf luid : Usually this cannot be changed much, if at all, for example
if air is the working fluid then the atmospheric air temperature is
essentially a constant.
• T |surf ace : Similar to Tf luid , the device temperature and thus the
resulting surface temperature is usually constrained to a small op-
erating range.
• h: This can be increased to a degree, however, it is still highly

constrained, for example adding a blower to increase convection
may be possible, but it may be impractical due to weight, size
restrictions, electrical considerations, aerodynamics, etc.
• A: It’s easy to increase this by huge margins using fins!

Of course, to properly analyze such systems, we must first have the ap-
propriate equation describing the conservation of energy. Here, we develop
the so–called fin equation, which is a special case of the general conduction
equation that explicitly incorporates a convection boundary condition. In
the usual fashion, we analyze a differential length dx of a fin as shown in
Fig. 3.12. Here, energy is conducted along the fin in the x direction and dis-
sipated by convection along the entire radial boundary. Conduction in this
problem is actually two–dimensional, both in the radial direction and along
the axis. However, as with the wire problem in the previous example, tem-
perature gradients are small in the radial direction, so that conduction will
be correspondingly small enough to be neglected. We therefore assume the
dAs
qx d q conv
Ac
convection
cooling around a x
fin
x dx q x + dx
x + dx
Figure 3.12. Differential fin element.
conduction is strictly one–dimensional along the axial x direction. We will

further assume steady–state conditions, constant conductivity, and constant
convection coefficient. The local cross–sectional area at x is A c , however, we
assume this can vary along the axis, so we properly write this as a function
of x, as in Ac (x). The local boundary area exposed to convection is dA s ,
which is a constant for any particular element.
Heat is conducted into the element at x at a rate of q x and is conducted
out of the element at x + dx at a rate qx+dx . Heat is convected away at the
radial boundary at a rate dqconv . For energy to be conserved, we write the
equation
(3.34) qx = qx+dx + dqconv .
(This is clearly a form of Eq. (1.1) on pp. 3: Ėstored = Ėin + Ėgen − Ėout ,
where Ėstored = Ėgen = 0.) We immediately recognize qx from Fourier’s
Law in Eq. (2.1) on pp. 7, which can be written for this problem as
dT
(3.35) qx = −k Ac (x) ,
dx
where Ac (x) is again the cross–sectional area of the fin that varies with x.
We can expand qx+dx according to a truncated 1–term Taylor series 3.3 to
obtain

dqx dT d dT
(3.36) qx+dx = qx + dx = −k Ac (x) − k Ac (x) dx ,
dx dx dx dx
where we have removed k from under the derivative in the second term,
since it is assumed constant. Finally, we write the convection term once
3.3Recall we did this with the derivation of the conduction equation in Chapter 2 for
flux terms as well.
again according to Newton’s Law of Cooling as

(3.37) dqconv = h dAs (T − T∞ ) ,
where T∞ is the temperature of the fluid absorbing the heat. We can
now substitute these components back into our original conservation law
in Eq. (3.34) to obtain

d dT
(3.38) h dAs (T − T∞ ) − k Ac (x) dx = 0 .
dx dx
If we divide by k dx and change sign, we get the canonical form

d dT h dAs
(3.39) Ac (x) − (T − T∞ ) = 0 .
dx dx k dx
This equation is the generalized fin equation for one–dimensional conduction.
It is a second–order equation which requires 2 boundary conditions to solve
it.
Notice that a non–trivial aspect of Eq. (3.39) is that we must know the
function Ac (x) before we can solve the problem. That is, Eq. (3.39) depends
on the geometric attribute of how the cross–section varies along the fin. Let
us examine the most straightforward case: a uniform cross–section. Under
these conditions, the total surface area A s is simply the perimeter of the
cross–section P multiplied by the length x
As = P x ,
so that we obtain dAs /dx = P by simple differentiation. Moreover, for
constant cross–section Ac (x) → Ac is a constant and can be moved outside
of the differential term in Eq. (3.39). With these two observations, Eq. (3.39)
simplifies to
d2 T hP
(3.40) − (T − T∞ ) = 0 .
dx2 k Ac
This is a second–order linear non–homogeneous equation, however, it’s not
as easily integrated as the standard conduction equation because it contains
both the second derivative of T and T itself. We recall from differential
equations that the typical method of handling this type of equation is (1)
solve the homogeneous form then (2) solve to get the particular solution,
then (3) sum to get the general solution. However, in this case we are lucky
that a change of variables3.4 can be made to reduce the problem to a simple
second–order homogeneous equation: θ(x) = T (x) − T ∞ . The result is
d2 θ
(3.41) − m2 θ = 0 ,
dx2
3.4Since T is a constant, derivatives of T can be recast directly in terms of θ, for
∞
example
dT d(θ + T∞ ) dθ dT∞ dθ
= = + = .
dx dx dx dx dx
Further derivatives clearly show the same behavior.
where
hP
(3.42) m2 =
k Ac
is a composite parameter made up of the physical and geometric attributes
of the problem. The general solution for Eq. (3.41) can be written in the
form of exponentials as
(3.43) θ(x) = C1 emx + C2 e−mx .
Likewise, it could also be written in terms of the hyperbolic trigonometric
functions according to the identities
eβ + e−β
(3.44) cosh β =
2
and
eβ − e−β
(3.45) sinh β = .
2
Of course, now we are back in the old position of having to use boundary
conditions for an actual problem to determine the constants of integration.
We once again need 2 boundary conditions, one at the base of the fin x = 0,
and one at the end of the fin at x = L (Fig. 3.13). The first boundary
x=0
x=L
Figure 3.13. Boundary locations for fins.
condition is rather straightforward: at the base of the fin we will assume

that the temperature is known, i.e. T (0) = T 0 . In our derived variable, this
translates to θ(0) = T (0) − T∞ = T0 − T∞ = θ0 . If we substitute this into
Eq. (3.43), we find that one of our boundary equations will always be
(3.46) θ(0) = θ0 = C1 + C2 .
At the end of the fin, there are 4 cases that are of interest here. Listed in
order of mathematical difficulty, they are:
• Fin is very long: Fins convect heat to the surrounding fluid along
their length. If we assume that the fin tip is very long, i.e. L →
∞, eventually all of the heat will be convected away, so that the
temperature at the very end must be equal to T ∞ . That is, there is
no longer any temperature gradient because there is no more heat
to be transferred. According to our original substitution θ(x) =
T (x) − T∞ , this means that θ(∞) → 0. If we substitute this into
Eq. (3.43), it is clear that the first term would be unbounded, so
we must conclude that C1 = 0. We can then use the boundary
condition at the base to find C2 = θ0 , so that the solution is
(3.47) θ(x) = θ0 e−mx .
• Negligible fin tip convection: In this case, we assume the end of the
fin is insulated, so that q = 0 at x = L. Writing this in terms of
Fourier’s Law, we obtain a boundary condition of

dθ
= 0.
dx x=L
We can take the derivative of Eq. (3.43)
dθ
= C1 m emx − C2 m e−mx
dx
and substitute this into the boundary condition to obtain the equa-
tion
C1 emL − C2 e−mL = 0 .
This equation, along with the boundary condition from x = 0 in
Eq. (3.46) allows us to solve for the constants. For example, we
have C1 = θ0 − C2 from Eq. (3.46), which can be substituted into
the above equation
0 = (θ0 − C2 ) emL − C2 e−mL
θ0 emL − C2 emL + e−mL

=
so that C2 can be solved as
emL
C2 = θ 0 .
emL + e−mL
We can substitute this right back into Eq. (3.46) to solve for C 1 as
emL emL

C1 = θ0 − θ0 mL = θ 0 1 −
e + e−mL emL + e−mL
which, with a little algebra, simplifies to
e−mL
C1 = θ 0 .
emL + e−mL
Finally, we can substitute C1 and C2 into Eq. (3.43) to obtain the

exact solution
e−mL mx emL
θ(x) = θ0 e + θ 0 mL e−mx
emL + e−mL e + e−mL
e−mL emx + emL e−mx
= θ0
emL + e−mL
e m(L−x) + e−m(L−x)
= θ0 .
emL + e−mL
According to the identity for cosh in Eq. (3.44), we can write this
result in the more simple form
cosh m (L − x)
(3.48) θ(x) = θ0 ,
cosh mL
which is the exact solution.
• Fin tip temperature is known: That is, we have a standard Dirichlet

boundary condition at x = L, similar to that at the base x = 0
θ(L) = θL .
The procedure to evaluate C1 and C2 for this case is identical to
what we have seen so far, although the algebra is somewhat more
involved. The final solution turns out to be
θL sinh mx + θ0 sinh m(L − x)
(3.49) θ(x) = .
sinh mL
• Convection from fin tip: Rate of energy transfered to the fluid by
convection at the tip of the fin equals the rate at which energy
arrives at the tip via conduction. We write the standard equation

dT
h Ac T (L) − T∞ = −k Ac ,
dx x=L
which can be simplified using our change of variables to

dθ
h θ(L) = −k .
dx x=L
This case is the most tedious to solve. We find
cosh m(L − x) + λ sinh m(L − x)
(3.50) θ(x) = θ0 ,
cosh mL + λ sinh mL
where λ = h/(mk) is a constant that depends upon the character-
istics of the boundary condition, specifically h and k.
In each of these four cases, we now know the exact temperature distri-
bution for the entire fin in terms of our modified variable θ(x). In principle,
it should now be possible to calculate the total amount of heat transfered
by the fin to its surroundings qf . This is the quantity of primary interest
for many real–world problems. But how do we do this? As is often the case,
there is an easy way, and a hard way.
Refer to Fig. 3.14. One of our original assumptions was that the problem
temperature
convection
q fin
f
convection
x
Figure 3.14. Schematic and qualitative temperature distribu-

tion in a constant cross–section fin. Magnitude of temperature
gradient decreases with increasing x as a consequence of the reduc-
tion in q with increasing x due to continuous convection loss from
the fin surface.
is steady state, so that the energy entering the fin via conduction at x = 0
must be equal to the energy convected by the fin to its surroundings over its
total area, i.e. over 0 ≤ x ≤ L. The two possibilities may already be clear
at this point.
The hard way is to integrate Newton’s Law of Cooling about the entire
surface area of the fin
Z Z
qf = h (T (x) − T∞ ) dAs = h θ(x) dAs .
As As
By definition, this gives us the entire energy transferred to the surround-
ings. While this task is somewhat straightforward for the infinite fin using
Eq. (3.47), the other cases are rather more difficult, especially since the fin
tip convection must be considered3.5. Conversely, the easy way is to sim-
ply evaluate the total heat transferred into the fin by way of Fourier’s Law
applied at the base of the fin x = 0

dθ
(3.51) qf = −k Ac .
dx x=0
Since we know the exact temperature distribution, this is a simple matter
of taking a first derivative. For example, in the case of the infinite fin, we
find
dθ
= −θ0 m e−mx ,
dx
3.5The exception is of course the case of the insulated fin tip.
3.6. FIN PERFORMANCE METRICS 34
which can be evaluated at x = 0 and substituted into Eq. (3.51) to yield

qf = −k Ac (−θ0 m) = θ0 k Ac m .
Recalling the original definition of parameter m from Eq.(3.42), we can write
the result in the more intuitive form
p
(3.52) q f = θ0 h P k A c .
This equation indicates that the temperature difference at the base is the
strongest parameter of the problem. In particular, doubling the temperature
difference will double the heat transfer. Conversely, the other four variables
are not as significant, since they appear as square roots. For example,
to double the heat transfer, the convection coefficient would have to be
increased by a factor of 4, and so forth.
Total heat transfer for the other three cases can be worked out in a
similar
√ fashion. If we write the shorthand quantity in Eq. (3.52) as ξ =
θ0 h P k Ac , we find for the adiabatic (insulated) fin tip
(3.53) qf = ξ tanh mL ,
for the prescribed fin tip temperature boundary condition I&D Ex. 3.8
cosh mL − θL /θ0 pp 133
(3.54) qf = ξ ,
sinh mL
and for the convective fin tip boundary condition
sinh mL + λ cosh mL
(3.55) qf = ξ ,
cosh mL + λ sinh mL
where again λ = h/(mk).
3.6. Fin Performance Metrics

The fact that heat must be conducted along the fin means that the fin
itself represents a thermal resistor in the context of the circuit analogy. In
other words, a conductor has been added to an original un–finned surface
and this represents another resistance. So how do we know that a fin actually
helps increase the heat transfer? We can evaluate this by using a measure
of fin effectiveness, i.e. the ratio of the finned heat transfer rate, q f , to the
heat transfer that would occur without the fin, h A c θ0
qf
(3.56) εf = ,
h A c θ0
where Ac is cross–sectional area of the virtual fin and θ 0 represents the
temperature difference at that location. Generally, the use of fins can be
justified if there would be at least a 100 % increase in heat transfer, i.e.
εf ≥ 2 (Incropera and Dewitt, 2002).
If we make what might be a big assumption in some cases, i.e. the
coefficient of convection h stays the same whether the fin is present or not, we
can compute εf for a variety of cases. For example, we can insert Eq. (3.52)
for the infinite fin into Eq. (3.56) to find
√ r
θ0 h P k A c kP
(3.57) εf = = .
h A c θ0 h Ac
What does this say generally about fins?
• Fin performance is enhanced by using a high conductivity material.
Of course, this is almost obvious.
• An effective fin has a high P/Ac ratio, that is the perimeter is large
compared to the cross–sectional area. For example, assume the
fin has a square cross–section of side length b, then P = 4b and
Ac = b2 , therefore P/Ac = 4/b, which means the ratio, thus the
effectiveness improves as b becomes smaller. Therefore, lots of thin,
closely–spaced fins promote heat transfer, with the provision that
the space between them is still enough to maintain good convection.
• The use of fins is justified for conditions when h itself is small,

because the increase in heat transfer can be many fold.
Typically h for gases is much lower than h for liquids (e.g. Incropera and
Dewitt, 2002, Table 1.1, pp. 8). In particular, unforced convection for a gas
is quite low. That’s why for cases like the small lawn mower engine, which
does not have an active cooling system, there are always fins to augment
heat removal. Heat exchanger design also follows such logic. Automotive
radiators have lots of fins on the air side (low h), but no fins on the anti–
freeze coolant side (high h).
Another metric is fin efficiency, i.e. the ratio of the actual heat transfer,
qf , to the maximum possible heat transfer, hA f θ0 , which would take place if
the entire fin was at the base temperature, so that convection was maximized
all along the fin length3.6. The form is
qf
(3.58) ηf = .
h A f θ0
Eqs. (3.52) through (3.55) can be used to compute the efficiency for constant–
area fins using the various types of boundary conditions we have discussed.
Efficiencies for more complicated fins, for example those having variable
cross–sectional areas, have been worked out and tabulated in references (e.g.
Özişik, 1985; Mills, 1999; Incropera and Dewitt, 2002).
Perhaps more useful is the concept of overall fin efficiency for an array of
fins. The definition is similar to Eq. (3.58) for a single fin, except quantities
are summed for all fins, i.e.
qtotal
(3.59) η0 = .
h Atotal θ0
3.6Recall from Fig. 3.14 that the gradient along a real fin lessens the heat transfer.
We can derive a relationship between η f for a single fin and the overall
efficiency for N fins as follows. First, the total area A total is simply the sum
of the areas of all the fins plus the remaining overall area of the unfinned
base Ab . For N fins this is
(3.60) Atotal = N Af + Ab .
According to Eq. (3.58), the heat transfer from one fin is η f hAf θ0 , so that
the heat transfer from N fins is N ηf hAf θ0 . We include the heat transfer
from the remaining unfinned base hAb θ0 , so that the total heat transfer is
(3.61) qtotal = N ηf hAf θ0 + hAb θ0 .
Now, assuming that h is taken as a single constant value for both the finned
surface and the remaining area of the base, we can substitute Eqs. (3.60)
and (3.61) into Eq. (3.59) to find
N ηf hAf θ0 + hAb θ0 N η f Af + A b
(3.62) η0 = = ·
h (N Af + Ab ) θ0 N Af + Ab
This equation is sometimes given in a slightly different form that can be
derived by adding and subtracting 1 from η f as
N (ηf + 1 − 1) Af + Ab [N Af + Ab ] + (ηf − 1) N Af
η0 = = ,
N Af + Ab N Af + Ab
which simplifies to I&D Ex. 3.9
Atotal − (1 − ηf ) N Af N Af pp 144
(3.63) η0 = = 1− (1 − ηf ) .
Atotal Atotal
Eq. (3.63) can be used to calculate the total heat transfer for a fin array.
CHAPTER 4
Transient Conduction
4.1. Generalities
There are of course numerous engineering situations where unsteady ef-
fects are important. For example, the annealed steel quenching problem
that we alluded to in Chapter 1. Consider also solar systems whose input is
a function of the sun’s position, which changes as a function of time, or the
start–up and shut–down of any generic thermal system, for instance inter-
nal combustion engines. Mathematically, these represent fairly complicated
conduction problems, i.e. whereas steady 1–D conduction problems can be
reduced to an ordinary differential equation, transient problems (in general)
result in a partial differential equation, which is typically much more difficult
to solve. Specifically, if we simplify the general equation of heat conduction,
Eq. (2.18) on pp. 12, to model a 1–D transient problem, we get
∂T ∂2T
(4.1) = α ,
∂t ∂x2
where α is the thermal diffusivity. This equation is quite a bit more difficult
to solve than what we’ve seen up to now.
4.2. Lumped Capacitance Analysis

That said, we’ll now reverse ourselves and look at a special case where
the problem can still be described by a fairly simple ordinary differential
equation. Consider the conceptual case where spatial gradients in a material
vanish, i.e. the temperature in the material is spatially uniform at any
instant in time4.1. By Fourier’s Law, it is implied that heat transfer with no
gradient can only take place if conductivity approaches infinity, a physical
situation that is not realistic within our current framework. Specifically
dT
q 00 = −k 6= 0
dx
if dT /dx → 0, but k → ∞. So although this situation is not actually pos-
sible, it may be closely approximated if the resistance to conduction within
the material is low compared to the resistance of energy transfer from the
boundary of the material to its surroundings. In other words, the approxima-
tion may be valid if the convection at the boundary is the “rate–determining
4.1For example, recall that the temperature in the electrical wire problem back in
Chapter 3 was very nearly constant.
37
4.2. LUMPED CAPACITANCE ANALYSIS 38
step”. The method we derive to treat this configuration is lumped capac-

itance analysis. Let us assume, for the moment, that temperature in a
conducting body is approximately constant. Gradients do not arise, so the
conventional conduction equation is not relevant. Instead, we can write
an overall energy balance, akin to Eq. (1.1) on pp. 3, except that we need
only account for the energy transfer that takes place at the boundary of the
conductor by the mechanism of convection. That is
Ėstored = −Ėout ,
where Ėout is the rate at which convection removes heat energy. This is once
again quantified by Newton’s law of Cooling as Ėout = hAsurf (T −T∞ ). The
rate at which heat energy stored decreases is E stored = ρV cdT /dt. This gives
dT
ρV c = −hAsurf (T − T∞ ) .
dt
Or, making the standard change of variables as introduced in Chapter 3,
θ(t) = T (t) − T∞ , this becomes
dθ hAsurf
(4.2) = − θ.
dt ρV c
Eq. (4.2) is a so–called separable differential equation, which can be con-
veniently solved as follows. First, re–arrange Eq. (4.2) so the variables are
“separated” as
dθ hAsurf
= − dt
θ ρV c
then integrate over a finite time
Z θ Z t
dθ 0 hAsurf
= − dt0 ,
θ0 θ ρV c
0
0
where t0 and θ 0 are variables of integration. Note that on the left hand side,
we are explicitly using limits that correspond to the initial condition (at
t = 0) and to the condition at the time t of interest, i.e.
θ|t0 =0 = θ0 and θ|t0 =t = θ(t) = θ .
Carrying out the integration, we find
θ hAsurf 0 t
ln θ 0 θ0 = − t 0 ,
ρV c
which evaluates to
hAsurf
(ln θ − ln θ0 ) = − t.
ρV c
Using the logarithm identity (ln a − ln b) = ln(a/b), we can write this ex-
pression more conveniently as

θ hAsurf
ln = − t.
θ0 ρV c
4.3. APPLICABILITY OF LUMPED CAPACITANCE 39
Finally, we can exponentiate both sides to obtain the more useful form

hAsurf
(4.3) θ = θ0 exp − t .
ρV c
Substituting back the actual temperatures, we can also write the alternative,
but equivalent form

T (t) − T∞ hAsurf
(4.4) = exp − t ,
T (0) − T∞ ρV c
where T (0) is the initial temperature of the conductor.
According to Eq. (4.4), temperature response is a simple exponential
when spatial gradients are small enough to be neglected. We can write this
equation as
T (t) − T∞
(4.5) = e−t/τ ,
T (0) − T∞
where
ρV c 1
(4.6) τ = = (ρV c)
hAsurf hAsurf
is a “time constant”. Notice that the first term is the standard form for
the convective resistance we examined back in Chapter 3. Increasing this
resistance, i.e. decreasing either h, A surf , or both, increases the response
time of the system. The second term, ρV c, is the lumped thermal capacitance
of the conductor. Likewise, any increase of this term increases response time.
In engineering systems, we would be interested not only in T (t), but also
in how much heat was transferred over a particular time period. Determining
the total energy transferred over some period of time is simply a matter of
integrating Newton’s Law of Cooling over a particular time range
Z t Z t
0
Q = q dt = hAsurf θ dt0 ,
0 0
which gives

(4.7) Q = ρV c θ0 1 − e−t/τ .
4.3. Applicability of Lumped Capacitance

The lumped capacitance method provides a very convenient analysis of
problems where gradients are negligible, with solutions for temperature and
total heat transfer given respectively by Eqs. (4.3) and (4.7). Aside from
the qualitative constraint about gradients, how do we actually determine
whether such analysis is applicable? Recall, we know conceptually that this
method is valid if the tendency for energy transfer via conduction within the
material is much greater than the tendency for energy transfer via convection
at the boundary. In other words, the resistance to conduction within the
material must be low compared to the resistance of energy transfer from the
boundary of the material to its surroundings. We can quantify this using a
low Biot (<<1)

conduction convection moderate Biot (~1)
high Biot (>>1)
Ts1
Ts2
Ts2
Ts2 Too
x=0 x=L
Figure 4.1. Diagram of relative magnitudes of temperature

gradients for various Biot numbers.
simple conduction–convection energy equation written at a boundary, e.g.

at x = L in Fig. 4.1. Assuming no heat generation, we use a straight–line
gradient to represent dT /dx in Fourier’s Law, so that we write
kA(Ts1 − Ts2 )
= hA(Ts2 − T∞ ) .
L
Re–arranging terms, we find
Ts1 − Ts2 hL
(4.8) = = Bi ,
Ts2 − T∞ k
where Bi is called the Biot number . We emphasize that k is the thermal
conductivity of the conductor, not the surrounding fluid 4.2. The Biot Num-
ber is a dimensionless parameter that describes the ratio of the temperature
drop (gradient) in the solid relative to the temperature drop (gradient) from
the surface to the fluid. Recalling our assumption of small temperature gra-
dient in the material, we would expect lumped capacitance analysis to be
valid when the Biot Number is very small, i.e. Bi 1. Notice further
Eq. (4.8) provides a straightforward way to evaluate Bi based on measur-
able parameters of the problem without even having to know what these
gradients are in advance. Mathematically, we can restate the simplification
as T (x, t) → T (t) for Bi 1. It follows that the very first thing one should
do when confronted with a transient temperature problem is check the Biot
Number! Convention has it that lumped capacitance can be used when
4.2We will find a dimensionless expression similar to Eq. (4.8) called the Nusselt num-
ber arises in convection, except where k in that case is the conductivity of the surrounding
fluid.
Bi < 0.1 (Incropera and Dewitt, 2002). Error in neglecting gradients is

then less than 5 % (Özişik, 1985).
The values for k and h are clear, as is the value of L if the geometry is
simple, e.g. as in Fig. 4.1. However, what is the length scale L if the geome-
try is more complicated? Often, we resort to using a so–called characteristic
length scale Lc which can be defined as, for example, the ratio of the volume
to the surface area in the direction of heat transfer of the conductor
V
Lc = .
Asurf
For example, this convention can be applied to some of the most common
geometries, e.g. as shown in Fig. 4.2.
2L
x R
H R
sphere: radius R
slab: width 2L
cylinder: radius R
Figure 4.2. Common geometries for which characteristic

length scales can be determined.
• Slab: V = 2L × H × D, where D denotes depth. Area is both

sides in the direction of heat transfer: A surf = 2(H × D), so that
Lc = L.
• Cylinder: V = πR2 H and area is the surface area in the (radial)
direction of heat transfer 2πRH, so that L c = R/2.
• Sphere: V = 4πR3 /3 and A = 4πR2 giving Lc = R/3.
Specific problems may observe other conventions, for instance the length
scale might be chosen to correspond to the dimension where the maximum
∆T occurs. This may be more conservative. I&D Ex. 5.1
pp 245
4.4. CASTING THE GENERAL ONE–DIMENSIONAL PROBLEM 42
4.4. Casting the General One–Dimensional Problem

The condition Bi 1 clearly excludes many cases of interest. For
configurations where Lumped Capacitance analysis is not valid, we must go
back to the governing equation of conduction, Eq. (4.1), which we recall is
∂T ∂2T
= α ,
∂t ∂x2
where α is the thermal diffusivity. This is a partial differential equation
rather than an ordinary differential equation, and is therefore more difficult
to solve. In general, we obtain the solution T (x, t) according to the following
procedure.
• Obtain the general solution via advanced calculus techniques, in-
cluding separation of variables and integral transforms ( Özişik,
1980).
• Specify 1 initial condition. This will be used to evaluate the con-
stant of integration resulting from the first–order temporal term.
We recall the basic form from Lumped Capacitance analysis, al-
though, now instead of simply being constant, the initial tempera-
ture may be some function of the spatial variable
T (x, 0) = T0 (x) .
• Specify 2 boundary conditions. These will be used to evaluate the
2 constants of integration resulting from the second–order spatial
term. We recall this from steady 1–D conduction analysis, for ex-
ample an insulated surface at the left and a convective boundary
condition for the surface on the right:

dT
= 0
dx x=0
and
dT
h [T (L, t) − T∞ ] = −k .
dx x=L
Therefore, the problem is a multivariate function
(4.9) T = T (x, t, T0 , T∞ , L, k, α, h) .
The physical manifestation of this problem is shown in Fig. 4.3. The “insu-
lated” boundary condition is equivalent to a “symmetry” condition at x = 0.
That is, no heat is conducted across the x = 0 boundary, so the temperature
profile is symmetric about this point.
Before delving into this problem, it is worthwhile to mention the stan-
dard practice of non–dimensionalizing the system, which was probably cov-
ered in fluid mechanics4.3. One of the primary advantages of this process
is to reduce the number of relevant variables in the problem. For example,
4.3Panton (1984) has an extensive discussion of dimensional analysis, including proofs
of the relevant theorems.
x*
convection
Figure 4.3. Diagram of transient problem.
recall how 3 separate variables in a standard flow problem: characteristic

velocity and length, u0 and L0 , and kinematic viscosity, ν, can be combined
into a single group, the Reynolds number, Re. In a typical study of drag,
D, for an aerodynamic body, instead of plotting families of curves for u 0 ,
L0 , and ν versus D, we simply plot Re versus D. So what are the relevant
non–dimensional groups in our 1–D conduction problem? Going through
the non–dimensionalization process using the Buckingham–Π Theorem, we
find a number of relevant “groups”:
• Dimensionless temperature: According to our now–standard vari-
able substitution θ = T − T∞ , we can write a dimensionless tem-
perature as
T − T∞ θ
θ∗ = = ,
T0 − T ∞ θ0
which clearly implies the limits 0 ≤ θ ∗ ≤ 1.
• Dimensionless spatial coordinate: Using the characteristic length
dimension, we obtain
x
x∗ = .
L
• Dimensionless time coordinate: Time may be non–dimensionalized
using the characteristic length dimension and the thermal diffusiv-
ity, as in
αt
t∗ = 2 = F o ,
L
where F o is commonly called the Fourier Number 4.4.

To apply the same dimensionless treatment to the governing equation,
Eq. (4.1), we first determine how derivatives behave in the dimensionless
realm using the Chain Rule of Calculus, as in
∂ ∂ ∂t∗ α ∂
= = 2 ∗
∂t ∂t∗ ∂t L ∂t
and
∂ ∂ ∂x∗ 1 ∂
= = .
∂x ∂x∗ ∂x L ∂x∗
The latter expression leads us to the required second derivative in x, as in
∂2 1 ∂2

∂ ∂ ∂ 1 ∂ 1 ∂ ∂x∗ ∂
= = = = .
∂2x ∂x ∂x ∂x L ∂x∗ L ∂x∗ ∂x ∂x∗ L2 ∂ 2 x∗
Using these expressions, Eq. (4.1) can be written in the dimensionless form
∂θ ∗ ∂ 2 θ∗
(4.10) =
∂t∗ ∂x∗ 2
Likewise, the initial condition becomes
(4.11) θ ∗ (x∗ , 0) = 1
and boundary conditions become

dθ ∗
(4.12) = 0
dx∗ x∗ =0
and

∗ ∗ dθ ∗
(4.13) −Bi · θ (1, t ) = .
dx∗ x∗ =1
The number of relevant variables has been dramatically reduced as compared
to the dimensional form of the problem stated in Eq. (4.9). That is, the
problem is now
(4.14) θ ∗ = θ ∗ (x∗ , t∗ , Bi) ,
where the single parameter, the Biot number, arises in the boundary condi-
tions.
4.4This is the unfortunate nomenclature used in most texts. The Fourier number is
not a dimensionless “number” in the same sense as, for example the Reynolds number in
fluid mechanics. It is a function of time, although it does not have any units. Therefore,
it would be more appropriately referred to as a dimensionless coordinate. However, as
a first approximation, it does also connote the relative effectiveness with which a body
conducts and stores energy (e.g. Incropera and Dewitt, 2002, pp. 256). In that sense, it
could also be loosely interpreted as a “dimensionless number”.
4.5. TRANSIENT ANALYSIS IN ONE DIMENSION 45
4.5. Transient Analysis in One Dimension

Eqs. (4.10) through (4.13) represent a linear partial differential system.
As mentioned, the solution process involves advanced techniques, including
separation of variables or integral transforms ( Özişik, 1980). At this point,
rather than examining details of the actual solution process for this system,
let us use the well–established solution directly. The solution takes the form
of an infinite series
∞
2 ∗
X
(4.15) θ∗ = Cn e−ζn t cos ζn x∗ ,
n=1
where the mode coefficients are

4 sin ζn
(4.16) Cn =
2ζn + sin (2ζn )
and ζn are eigenvalues given by the roots of the transcendental equation
(4.17) ζn tan ζn = Bi .
Several roots for various values of Bi have been tabulated (e.g. Incropera
and Dewitt, 2002, Appendix B.3, pp. 936).
This solution is quite a bit more complex than what we have dealt with so
far, especially since the eigenvalues are not given by an explicit relationship.
However, except for small values of the Fourier Number (t ∗ < 0.2), the series
can be approximated by the first term only. Thus, for t ∗ = F o > 0.2, the
solution can be taken to be
2 ∗
(4.18) θ ∗ = C1 e−ζ1 t cos ζ1 x∗ ,
where
4 sin ζ1
(4.19) C1 = .
2ζ1 + sin (2ζ1 )
We can express the solution equivalently as
(4.20) θ ∗ = θ0∗ cos ζ1 x∗ ,
where θ0∗ is the time–dependent temperature at the mid–plane x = 0
2 ∗
(4.21) θ0∗ = C1 e−ζ1 t .
A condensed summary of coefficients C 1 and the eigenvalues ζ1 is provided
in Table 4.1. Extensive tabulations are available in reference texts (e.g.
Incropera and Dewitt, 2002, pp. 258).
To calculate the total amount of heat transferred over some interval, we
again resort to our basic conservation equation
∆Estored = Ein − Eout ,
where Q = Eout , Ein = 0, and ∆Estored = E(t) − E(0). Substituting these
quantities, we can integrate the expression over the domain to obtain the
Table 4.1. Coefficients and eigenvalues for the one–term

approximate solution of transient heat conduction in the rect-
angular coordinate system.
Bi ζ1 C1 Bi ζ1 C1 Bi ζ1 C1
0.01 0.0998 1.0017 0.2 0.4328 1.0311 2.0 1.0769 1.1795
0.02 0.1410 1.0033 0.25 0.4801 1.0382 3.0 1.1925 1.2102
0.03 0.1732 1.0049 0.3 0.5218 1.0450 4.0 1.2646 1.2287
0.04 0.1987 1.0066 0.4 0.5932 1.0580 5.0 1.3138 1.2402
0.05 0.2217 1.0082 0.5 0.6533 1.0701 6.0 1.3496 1.2479
0.06 0.2425 1.0098 0.6 0.7051 1.0814 7.0 1.3766 1.2532
0.07 0.2615 1.0114 0.7 0.7506 1.0919 8.0 1.3978 1.2570
0.08 0.2791 1.0130 0.8 0.7910 1.1016 9.0 1.4149 1.2598
0.09 0.2956 1.0145 0.9 0.8274 1.1107 10.0 1.4289 1.2620
0.10 0.3111 1.0160 1.0 0.8603 1.1191 20.0 1.4961 1.2699
∞ 1.5707 1.2733
total heat transfer, as in

Z
Q = − [E(t) − E(0)] = − ρ c [T (t) − T0 ] dV .
vol
Let us introduce a (non–obvious) factor Q 0 to non–dimensionalize this equa-

tion based on the initial temperature difference, i.e. the temperature differ-
ence at t = 0
Q0 = ρ c V [T0 − T∞ ] .
The expression Q0 is not only the initial energy present, it is also the max-
imum amount of energy which could be transferred as t → ∞. The ratio
Q/Q0 is therefore the fraction of total energy transferred over an interval
0 → t. We find
Q −ρ c [T (t) − T0 ] dV − [T (t) − T0 ] dV
Z Z
= = ,
Q0 vol ρ c [T0 − T∞ ] V vol [T0 − T∞ ] V
which can be manipulated further by adding and subtracting T ∞ in the
numerator to obtain
Z
Q T (t) − T∞ T0 − T∞ dV
= − − .
Q0 vol T0 − T ∞ T0 − T ∞ V
Using our definition of θ ∗ , this simplifies to
Q 1
Z
= (1 − θ ∗ ) dV .
Q0 V vol
Integration can be carried out by substituting the one–term approximate
solution, preferably Eq. (4.20), into this expression. Recall, the length of
our dimensionless domain varies as 0 ≤ x ∗ ≤ 1 from the mid–plane to the
edge. If we take w ∗ and l∗ to be dimensionless width and depth of our

domain, we have dV = w ∗ l∗ dx∗ and V = w∗ l∗ · 1 = w∗ l∗ . We thus find
Z 1 Z 1
Q 1 ∗ ∗ ∗ ∗ ∗
= ∗∗ (1 − θ0 cos ζ1 x ) w l dx = (1 − θ0∗ cos ζ1 x∗ ) dx∗ .
Q0 w l 0 0
Again, recall that θ0∗ only describes the time–dependent temperature behav-
ior at the mid–plane, so it is not a function of dx ∗ . Also, the eigenvalue ζ1
is a constant. We therefore have a simple “cosine” integration, which yields
1
Q ∗ θ0∗
∗
= x − sin ζ1 x ,
Q0 ζ1 0
which evaluates to
Q θ∗
(4.22) = 1 − 0 sin ζ1 .
Q0 ζ1
These results can, of course, also be applied to a slab of thickness L with
an insulated boundary on the left at x ∗ = 0, since the mathematical problem,
specifically the boundary conditions, are exactly the same as the problem
we have just examined. Also, our result can also be applied to problems
where there is a step change of the surface temperature on free boundary at
x∗ = 1 to a new value Ts . Conceptually, we would realize the same result if
we allowed a step change in fluid temperature T ∞ → Ts and prescribed an
infinite Biot number4.5, so that the change in fluid temperature would also
be immediately transmitted to the surface at x ∗ = 1. Thus, step change
problems of this sort are handled by allowing Bi → ∞. I&D Ex. 5.4
For other systems, e.g. the infinitely long cylinder and the sphere, similar pp 262
approaches can be taken to determine exact solutions. These are even more
complicated than slab problems because of the additional considerations that
non–rectangular coordinate systems place on the problem. For example, the
solution for radial one–dimensional transient conduction is
∞
X 2 J1 (ζn ) 2 ∗
∗
θ = 2 2 e−ζn t J0 (ζn r ∗ ) ,
n=1
ζn J0 (ζn ) + J1 (ζn )
where t∗ = αt/r02 and J0 and J1 are Bessel functions of the first kind. Here,
eigenvalues are the roots of
J1 (ζn )
ζn = Bi .
J0 (ζn )
So, we find that the problems are more difficult, but their solutions appear
to be more difficult to evaluate, as well. Luckily, we find that the one–term
approximation is again valid for t∗ = F o > 0.2, so that tabulated values
(e.g. Incropera and Dewitt, 2002, pp. 258) can be used directly.
4.5Recall from Eq. (4.8) the form of the Biot number is Bi = hL/k, so that an infinite
Biot number is equivalent to prescribing an infinite convection coefficient h.
4.6. THE SIMILARITY TECHNIQUE 48
4.6. The Similarity Technique

Another geometry which can be studied analytically is that of a semi–
infinite solid, which serves as a useful idealization for many problems, e.g.
any body which is very “thick” Fig. 4.4. This sort of problem is described
Figure 4.4. Diagram of semi–infinite domain.
by the same equation and initial conditions as the other problems, however,
boundary conditions are special. We still require 2 boundary conditions
because of the second–order spatial derivative, but there is clearly only 1
surface to specify a boundary condition at, i.e. x = 0. There is no boundary
at x → ∞. However, we can use the conceptual boundary condition of
T (x → ∞, t) = T0 ,
which says simply that for some value x that is large enough, the temper-
ature will always be the initial temperature. Physically, this makes sense
because if truly x → ∞, then we can always pick x large enough such that
it is so far away from the boundary at x = 0 that temperature variations
cannot be perceived. For the surface, we can specify any of the 3 standard
types of boundary conditions we’ve studied.
The mathematical significance of this problem is that there is a very
clever trick to simplify the partial differential equation, Eq. (4.1), to an
ordinary differential equation. This trick is known as a similarity transform
and effectively converts the problem of two independent variables, x and t,
into a problem of just one independent variable, the similarity transform
variable η. Such a transform is often seen in fluid mechanics and may be
possible for any problem that doesn’t have naturally–identifiable scales, for
example problems where there’s no clear length scale because of infinite
dimensions. Determining what the similarity transform variable is for a
4.6. THE SIMILARITY TECHNIQUE 49
specific problem is not trivial! In this case, the correct variable is

x
η = √ ·
4αt
Let us apply this to Eq. (4.1).
First, we use the Chain Rule to determine what the derivatives are with
respect to the new (transformed) variable η, specifically
∂T dT ∂η 2 x α 4 α t dT x dT
= = − 3/2
= − √
∂t dη ∂t (4 α t) 4 α t dη 2 t 4 α t dη
and
∂T dT ∂η 1 dT
= = √ ·
∂x dη ∂x 4 α t dη
Another derivative in x yields
∂2T 1 d2 T

d ∂T ∂η d 1 dT 1
= = √ √ = ·
∂x2 dη ∂x ∂x dη 4 α t dη 4αt 4 α t dη 2
Substituting these into Eq. (4.1), we find
x dT 1 d2 T
− √ = α ,
2 t 4 α t dη 4 α t dη 2
which can be simplified to
d2 T 4xt dT
= − √ ,
dη 2 2 t 4 α t dη
and finally
d2 T dT
(4.23) 2
= −2η .
dη dη
Note that Eq. (4.23) is now an ordinary differential equation (second–order),
having the single independent variable η.
The most straightforward method to solve this equation is to make the
variable substitution
dT
ξ= ,
dη
so that Eq. (4.23) is written as

d dT dT
= −2η
dη dη dη
and recast according to the new variable as
dξ
(4.24) = −2ηξ.
dη
Eq. (4.24) is separable, so that it can be written as
dξ
= − 2 η dη
ξ
4.7. MULTI–DIMENSIONAL TRANSIENT CONDUCTION 50
and then integrated once to obtain ln ξ = −η 2 + C0 , where C0 in an inte-

gration constant. We can write an equivalent equation in exponential form
as
2 +C 2
ξ = e−η 0
= C1 e−η ,
where C1 replaces the equivalent constant exp(C 0 ). Recalling our variable
substitution, we can now write this equation in the form
dT 2
(4.25) = C1 e−η .
dη
Eq. (4.25) is the result of doing one integration on our second–order Eq. (4.23).
While this equation appears to be in good shape to perform the second in-
tegration, i.e.
Z
2
(4.26) T = C1 e−η dη + C2 ,
the process is actually not trivial. If, for example, we prescribe boundary
conditions of the first–kind, e.g. T = T s at x = 0, we obtain the solution
T − Ts
(4.27) = erf η ,
T0 − T s
where erf is the Gaussian Error Function. This function is tabulated in
reference texts (e.g. Incropera and Dewitt, 2002, Appendix B.2, pp. 935).
The surface heat flux can be obtained by applying Fourier’s Law, from which
one obtains
Ts − T 0
(4.28) qs00 = k √ ·
παt
Although somewhat more involved, solutions for boundary conditions of the
second and third kind can be obtained as well (Incropera and Dewitt, 2002).
4.7. Multi–dimensional Transient Conduction

Transient problems that depend upon more than 1 dimension are often
encountered in practice. If there is no energy generation, it is possible to
combine 1–D solutions that we have looked at to construct, for example, 2–D
solutions for transient problems. In particular, we can show that if the one–
term approximation is valid, then multi–dimensional solutions are merely the
simple products of corresponding one–dimensional solutions. This approach
is called the method of product solution, and is applicable only if the solution
of the general problem can be shown to be equivalent to the product of 2
one–dimensional transient heat conduction problems. What is the theory
behind this?
Consider a rectangular region of dimension 2L 1 ×2L2 , where −L1 ≤ x ≤
L1 and −L2 ≤ y ≤ L2 and where the initial temperature is T 0 (Fig. 4.5). The
y=L2 y
x
x = − L1
x=L1
y = − L2
Figure 4.5. Two–dimensional heat conduction domain.
problem is driven by convection on all 4 faces at an ambient temperature of

T∞ . The dimensionless temperature is defined as
T (x, y, t) − T∞
θ(x, y, t) = ,
T0 − T ∞
where θ(x, y, t) is governed by the multi–dimensional equation
∂θ ∂2θ ∂2θ
(4.29) = + .
∂t ∂x2 ∂y 2
Eq. (4.29) is obtained via standard non–dimensionless techniques 4.6 shown
in § 4.4. We will examine the product solution technique for the governing
equation in 4.29 with the realization that boundary conditions are treated
in a similar fashion.
Let us now assume that we can represent θ(x, y, t) in the form of the
product θ(x, y, t) = θ1 (x, t)·θ2 (y, t). The components θ1 (x, t) and θ2 (y, t) are
one–dimensional solutions in the x and y coordinate directions, respectively,
and are further assumed to be given by the one–dimensional approximation.
That is, we assume these functions satisfy
∂θ1 ∂ 2 θ1
(4.30) =
∂t ∂x2
and
∂θ2 ∂ 2 θ2
(4.31) =
∂t ∂y 2
in the x and y directions4.7. Now, substitute θ(x, y, t) = θ1 (x, t) · θ2 (y, t) into
Eq. (4.29). For various components we obtain
∂θ ∂θ1 · θ2 ∂θ1 ∂θ2
= = θ2 + θ1
∂t ∂t ∂t ∂t
4.6We have omitted the star notation indicating dimensionless variables as shown in
§ 4.4 for brevity. Independent variables in Eq. (4.29) are taken to be dimensionless.
4.7Compare these equations to Eq. (4.10).
and
∂2θ ∂ 2 θ1 · θ 2

∂ ∂θ1 ∂θ2
2
= = θ2 + θ1 ,
∂x ∂x2 ∂x ∂x ∂x
so that
∂2θ ∂θ2 ∂θ1 ∂ 2 θ1 ∂θ1 ∂θ2 ∂ 2 θ2
= + θ 2 + + θ 1 .
∂x2 ∂x ∂x ∂x2 ∂x ∂x ∂x2
However, recalling that ∂θ2 /∂x = 0 since θ2 is not a function of the x
coordinate, this term simplifies to
∂2θ ∂ 2 θ1
= θ 2 .
∂x2 ∂x2
According to a similar procedure, we see that
∂2θ ∂ 2 θ2
= θ 1 .
∂y 2 ∂y 2
Substituting these terms into Eq. (4.29) and moving everything to the left
hand side, we get
∂ 2 θ1 ∂ 2 θ2

∂θ1 ∂θ2
(4.32) θ2 − + θ1 − = 0.
∂t ∂x2 ∂t ∂y 2
Eq. (4.32) is the original governing equation written in terms of our product
solution hypothesis. If it is valid, then either θ 1 and θ2 must both vanish, or
the terms within the brackets must both vanish. Now, θ 1 and θ2 do not gen-
erally vanish, otherwise the problem is trivial. Examining this equation more
closely we see that the terms in brackets are nothing more than Eqs. (4.30)
and (4.31). Thus, the terms in brackets do in fact vanish, and the original
hypothesis of the product solution is valid. As mentioned above, boundary
conditions can be treated similarly. Note how the presence of a heat gener-
ation term would prevent us from drawing our final conclusion since neither
of the equations for θ1 and θ2 would necessarily be zero. Of course, from
our one–dimensional work, we know various solutions corresponding to θ 1
and θ2 and the rules under which they (and their simplifications) can be
applied. So, once again, we’ve managed to find a significant simplification I&D Ex. 5.7
for a special case, i.e. two–dimensional problem with no heat generation. pp 277
Once it is determined that the product solution is valid, solving a prob-
lem proceeds by treating the one–dimensional problems individually. In
particular
• calculate individual Biot numbers for each coordinate direction us-
ing the appropriate length scale for each coordinate
• likewise, calculate individual Fourier numbers (dimensionless times)
for each coordinate direction
• compute individual one–dimensional solutions, e.g. θ 1 and θ2 , for
the required conditions of the two–dimensional problem
• take the product of the one–dimensional expressions as the final
solution of the problem
CHAPTER 5
Introduction to Convection
So far we have only considered convection to the extent that it supplies a

boundary condition for conduction problems. However, in many cases, con-
vection, driven by bulk motion of the participating medium, is the dominant
mode of heat transfer, even though there remains some level of conduction.
In this chapter, we focus on the underlying physics before actually develop-
ing calculation methods for quantifying the heat transfer.
Consider a fluid moving over a flat plate at free–stream velocity u ∞ and
free–stream temperature T∞ , where the surface of the plate is at temperature
Ts (Fig. 5.1). If Ts 6= T∞ then we can write the heat transfer for a differential
u oo Too
Ts x
dx
Figure 5.1. Flat–plate configuration showing boundary layer.
surface area using the usual Newton’s Law of Cooling

dq = h (Ts − T∞ ) dAs ,
where h is formally defined as the local coefficient of heat transfer or the
local convection coefficient. Whereas before we assume that h is a constant,
we know that, in reality, h will vary from point to point on the surface
because the flow conditions will vary. Clearly, the total heat transfer can be
obtained by integrating over the surface, as in
Z
q = h (Ts − T∞ ) dAs .
As
One of our basic assumptions will be that both T ∞ and Ts are constant, so
that their difference, Ts − T∞ , is also constant. The integration can then be
53
5.1. BOUNDARY LAYER INTRODUCTION 54
simplified as
Z
q = (Ts − T∞ ) h dAs .
As
Let us separately define an average heat transfer coefficient by again apply-
ing Newton’s Law of Cooling
q = h̄ (Ts − T∞ ) As .
From these two equations, it is clear that we can write the average convection
coefficient in terms of its local value as
1
Z
(5.1) h̄ = h dAs .
A s As
So we see that what we were really talking about all along in conduction
was an averaged value of the convection coefficient. Now the task becomes
actually figuring out what this is for specific problems.
For the special case of the flat plate in Fig. 5.1, the average convection
coefficient can be simplified. If we take the length of the plate as L and the
width as W , then As = W × L and dAs = W × dx. Integrating over the
length of the plate from the leading edge 0 → L, we find
Z L
1
(5.2) h̄ = h dx .
L 0
Determining convection coefficients is viewed as the “convection prob-
lem”. However, the problem is not trivial, since now we have the additional
complexity of fluid motion. So, not only is geometry, specific heat, conduc-
tivity, etc. important, we must now also worry about flow conditions, fluid
density, viscosity, etc.. Much of this depends upon the concept of boundary I&D Ex. 6.1
layer theory, which can be used to describe flow regions in the neighborhood pp 329
of a solid surface.
5.1. Boundary Layer Introduction

Recall the concept of the boundary layer from fluid mechanics. The drag
imposed by the fixed no–slip surface acts through the mechanism of shear
stress, τ , which gradually decreases as distance y from the plate increases.
At a distance defined as y = δ the drag effect becomes negligible. Therefore,
the velocity along the surface, u, gradually increases from zero at the surface
(no–slip boundary conditions) to a value of u ∞ , the free–stream velocity.
The convention used to define the boundary layer is δ = y at which u =
0.99u∞ . Thus the flow problem is defined by two distinct regions:
• The boundary layer in which velocity gradients and shear stresses
are significant.
• The region outside the boundary layer in which gradients and shear
stresses are negligible.
These observations refer to the fluid problem, so this boundary layer is

termed the velocity boundary layer. The primary factor, τ , is defined for a
Newtonian fluid according to Newton’s Law of Viscosity

∂u
τ = µ
∂y y=0
at the plate’s surface, where µ is the material property known as the dynamic
viscosity.
Just as the velocity boundary layer develops for the fluid problem, a
thermal boundary layer develops if the fluid free stream and the surface
temperatures differ (Fig. 5.2). Specifically, assume T s > T∞ . The incoming
Too
Too
y
δt
x
Ts
Figure 5.2. Thermal boundary layer on a flat plate.
flow has a uniform temperature distribution T ∞ , however, particles at the

plate surface attain thermal equilibrium at the plate’s higher surface tem-
perature Ts . In turn, these particles exchange energy with those near them,
etc., which is the same basic dynamic process familiar from the action of
viscosity. Therefore, a temperature profile develops, in which temperature
gradients within the thermal boundary layer, δ t , are significant, and outside
they are negligible. The layer itself is defined as y at which
Ts − T
= 0.99 .
Ts − T ∞
Note the correspondence with the factor of 0.99 for the velocity boundary
layer. The behavior of the thermal boundary layer is qualitatively similar
to the velocity boundary layer in that the thickness grows as x increases.
At the plate’s surface, the no–slip condition indicates there is no fluid
motion. It follows that heat transfer between the plate and the fluid imme-
diately adjacent to it must result solely from conduction. That is, we can
write heat transfer directly from Fourier’s Law of Conduction as

00 ∂T
q = −kf ,
∂y y=0
where kf is the thermal conductivity of the fluid. By combining this with

a local heat flux derived from Newton’s Law of Cooling, q 00 = h (Ts − T∞ ),
we see

kf ∂T
(5.3) h = − .
Ts − T∞ ∂y y=0
From this, we make the following interesting observation: T s − T∞ and kf
are constants, therefore variation in h depends upon ∂T /∂y. Because δ t
increases with x, the gradient decreases with x, therefore h and q 00 decrease
with x. The integrals in Eqs. (5.1) and (5.2) are expected to be non–trivial
since the local value of h in the integrand is not constant.
From fluid mechanics, we should recall that the physical structure of
the boundary layer depends upon whether it is laminar or turbulent. The
laminar regime is characterized by orderly flow in which streamlines can be
easily identified, while turbulent flow is chaotic and irregular. Chaotic mix-
ing in turbulent boundary layers increases momentum and energy transfer,
so that convection and shear stress near the plate are generally higher than
in the laminar case. The turbulent velocity profile is thus sharper than its
laminar counterpart (Fig. 5.3). The relevant parameter for boundary layers
FLOW
laminar
turbulent
Figure 5.3. Qualitative characterization of velocity profiles in

laminar and turbulent boundary layers.
is the Reynolds number , defined as

u∞ x
(5.4) Rex = ,
ν
where x is distance traveled as measured from the leading edge, ν is the fluid
kinematic viscosity, and u∞ is the free–stream velocity of the flow. Transi-
tion from a laminar to a turbulent boundary layer occurs at approximately
Rex ∼ 5 × 105 .
5.2. GOVERNING EQUATIONS 57
5.2. Governing Equations

From fluid mechanics, we recall further that the flow physics are governed
by the Navier–Stokes Equations, which describe conservation of mass and
conservation of momentum. If we assume steady flow over the flat plate in
the (x, y) domain as shown in Figs 5.1 and 5.2, we can write these equations
as
∂u ∂v
(5.5) + = 0,
∂x ∂y
2
∂2u

∂u ∂u 1 ∂P ∂ u
(5.6) u + v = − + ν + ,
∂x ∂y ρ ∂x ∂x2 ∂y 2
and
∂2v ∂2v

∂v ∂v 1 ∂P
(5.7) u + v = − + ν + ,
∂x ∂y ρ ∂y ∂x2 ∂y 2
where (u, v) are the velocity components in the (x, y) coordinate directions,
respectively. Eq. (5.5) represents the conservation of mass, while Eqs. (5.6)
and (5.7) are conservation of momentum in the (x, y) coordinate directions,
respectively.
These equations are typically derived using a differential approach much
like that introduced in Chapter 2 for the conduction equation 5.1. Expanding
this approach for thermal energy to include not only conduction, but also
convection and the mechanisms which degrade mechanical energy into heat,
we can derive the following equation for the conservation of energy
2
∂2T

∂T ∂T ∂ T ν
(5.8) u + v = α 2
+ 2
+ Φ,
∂x ∂y ∂x ∂y cp
where " 2 2 # 2
∂u ∂v ∂v ∂u
Φ = 2 + + +
∂x ∂y ∂x ∂y
is the viscous energy dissipation function. The left hand side of Eq. (5.8)
represents the net outflow of energy from the control volume due to convec-
tion5.2, while the term in brackets is the net rate at which energy is con-
ducted into the control volume5.3. The last term, Φ, denotes thermal energy
that arises via the action of viscous (frictional) forces degrading mechanical
energy.
This system of equations is, in fact, more comprehensive than what is
actually necessary to model flow over the flat plate because certain gradients
in the streamwise direction are very small compared to those in the normal
direction. We can invoke the so–called boundary layer approximation to
5.1Recall that Eqs. (5.6) and (5.7) arise from applying Newton’s Second Law.
5.2That is, the energy convected into the control volume minus the energy convected
out.
5.3Recall this interpretation from Eq. (2.19) on pp. 12.
derive a simplified system. The simplification process proceeds by examining

the order of magnitude of each of the terms (Schlichting, 1979).
• Assume in continuity Eq. (5.5) that ∂u/∂x is of order unity, which
we write as ∂u/∂x ∼ [1]. Since there are no parameters in this
equation and we know that the flow must develop along the axis, i.e.
∂/∂x 6= 0, the term ∂v/∂y must be the same order of magnitude.
That is ∂v/∂y ∼ [1].
• The free–stream flow is of order unity, i.e. u ∼ [1].
• The flow remains predominantly oriented along the plate so that
the vertical velocity component is much smaller than the horizontal
component, v u. Therefore, we write v ∼ [], where 1. That
is, is a small number.
• Since changes in the vertical direction happen over a very short
distance, i.e. the boundary layer thickness, then gradients in the
vertical direction must be much larger than gradients along the
plate, i.e. ∂/∂x ∂/∂y. Since u ∼ [1] and ∂u/∂x ∼ [1], we can say
that ∂/∂x ∼ [1]. Since ∂/∂x ∂/∂y, we can write ∂/∂y ∼ [1/].
Note that we could also infer this from the fact that ∂v/∂y ∼ [1],
but that v ∼ []. Thus, we see ∂ 2 u/∂ 2 x ∼ [1], ∂v/∂x ∼ [] and
∂ 2 v/∂ 2 x ∼ []. Moreover, ∂u/∂y ∼ [1/], ∂ 2 u/∂ 2 y ∼ [1/2 ], and
∂ 2 v/∂ 2 y ∼ [1/].
• In the boundary layer, inertial and viscous effects must be of the
same overall order of magnitude. If we were to rewrite the Eqs. (5.5)
through (5.7) in dimensionless form, i.e. using the Reynolds num-
ber as defined in Eq. (5.4), we would see that Re ∼ [1/ 2 ] for this
to be the case.
Following the last point, let us rewrite Eqs. (5.5) through (5.7) in dimen-
sionless form along with the orders of magnitude for each term we have
deduced above. Showing the orders of magnitude in square brackets beside
each term, we find
1 2 ∂2u ∂ 2 u −2

∂u ∂u −1 ∂P
(5.9) u [1]·[1] + v []·[ ] = − + [ ] [1] + [ ]
∂x ∂y ∂x Re ∂x2 ∂y 2
and
∂2v ∂ 2 v −1

∂v ∂v ∂P 1 2
(5.10) u [1] · [] + v [] · [1] = − + [ ] [] + [ ]
∂x ∂y ∂y Re ∂x2 ∂y 2
for the momentum equations. As we said above, the entire continuity equa-
tion is of order unity, i.e.
∂u ∂v
(5.11) [1] + [1] = 0 .
∂x ∂y
We did not write down the orders of magnitude for the pressure terms in
Eqs. (5.9) and (5.10). We can infer that the pressure gradient in Eq. (5.10)
should be the same order of magnitude as the equation itself. In other
words, changes in pressure across the boundary layer should be of compa-

rable size to the other terms in the equation. Specifically, ∂P/∂y ∼ [],
since Eq. (5.10) is of overall order []. This leads us to conclude P ∼ [ 2 ]
across the boundary layer (in the vertical direction), since ∂P/∂y ∼ [] and
∂/∂y ∼ [1/]. Therefore, P is approximately only a function of x, so that
we assume the pressure term in Eq. (5.9) is of the same overall magnitude
as the equation itself, i.e. order [1].
We can now make some simplifying observations. First, ∂ 2 u/∂x2 in
Eq. (5.9) is very small compared to the other terms in that equation, so it
can be dropped. Also, the entire Eq. (5.10) is small compared Eq. (5.9),
so that it may be neglected. These deductions show that Eq. (5.9) through
(5.11) can be simplified to
∂u ∂u 1 ∂P ∂2u
(5.12) u + v = − + ν
∂x ∂y ρ ∂x ∂y 2
and
∂u ∂v
(5.13) + = 0.
∂x ∂y
We can also apply the same order of magnitude analysis for energy in
Eq. (5.8). We assume
• temperature is of order unity, T ∼ [1]
• similar to the momentum equations, we see that the convective
terms will be of order unity, so that the viscous terms should also be
of overall order unity — since the largest viscous term is ∂ 2 T /∂ 2 y ∼
[1/2 ] (similar again to the momentum equations), we must have
α ∼ [2 ] for the overall viscous order to be unity.
• the dissipation term Φ should also be of overall order of magnitude
unity — the largest term5.4 here is (∂u/∂y)2 ∼ [1/2 ], so that we
see ν/cp ∼ [2 ]
We can then write the energy equation as
2
∂ 2 T −2

∂T ∂T 2 ∂ T ν
(5.14) u [1]·[1]+v −1
[]·[ ] = α[ ] 2
[1] + 2
[ ] + [2 ]Φ,
∂x ∂y ∂x ∂y cp
where
" 2 2 # 2
∂u ∂v ∂v ∂u −1
Φ = 2 [1] + [1] + [] + [ ] .
∂x ∂y ∂x ∂y
Dropping terms of order [] and higher, this simplifies to
2
∂T ∂T ∂2T ν ∂u
(5.15) u + v = α 2
+ .
∂x ∂y ∂y cp ∂y
Eqs. (5.12), (5.13), and (5.15) are the so–called boundary layer equations.
5.4The basic term ∂u/∂y is clearly of order [1/], but squaring this term raises it to
2
[1/ ].
5.3. DIMENSIONLESS PARAMETERS 60
Reiterating some of the interesting characteristics of these equations:

• The pressure does not vary in the y direction.
• Since the pressure is constant in the y direction across the boundary
layer, its value is equivalent to the value in the free–stream.
• Thus P = P (x) and ∂P/∂x → dP/dx, which further implies that
P (x) can be determined from free–stream conditions so that dP/dx
is a known quantity.
• No terms drop out of the continuity equation.
5.3. Dimensionless Parameters

Unfortunately, we are still faced with the fact that this system of equa-
tions is non–linear and cannot be solved with the mathematical techniques
that we have utilized so far. There are advanced techniques that can be
employed (e.g. similarity approaches that we outlined earlier), but usually
a full–blown numerical approach is used. So why do we bother to study this
problem here? Aside from developing an appreciation for the physical pro-
cesses, our primary goal is to identify relevant non–dimensional parameters
for the problem (much like we did for conduction) and to develop some im-
portant analogies between momentum and heat transfer. For the moment,
we will neglect the viscous dissipation term.
As usual, let us define characteristic length, L, velocity, u ∞ , and tem-
perature, T∞ , scales for the problem at hand to non–dimensionalize the
variables as
x y
x∗ = , y∗ = ,
L L
u v
u∗ = , v∗ = ,
u∞ u∞
and
T − Ts
T∗ = ·
T∞ − T s
Performing the non–dimensionalization on the equations according to the
results we developed in §4.4 on pp. 42, we find
∂u∗ ∂v ∗
+ = 0,
∂x∗ ∂y ∗
∂u∗ ∗ ∂u
∗ dP ∗ ν ∂ 2 u∗
u∗ + v = − + ,
∂x∗ ∂y ∗ dx∗ u∞ L ∂y ∗2
and
∂T ∗ ∗ ∂T
∗ α ∂2T ∗
u∗ + v = .
∂x∗ ∂y ∗ u∞ L ∂y ∗ 2
From these equations, we can identify two non–dimensional parameters
• Reynolds number: defined as Re = u∞ L/ν
• Prandtl number: defined as P r = ν/α
The Prandtl number is clear in light of the following from the energy equa-
tion
u∞ L u∞ L ν u∞ L ν
= = = Re · P r .
α α ν ν α
Thus, the final non–dimensionalized form of the equations can be written
∂u∗ ∂v ∗
(5.16) + = 0,
∂x∗ ∂y ∗
∂u∗ ∗ ∂u
∗ dP ∗ 1 ∂ 2 u∗
(5.17) u∗ + v = − + ,
∂x∗ ∂y ∗ dx∗ Re ∂y ∗ 2
and
∂T ∗ ∂T ∗ 1 ∂2T ∗
(5.18) u∗ + v∗ = .
∂x ∗ ∂y ∗ Re · P r ∂y ∗ 2
The Reynolds number is familiar from fluid mechanics. It can be interpreted
as the ratio of inertial forces to viscous forces, so it serves as a direct metric
of how well frictional dissipation is able to dampen perturbations in a flow.
This of course bears directly on whether a particular flow is laminar or
turbulent.
Conversely, the Prandtl number is new — it appears only in the energy
equation. Taking a closer look, we see that P r is a direct property of the
fluid itself. It does not contain any non–fluid properties, as for example the
velocity and length scales in the Reynolds number. We can therefore speak
of “the Prandtl number of a fluid”. How do we interpret this parameter
physically? The Prandtl number is the ratio of kinematic viscosity ν to
thermal diffusivity α, the former of which indicates the rate of momentum
diffusion (from the viscous terms in the momentum equations) and the latter
of which is the rate of thermal diffusion
ν momentum diffusion
Pr = = ·
α thermal diffusion
We can say equivalently that P r represents the ratio of the effectiveness
of diffusional energy transport in the velocity boundary layer versus the
thermal boundary layer. In laminar flows, we can thus infer
δ
Pr ∼ ,
δt
that is, P r gives some approximate indication of the thickness of the velocity
boundary layer versus the thermal boundary layer 5.5 (Table 5.1).
Of course, we still have not lost anything in terms of non–linearity of
the system in Eqs. (5.16) through (5.18), however, just as previously, the
5.5This is not the case in turbulent flows where transport is a function of turbulent
mixing, as well as diffusion.
Table 5.1. Approximate boundary layer relationships for

various fluid types
Fluid Pr relationship
gases ∼1 δ ≈ δt
liquid metals 1 δ δt
oils 1 δ δt
number of relevant variables to any problem has been reduced. For example,
the velocity distribution should be of the form

∗ ∗ ∗ ∗ dP ∗
(5.19) u = u x , y , Re, ,
dx∗
where dP ∗ /dx∗ is considered to be determined by the geometry of the prob-
lem and the free–stream conditions and is considered a known parameter as
mentioned previously.
What other dimensionless parameters arise for such configurations? Let
us consider frictional drag on the body surface that arises because of viscous
shear stress. Recall from fluid mechanics that the stress is defined as

∂u µ u∞ ∂u∗
τ = µ = .
∂y
y=0 L ∂y ∗ ∗y =0
This expressions allows us to derive the following form of the skin friction
factor , defined by Cf = τ /(ρu2∞ /2), as

2 ∂u∗
Cf = .
Re ∂y ∗ y∗ =0
Eq. (5.19) indicates that ∂u∗ /∂y ∗ evaluated at y ∗ = 0 is only a function of
(x∗ , Re, dP ∗ /dx∗ ). Therefore, we arrive at the significant conclusion that
the skin friction factor, an important engineering quantity, is only governed
by three “universal” parameters according to the functional form

∗ dP ∗
(5.20) Cf = Cf x , Re, .
dx∗
Likewise, we see that T ∗ is essentially a function of the form

∗ ∗ ∗ ∗ dP ∗
(5.21) T = T x , y , Re, P r, ,
dx∗
so that it has the same dependencies of u ∗ , except with the addition of the
Prandtl number as another parameter. Recall that we cast the convection
coefficient in Eq. (5.3) on pp. 56 as

kf ∂T
h = − ,
Ts − T∞ ∂y y=0
5.4. REYNOLDS–COLBURN ANALOGY 63
which, we can recast with dimensionless variables as

kf T∞ − Ts ∂T ∗ kf ∂T ∗
h = − = .
L Ts − T∞ ∂y ∗ y∗ =0 L ∂y ∗
y ∗ =0
This expression leads directly to the definition of the Nusselt number

∂T ∗ hL
Nu = = ,
∂y y∗ =0
∗ kf
which is essentially the non–dimensional temperature gradient at the wall.
As with the skin friction factor, y ∗ does not come into play at the wall, i.e.
at y ∗ = 0, so that the Nusselt number depends only upon variables as

∗ dP ∗
(5.22) N u = N u x , Re, P r, .
dx∗
The Nusselt number, as we have written it here, depends on the local con-
vection coefficient h, so that we refer to this form as the “local Nusselt
number”. We can also write an average Nusselt number by simply substi- I&D Ex. 6.5
tuting the average convection coefficient h̄. Note that, while the form of pp 351
the Nusselt number and the Biot number appear the same, k is that of the
fluid for the Nusselt number (convection) and that of the solid for the Biot
number (conduction).
5.4. Reynolds–Colburn Analogy

Now that we have introduced Cf , N u, Re, and P r as important dimen-
sionless parameters in convection, we will develop an analogy among these
parameters, so that we can get the answer to a convection problem by solv-
ing a fluid mechanics problem5.6. Here, we will derive the Reynolds–Colburn
analogy for laminar flow along a flat plate, which has the functional form
N u = N u (Cf , Re, P r) .
We will be able to calculate the Nusselt number from the Prandtl and
Reynolds numbers and the skin friction factor. This is particularly use-
ful, since measuring Re and Cf for the fluid mechanics problem is more
straightforward than measuring N u directly.
The rigorous development of this result relies on the exact solution of
the boundary layer problem, as posed by Eqs. (5.16) through (5.18), where
the the pressure gradient along the plate is assumed to vanish. We are not
in a mathematical position to derive this solution, but we will rather use
the approximate solution as a surrogate, so that the main concept can at
least be illustrated5.7. Assume, according to Figs. 5.1 and 5.2, that u ∞ is
5.6Recall in Chapter 3 where the circuit analogy allowed us to get the answer to a
heat conduction problem by solving an electrical circuit problem.
5.7Most textbooks write down the exact solution directly in the form of the required
dimensionless parameters (using the “please believe this” approach), so that the analogy
is trivially obvious. We start with the approximate solution for velocity and temperature
distribution. This is only a “please believe this temporarily” approach, as we will actually
free–stream velocity, y is the coordinate normal to the plate, δ is the velocity

boundary layer thickness, and δt is the thermal boundary layer thickness.
We start with an approximate solution for the boundary layer problem
r
∗ u 3 y 1 y 3 280 νx 4.64 x
(5.23) u = = − where δ = ≈ ,
u∞ 2 δ 2 δ 13 u∞ Rex
and
1 y 3

T − Ts 3 y 4.52 x
(5.24) T ∗ = = − where δt = 1/2
·
T∞ − T s 2 δt 2 δt Rex P r 1/3
Deriving the analogy is matter of manipulating the solutions appro-
priately. Writing the skin friction factor in the usual fashion as C f =
τ / 0.5ρu2∞ , and substituting τ = µ ∂u/∂y = ν ρ ∂u/∂y at y = 0 from

Newton’s Law of Viscosity, we can write

2 ν ∂u
Cf = 2 .
u ∂y
∞ y=0
From the solution for the velocity profile in Eq. (5.23), we can solve for the
velocity gradient at the wall as

∂u 3 u∞
= ,
∂y y=0 2δ
so that a substitution yields Cf as
r r
2 ν 3 u∞ 3ν 3ν 13 u∞ 117 ν 0.646
Cf = 2 = = = = 1/2
·
u∞ 2 δ u∞ δ u∞ 280 ν x 280 u∞ x Rex
Again, the constant 0.646 represents the approximate solution and is very
close to the value for the exact solution of 0.664, which we will now use. We
1/2
therefore write Cf = 0.664/Rex , or, in a more useful form
Cf 0.332
(5.25) = 1/2
·
2 Rex
This is one–half of what is needed to complete the analogy.
Referring back to Eq. (5.3) on pp. 56, we insert the dimensionless tem-
perature into the gradient to obtain

kf ∂T ∂T ∗
h = − = kf ,
Ts − T∞ ∂y y=0 ∂y y=0
From the solution of the temperature profile in Eq. (5.24), we can solve for
the temperature gradient at the wall as

∂T ∗ 3
= ,
∂y 2 δt
y=0
derive this solution in the following chapter. The only non–rigorous component of the
process is where we substitute a certain constant derived from the exact solution for its
counterpart from the approximate solution. This will be pointed out.
so that substitution yields h as

3 kf
h = ·
2 δt
Plugging in δt , we find
1/2
0.332 kf Rex P r 1/3
h ≈ ,
x
from which we can manipulate to write in terms of the Nusselt number as
hx
(5.26) Nu = = 0.332 Re1/2
x Pr
1/3
.
kf
Combining Eqs. (5.25) and (5.26), we can write
Cf Re P r 1/3
(5.27) Nu = ,
2
which is the Reynolds–Colburn analogy. This equation is applicable for lam-
inar flat plate flow in the range of approximately 0.6 < P r < 60 (Incropera
and Dewitt, 2002).
CHAPTER 6
External Convection
Now we would like to focus on convection problems of external flow

more closely. This includes cases such as flow over a flat plate, airfoil,
etc. We will concentrate on forced convection, meaning that a pressure
gradient or some other externally–applied force powers the flow (as opposed
to buoyancy effects). Once again, the primary objective is to determine the
convection coefficient, which may be found via the Nusselt number N u x =
N ux (x∗ , Rex , P r). Of course, there are a number of approaches for doing
this
• theoretical: apply analytical techniques to solve governing equa-
tions (this is the approach we exploited for all of our conduction
problems)
• empirical: perform experimental measurements under controlled
conditions and correlate data in terms of relevant dimensionless
parameters
• numerical: approximate governing equations with discrete expres-
sions and use computers to solve iteratively
As we have discussed in Chapter 5, convection problems are usually
non–linear, so the success of theoretical approaches will be limited to a few
specific cases. Conversely, empirical techniques lend themselves well to these
sorts of problems. We have already determined the relevant dimensionless
groups for the typical convection problem. Given sufficient measurement
data, results can be cast in the form of correlations, for example, equations
of the form
N ux = C Rem n
L Pr ,
where N ux is the averaged Nusselt number, and C, m, and n are problem–

specific constants that must be determined.
However, this sort of approach presents an immediate and obvious prob-
lem. It is well–known that fluid properties, for example viscosity and con-
ductivity, vary according to temperature. How do we properly account for
such variations in a correlation equation? One method is to evaluate prop-
erties at the so–called film temperature
Ts + T ∞
(6.1) Tf = ,
2
66
6.1. LAMINAR FLOW OVER A FLAT PLATE 67
which is simply the average of the temperature extremes. Another possibility

is the incorporation of additional correlation terms, for example of the form
(µ∞ /µs )r , to account for such variations.
6.1. Laminar Flow Over A Flat Plate

Before delving further into correlations, we will discuss cases that can
be treated purely via theory. Flow over a flat plate is one such configuration
(Figs. 5.1 and 5.2). Assuming no pressure gradient along the plate and
no viscous dissipation, the boundary layer form of the equations given by
Eqs. (5.12), (5.13), and (5.15) on pp. 59 can be written as
∂u ∂u ∂2u
(6.2) u + v = ν ,
∂x ∂y ∂y 2
∂u ∂v
(6.3) + = 0,
∂x ∂y
and
∂T ∂T ∂2T
(6.4) u + v = α ·
∂x ∂y ∂y 2
In Chapter 5, we derived the Reynolds–Colburn analogy using analytical
solutions, both approximate and exact, to this problem. Where do these
solutions come from? The exact solution is not trivial, in fact it is beyond
the mathematical capabilities we have developed thus far. Again, this is a
result of the non–linearity of the system in Eqs. (5.16) through (5.18) on
pp. 61. We can cast the solution procedure, but will be unable to follow it
to completion.
The similarity technique introduced in §4.6 can be used to reduce the
partial differential system to an ordinary differential equation. As we men-
tioned previously, one should suspect that a similarity approach may work
when no inherent scales can be identified in the problem. In this case, the
transformation variables
r
u∞
η = y and
νx
r
ψ νx
f (η) =
u∞ u∞
are appropriate, where ψ = ψ(x, y) is the stream function, which is defined
as
∂ψ ∂ψ
(6.5) u = and v = − ·
∂y ∂x
Transforming the momentum equation appropriately, we find it becomes
2f 000 + f f 00 = 0 ,
6.1. LAMINAR FLOW OVER A FLAT PLATE 68
where the prime symbol indicates the regular derivative of a uni–variate

function. Unfortunately, unlike the conduction problem in §4.6, this ordi-
nary differential equation is still non–linear. This equation can be solved in
a number of ways6.1 to yield the solution given in Table 6.1. The resulting
Table 6.1. Boundary layer solution
η f f 0 = u∗ f 00
0 0 0 0.332
0.4 0.027 0.133 0.331
0.8 0.106 0.265 0.327
1.2 0.238 0.394 0.317
1.6 0.420 0.517 0.297
2.0 0.650 0.630 0.267
2.4 0.922 0.729 0.228
2.8 1.231 0.812 0.184
3.2 1.569 0.876 0.139
3.6 1.930 0.923 0.098
4.0 2.306 0.956 0.064
4.4 2.692 0.976 0.039
4.8 3.085 0.988 0.022
5.2 3.482 0.994 0.011
5.6 3.880 0.997 0.005
6.0 4.280 0.999 0.002
6.4 4.679 1.000 0.001
6.8 5.079 1.000 0.000
boundary layer thickness is

5x
(6.6) δ = √
Rex
and the resulting skin friction coefficient is
0.664
(6.7) Cf = √ ·
Rex
Conversely, if we take a closer look at the energy equation in Eq. (5.18)
on pp. 61, we will see that it is linear, because the velocity components are
known. Using the similarity variables, we can transform this equation into
Pr
T ∗ 00 + f T∗ 0 = 0 ,
2
where T ∗ = T ∗ (η) and the prime symbol again denotes the regular derivative
of a uni–variate function. This can be solved, for example by numerical
6.1Numerically, this equation is fairly straightforward to solve, e.g. with the so–called
“shooting method”. Conversely, a series expansion technique can also be used to solve to
any desired degree of precision.
6.2. KARMAN–POHLHAUSEN INTEGRAL SOLUTION 69
integration for a specific Prandtl number, to yield T ∗ . The dimensionless

results can be cast as
hx x
(6.8) N ux = = 0.332 Re1/2
x Pr
1/3
for P r ≥ 0.6 .
kf
A number of additional results are shown in Incropera and Dewitt (2002),
in particular the local Nusselt number N u x in Eq. (6.8) can be integrated
to obtain an averaged value of
(6.9) N u x = 2 N ux .
6.2. Karman–Pohlhausen Integral Solution

The fact that the exact solution in the previous section is difficult from
a mathematical standpoint suggests that an approximate solution might be
useful. T. von Karman (1921) and K. Pohlhausen (1921) derived just such
a solution in the form of an integral method that can be solved to various
degrees of accuracy. In fact, we have already used this solution in the form
of Eqs. (5.23) and (5.24) on pp. 64 for the Reynolds–Colburn analogy. First,
we start with the continuity equation
∂u ∂v
(6.10) + = 0
∂x ∂y
and integrate it across the boundary layer:
Z δ Z δ
∂u ∂v
(6.11) dy + dy = 0 .
0 ∂x 0 ∂y
Now note that v = 0 at y = 0 (as part of the no–slip boundary condition)
and that the second part of the equation is simply dv because of the chain
rule. This gives:
Z δ
∂u
(6.12) v |y=δ = − dy .
0 ∂x
Now, we also integrate the boundary layer form of the momentum equa-
tion (6.2) over the boundary layer
Z δ Z δ Z δ 2
∂u ∂u ∂ u
(6.13) u dy + v dy = ν 2
dy ,
0 ∂x 0 ∂y 0 ∂y
which gives, after integrating the second term on the left hand side by
parts6.2 and recognizing the right hand side can be integrated directly,
∂u y=δ
Z δ Z δ
∂u y=δ ∂v
(6.14) u dy + uv |y=0 − u dy = ν .
0 ∂x 0 ∂y ∂y y=0
6.2If integrating by parts, we can consider ∂u

dy
as simply du by the Chain Rule, which
R ∂y R R
means we are evaluating vdu. Applying integration by parts, we see vdu = uv − udv.
However, ∂v∂y
dy = dv, again by the Chain Rule. This gives the second and third terms
on the left hand side of Eq. (6.14).
The first term remains the same, however, to further develop Eq. (6.14), we
make the following observations. The second term can be evaluated using
Eq. (6.12) for v and observing that u = 0 at y = 0 and u = u 0 at y = δ.
In the third term, we utilize the original continuity equation (6.10) to swap
∂v/∂y = − ∂u/∂x. The fourth term is evaluated using the fact that ∂u/∂y
is zero at y = δ. These modifications give
Z δ Z δ Z δ
∂u ∂u ∂u ∂u
(6.15) u dy − u0 dy + u dy = −ν ,
0 ∂x 0 ∂x 0 ∂x ∂y y=0
which can be simplified to
Z δ Z δ
∂u ∂u ∂u
(6.16) u0 dy − 2u dy = ν .
0 ∂x 0 ∂x ∂y
y=0
The u0 can be taken directly under the integral and the u can be “integrated”
in6.3, which yields:
Z δ
∂ ∂u
(6.17) (u0 · u − u · u) dy = ν .
0 ∂x ∂y y=0
Now, the right hand side is basically a term that is evaluated, and the
differential on the left hand side can be taken outside the integral (since the
term is an integral w.r.t. y) and done later, therefore ∂/∂x → d/dx, which
yields:
Z δ
d ∂u
(6.18) (u0 − u) · u dy = ν .
dx 0 ∂y y=0
Eq. (6.18) is an integral form of the boundary layer equations. If we examine
this equation carefully, we see that the right hand side is basically the shear
stress at the surface of the plate. That is,
Z δ
d
(6.19) (u0 − u) · u dy = τw .
dx 0
This quantity is unknown because we do not know the velocity profile u.
Similarly, the left hand side also cannot be evaluated because it contains
u. We apparently have an implicit equation in u that cannot be solved in
closed form.
However, we can apply reasonable approximations for u. For example,
let us assume that u is described by a polynomial 6.4. We know that the flow
is developing, therefore it is a function of both x and y, i.e. u = u(x, y).
However, we would like to use a polynomial that is a function of just a single
6.3i.e. 2u ∂u = ∂u2
∂x ∂x
6.4Here is where the idea of various degrees of accuracy enters into the problem.
Specifically, lower–order polynomials will, in general, yield less accurate results as com-
pared to higher–order polynomials. The order is limited by how many boundary conditions
we can identify because these must be used to evaluate the polynomial coefficients.
variable. Let us then assume the following: choose a third–order polynomial

having 4 constants6.5
u y y 2 y 3
(6.20) = a1 + a2 + a3 + a4 .
u0 δ δ δ
Eq. (6.20) appears to depend only upon y, however, since the boundary
layer thickness depends upon x, i.e. δ = δ(x) as shown in Eq. (6.6), our
solution for u/u0 is a function of both x and y. Let us describe the boundary
conditions that we will use to evaluate the coefficients in Eq. (6.20):
• No–slip boundary conditions at the wall give
(6.21) u |y=0 = 0 .
• The gradient of velocity vanishes at the edge of the boundary layer.
In other words, the rate of change of u with respect to y goes to
zero at y ≥ δ. This gives

∂u
(6.22) = 0.
∂y y=δ
• The velocity at the edge of the boundary layer is known to be u 0 ,

which gives
(6.23) u |y=δ = u0 .
• While these 3 boundary conditions are fairly obvious, a fourth can
be deduced with a little effort. Looking back to the boundary layer
momentum equation (6.2) we see that this can be evaluated at
y = 0 using, once again, the concept that there is no–slip at the
plate surface. Therefore, plugging in u = v = 0 at y = 0, we find
∂ 2 u

(6.24) = 0.
∂y 2 y=0
Now, we have four unknowns: a1 . . . a4 , and four equations with which to

evaluate them: (6.21) through (6.24). Carrying this out, we find a 1 = a3 =
0, a2 = 1.5, and a4 = −0.5, which gives
u 3 y 1 y 3
(6.25) = − .
u0 2 δ 2 δ
Note that this equation is the first part of the solution shown back in
Eq. (5.23). We now know the form of u/u 0 , but since we do not know
how δ varies, we still do not know the explicit solution for u/u 0 ! This is
where we will now apply the integral form of the boundary layer equations
we derived: i.e. Eq. (6.18).
We use Eq. (6.25) to now evaluate the terms in Eq. (6.18). Let us show
this explicitly:
6.5There are three boundary conditions that can be easily identified, and a fourth
that can be deduced without too much difficulty, as we shall see shortly.
• From Eq. (6.25), we write u as

1.5 0.5 3
u = u0 y − 3y
δ δ
• Starting inside the integral on the left hand side we can then write

1.5 0.5 3
u0 − u = u 0 1 − y + 3y
δ δ
so that

1.5 0.5 1.5 0.5
(u0 − u) · u = u20 1 − y + 3 y3 y − 3 y3
δ δ δ δ
• We can then evaluate the integral on the left hand side of Eq. (6.18)
as
Z δ Z δ
2 1.5 2.25 2 0.5 3 1.5 4 0.25 6
(u0 − u) · u dy = u0 y− 2 y − 3 y + 4 y − 6 y dy
0 0 δ δ δ δ δ
which is
1 7 δ

2 3 2 3 3 1 4 3 5
u0 y − 2y − 3y + y − y
4δ 4δ 8δ 10δ 4 28δ 6 y=0
which simplifies to
39 2
u δ
280 0
• The right hand side of Eq. (6.18) is easily evaluated as
3 νu0
2 δ
• Eq. (6.18) can then be written as

d 39 2 3 νu0
u0 δ =
dx 280 2 δ
• We note that δ is the only term on the left hand side that is a func-
tion of x. Everything else is constant with respect to x, therefore
we can write the equation as

39 2 dδ 3 νu0
u =
280 0 dx 2 δ
which is a simple separable differential equation that can be written
as
140 ν
δ dδ = dx
13 u0
• This equation can be integrated along the flow direction, that is,
in x as
140 ν
Z Z
δ dδ = dx
13 u0
from which we find

δ2 140 ν x
= + C0
2 13 u0
where C0 is a constant.
• Recalling the fact that the boundary layer thickness is zero at the
leading edge of the plate, that is δ(0) = 0, we find C 0 = 0 so that
δ2 140 ν x
=
2 13 u0
which can be written as
r r r
280 ν x 280 ν x 2 ν
(6.26) δ = = 2
x = 4.640955 x
13 u0 13 u0 x u0 x
Note that this equation is the second and concluding part of the solution
shown back in Eq. (5.23). We see that the second part of Eq. (6.26) is simply
the inverse of the square root of the Reynolds number based on x. We can
therefore write the boundary layer solution as
4.64 x
(6.27) δ ≈
Re1/2
We have now solved the problem, since Eq. (6.27), along with Eq. (6.25) de-
scribe the complete velocity profile. Note that the constant 4.64 is remark-
ably close to the value of 5.0 that was obtained from the full non–linear (and
much more difficult) exact solution! Using Eq. (6.19), we can also evaluate
the shear stress and the wall skin friction coefficient C f , for example
τw 0.646
(6.28) Cf = 2 = √ ,
ρ u0 /2 Rex
which is, once again, quite comparable to the exact solution in Eq. (6.7).
We assumed a third–order polynomial for the velocity profile in Eq. (6.20).
Of course, other forms could have been assumed as well. Table 6.2 compares
the exact solution to several approximate ones, including the cubic polyno-
mial we have used here. The procedure to solve Eq. (6.4) for temperature is
Table 6.2. Laminar boundary layer solutions

√ √
solution δ Rex /x Cf Rex
exact 5.0 0.664
linear 3.46 0.578
parabolic 5.48 0.730
cubic 4.64 0.646
sinusoidal 4.79 0.655
somewhat similar, but more tedious and leads to Eq. (5.24). Development
of this result is shown in a number of reference texts (e.g. Burmeister, 1983,
§8.3).
6.3. EMPIRICAL CORRELATIONS 74
6.3. Empirical Correlations

Now that we have comprehensively studied laminar flow over a flat plate,
we are compelled to admit our inability to examine more complicated con-
figurations from a purely theoretical perspective. However, we have charac-
terized boundary layer–type problems in terms of the relevant dimensionless
parameters and this provides a basis for correlating experimental measure-
ments for various configurations. We do not delve into the details of making
such measurements. Rather, we simply catalog some of the notable for-
mulae for basic problems. Various texts have more extensive listings (e.g
Özişik, 1985; Incropera and Dewitt, 2002). These texts also list the original
references for the correlations presented below.
It is important to understand the limitations of these formulae. They
are not general solutions! They are simply relationships that have been
found to be reasonably accurate in some limited range of the parameters 6.6.
Therefore, when solving problems, it is always important to make sure its
parameters correspond to the correlation that is chosen. Unless otherwise
noted, these results are based on evaluating fluid properties at the film
temperature.
6.3.1. Flat Plate. Theoretical results for the laminar flat plate were
restricted to P r > 0.6, however, recall from Table 5.1 that liquid metals
can have substantially lower Prandtl numbers. In this case, the thermal
boundary layer develops very rapidly, so that δ t δ. We can assume then
that u = u∞ over the whole thermal boundary layer and rework the thermal
boundary layer solution. This yields
(6.29) N ux = 0.565 P e1/2
x P r ≤ 0.05 and P ex ≥ 100 ,
where P ex = Rex P r is the Peclet number . A correlation valid for laminar
flow and all P r is given by
1/2
0.3387 Rex P r 1/3
(6.30) N ux = 1/4 P ex ≥ 100 .
1 + [0.0468/P r] 2/3
For turbulent flow, the correlation
(6.31) N ux = 0.0296 Re4/5
x Pr
1/3
0.6 ≤ P r ≤ 60 and Rex ≤ 107
is reasonably accurate.
Eqs. (6.29) through (6.31) are valid for the local Nusselt number. Corre-
lations for the averaged Nusselt number are somewhat more difficult because
they have to account for the fact that a boundary layer initially has a lam-
inar behavior, undergoes transition, and then becomes turbulent 6.7. One
6.6Accuracy may be on the order of about 20 %.
6.7Of course, if the area of interest is strictly laminar or turbulent, the appropriate
averaged Nusselt number can be obtained by direct integration according to Eq. (5.2) on
pp. 54.
way to handle this is to assume that transition occurs at a specific critical

location along the plate xc , which divides the plate into laminar and tur-
bulent regions. According to Eq. (5.2) on pp. 54, we could then model the
overall convection coefficient as
Z xc Z L
1
(6.32) h̄ = hlam dx + hturb dx .
L 0 xc
Using hlam based on Eq. (6.8) and hturb based on Eq. (6.31), we find I&D Ex. 7.1

4/5
pp 400
(6.33) N uL = 0.037 ReL − 871 P r 1/3 ,
where the limits are 0.6 ≤ P r ≤ 60 and the end of the plate at x = L is
assumed to lie in the turbulent zone, i.e. 5 × 10 5 ≤ Re ≤ 108 . (We assume
that transition occurs at Re = 5 × 105 .)
6.3.2. Cylinders in Cross–Flow. Engineering applications frequently

involve cross–flow over one or more cylinders (Fig. 6.1). As you might recall
laminar boundary layer turbulent boundary layer

large wake small wake
Figure 6.1. Laminar and turbulent boundary layer formation

for cross–flow over a cylinder.
from fluid mechanics, flow separation occurs at some point where the bound-
ary layer detaches from the surface, forming a low–pressure wake. The flow
is characterized by the Reynolds number
u∞ D
ReD = ,
ν
where u∞ is the velocity upstream of the cylinder and D is the diameter,
and by the coefficient of drag
FD
CD = ,
Af ρ u2∞ /2
where FD is the measured drag force and Af is the projected frontal area
normal to the flow. The character of the boundary layer (laminar or turbu-
lent) strongly influences where separation occurs, and thus the magnitude of
CD (Fig. 6.2). There is a significant decrease as transition and turbulent flow
become important at Reynolds numbers above 200,000 to 500,000. These
1000
100
drag coefficient
10
0.1
0.01
0.1 1 10 100 1000 10^4 10^5 10^6
Reynolds number
Figure 6.2. Coefficient of drag for cross–flow over a circular

cylinder.
features strongly influence the convection characteristics (e.g. Incropera and

Dewitt, 2002, Fig. 7.9, pp. 410).
An overall Nusselt number, valid over all Reynolds numbers Re D for
which data are available, has been proposed as
1/2
" 5/8 #4/5
0.62 ReD P r 1/3

ReD
(6.34) N uD = 0.3 + 1/4 1 + ·
282, 000
1 + [0.4/P r]2/3
The restriction on the Prandtl number is roughly Re D P r > 0.2. Again,
fluid properties are evaluated at the film temperature. Incropera and Dewitt
(2002) list several additional correlation equations that are appropriate for
groups of more than a single cylinder.
6.3.3. Spheres. Boundary layer effects for flow over spheres are sim-
ilar to those described for cylinders. The primary correlation for averaged
Nusselt number is
1/4

1/2 2/3

0.4 µ
(6.35) N uD = 2 + 0.4 ReD + 0.06 ReD Pr ,
µs
where all properties are evaluated at T ∞ , except µs , which is evaluated at the
sphere surface temperature The restrictions on Eq. (6.35) are approximately I&D Ex. 7.5
0.71 < P r < 380 and 3.5 < ReD < 7.6 × 104 (Incropera and Dewitt, 2002). pp 415
CHAPTER 7
Internal Convection
In Chapter 6, we examined convection in external flows, where the

growth of the boundary layer is not constrained. Here, we shall study con-
figurations of internal flow in which the boundary layer is not permitted to
grow without limitation, e.g. pipes, ducts, etc. Now, in addition to the
consideration of laminar versus turbulent flow, we must be concerned with
whether the flow is developing or fully developed 7.1. We will find that, unlike
for external flows, certain internal flow configurations can be solved exactly
in closed form since non–linear terms in the momentum equations may dis-
appear. These cases, as you probably remember from fluid mechanics, are
always those in which the flow is fully developed and thus the velocity vector
is only a function of the streamwise velocity component. For configurations
where the flow remains in a developing state, solution procedures will be
somewhat limited and we shall be compelled to rely more on empirical data
and correlations.
7.1. Laminar Pipe Flow

Let us start with the example of fully developed laminar flow in a circular
pipe (Fig. 7.1). As fluid enters the pipe, a boundary layer begins to form
boundary layer
inviscid core
boundary layer
x
developing flow (hydrodynamic entrance region) fully developed flow
Figure 7.1. Flow in a pipe with developing and fully developed

regions.
and grow inward from the periphery, eventually merging so as to encompass

the entire cross–sectional area of the pipe. The distance x f d at which this
7.1Recall from fluid mechanics that flow along the x coordinate direction is developing
if ∂/∂x 6= 0, whereas it is fully developed if ∂/∂x = 0 (with the exception of the pressure
gradient, which need not vanish).
77
7.1. LAMINAR PIPE FLOW 78
occurs is the hydrodynamic entry length. For x ≥ x f d , the flow is fully

developed, meaning quantities7.2 no longer change as a function of x, and
viscous effects are non–negligible throughout the entire cross–section of the
pipe.
The Reynolds number characterizing pipe flows is based on the pipe
diameter D and the average, or mean velocity over the cross–section ū
ū D
ReD = ·
ν
Transition from laminar to turbulent flow is usually in the neighborhood of
ReD ≈ 2300. For laminar flow, xf d is on the order
xf d ≈ 0.05 ReD D ,
while for turbulent flow it is in the approximate range
10 D < xf d < 60 D .
We might recall from fluid mechanics that for fully developed laminar
flow, the velocity vector simplifies to u = u (r). That is, the radial and az-
imuthal velocity components vanish, leaving only the streamwise component
u. Consequently, the governing equation for u is obtained by simplifying the
streamwise momentum equation

µ d du dP
(7.1) r = .
r dr dr dx
This is a second–order ordinary differential equation — recall that the flow is
driven by the pressure gradient dP/dx, which is taken as a known quantity.
Therefore, we simply integrate twice to obtain the velocity distribution
1 dP r 2

(7.2) u(r) = + C1 ln r + C2 ,
µ dx 4
where C1 and C2 are constants of integration. As usual, we must use bound-
ary conditions to evaluate them. One boundary condition is immediately
available from the no–slip requirement, u = 0 at r = r 0 , where r0 is the pipe
radius. However, this is a situation in which the second boundary condition
is not completely obvious. It can be derived from either of the following
physical observations:
• the flow is well–behaved, so that the natural log term must remain
finite at r = 0
• because of symmetry of the overall problem, the velocity profile
must also be symmetric about r = 0
Both these observations yield the same result that C 1 = 0. For example,
u|r=0 = 0 + C1 ln 0 + C2 = finite C1 = 0 .
7.2Again, except for the pressure gradient, which cannot vanish, otherwise there would
be no driving force to sustain the flow.
7.1. LAMINAR PIPE FLOW 79
Applying no–slip, we find

r02

1 dP
u|r=r0 = + C2 = 0 ,
µ dx 4
so that
1 dP r02

C2 = −
µ dx 4
Substituting C1 and C2 into the general solution given by Eq. (7.2), we find
the exact solution is
1 dP r 2 1 dP r02

u(r) = − ,
µ dx 4 µ dx 4
or with further simplification
" 2 #
1 dP r
(7.3) u(r) = − r02 1− .
4µ dx r0
Note that the pressure gradient must be negative to drive the flow in the
positive x direction, i.e. the flow moves from high pressure to low pressure.
Volume flow rate can be calculated by integrating the velocity profile as
π r04 dP
(7.4) Q = −
8 µ dx
as shown in Appendix B. Again, the negative sign in Eq. (7.4) is com-
pensated for by the negative pressure gradient dP/dx. According to the
definition Q = ūA, we immediately have the average, or mean velocity as
Q Q r02 dP D 2 dP
(7.5) ū = = = − = − .
A π r02 8 µ dx 32 µ dx
We have rewritten the result in terms of the pipe diameter D in the last
expression. From this result, we see that the exact solution in Eq. (7.3) can
be expressed in the equivalent form
" 2 #
r
(7.6) u(r) = 2 ū 1 − .
r0
Also, the mass flow rate ṁ = ρūA = ρQ is
ρ π r04 dP
(7.7) ṁ = −
8 µ dx
Furthermore, Eqs. (7.3) and (7.6) indicate that the maximum velocity, which
occurs at r = 0, is
r02 dP
(7.8) umax = u(0) = − = 2 ū .
4 µ dx
Now that the fluid mechanics problem is solved, we can turn our atten-
tion to the heat transfer problem. The immediate trouble we see is that,
unlike for external flows, there’s no fixed free–stream reference temperature
7.2. THE CASE OF CONSTANT HEAT FLUX 80
T∞ . The convention is instead to use a mean or “bulk” temperature defined

as Z
ρ u cv T dAcs = ṁ cv Tm .
Acs
From this expression, we see that Tm is based on the thermal energy trans-
ported as the flow moves through a cross–section of the pipe 7.3. Let us
assume constant values for density ρ and specific heat c v . Moreover, dAcs =
2 π r dr and Acs = π r02 , which gives
Rr
ρ cv 0 0 u T 2 π r dr
Tm = ,
ρ ū π r02 cv
or simplifying
r0
2
Z
(7.9) Tm = u T r dr .
ū r02 0
This is now a suitable quantity to use in our definition of the convective
heat transfer coefficient, which we define for this problem as ( Özişik, 1985,
pp. 243)
(7.10) qs00 = h (Tm − Ts ) ,
where Ts and qs00 are the temperature and heat flux at the surface of the
pipe, respectively.
7.2. The Case of Constant Heat Flux

Note that unlike T∞ for external flows, Tm as defined by Eq. (7.9) is
generally not constant, but varies with axial distance. If the fluid is be-
ing heated (Ts > Tm ), then Tm will increase along the flow direction, and
vice versa. If that is the case, then clearly dT m /dx 6= 0, so it appears that
we can never realize the fully developed condition for the convection prob-
lem. Fortunately, this apparent contradiction is resolved if we work with
dimensionless temperature. Let us define
T (r, x) − Ts (x)
(7.11) θ = .
Tm (x) − Ts (x)
We then take fully developed thermal conditions to be defined by ∂θ/∂x = 0,
which can be obtained for the case of constant wall heat flux q s00 .
Let us now deduce some consequences laid out by Eq. (7.11). If ∂θ/∂x =
0, then θ cannot be a function of x for thermally fully developed flow. It
follows that ∂θ/∂r cannot likewise be a function of x. We can evaluate

∂θ ∂ T (r, x) − Ts (x) 1 ∂T
= = ,
∂r r=r0 ∂r Tm (x) − Ts (x) Tm (x) − Ts (x) ∂r
r=r0 r=r0
7.3That is, ρu is the mass flux and c T is the energy per unit mass. Integrating
v
their product, the energy flux, over the cross–section gives the rate at which energy is
transported through the cross–section.
and by the same argument conclude that this result is not a function of x.
In other words, this expression remains fixed as the flow proceeds in the x
direction. If we substitute surface heat flux from Newton’s Law of Cooling
in Eq. (7.10), qs00 /h = (Tm − Ts ), and from Fourier’s Law of Conduction,
∂T /∂r = −qs00 /k at r = r0 then we find
" #
1 ∂T h qs00 h
= 00 − = ,
Tm (x) − Ts (x) ∂r r=r0
qs k r=r0 k
from which we must conclude that the local convection coefficient h is inde-
pendent of x for fully developed flow 7.4.
Furthermore, Eq. (7.10) indicates that for constant h and constant q s00 ,
the difference Ts − Tm must be a constant that does not vary with x. We
already know that Ts and Tm change with x, so that, if their difference is a
constant, the rates of change of these two temperatures must be the same
in the fully developed regime, i.e.
dTm dTs
(7.12) = ·
dx dx
Let us go one step further. We evaluate ∂θ/∂x using θ as defined in
Eq. (7.11). Our initial assumption of fully developed conditions requires
∂θ/∂x = 0. Evaluating, we find

∂θ ∂ T (r, x) − Ts (x)
=
∂x ∂x Tm (x) − Ts (x)
∂
= [T (r, x) − Ts (x)] [Tm (x) − Ts (x)]−1
∂x
1 ∂ [T (r, x) − Ts (x)]
=
Tm (x) − Ts (x) ∂x
T (r, x) − Ts (x) ∂ [Tm (x) − Ts (x)]
−
[Tm (x) − Ts (x)]2 ∂x
= 0,
The second term vanishes since ∂/∂x of [T m (x) − Ts (x)] was just shown to
be 0 in Eq. (7.12). This leaves
1 ∂ [T (r, x) − Ts (x)]
= 0,
Tm (x) − Ts (x) ∂x
where Tm (x) − Ts (x) is, again, a non–zero constant according to Eq. (7.10).
We can therefore write more concisely
∂ [T (r, x) − Ts (x)]
= 0,
∂x
7.4Since the thermal conductivity k is constant, h must not be a function of x if the

expression as a whole cannot be a function of x.
which leads us to
∂T (r, x) dTs (x)
= ,
∂x dx
where we have changed ∂ to a plain derivative, since T s is only a function
of x. In light of Eq. (7.12), we now conclude
∂T (r, x) dTm
(7.13) = ·
∂x dx
In other words, the streamwise temperature gradient anywhere in the cross– I&D Ex. 8.1
section is equal to the streamwise gradient of the mean temperature. This pp 475
result will be used in solving for the temperature distribution T (r, x) in the
pipe and, ultimately, the Nusselt number for laminar fully developed flow.
We now address the actual problem of laminar fully developed pipe flow
having a constant applied heat flux q s00 at the pipe surface r = r0 . We assume
constant properties and no energy generation. Derivation of the convection
energy equation in cylindrical coordinates is beyond our present scope, but
it is shown in numerous reference texts (e.g. Bejan, 1984, Chapter 1). For
steady flow in (x, θ, r) coordinates (x is in the streamwise axial direction),
we have
2
1 ∂2T

∂T vθ ∂T ∂T ∂ T 1 ∂ ∂T
(7.14) u + + wr =α + 2 + r ,
∂x r ∂θ ∂r ∂x2 r ∂θ 2 r ∂r ∂r
where u is again the axial streamwise velocity component, and v θ and wr are
the azimuthal and radial velocity components. According to our hydrody-
namic solution, u = u (r) is given by Eq. (7.3), while v θ = wr = 0. Also, the
problem is symmetric, so that ∂/∂θ = 0. We will further assume 7.5 that the
mean temperature Tm is a linear function of x, so that ∂T /∂x is a constant
in Eq. (7.13) and therefore ∂ 2 T /∂x2 = 0. Eq. (7.14) then simplifies to

dTm 1 ∂ ∂T
(7.15) u(r) = α r .
dx r ∂r ∂r
If we examine this equation more closely, we notice that since dT m /dx is
constant, the equation can be written as an ordinary differential equation

dTm 1 d dT
(7.16) u(r) = α r .
dx r dr dr
Recasting this equation in dimensionless temperature θ, as defined by Eq. (7.11),
we write

dTm 1 d dθ
(7.17) u(r) = α (Tm − Ts ) r ,
dx r dr dr
which can be re–arranged as

1 d dθ u(r) dTm
(7.18) r = .
r dr dr α (Tm − Ts ) dx
7.5This can be shown according to an energy balance (e.g. Incropera and Dewitt,
2002, §8.3.1).
Using the solution for u(r) expressed in terms of the average velocity ū
in Eq. (7.6), we can write
" 2 #
d dθ r
(7.19) r = Ar 1− ,
dr dr r0
where A is the constant defined by
2 ū dTm
A = .
α (Tm − Ts ) dx
Eq. (7.19) can be integrated once to get
∂θ A r2 A r4
r = − + C1
∂r 2 4 r02
and then again to obtain
A r2 A r4
θ = − + C1 ln r + C2 ,
4 16 r02
where C1 and C2 are constants of integration.
The appropriate boundary conditions are similar to the hydrodynamic
problem. Temperature has a given value T s at the wall, r = r0 , so that θ = 0
at r = r0 . A boundary condition at r = 0 can be deduced from either of
• the temperature is well–behaved, so that the natural log term must
remain finite at r = 0
• because of symmetry of the overall problem, the temperature profile
must also be symmetric about r = 0
Both these observations yield the same result that C 1 = 0. Applying the
remaining condition, we find
3
C2 = − A r02 ,
16
which enables us to write θ as
A r2 A r4 3
θ = − 2 − A r02 .
4 16 r0 16
A equivalent, but more convenient form is
" 4 2 #
2 3 1 r 1 r
(7.20) θ = − A r0 + − .
16 16 r0 4 r0
The unknown value of A can be determined by employing the definition
of the bulk mean temperature in Eq. (7.9). Let us write this in dimensionless
form by forming θ on both sides (subtract T s , then divide by Tm −Ts ), which
yields
Z r0
Tm − T s 2 T − Ts
(7.21) = 2 u r dr .
Tm − T s ū r0 0 Tm − T s
Converting to dimensionless notation, this equation is

Z r0
2
(7.22) 1 = u θ r dr .
ū r02 0
Substituting in for u and θ and integrating, we obtain
96
(7.23) A r02 = − ,
11
as shown in Appendix B. Using this result, we can complete Eq. (7.20) as
" 4 2 #
96 3 1 r 1 r
(7.24) θ = + − .
11 16 16 r0 4 r0
This is the dimensionless temperature profile for the forced convection in a
circular tube for constant wall heat flux.
Is it possible to develop this result even further? Fourier’s Law written
at the pipe’s surface assumes the form

00 ∂T
qs = − k f .
∂r
r=r0
Combining this with the expression for Newton’s Law of Cooling in Eq. (7.10),
we can express the convection coefficient as

kf ∂T
h = − .
Tm − Ts ∂r r=r0
Let us now express h in terms of the dimensionless temperature θ from
Eq. (7.11), where T (r, x) = θ [Tm (x) − Ts (x)] + Ts (x). We find

∂θ
(7.25) h = − kf .
∂r r=r0
Using the solution for θ in Eq. (7.24) to evaluate ∂θ/∂r at r = r 0 , we find

∂θ 96 96 48
= − = − = − .
∂r
r=r0 44 r0 22 D 11 D
Substituting into Eq. (7.25) yields
48
h = kf
11 D
from which we can identify the Nusselt number as
hD 48
(7.26) Nu = = ≈ 4.364 .
kf 11
The case of a constant surface temperature can be examined, however
the mathematics of the problem are appreciably more difficult (Kays and
Crawford, 1980, pp. 93–97). The result is similarly straightforward
(7.27) N u ≈ 3.658 .
7.3. THE COUETTE PROBLEM 85
7.3. The Couette Problem

Let us now look at another internal configuration known as the Couette
flow problem (Fig. 7.2). Here, two infinitely long flat plates a distance L
y T1
u1
T0 x
Figure 7.2. Schematic of the Couette flow problem.
apart enclose a working fluid characterized by constant density ρ, viscosity

µ, and thermal conductivity k. One plate translates in its own plane past
the other with a constant axial velocity u 1 . By convention, we specify the
top plate as the moving one (y = L) and anchor our coordinate origin to
the bottom plate at y = 0. The top plate is at a temperature of T 1 , while
the bottom plate has a temperature of T 0 . As with the pipe problem, we
assume the flow is fully developed, i.e. ∂/∂x = 0, however, there is no
pressure gradient. Rather, the top plate replaces pressure as the driving
force. This configuration is a good model for a number of applications, for
example a journal in an oil–filled bearing.
The problem can be framed in terms of the Navier–Stokes equations
introduced in Eqs. (5.5) through (5.8) on pp. 57. For the fully developed
flow assumption, we solve this problem as follows. The first term in the
continuity equation
∂u ∂v
+ = 0
∂x ∂y
must vanish since the flow is fully developed, which leaves ∂v/∂y = 0. At
most, v could then be a function of x, but since the flow is fully developed v
must be a constant. However, no–slip boundary conditions dictate v = 0 at
both y = 0 and y = L, so the only admissible constant must be v = 0. As
a consequence, the y–momentum equation, Eq. (5.7), vanishes altogether,
while the only non–vanishing term in the x–momentum equation, Eq. (5.6),
is ∂ 2 u/∂y 2 . Clearly, since u is the only non–zero velocity component, and it
is not a function of x (because of fully developed flow conditions), it must be
the case that u = u(y), so that Eq. (5.6) simplifies to the ordinary equation
d2 u
= 0.
dy 2
Integrating twice, we find
u(y) = C1 y + C2 ,
where C1 and C2 are evaluated from no–slip boundary conditions u| y=0 = 0

and u|y=L = u1 . We find
u1 y
u(y) = .
L
With these results, the energy equation, Eq. (5.8) on pp. 57, simplifies
to
d2 T µ du 2

= − .
dy 2 k dy
Notice that we wrote this as an ordinary equation rather than a partial
one, since temperature is not a function of x, i.e. T = T (y). The term
on the right hand side represents viscous dissipation, which describes how
mechanical energy is degraded to heat by the mechanism of friction. In the
current problem, this can be viewed as a source term, because we know the
solution for u and can evaluate this term explicitly. The result is
d2 T µ u21
= − .
dy 2 k L2
Once again, we can perform straightforward integration to obtain the general
solution
µ u21 2
T (y) = − y + C3 y + C 4 .
2 k L2
The boundary conditions T |y=0 = T0 and T |y=L = T1 are used to determine
constants C3 and C4 , so that the exact solution is
µ u21

y y
(7.28) T (y) − T0 = (T1 − T0 ) + 1− .
L 2k L
First, let us examine the case of T0 6= T1 . We will assume T1 > T0 ,
although the opposite case would yield analogous results. Dividing by T 1 −
T0 , we obtain
µ u21

T (y) − T0 y 1 y
= 1 + 1− ,
T1 − T 0 L 2 k (T1 − T0 ) L
or, written in dimensionless terms

∗ P r Ec ∗
(7.29) θ = y 1 + (1 − y ) ,
2
where θ is dimensionless temperature, y ∗ = y/L is the dimensionless coor-
dinate,
cp µ cp µ ρ 1 µ ν
Pr = = = =
k k ρ α ρ α
is the Prandtl number, and
u21
Ec =
cp (T1 − T0 )
is the Eckert number.
Note that the case of P r Ec = 0 corresponds to the no–flow condition,
and thus pure conduction. Thus, the temperature profile is a straight line
connecting T0 and T1 (Fig. 7.3). For other cases, viscous dissipation distorts
the straight–line relationship as shown by plotting θ versus y ∗ in Fig. 7.3.
An interesting aspect of this problem is the heat flux at the top wall, defined
0.8
dimensionless "y"
0
=
0.6 Ec 2
* =
Pr Ec
*
0.4 Pr =4
c
*E
Pr
0.2 8
Ec =
Pr *
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
dimensionless temperature
Figure 7.3. Temperature distribution in Couette flow with vis-

cous dissipation for T1 > T0 .
by
dT
q = −k .
dy y=L
From Eq. (7.28), we evaluate the derivative to obtain
µ u21 1

dT T1 − T 0 2 y
= + − 2
dy y=L L 2k L L y=L
T1 − T 0 µ u21
= −
L 2kL
µ u21

T1 − T 0
= 1 −
L 2 k (T1 − T0 )

T1 − T 0 P r Ec
= 1− ,
L 2
which means that the heat flux is

T1 − T 0 P r Ec
(7.30) q = −k 1− .
L 2
There are clearly three cases of interest, depending upon the sign of the
(1 − P r Ec/2) term (again, assuming T 1 > T0 ):
• P r Ec < 2, which makes the right hand side negative and therefore
heat flows in the −y direction, or from the upper wall into the
liquid. This is what we would intuitively expect, given T 1 > T0 .
• P r Ec > 2, which makes the right hand side positive and therefore
heat flows in the +y direction, or from the liquid into the wall, even
though the upper wall is at a higher temperature than the lower
wall. Not intuitive.
• P r Ec = 2, which makes the right hand side vanish and thus there
is no heat transfer at the upper wall. The upper wall behaves as
an insulated boundary (q 00 = 0), even though there is no actual
insulation. Also not intuitive.
The opposite case from the above is T 0 = T1 . Here, Eq. (7.28) reduces
to
µ u21 y y
T (y) − T0 = 1− .
2k L L
Non–trivial temperature distributions are due strictly to viscous dissipation
effects. This configuration is symmetric about y = L/2. In fact, the maxi-
mum temperature occurs at the middle of the channel and can be computed
by setting y = L/2 to get
µ u21
Tmax − T0 = .
8k
Though not identical, this case is qualitatively similar to heat generation for
conduction heat transfer in §3.4.

Other problems related to internal convection are more complicated than
our present level of theory will allow us to handle. As with external config-
urations, we must defer to empirically–based correlations using the dimen-
sionless parameters we have developed.
7.4.1. Turbulent Flow in Circular Tubes. If the flow is fully devel-

oped, both hydrodynamically and thermally, the Dittus–Boelter correlation
can be applied
4/5
(7.31) N uD = 0.023 ReD P r n ,
where n = 0.4 for Ts > Tm (heating) and n = 0.3 for Ts < Tm (cooling). The
experimentally established range of applicability according to Incropera and
Dewitt (2002) is 0.7 ≤ P r ≤ 160, ReD > 104 , and x > 10 D. All properties
are evaluated at Tm .
Eq. (7.31) is appropriate when Ts is not too different from Tm , however,
if there is appreciable variation, the Sieder–Tate correlation should be used
0.14
4/5 1/3 µ
(7.32) N uD = 0.027 ReD P r .
µs
The limits are identical to Eq. (7.31), except that the Prandtl number can
be extended up to P r ≈ 16, 700. For Eq. (7.32), all properties are evaluated
at Tm , except µs , which is evaluated at Ts . Eqs. (7.31) and (7.32) can be
reasonably applied for both the constant temperature and constant heat flux
boundary conditions.
It is emphasized that both Eqs. (7.31) and (7.32) should only be used
for fully turbulent problems. The transition regime between laminar and
fully turbulent flow, 2000 < ReD < 104 , entails additional considerations.
Incropera and Dewitt (2002, §8.5) discuss additional results for these cases.
7.4.2. Non–Circular Tubes. Many engineering applications involve
tubes or ducts having non–circular cross sections. Even for laminar flow,
theoretical approaches are beyond the mathematics we have discussed thus
far. For example, for a rectangular cross section, the momentum equation
cannot be reduced to an ordinary differential equation like what we obtained
for circular pipes in Eq. (7.1). However, results we have obtained for circu-
lar tubes can be applied as an initial approximation to non–circular cross
sections using the hydraulic diameter as a length scale
4 Ac
(7.33) Dh = ,
P
where Ac is the cross–sectional area of the conduit and P is the wetted
perimeter7.6. The value of Dh can be computed directly from geometry, for
example, for a rectangular duct of height a and width b, we obtain
4ab 2ab
(7.34) Dh = = ,
2 (a + b) a+b
Thus, Dh should be used to calculate the non–circular analogs of the dimen-
sionless parameters, e.g. ReDh and N uDh .
For fully turbulent flows, the Dittus–Boelter and Sieder–Tate correla-
tions can be reasonably used in conjunction with the hydraulic diameter.
Results are less accurate for laminar flows, especially in geometries having
sharp corners. Here, Nusselt numbers based on exact solutions should be
used. For example, Table 7.1 shows selected results for channels of rectan-
gular cross section arranged according to the aspect ratio φ = b/a, where
b > a. Kays and Crawford (1980) consider such problems in greater detail.
Table 7.1. Nusselt numbers for rectangular channels of

various aspect ratios (Kays and Crawford, 1980).
φ N uDh (const qs00 ) N uDh (const Ts )

1 3.61 2.98
2 4.12 3.39
3 4.79 3.96
4 5.33 4.44
7.6Eq. (7.33) contains a factor of 4 so that the hydraulic diameter of a circular pipe
is equal to its geometric diameter, i.e. Dh = 4 πD2 /4 / (πD) = D.
` ´
CHAPTER 8
Natural Convection
Up until now we have considered cases where convection is a result of

the forced motion of a fluid, e.g. a pressure gradient, moving boundary, etc.
However, there are many cases where initial heat transfer causes the local
temperature of a fluid to change and thus its density to change as well. This
causes a “natural” fluid motion which in turn gives rise to natural, or free
convection. Thus, free convection arises because of buoyancy forces. A few
notable situations are refrigerator coils, piping, electronics, steam radiators
in old homes, and flow along window panes.
Generally, the density of a fluid decreases with increasing temperature
because of volumetric expansion, which may induce a natural convection
flow. This actually depends upon the configuration of the problem. In par-
ticular, an instability must be able to occur. For example, imagine 2 infinite
parallel horizontal plates which are both fixed. If the top plate is hotter
than the lower one (Fig. 8.1, left), the configuration is inherently stable be-
cause the lighter fluid is already above the cooler heavier fluid. There is no
hot cold
g g
y y
cold hot
Figure 8.1. The concept of instability as a necessary condition

for natural convection.
tendency for the system to move away from this state of equilibrium and
any heat transfer between the plates will be a result purely of conduction.
However, if the bottom wall is hotter (Fig. 8.1, right), the hot light fluid
will initially be below the cooler heavier fluid. This “top heavy” system will
tend to flow, with the heavy fluid falling by the action of gravity and the
light fluid correspondingly rising. A circulation pattern will arise, so that
heat transfer between the plates is effected primarily by natural convection.
90
8.1. WALL BOUNDED CONVECTION ON A VERTICAL FLAT SURFACE 91
8.1. Wall Bounded Convection on a Vertical Flat Surface

Let us start by considering the familiar laminar boundary layer situ-
ation, except this time the flat plate is placed vertically. If we consider
the problem first on a conceptual basis, we quickly find that the boundary
layer velocity profile will be fundamentally different from the standard case
we are accustomed to, i.e. as depicted in Fig. 5.1. While we still have a
no–slip boundary condition on the plate itself, the fluid far away from the
plate (y → ∞) is at rest, rather than moving at some non–zero freestream
velocity. If we assume that the plate temperature T s is greater than the
fluid temperature T∞ , lighter fluid will rise, causing an upward convection
pattern. The resulting velocity profile would look qualitatively like the one
plotted in Fig. 8.2.
g x
vertical plate
Figure 8.2. Boundary layer profile on a vertical flat plate.
We are again assuming incompressible flow with one important excep-

tion: we must allow for the buoyancy force which drives the flow. It this
term, the density is not constant! Using the boundary layer approximation,
we write the resulting equation in the streamwise direction x as
∂u ∂u 1 dP ∂2u
u + v = − − g + ν ,
∂x ∂y ρ dx ∂y 2
where g represents the gravity vector. There is no body force in the y
direction (i.e. horizontally), therefore the x pressure gradient at any point
within the boundary layer is the same as that in the quiescent region far
outside the boundary layer. Since there is no fluid motion in this area,
vertical pressure gradient is prescribed simply by hydrostatic considerations,
∂P/∂x = −ρ∞ g, where ρ∞ is the density of quiescent fluid far from the plate.
Substituting back into the equation gives
∂u ∂u g ∂2u
u + v = (ρ∞ − ρ) + ν .
∂x ∂y ρ ∂y 2
8.1. WALL BOUNDED CONVECTION ON A VERTICAL FLAT SURFACE 92
The first term on the right hand side is the buoyancy force and in this
case it is the driving force of the flow. If temperature is the only factor
affecting density8.1, this term can be described by a thermodynamic property
known as the volumetric thermal expansion coefficient

1 ∂ρ
β = − ,
ρ ∂T P
which describes the change of density as a function of temperature at con-
stant pressure. This can be approximated by a simple finite difference ex-
pression
1 ρ∞ − ρ
β ≈ − ,
ρ T∞ − T
which implies
ρ∞ − ρ ≈ ρ β (T − T∞ ) .
This idealization is the so–called Boussinesq approximation for density vari-
ation. Substituting, the streamwise momentum equation becomes
∂u ∂u ∂2u
(8.1) u + v = g β (T − T∞ ) + ν ,
∂x ∂y ∂y 2
which nicely shows how the driving buoyancy force is related to a tempera-
ture difference.
Buoyancy effects are strictly limited to the momentum equation, there-
fore the remaining equations are unaffected. Specifically, for conservation of
mass, we have the usual
∂u ∂v
(8.2) + = 0
∂x ∂y
and for conservation of energy, we likewise have
∂T ∂T ∂2T
(8.3) u + v = α .
∂x ∂y ∂y 2
Eqs. (8.1) through (8.3) are the boundary layer form of the natural convec-
tion equations. Viscous dissipation has been neglected here, primarily due
to the small velocities associated with natural convection.
Examining these equations, we discover a major mathematical compli-
cation as compared to forced convection: T now appears in the momentum
equation in addition to u appearing in the energy equation. Therefore, this
system is fully coupled, as opposed to forced convection, which was uncou-
pled since T only appeared in the energy equation. We can no longer treat
the hydrodynamic problem independently. The equations must be solved
simultaneously.
8.1Density changes can arise via other means, as in for example supersonic flow. We
do not consider such cases here.
8.2. DIMENSIONLESS FORMULATION 93
The thermal expansion coefficient β is a material property. For an ideal

gas, the Ideal Gas Law ρ = P/RT gives ∂ρ/∂T = −P/RT 2 , so that

1 ∂ρ 1 P RT P 1
β = − = 2
= 2
= .
ρ ∂T ρ RT P RT T
For liquids and non–ideal gases, β must be obtained from appropriate tables.
8.2. Dimensionless Formulation

Let us now look at some of the non–dimensional parameters that are
relevant in natural convection. We introduce the usual scales, e.g. similar
to those described in §5.3
x y
x∗ = , y∗ = ,
L L
u v T − T∞
u∗ = , v∗ = , T∗ = ,
u0 u0 Ts − T ∞
where L is a characteristic length and u 0 is a characteristic velocity8.2 . Non–
dimensionalizing the equations according to the results we developed in §4.4,
we find, as usual, that the continuity equation remains the same in non–
dimensional form
∂u∗ ∂v ∗
(8.4) + = 0.
∂x∗ ∂y ∗
The momentum and energy equations become
∂u∗ ∗ ∂u
∗ g β (Ts − T∞ ) L ∗ 1 ∂ 2 u∗
(8.5) u∗ + v = 2 T + ,
∂x∗ ∂y ∗ u0 Re ∂y ∗2
and
∂T ∗ ∗ ∂T
∗ 1 ∂2T ∗
(8.6) u∗ + v = .
∂x∗ ∂y ∗ Re P r ∂y ∗ 2
There is a new coefficient we have not seen before in front of the buoyancy
term. However, we recall that u0 is not a readily prescribed scale, so we can
arbitrarily multiply by another dimensionless parameter to obtain a new
dimensionless parameter8.3. In this case, we multiply this quantity by the
square of the Reynolds number, Re = u 0 L/ν, to get the Grashof number
g β (Ts − T∞ ) L u20 L2
Gr = × ,
u20 ν2
which gives
g β (Ts − T∞ ) L3
(8.7) Gr = ·
ν2
8.2Characteristic velocity is not readily identifiable since conditions outside the bound-
ary layer are at rest and thus u∞ = 0. This is slightly different from the method of
non–dimensionalization shown in §5.3.
8.3Recall that this is permissible according to the Buckingham–Pi theorem from fluid
mechanics.
8.3. SIMILARITY SOLUTION FOR THE VERTICAL PLATE 94
The Grashof number plays the same role in free convection that the Reynolds
number plays in forced convection, i.e. Re is the ratio of inertial forces to
viscous forces, whereas and Gr is the ratio of buoyancy forces to viscous
forces. We can then recast momentum conservation in Eq. (8.5) as
∂u∗ ∗ ∂u
∗ 1 ∂ 2 u∗
(8.8) u∗ + v = Gr T ∗
+ ,
∂x∗ ∂y ∗ Re ∂y ∗ 2
The three conservation laws, Eqs. (8.4), (8.6), and (8.8), are functions of
the dimensionless parameters Re, P r, and Gr. We would therefore expect
solutions of the natural convection problem, given by the Nusselt number,
to be of the form
N u = N u (Re, P r, Gr) .
There is a subtle aspect to this interpretation. In formulating the Grashof
number, we were able to eliminate the velocity scale u 0 based on the obser-
vation that we could not necessarily characterize it appropriately. Yet, it
could be that an external forcing results in an explicit u 0 , so that a problem
would be a mixture of forced convection and natural convection. The above
presumption about the functional dependence of N u is valid when forced
and natural convection aspects are comparable, i.e.
Gr
∼ 1.
Re2
Looking back at the form of Gr in Eq. (8.5), this condition is equivalent to
g β (Ts − T∞ ) L ∼ u20 .
Consequently, if forced convection dominates a problem,
Gr
1 and N u = N u (Re, P r) .
Re2
That is, free convection effects are so small as to be neglected. Conversely,
if natural convection is dominant, we have
Gr
1 and N u = N u (P r, Gr) .
Re2
Here, the small forced convection effect would be neglected, although strictly
speaking, natural convection occurs exclusively only for Gr/Re 2 → ∞, for
which u0 = 0.
8.3. Similarity Solution for the Vertical Plate

As already noted, natural convection is a difficult problem to solve ana-
lytically because the equations are fully coupled. However, as with external
forced convection on the horizontal flat plate, there are no obvious scales for
the problem, so we suspect that a similarity transform 8.4 might be used. If
successful, this would reduce the partial differential problem to an ordinary
differential problem.
8.4The similarity technique was introduced in §4.6.
From Fig. 8.2, we deduce the boundary conditions

y=0 : u = v = 0, T = Ts
and
y→∞ : u → 0, T → T∞ ,
where Ts is the temperature of the plate. In this case, the transformation
variables
y Grx 1/4

η = and
x 4
4 1/4

ψ
f (η) =
4 ν Grx
are appropriate, where ψ = ψ(x, y) is the stream function defined in Eq. (6.5)
on pp. 67 and
g β (Ts − T∞ ) x3
Grx =
ν2
is the local Grashof number based on x.
Transforming the momentum equation appropriately, we find it becomes
2
f 000 + 3f f 00 − 2 f 0 + T ∗ = 0 ,
where the prime symbol indicates the regular derivative of f (η). The energy
equation becomes
T ∗ 00 + 3 P r f T ∗ 0 = 0 .
Like the horizontal flat plate problem, the momentum equation remains
non–linear. Moreover, the two equations are still coupled to each other. i.e.
f and T ∗ appear in both. In this case, a numerical method is typically
invoked to complete the solution.
It can be shown (e.g. Incropera and Dewitt, 2002, pp. 541–542) that the
general form for the local Nusselt number is
Grx 1/4

hx
(8.9) N ux = = F (P r) ,
k 4
where F (P r) is a function of the Prandtl number as determined by the
solution of the above problem. A correlation, purported to be accurate to
within about 0.5 % (Incropera and Dewitt, 2002, pp. 542) is given by the
expression
0.75 P r 1/2
(8.10) F (P r) = 1/4 ·
0.609 + 1.221 P r 1/2 + 1.238 P r
The result can also be used to calculate the overall (averaged) Nusselt num-
ber according to the definition given by Eq. (5.2) on pp. 54 as
k Grx 1/4
Z L Z L Z L
1 1 N ux k 1
h̄ = h dx = dx = F (P r) dx .
L 0 L 0 x L 0 x 4
Substituting the local Grashof number, and observing that a number of

factors in the argument, including F (P r) are not functions of x, and so can
be taken outside the integral, we find
L 1/4
g β (Ts − T∞ ) x3

F (P r) k 1
Z
h̄ = dx
41/4 L 0 x ν2
F (P r) k g β (Ts − T∞ ) 1/4 L 1 3/4
Z
= x dx
41/4 L ν2 0 x
F (P r) k g β (Ts − T∞ ) 1/4 L 1
Z
= dx
L 4 ν2 0 x
1/4
F (P r) k g β (Ts − T∞ ) 1/4 4 3/4 L

= x
L 4 ν2 3
0
3 1/4
4 F (P r) k g β (Ts − T∞ ) L
=
3 L 4 ν2
4 k GrL 1/4

= F (P r)
3 L 4
4 k
= N uL ,
3 L
where the last line results directly from the form of the solution given in
Eq. (8.9). Moving k and L over to the left hand side, we find the average
Nusselt number to be
h̄ L 4
(8.11) N uL = = N uL ,
k 3
where N uL is the local Nusselt number at the end of the plate, i.e. at x = L.
As with forced flow, turbulent effects can arise in natural convection.
We mentioned above that the Grashof number, defined by Eq. (8.7), can
be taken as the ratio of buoyancy effects versus viscous effects. The inter-
play between these two phenomena is very similar to that defined by the
Reynolds number for forced convection, i.e. viscous forces tend to dissi-
pate disturbances which could grow large enough to de–stabilize the flow.
We would therefore expect that Gr serves as a direct measure of the ten-
dency toward turbulent flow for natural convection in the same sense as Re
for forced convection. However, data correlate more strongly to a closely–
related parameter call the Rayleigh number , which is simply the product of
the Grashof and Prandtl numbers. The critical Rayleigh number for turbu-
lent flow is approximately. I&D Ex. 9.1
pp 543
g β (Ts − T∞ ) x3
(8.12) Rax = Grx P r = ≈ 109 .
να
Gebhart et al. (1988) discuss this in greater detail.

There is little more than can be done from an analytical standpoint with
natural convection because of the complexity of the problem, so we tend to
focus more on empirically correlated results. As with turbulent flow, most
of the data correlate more appropriately to Ra rather than Gr. Properties
are again evaluated at the film temperature, defined by Eq. (6.1) on pp. 66.
8.4.1. Vertical Plate. The Churchill–Chu correlation can be applied

over the entire range of Ra
 2
1/6
0.387 RaL
(8.13) N uL = 0.825 + 8/27  ,
 
1 + [0.492/P r] 9/16
although somewhat better accuracy for strictly laminar flow is given by

1/4
0.67 RaL
(8.14) N uL = 0.68 + 4/9 where RaL < 109 .
9/16
1 + [0.492/P r]
These correlations are applicable to cases where the plate temperature T s

is constant. Incropera and Dewitt (2002, pp. 546) describe procedures to I&D Ex. 9.2
extrapolate these results to several other configurations. pp 546
8.4.2. Horizontal Cylinder. Heated horizontal cylinders are closely

related to the vertical plate configuration, except that the boundary layer
curves around the contour of the body rather than being strictly vertical
(Fig. 8.3). Development begins on the underside of the cylinder and the
boundary layer eventually detaches in the form of a plume that rises above
the cylinder. Churchill and Chu have formulated a correlation valid over a
Figure 8.3. Natural convection boundary layer on a horizontal

cylinder.
wide range of Ra
 2
1/6
0.387 RaD
(8.15) N uL = 0.6 + 8/27  , RaD < 1012 .
 
9/16
1 + [0.559/P r]
where the Rayleigh number is based on the pipe diameter D.
8.4.3. Rectangular Enclosures. Configurations we have discussed
thus far are primarily of the external flow type, but internal configurations
also arise in applications, for example in the form of various enclosures
(Fig. 8.4). Here, the bottom and top surfaces are adiabatic, while the ver-
cold
hot
H
Figure 8.4. Recirculating natural convection flow pattern in a

vertically–oriented enclosure. This particular configuration is an
excellent model for analyzing double–pane windows.
tical surfaces have specific temperatures, T 1 on the hot side and T2 on the
cold side, and are separated by a distance L. If the Rayleigh number is suffi-
ciently high, a recirculating flow pattern will form, similar to that shown in
the figure. For small RaL , the flow is weak and heat transfer occurs mainly
by conduction, i.e. N uL ∼ 1. For larger RaL , a number of correlations have
been devised, e.g.
P r RaL 0.28 H −0.25

(8.16) N uL = 0.22 , 103 < RaL < 1010 ,
0.2 + P r L
which is valid for aspect ratios 2 < H/L < 10 and P r < 10 5 . Other
correlations are available for larger aspect ratios ( Özişik, 1985; Incropera
and Dewitt, 2002).
CHAPTER 9
Heat Exchangers
Heat exchangers are devices that facilitate heat transfer between two
fluids at different temperatures without allowing them to mix. Usually, the
fluids are separated by a solid boundary, for example as in car radiators,
air conditioners, distillers, etc. These are called indirect contact exchangers.
However, if the fluids do not tend to mix naturally, a direct contact heat
exchanger may be used, e.g. a water chiller.
Heat exchangers are typically classified by flow arrangement and type
of construction. In parallel flow units, fluids enter at the same end and flow
in the same direction, while in counter–flow, the flow of the two fluids is in
opposite directions. Fig. 9.1 illustrates these arrangements for a simple heat
exchanger consisting of concentric tubes.
parallel flow
counter flow
Figure 9.1. Flow arrangements for concentric tube heat ex-

changer. One fluid flows within the innermost tube, while the other
flows in the gap between the two tubes.
The concentric tube heat exchanger is a good model for introducing a

number of concepts. According to convention, the inner tube is called the
“tube”, while the outer tube is referred to as the “shell”. Thus, Fig. 9.1
shows the simplest shell–and–tube heat exchanger architecture. However,
according to our results for fins in §3.5 we suspect that actual units might
not be as simple as that depicted in Fig. 9.1 because of the limited surface
area of the tube. Instead, units routinely utilize many tubes within a single
shell, so that the area for heat transfer is maximized (Fig. 9.2). Variations
99
9.1. THE THUMBNAIL DIAGRAM 100
shell side inlet
tube side tube side

inlet outlet
shell side outlet
Figure 9.2. Shell–and–tube heat exchanger with one tube pass

and one shell pass. Tube side enters and exits through manifolds.
on this basic architecture are shown in reference texts (e.g. Özişik, 1985;
Mills, 1999; Incropera and Dewitt, 2002).
It is also common to classify heat exchangers according to the area avail-
able for heat transfer per unit volume of the unit, i.e.
total area available for heat transfer
ϕ = ·
total volume of heat exchanger
Units having ϕ > 700 m−1 are arbitrarily referred to as being compact heat
exchangers, regardless of their design architecture. Table 9.1 summarizes ϕ
for several classes of heat exchangers.
Table 9.1. Approximate area density for various classes of

heat exchangers (Özişik, 1985).

Type ϕ m−1
industrial shell–and–tube 70 – 500
automotive radiators 1000
precision specialty units 5000 – 10,000
human lung 20,000
9.1. The Thumbnail Diagram

In analyzing heat exchangers, we will find the concept of the thumbnail
diagram to be especially useful. The diagram is simply a plot of the temper-
atures of the two streams, hot side (H) and cold side (C), as functions of the
distance from an origin. For example, Fig. 9.3 shows a simple tube–type unit
of length L running in parallel flow mode. Assuming the origin to be the
inlet side at x = 0, temperatures of the two streams are shown, along with
their flow directions. Although direction arrows are not mandatory, they
will be extremely helpful for the “bookkeeping” aspect of heat exchanger
calculations.
Fig. 9.3 shows a number of interesting concepts for the parallel flow
arrangement:
9.1. THE THUMBNAIL DIAGRAM 101
parallel flow
temperature
H
∆T
C
x=0 x=L
Figure 9.3. Thumbnail diagram for a simple tube–type heat

exchanger using the parallel flow arrangement.
• The outlet temperature of the cool side cannot exceed that of the
hot side. At most, they can be equal, otherwise an un–physical
temperature gradient exists. Therefore, the effectiveness of parallel
flow is limited and not typically used for heat recovery.
• The temperature gradient between the streams, ∆T , is not con-
stant. Therefore the local heat flux q 00 varies with x.
• If we assume that the wall temperature of the material separat-
ing the streams is the average between the two streams, then wall
temperature is roughly constant. This might be an advantageous
design aspect if the wall material is sensitive to stresses induced by
temperature differences along x.
Fig. 9.4 shows thumbnail diagrams for three other cases: counter–flow,
and condensing and boiling heat transfer. For the counter–flow design, the
H
H
temperature
temperature
temperature
C
H
C
C
0 L 0 L 0 L
counter−flow condensing boiling
Figure 9.4. Thumbnail diagrams for the counter–flow arrange-

ment, and for condensing and boiling heat transfer.
9.2. OVERALL HEAT TRANSFER COEFFICIENT 102
exit temperature of the cold fluid can in fact be higher than that of the
hot fluid. This makes the counter–flow design more attractive than parallel
flow units if the design requirements permit such a choice. Clearly, the wall
temperature in these units is not constant, which might cause large thermal
stresses that may eliminate certain materials from design consideration. Al-
though not necessarily the case, ∆T might also be constant along x for the
counter–flow arrangement.
Although we do not discuss the physics of condensation and boiling heat
transfer per se, we can examine their application in heat exchangers utilizing
the simple fact that temperature is constant during a phase change. It should
be clear that the flow direction of the phase change stream is irrelevant
because of its constant temperature. For the condenser, heat energy is
liberated from the condensate and absorbed by the cold side, causing its
temperature to rise9.1. For a boiling heat exchanger, the reverse is true: the
hot side transfers energy to boil the fluid on the cold side and cools in the
process.
Thumbnail diagrams can be drawn for more complicated configurations
as well. For example, Fig. 9.5 illustrates a shell–and–tube exchanger having
one shell pass and two tube passes, along with its corresponding thumbnail
diagram.
tube side inlet
shell side shell side outlet

inlet
tube side outlet
H
temperature
Figure 9.5. Thumbnail diagram for one shell pass, two tube
pass heat exchanger. We have assumed the tube side contains the
cold fluid.
9.2. Overall Heat Transfer Coefficient

We have devoted considerable attention to obtaining convection coeffi-
cients in Chapters 5 through 8. While these problems dealt strictly with
single flows, the heat exchanger presents us with two flows to analyze. How
9.1A cold can of beer on a warm humid day is good example of this phenomenon.
Water vapor condenses on the outside of the can, transferring heat to the beer, causing it
to warm up.
9.2. OVERALL HEAT TRANSFER COEFFICIENT 103
should this be handled? As we discussed in Chapter 3, the circuit analogy

can be used to obtain overall effective thermal resistances. The same con-
cept can be applied here to calculate an overall heat transfer coefficient for
a heat exchanger. The general form of this coefficient is written 9.2
1
(9.1) U = ,
At Rt
where At is the total area available for heat transfer and R t is the effective
(overall) thermal resistance. Calculating U is clearly a matter of first deter-
mining Rt . For a simple shell–and–tube heat exchanger, thermal resistance
is the sum of three basic resistive components in series (Fig. 9.6):
• convective resistance on the tube side
• conductive resistance of the wall between the two flow streams
• convective resistance on the shell side
tube side convective resitance
conductive resistance
shell side convective resistance
Figure 9.6. Resistive components involved in calculating the

overall thermal resistance.
We can write the overall resistance as

1 1
(9.2) Rt = + Rc + ,
Ai hi Ao ho
where subscripts i and o designate inner and outer, respectively, and R c is
the conductive resistance for a cylindrical surface derived in Eq. (3.22) on
pp. 22 as
ln (ro /ri )
Rc = ·
2πLk
Notice that we can use either Ai or Ao as the total area available for heat
transfer At in Eq. (9.1). It makes no difference which one is used, as long
9.2By convention, U is the symbol for overall convection coefficient for heat exchang-
ers, rather than some form of h which is what we might have expected. This notation is
the standard (e.g. Özişik, 1985; Mills, 1999; Incropera and Dewitt, 2002).
9.3. LMTD ANALYSIS 104
as the basis is specified9.3. Using Ao = 2 π ro L as the basis, we obtain

1 1
Uo = = ,
Ao Rt Ao Ai1hi + ln(ro /ri )
+ 1
2πLk A o ho
which, after simplifying yields

1
(9.3) Uo = ro ln(ro /ri )
·
ro 1
r i hi + k + ho
Similarly, for Ai = 2 π ri L as the basis, we get

1
(9.4) Ui = ri ln(ro /ri )
·
1 ri
hi + k + r o ho
If the wall thickness is minimal9.4 and the thermal conductivity is high, tube
wall resistance can be neglected to obtain
1
(9.5) U = 1 1 ·
hi + ho
In the course of normal operation, inner surfaces of a heat exchanger

can become coated with deposits that leach out of the working fluids and/or
corrode due to reaction with the fluid. These factors present additional
resistance to heat transfer that can be modeled via a fouling factor . Eq. (9.2)
can be modified as
Fi 1 1 Fo
(9.6) Rt = + + Rc + +
Ai Ai hi Ao ho Ao
to consider fouling, where Fo and Fi are the fouling factors for the outer and
inner surfaces, respectively. Generally, performance is gradually degraded
over time and costs are increased because of maintenance requirements and
down time. The fouling factor is generally a known quantity based on the
working fluid (Table 9.2).
9.3. LMTD Analysis

At this point we probably already suspect that the somewhat exacting
analysis methods we have applied to previous problems will not be possible
for heat exchangers. The flow physics of these units is simply too complex
to obtain analytical solutions, e.g. because of turbulence, developing flow,
separation, etc. Moreover, diverse geometry and architecture preclude gen-
eralizing results into a few relevant correlations. Instead, we will take what
9.3Calculating the overall heat transfer will involve multiplying by A using a form of
t
Newton’s Law of Cooling (see Eq. (9.7)), so that, as long as consistency is preserved with
respect to which area is used, the effect will cancel itself out.
9.4If wall thickness is small, then r ≈ r , which means that ln (r /r ) = ln (1) = 0,
i o o i
resulting in a vanishing wall resistance term.
Table 9.2. Some representative fouling factors (TEMA, 1978).
Fouling Factor m2 K/W

Fluid
seawater below 50 o C 0.0001
seawater above 50 o C 0.0002
fuel oil 0.0009
machine oils 0.00018
vegetable oils 0.00053
most refrigerants 0.0002
steam (non oil bearing) 0.0001
alcohol vapor 0.00009
industrial air 0.0004
may be thought of as an approximate integral approach in which the anal-

ysis is only dependent upon temperatures at the inlets and outlets and the
overall convection coefficient.
Let us cast our analysis in the form of Newton’s Law of Cooling
(9.7) q = At Um ∆Tm ,
where At and Um are the total area available for heat transfer and the overall
convection coefficient, respectively, and ∆T m is a temperature difference that
must be determined. In general, the temperature difference between the hot
and cold fluids is not a constant, as suggested by Figs. 9.3 and 9.4. How,
then do we determine ∆Tm ?
Here, we will introduce the method of the “Log Mean Temperature Dif-
ference” (LMTD) for solving heat exchanger problems. This procedure al-
lows us to calculate ∆Tm , which can be thought of as an appropriately av-
eraged temperature difference between the two flow streams. It is probably
no surprise that ∆Tm depends upon the heat exchanger configuration, flow
arrangement, etc. To illustrate the concept, we will develop ∆T m for the
simple tube–type unit shown in Fig. 9.3. There are a number of additional
assumptions we must make to implement this method
• the unit is insulated such that no heat is exchanged with its sur-
roundings; heat transfer only takes place between the hot and cold
streams within the unit
• specific heats of both fluids are constant
• overall heat transfer coefficient is constant
• potential and kinetic energy changes can be neglected
We start by adding some detail to the thumbnail diagram (Fig. 9.7).
As usual, let x be the distance along the heat exchanger, A be the area,
ṁc and ṁh be the mass flow rates of the cold and hot sides, respectively,
∆T = Th − Tc = ∆T (x) be the local temperature difference between the hot
and cold sides, and U = U (x) be the local overall heat transfer coefficient.
Each stream experiences a differential change in temperature as a result of
T Th + d T h hot side
h
Tc dq Tc + d T c cold side
dA
Th,i dq
Th hot side T
h,o
dTh
∆ T0 ∆T ∆T
L
Tc cold side Tc,o
dTc
Tc,i
x=0 x=L
Figure 9.7. Differential analysis for LMTD calculation of the

parallel flow simple tube–type heat exchanger.
interacting over an area dA. We can write a differential form of Newton’s

Law at x as
(9.8) dq = dA U ∆T ,
where dq is the thermal energy transferred from the hot stream to the cold
stream through area dA. The differential energy transfer can also be ex-
pressed in terms of the change of temperature of each stream over dA.
Specifically, the heat lost by the hot side results from a change in tem-
perature of (Th + dTh ) − Th and is thus equal to
(9.9) dq = − ṁh ch dTh ,
while for the cold side we have
(9.10) dq = ṁc cc dTc ,
where ch and cc are the fluid specific heats. From the local temperature
difference ∆T = Th − Tc , we see that d(∆T ) = dTh − dTc , so that we can
use Eqs. (9.9) and (9.10) to write

dq dq 1 1
(9.11) d(∆T ) = − − = − dq + ·
ṁh ch ṁc cc ṁh ch ṁc cc
Since the mass flow rates and specific heats are assumed to be constant, we
let the last term in brackets be represented by the constant C 0 , i.e.
(9.12) d(∆T ) = − dq C0 .
Substituting dq from Eq. (9.8), we find

(9.13) d(∆T ) = − dA U ∆T C0 ,
or, dividing by ∆T ,
d(∆T )
(9.14) = − dA U C0 .
∆T
Eq. (9.14) can be integrated over the whole length of the heat exchanger
Z ∆Tx=L Z At
d(∆T )
(9.15) = − C0 U dA .
∆Tx=0 ∆T 0
Notice that C0 comes outside the integral since it is a constant. The right
hand side can be multiplied by At /At , i.e.
Z ∆Tx=L Z At
d(∆T ) 1
(9.16) = − C 0 At × U dA .
∆Tx=0 ∆T At 0
Let us now define the average overall heat transfer coefficient for the whole
heat exchanger in the usual fashion as
Z At
1
(9.17) Um = U dA ,
At 0
so that by integrating the left hand side of Eq. (9.16), we get

∆Tx=L ∆Tx=L
(9.18) ln ∆T = ln = − U m At C0 ,

∆Tx=0 ∆Tx=0
or rewriting to remove the minus sign, it becomes

∆Tx=0
(9.19) ln = U m At C0 .
∆Tx=L
Eq. (9.19) is an interesting intermediate result, but is not yet in the form
of Newton’s Law of Cooling in Eq. (9.7). We can go back to Eq. (9.12) and
integrate this over the whole heat exchanger as
Z ∆Tx=L Z q
(9.20) d(∆T ) = − C0 dq 0 ,
∆Tx=0 0
from which we find that the total heat transfer q can be expressed as
∆Tx=L − ∆Tx=0 = − C0 q .
Solving for C0 , we obtain
∆Tx=0 − ∆Tx=L
C0 = ,
q
which can be substituted into Eq. (9.19)

∆Tx=0 ∆Tx=0 − ∆Tx=L
(9.21) ln = U m At ·
∆Tx=L q
Eq. (9.21) can be inverted as

∆Tx=0 − ∆Tx=L
(9.22) q = U m At
ln (∆Tx=0 /∆Tx=L )
Therefore, we see based on our original assumption in Eq. (9.7) about the
form of the solution that
∆Tx=0 − ∆Tx=L
(9.23) ∆Tm = .
ln (∆Tx=0 /∆Tx=L )
This is the log mean temperature difference.
Now, what happens if ∆Tx=0 = ∆Tx=L ? It appears that we get a zero
over zero situation, i.e. the numerator is 0 and the denominator is ln 1 =
0. Mathematically, this is of course undefined and must be resolved. We
can apply L’Hospitals rule, which says essentially that both numerator and
denominator are differentiated until the limit of both can be meaningfully
evaluated. In essence, we want to evaluate
∆T0 − ∆TL
lim ,
∆T0 →∆TL ln (∆T0 /∆TL )
where we have shortened the notation for the temperatures. For the numer-
ator, we evaluate
d h i
∆T0 − ∆TL = 1 − 0 .
d∆T0
Recall that derivatives of the natural log function are defined as
dh i 1 df
ln f (x) = ,
dx f (x) dx
so that for the denominator f (∆T0 ) = ∆T0 /∆TL we evaluate

d ∆T0 ∆TL 1 1
ln = = .
d∆T0 ∆TL ∆T0 ∆TL ∆T0
Reformulating the ratio, we now have
1 − 0
lim = ∆T0 ,
∆T0 →∆TL 1/∆T0
so in fact, the limit is ∆T0 . Therefore, in the limiting case of ∆T 0 = ∆TL ,
the log mean temperature difference ∆T m = ∆T0 = ∆TL .
We have mentioned in a qualitative sense that the counter–flow arrange-
ment is somewhat “better” than parallel flow. Some quantitative light can
be shed on this matter by estimating ∆T m for each case. Assuming equal At
and Um , Eq. (9.7) indicates that heat transfer q depends upon ∆T m . The
expression given by Eq. (9.23) entails a subtle bookkeeping task of using the
appropriate temperatures to compute ∆T x=0 and ∆Tx=L . For example, the
parallel flow situation in Fig. 9.7 implies
∆Tx=0 = Th,i − Tc,i and ∆Tx=L = Th,o − Tc,o ,
9.4. CORRECTION FACTORS FOR COMPLEX CONFIGURATIONS 109
so that
(Th,i − Tc,i ) − (Th,o − Tc,o )
∆Tm, parallel =
ln ([Th,i − Tc,i ] / [Th,o − Tc,o ])
(Th,i − Th,o ) − (Tc,i − Tc,o )
=
ln ([Th,i − Tc,i ] / [Th,o − Tc,o ])
∆Thot side + ∆Tcold side
=
ln (large number/small number)
Here, subscript “i” refers to an inlet temperature, while subscript “o” refers
to an outlet temperature. Although the numerator is large, the denominator
is not terribly small, i.e. it is the log of a rather large number.
Were we to duplicate the LMTD analysis for the counter–flow arrange-
ment, we would once again realize Eq. (9.23), except that the bookkeeping
aspect dictates
∆Tx=0 = Th,i − Tc,o and ∆Tx=L = Th,o − Tc,i ,
which can also be deduced from Fig. 9.4. We can then estimate ∆T m as I&D Ex. 11.1
(Th,i − Tc,o ) − (Th,o − Tc,i ) pp 655
∆Tm, counter−f low =
ln ([Th,i − Tc,o ] / [Th,o − Tc,i ])
(Th,i − Th,o ) − (Tc,o − Tc,i )
=
ln ([Th,i − Tc,o ] / [Th,o − Tc,i ])
∆Thot side − ∆Tcold side
=
ln (a number which may be close to unity)
Here, the numerator can be fairly small, but the denominator is the log of
an argument that can be close to unity. Therefore, the denominator is very
small, giving a larger LMTD temperature difference for the same input and
output values (Table 9.3).
Table 9.3. Representative LMTD temperatures for par-

allel and counter–flow heat exchangers (all temperatures in
o C)
Type Th,i Th,o Tc,i Tc,o ∆Tm

parallel 100 80 60 78 12.69
counter–flow 100 80 60 78 20.98
9.4. Correction Factors for Complex Configurations

The LMTD analysis we developed is valid for simple parallel flow and
counter flow units, but not for more complex multipass units (e.g. Fig. 9.5),
or cross flow units (Fig. 9.8). In the latter design, fluids can either be
“un–mixed”, meaning that transverse motion in a stream is not allowed, or
“mixed” meaning that a stream can mix laterally.
9.5. –NTU ANALYSIS 110
tube−type cross−flow design

tube side is un−mixed
plate−type cross−flow design
shell side can mix
both fluids un−mixed
Figure 9.8. Cross flow heat exchanger designs for both fluids
un–mixed (left) and one fluid mixed, one fluid un–mixed (right).
While such configurations have been analyzed, the results are usually
considered too complicated for practical use. Instead, these results have
been cast as correction factors to calculate appropriate LMTD temperatures.
The form of the corrected LMTD temperature is
∆Tm, corrected = ∆Tm F ,
where F is the appropriate correction factor and ∆T m is computed on the
basis of counterflow conditions! References (e.g. Özişik, 1985; Incropera
and Dewitt, 2002) typically present F graphically. Parameters in these I&D Ex. 11.2
graphical results depend by convention on the following nomenclature: t is pp 657
the temperature on the tube side (hot or cold), while T is the temperature of
the opposite (shell) stream. If the temperature change of one of the streams
is negligible, as in a phase change, then F = 1. Generally, an exchanger
with both sides being unmixed will give somewhat better performance than
if one side is mixed because the streamwise temperature gradient is more
conserved in the former.
9.5. –NTU Analysis

For the LMTD method to be applicable, we must be able to compute
∆Tm , therefore we must know both the outlet and inlet temperatures for
each stream. However, what if there is a situation in which the outlet
temperatures are not known? For example, we may have a design problem
where the inlet temperatures, fluid properties, and flow rates are known, but
the outlet temperatures are not. Using LMTD analysis in this case would
require a tedious iteration procedure to make sure that the heat transferred
in the heat exchanger equals the heat energy gained by the cold side and
lost by the hot side:
• Assume outlet temperatures and compute ∆T m

• Determine q from Eq. (9.7), i.e. q = A t Um ∆Tm
• Compute the outlet temperatures using q and the net temperature
change of each stream
• Compare the outlet temperatures with those assumed in first step
and adjust as necessary
• Repeat until suitable convergence is achieved
In this case, there is a better analysis method termed the “Effectiveness
Method”, or commonly known as the –NTU method. First, we define
the effectiveness of a heat exchanger as the actual heat transfer over the
maximum possible heat transfer
q
(9.24) = .
qmax
Before we delve into the –NTU method itself, let use also define heat ca-
pacity rates for the hot and cold side as
Ch = ṁh cp, h and Cc = ṁc cp, c .
According to what we have already concluded, q max would be obtained
using a counter–flow architecture having an infinite length. In this configura-
tion, one of the fluids would experience the maximum possible temperature
change
(9.25) ∆Tmax = Th, i − Tc, i ,
where Th, i and Tc, i are again the inlet temperatures of the hot and cold
streams, respectively. In this case the maximum heat transfer would be
(9.26) qmax = Cmin (Th, i − Tc, i ) ,
where
(9.27) Cmin = min [Ch , Cc ]
is the minimum of the two heat capacity rates. Can we prove that Eqs. (9.25)
through (9.27) always represent the maximum temperature change of one of
the streams?
There are only two possible cases: the cold side is the “minimum” side,
or the hot side is the “minimum” side (Fig. 9.9). If the cold side is the
minimum, then conservation of energy, written as
q = Cc ∆Tc = Ch ∆Th ,
dictates that the cold side undergoes a larger change in temperature than
the hot side, ∆Tc > ∆Th , so that the cold side outlet temperature would
be heated up to the value of the hot side inlet temperature, T c, o = Th, i .
Therefore, the maximum energy transfer would be expressed as
q = Cc (Tc, o − Tc, i ) = Cc (Th, i − Tc, i ) where Cc = Cmin .
Conversely, if the hot side is the minimum, conservation of energy requires
the hot side to experience the larger change in temperature, ∆T h > ∆Tc ,
T h, i
T c, o = T h, i
temperature
temperature
T h, o
T c, o
T c, i
T c, i = T h, o
cold side is the "minimum" hot side is the "minimum"
Figure 9.9. Diagrammatic representation of the concept of the

“minimum” side for heat exchangers.
so that the hot side outlet temperature would be cooled to the value of the
cold side inlet temperature, Th, o = Tc, i . Now, maximum energy transfer
would be written
q = Ch (Th, i − Th, o ) = Ch (Th, i − Tc, i ) where Ch = Cmin .
This proves the above concept.
Based on this exercise, it should be clear that that q max is not based on
the maximum ṁcp . If it were, then according to conservation of energy, the
fluid with the minimum ṁcp would undergo a temperature change greater
than the maximum ∆T , which is not possible. According to Eq. (9.24),
computing the heat transfer q depends on knowing . How do we determine
this?
The parameter depends upon the heat exchanger design and flow ar-
rangement. Here, we illustrate the general procedure for deriving assuming
a parallel flow single pass arrangement 9.5. This derivation refers, once again
to Fig. 9.7. From the definition in Eq. (9.24), we get
q
= .
Cmin (Th, i − Tc, i )
Also, based upon the total heat transferred, we have
q = Ch (Th, i − Th, o ) = Cc (Tc, o − Tc, i ) .
By definition, we then have
Ch (Th, i − Th, o ) Cc (Tc, o − Tc, i )
(9.28) = = .
Cmin (Th, i − Tc, i ) Cmin (Th, i − Tc, i )
9.5It is important to realize that q

max is always based on the counter–flow arrange-
ment, but the analysis is applied to whatever specific architecture is used, for example as
shown here for the parallel flow arrangement.
Now, recall Eq. (9.19) that was used in the derivation of the LMTD analysis
method, again for the parallel flow arrangement

∆Tx=0 Th, i − Tc, i
ln = ln = U m At C0 ,
∆Tx=L Th, o − Tc, o
where we recall that
1 1 1 1
C0 = + = + ·
ṁh ch ṁc cc Ch Cc
Taking the exponential, we can write
Th, i − Tc, i
= e Um A t C0 ,
Th, o − Tc, o
which inverts as
Th, o − Tc, o
= e − Um At C0 .
Th, i − Tc, i
From Eq. (9.28), we can solve for Th, o to get
Cmin (Th, i − Tc, i )
Th, o = Th, i −
Ch
Cc (Tc, o − Tc, i ) Cmin (Th, i − Tc, i )
= Th, i −
Cmin (Th, i − Tc, i ) Ch
Cc (Tc, o − Tc, i )
= Th, i − ,
Ch
which can be used to eliminate Th, o from the exponential equation. This
yields
Cc (Tc, o − Tc, i )
Th, i − Ch − Tc, o
= e − Um At C0 .
Th, i − Tc, i
Add and subtract Tc, i to the numerator of the left hand side and we can
show that the expression becomes
C (T −T )
Th, i − Tc, i Tc, i − Tc, o − c c,Coh c, i
+ = e − Um At C0
Th, i − Tc, i Th, i − Tc, i

Cc Cc
Tc, i 1 + C h
− T c, o 1 + Ch
1 + =
Th, i − Tc, i

Cc Tc, i − Tc, o
1 + 1 + = e − Um At C0
Ch Th, i − Tc, i
Finally, this can be written as
Tc, i − Tc, o e − Um At C0 − 1
=
Th, i − Tc, i 1 + C Cc
h
or
Tc, o − Tc, i 1 − e − Um At C0
=
Th, i − Tc, i 1 + CCc
h
Referring back to Eq. (9.28), we see that the above expression can be directly
substituted to obtain , i.e.
C c 1 − e − Um At C0
=
Cmin 1 + CCc
h
1 − e − Um At C0
= Cmin Cmin
·
Cc + Ch
h i
1 − exp − Um At C1h + C1c
= ·
Cmin C1h + C1c
h i
1 − exp − UCmmin At
Cmin C1h + C1c
= ·
Cmin C1h + C1c
Such expressions are normally written in a short–hand form, where we define
the number of transfer units (NTU) as
Um At
NT U =
Cmin
and the ratio of the minimum to maximum heat capacity rates as
Cmin
C1 = ·
Cmax
It should be clear that, regardless of which side is the minimum side and
which is the maximum,

1 1
Cmin + = 1 + C1 ,
Ch Cc
which allows us to finally define for the parallel flow arrangement as
1 − e−N T U (1 + C1 )
(9.29) = ·
1 + C1
This equation is quite significant in heat exchanger analysis, because one
need not know the outlet temperatures. It suffices only to know the heat
capacity rates9.6, the area, and the overall convection coefficient to calculate
. Moreover, because of the way is defined, the total heat transfer can be
computed from a knowledge of only the inlet temperatures of both streams.
9.6There is, however, one subtle issue: the specific heats may depend upon tempera-
ture. If we were to use the average temperature of a stream to approximate specific heat,
we would need some a priori estimate of the exit temperature.
The procedure for analyzing other heat exchanger designs, e.g. counter–
flow, multiple passes, etc., is similar but more complicated as compared to
what we have shown here. If we simply reverse the design in our simple
tube–type heat exchanger for counter–flow conditions, we find 9.7
1 − e−N T U (1 − C1 ) / 1 − C1 e−N T U (1 − C1 )

: C1 < 1
(9.30) =
N T U/ (N T U + 1) : C1 = 1
Another notable configuration arises when there is a phase change (boil-
ing or condensation), in which case ∆T of the phase change side essentially
vanishes (Fig. 9.4). According to the fact that there is a finite thermal
energy transfer, conservation of energy implies that ṁc p can be taken as in-
finite. Therefore, the phase change side is always the “maximum” side and
the heat capacity ratio can be taken as zero, i.e. C 1 = 0. Moreover, heat
exchanger performance is independent of architecture and flow arrangement!
The analysis gives
(9.31) = 1 − e−N T U .
Reference texts (e.g. Incropera and Dewitt, 2002, Table 11.3, pp. 662–663)
furnish –NTU relationships for a number of configurations and the inverted
relations as well9.8. I&D Ex. 11.3
pp 665
9.7The formula for this case in Incropera and Dewitt (2002, Table 11.3, pp. 662) is
incorrect.
9.8That is, NTU as a function of is provided for the cases that are difficult to invert.
CHAPTER 10
Introduction to Radiation
Up to now, we have considered our subject for the following general

situation: heat transfer occurs when there is a temperature gradient in a
participating medium. In conduction, the medium is not moving, while in
convection, motion is the primary driver. Conversely, heat transfer by ther-
mal radiation requires a temperature gradient but not participating medium.
All bodies at a temperature above absolute zero emit thermal radiation.
Consider for example a hot object in an evacuated container whose walls
are cold. Conduction and convection are not possible, so heat is transferred
strictly by radiation. Heat is transferred from the Sun to the Earth via
radiation. However, the actual mechanism of radiation transfer is not yet
fully understood. Maxwell’s electromagnetic theory and Planck’s quantum
theory have both been used toward this goal.
If radiation is treated as a wave, the radiation from a body of tempera-
ture T is considered emitted at all wavelengths from λ = 0 to λ = ∞. For
most thermal engineering applications, the range of interest is typically be-
tween λ ≈ 0.1 µm and λ ≈ 100 µm (Fig. 10.1). This portion of the spectrum
visible
ultraviolet microwave
X − rays infrared
γ −rays thermal radiation
1e−5 1e−4 1e−3 1e−2 0.1 1 10 100 1e3 1e4 1e5

wavelength (micrometers)
Figure 10.1. The electromagnetic spectrum.
is therefore usually referred to as the thermal radiation range. For example,

solar radiation from the Sun (T = 5760 K) is in the range λ ≈ 0.1 µm to
λ ≈ 3µm (Özişik, 1985). Between λ ≈ 0.4µm and λ ≈ 0.7µm is visible light.
For propagating radiation, the standard relationship between frequency ν
and wavelength λ is
c
λ = ,
ν
116
10.1. SOLID ANGLE AND RADIATIVE QUANTITIES 117
where c is the speed of light in the medium 10.1.

Radiative heat transfer is quite a different phenomenon than conduction
or convection. For example, in some materials, such as glass at certain tem-
peratures, the radiation emitted from a body is a volumetric phenomenon.
That is, it results as an integrated effect from the depths of the body all
the way to its surface. Such bodies are called semi–transparent. Conversely,
opaque bodies realize radiative effects mainly as a surface phenomenon. Ra-
diation from an interior molecule is largely absorbed by its neighbors, so that
any net emission results almost entirely from molecules near the surface of
the body, i.e. within about 1 µm (Özişik, 1985). We shall concentrate on
this model for our studies.
There are two additional points to mention before we proceed. First is
the spectral nature of radiation (Fig. 10.2). This means that the thermal
angular dependence
radiation intensity
wavelength
Figure 10.2. “Spectral” nature of radiation (left) implies a

dependency upon wavelength, while the “directional” nature (right)
implies dependence upon direction.
radiation emitted by a body encompasses a range of wavelengths, where the

magnitude of the radiation varies with the wavelength. Second, radiation
has a directional dependence: emissions are generally a function of direction.
Because there are a number material properties associated with these phe-
nomena, the “bookkeeping” aspect of radiation calculations is much more
substantial than with other modes of heat transfer we have discussed thus
far.
10.1. Solid Angle and Radiative Quantities

In order to quantify radiative heat transfer, we must first define a number
of fundamental entities. First, we consider a new geometric concept called
the solid angle, which will support the modeling of electromagnetic waves
10.1For a vacuum, c ≈ 2.998 × 108 m/s.
as rays that can be described according to straight–line optical principles.

The solid angle can be thought of as the three–dimensional analog of the
standard planar angle. For example, planar angle is defined as the unitless
ratio10.2 of an arc length that is swept out by a radial arm to the arm’s
length (Fig. 10.3). This is naturally defined in the context of a circle by the
r y
dAn
d An
d ω = d An / r 2
θ r
(r, θ , φ )
r
dθ = d s / r ds
dω
dA
x
definition of "solid angle" and
φ
comparison to conventional
planar angle
z spherical coordinate system
Figure 10.3. Definition of planar and solid angles and config-

uration of the solid angle in the spherical coordinate system.
differential expression10.3
ds
dθ =
·
r
Solid angle is instead defined in the context of a sphere as the ratio of a
subtended area and the square of the radius 10.4. We express this as the
differential
dAn
(10.1) dω = ·
r2
We consider solid angle in the context of an observer anchored at the origin
of a spherical coordinate system whose field of view encompasses a normally–
oriented area dAn a distance r away.
10.2Although we attach a “unit” of radians to the planar angle, the quantity is indeed
dimensionless. Angles expressed in the artificial system of degrees are not dimensionless.
10.3This clear from the observation that we can integrate this equation to obtain the
circumference as
Z θ0 =2π Z s0 =s
˛θ0 =2π
r dθ0 = ds , which gives s = r θ 0 ˛θ0 =0 = 2πr .
θ 0 =0 s0 =0
10.4The “unit” of the solid angle is the steradian, (abbreviated as sr ) which, like its
planar counterpart, is actually dimensionless.
Actually, we must take this a step further and define solid angle in terms
of differentials associated with the spherical coordinate system (Fig. 10.4).
Consider emission from a differential element dA along radius r in a specific
r sin θ r dθ
θ r sin θ d φ
r
dθ
dA
dφ
Figure 10.4. Relationship between solid angle and spherical

coordinates.
direction defined by angles θ and φ. When viewed from dA, the emitted
radiation passes through the differential element dA n on the boundary of
the hemisphere specified by radius r. By geometric arguments, we see
dAn = (r dφ sin θ) × (r dθ) = r 2 sin θ dφ dθ ,
so that the solid angle relationship is 10.5
dAn r 2 sin θ dφ dθ
(10.2) dω = = = sin θ dφ dθ .
r2 r2
Another geometric consideration in analyzing radiation emitted from dA
is its projected area on dAn . In other words, the question is how much of
dA can an observer anchored on dAn see? We should notice from Fig. 10.4
that when these areas are normal, i.e. θ = 0, our observer will see all of dA,
while at θ = π/2, surface dA cannot be seen at all. A little trigonometry
(Fig. 10.5) should convince us that
(10.3) dAn = dA cos θ .
The second concept is the fundamental quantity that describes radiative

heat transfer, which is called the radiation intensity. Using this concept,
10.5The solid angle for the entire hemisphere is
Z Z 2π Z π/2
ω = dω = sin θ dθ dφ = φ |2π
0 (− cos θ)|π/2
0 = 2π .
area 0 0
d An
d An
θ
dA dA θ
Figure 10.5. Area of the emitter projected onto the hemisphere

boundary.
we will be able to quantify the spectral and directional properties of radi-

ation, and integrate them into forms that are useful for engineering calcu-
lations. Again, the context is emission by differential surface dA which is
intercepted by differential surface dA n . Radiation intensity of the emitter,
Iλ, e , is defined as the rate at which radiation dq is emitted in the (θ, φ)
direction at wavelength λ per unit area of the emitter normal to this direc-
tion dAn = dA cos θ, per unit solid angle about this direction dω, per unit
wavelength interval dλ about wavelength λ
dq
(10.4) Iλ, e (λ, θ, φ) = ·
dA cos θ dω dλ
Intensity has units of W/ m2 sr µm . We can re–arrange Eq. (10.4) in terms

of the rate at which radiation leaves the emitter dA and is intercepted by
dAn as
dq
(10.5) dqλ = = Iλ, e (λ, θ, φ) dA cos θ dω ,
dλ
where dqλ is a shorthand notation and has units of W/µm. This expression
allows us to compute radiation heat transfer as a function of radiation in-
tensity, once it is known. Substituting our results for the solid angle from
Eq. (10.2) and casting the heat transfer as a flux quantity with respect to
dA, i.e. heat transfer per unit dA, we can write
(10.6) dqλ00 = Iλ, e (λ, θ, φ) cos θ sin θ dθ dφ .
This provides a basis for integrating dq λ00 to find qλ00 , the flux associated with
any particular finite solid angle. For example, the flux associated with the
hemisphere around dA is
Z 2π Z π/2
00
qλ = Iλ, e (λ, θ, φ) cos θ sin θ dθ dφ .
0 0
In other words, this expression gives the rate at which radiation at wave-
length λ is emitted in all directions. In more direct terms, it is simply the
rate at which radiation is emitted per unit area of the emitter. We define this
as the emissive power , or, more exactly, the spectral hemispherical emissive
power10.6
Z 2π Z π/2
(10.7) Eλ (λ) = Iλ, e (λ, θ, φ) cos θ sin θ dθ dφ ,
0 0
which has units of W/ m2 µm . The total hemispherical emissive power,

i.e. the rate at which radiation is emitted in all directions at all wavelengths
per unit area of the emitter, is obtained by integrating over λ
Z ∞ Z 2π Z π/2
(10.8) E = Iλ, e (λ, θ, φ) cos θ sin θ dθ dφ dλ ,
0 0 0
which has units of 2
W/m .
Although the directional distribution of surface radiation emission varies
according to the properties of specific surfaces, there is an idealization which
closely approximates conditions for many surfaces. This idealized form of
radiation is called diffuse radiation, which simply indicates that there is no
directional dependence. For example, for a diffuse emitter, the radiation
intensity is independent of direction, so that I λ,e (λ, θ, φ) → Iλ,e (λ). Since
Iλ,e (λ) is now constant with respect to directions, we see the emissive power
in Eq. (10.7) can be cast as
Z 2π Z π/2
Eλ (λ) = Iλ, e (λ) cos θ sin θ dθ dφ
0 0
Z 2π Z π/2
= Iλ, e (λ) cos θ sin θ dθ dφ
0 0
Z 2π Z π/2
= Iλ, e (λ) dφ cos θ sin θ dθ
0 0
2π 1 π/2
2

= Iλ, e (λ) φ sin θ

0 2 0
1
= Iλ, e (λ) 2π ×
2
(10.9) Eλ (λ) = π Iλ, e (λ) (diffuse emitter)
It can be similarly shown that E = πIe , where again E is the total emissive
power and Ie is the total radiation intensity. I&D Ex. 12.1
Now let’s look at some additional phenomena of interest in radiation pp 707
heat transfer. First is the concept of irradiation, i.e. incoming or incident
radiation. This is described by its own corresponding intensity, called the
incident spectral intensity, Iλ,i (λ, θ, φ), which, like the corresponding quan-
tity for emitters, is a function of wavelength and direction and is given per
unit area of the intercepting surface normal to this direction per unit solid
angle, per unit dλ. The corresponding flux is the irradiation. Specifically,
there is the spectral irradiation, G λ (λ), which is the rate at which radiation
10.6We include the subscript λ in the notation for E (λ) to indicate that this is a per
λ
unit wavelength dλ about λ quantity, synonymous with the way we have cast Eq. (10.5).
10.2. BLACKBODY RADIATION 122
of wavelength λ is incident on a surface per unit area of surface per unit dλ

about λ from all directions. Also, there is the total irradiation, G, which is
the rate at which radiation is incident per unit area from all directions at all
wavelengths. This phenomenon obeys all the equations we discussed above I&D Ex. 12.2
if we substitute Eλ → Gλ , E → G, Iλ,e → Iλ,i , etc. pp 710
Another phenomenon is that of radiosity, which accounts for all radi-
ation leaving a surface, i.e. what is emitted directly plus whatever inci-
dent radiation is reflected. Therefore, this is generally not the same as the
emissive power. We have all the corresponding quantities with their usual
definitions: spectral radiosity intensity, I λ,e+r (λ, θ, φ), where e + r indicates
emission plus reflection, spectral radiosity, J λ (λ), and total radiosity, J. All
corresponding relationships among these quantities are the same as those
derived above.
10.2. Blackbody Radiation

When describing radiation with regard to real surfaces, it is useful to de-
fine the ideal case, called a blackbody, for comparison. It has three idealized
properties
• A blackbody absorbs all incident radiation without reflecting, trans-
mitting, or scattering it, regardless of wavelength and direction
• For a given temperature and wavelength, no surface can emit more
energy than a blackbody — it is the ideal emitter
• While emitted radiation from a blackbody is a function of wave-
length and temperature, it is independent of direction — a black-
body is a diffuse emitter
Thus, a blackbody is defined as a perfect absorber and emitter. The term
“black” should be distinguished from its common usage regarding the ap-
pearance by visual observation of the blackness of a surface. The human
eye can detect the color black only in the visual range of the radiation spec-
trum, but the visible range of light is only a small portion of the range
of the thermal radiation spectrum (recall Fig. 10.1). For example, ice is
non–black to the eye, however, for long wavelength thermal radiation, it is
essentially black. Note that our definition of a true blackbody means that
these properties must exist over the entire spectrum, i.e. λ = 0 → ∞.
A blackbody, not surprisingly, is characterized by its own parameters,
for example, the spectral blackbody radiation intensity, I λ,b (λ, T ). Note
that this does not depend on direction because of the diffuse nature of the
radiation. This quantity was first determined by Max Planck for emission
into a vacuum as
2 h c2
(10.10) Iλ,b (λ, T ) = ,
λ5 eh c/(λ k T ) − 1
where h = 6.6256 × 10−34 J · sec is the universal Planck constant, k =
1.3805 × 10−23 J/K is the Boltzmann constant, c ≈ 2.998 × 10 8 m/sec is the
approximate speed of light in a vacuum, and T is the absolute temperature.

Because the blackbody is a diffuse emitter, we can immediately derive the
spectral blackbody emissive power from the relation E λ,b (λ, T ) = πIλ,b (λ, T )
to get
2 π h c2
Eλ,b (λ, T ) = 5 h c/(λ k T ) .
λ e − 1
Some texts (e.g. Özişik, 1985; Incropera and Dewitt, 2002) show this relation
in terms of the so–called first and second radiation constants, C 1 = 2πhc2
and C2 = hc/k giving10.7
C1
(10.11) Eλ,b (λ, T ) = .
λ5 e 2 T)
C /(λ − 1
Eq. (10.11) is known as Planck’s Distribution (Fig. 10.6) and has the fol-
1e+8
spectral emissive power (W/m**2 micrometer)
T = 50 K
T = 100 K
T = 500 K
1e+6 T = 1000 K
T = 2000 K
T = 5800 K
1e+4
1e+2
1e+0
1e-2
0.1 1 10 100
wavelength (micrometers)
Figure 10.6. Planck’s Distribution of blackbody radiation.
lowing notable properties

• Emitted radiation varies continuously with wavelength
• At any wavelength, the magnitude of emitted radiation increases
with temperature
• The bulk of radiation is emitted at shorter wavelengths for higher
temperatures
• The sun, approximated as a 5800 K blackbody, emits a significant
fraction of radiation in the visible region of the spectrum.
10.7These products yield C = 3.743 × 108 W µm4 /m2 and C = 14, 387 µm K.
1 2
The third point above has an interesting proof associated with it known as
Wien’s Displacement Law . If the Planck Distribution is differentiated with
respect to λ and set equal to zero then solved, the result is
(10.12) λ T ≈ 2897.8 µm · K .
This is shown in detail in Appendix C. The equation specifies the locus
of peak points for Eλ,b . Because the product λT is constant, the maxi-
mum spectral emissive power drifts toward shorter wavelengths for higher
temperatures.
What we are really interested in, i.e. what will enable us to solve prob-
lems, is the total blackbody emissive power. In principle, this can be derived
by integrating the spectral blackbody emissive power given by Eq. (10.11)
over all wavelengths
Z ∞
C1
(10.13) Eb (T ) = 5 eC2 /(λT ) − 1
dλ .
0 λ
Although this operation is not trivial 10.8 it can be shown that
(10.14) Eb (T ) = σ T 4 ,
where σ = 5.670 × 10−8 W/(m2 · K 4 ) is the Stefan–Boltzmann constant,
which depends upon the radiation constants C 1 and C2 . This equation is
known as the Stefan–Boltzmann Law and is important because it enables
calculation of the amount of radiation emitted in all directions and over all
wavelengths simply by knowing the blackbody temperature. Because the
blackbody is diffuse, we can back out the total blackbody intensity as
Eb σ T4
Ib = = .
π π
Often we need to know the fraction of radiative emission from a black-
body occurring over some finite wavelength range, e.g. λ = 0 → λ 1 . Rigor-
ously, this can be computed by simply integrating Eq. (10.11) appropriately,
e.g.
R λ1 Z λ1
0 Eλ,b dλ 1
f0→λ1 = R ∞ = Eλ,b dλ
0 Eλ,b dλ
σ T4 0
However, like the Stefan–Boltzmann Law, computation of this integral is
not trivial. Introducing the Planck Distribution from above, we see that
this gives
Z λ1
1 C1
f0→λ1 = 4 5 C /(λ T) − 1
dλ .
σT 0 λ e 2
Because this is so difficult to evaluate, we would like to leverage any ana-

lytical advantage that we can from this expression. We notice that f 0→λ1
is a function of both wavelength and temperature, however, in this case, a
clever reduction of two variables to one variable can be made.
10.8This integral can be evaluated in terms of the Riemann zeta function.
Let us use the “Integration by Substitution” Theorem from Calculus

(e.g. Gillman and McDowell, 1978, pp. 270), given by 10.9
b g(b)
du
Z Z
(10.15) f (u) dx = f (u) du ,
a dx g(a)
where u = g(x). We want to write f0→λ1 in terms of the combined variable

λT . Specifically, λ corresponds to x, and u maps to λT , therefore,
d (λ T )
du = d (λ T ) = dλ = T dλ ,
dλ
which leads to the substitution
d (λ T )
dλ → .
T
The upper limit is now λ1 → λ1 T . Therefore, the equation can be rewritten
as
Z λ1 T
1 C1 d (λT )
f0→λ1 =
σ T4 0

5
λ e C 2 /(λ T ) − 1 T
Z λ1 T
C1 1
= 5 d (λ T )
σ 0 (T λ) e 2 /(λ T ) − 1
C
This expression can be recast in terms of u = λT as

Z u1
C1 1
(10.16) f0→λ1 = du .
σ 0 u e 2 /u − 1
5 C
The form and difficulty are similar to Eq. (10.13). Reference texts have tab-
ulated f as a function of the combined variable λT . For example, Incropera
and Dewitt (2002, Table 12.1, pp. 716) and Appendix C (Table C.1)give
abbreviated summaries, while Dunkle (1954) and Siegel and Howell (2001)
contain more comprehensive data.
10.9This theorem can be proven in a straightforward fashion, for example, let F be

the anti–derivative of function f and u = g(x), so that
Z b Z b
du ´˛˛b
f g(x) g 0 (x) dx = F g(x) ˛
` ´ `
f (u) dx =
a dx a a
` ´ `` ´˛˛g(b)
´
Z g(b)
= F g(b) − F g(a) = F u ˛ = f (u) du ,
g(a) g(a)
which proves Eq. (10.15).

10.3. RADIATION CHARACTERISTICS OF REAL SURFACES 126
Eq. (10.16) can be applied more generally for arbitrary ranges of radia-
tion λ = λ1 → λ2 according to
R λ2
λ1 Eλ,b dλ
fλ1 →λ2 =
σ T4
R λ2 R λ1
0 Eλ,b dλ − 0 Eλ,b dλ
=
σ T4
R λ2 R λ1
0 Eλ,b dλ Eλ,b dλ
= 4
− 0
σT σ T4
(10.17) fλ1 →λ2 = f0→λ2 − f0→λ1
Note the special case for a range with a lower bound λ = λ 2 → ∞ corre-
sponds to I&D Ex. 12.3
R∞ pp 717
λ2 Eλ,b dλ
fλ2 →∞ = I&D Ex. 12.4
σ T4
R∞ R λ2 pp 719
0 Eλ,b dλ − 0 Eλ,b dλ
=
σ T4
R∞ R λ2
0 Eλ,b dλ Eλ,b dλ
= 4
− 0
σT σ T4
R λ2
σ T4 Eλ,b dλ
(10.18) fλ2 →∞ = 4
− 0 = 1 − f0→λ2
σT σ T4
10.3. Radiation Characteristics of Real Surfaces
10.3.1. Emissivity. Now that we have developed the concept of an
ideal surface (such that we have a basis of comparison), let us now look at
characteristics of real surfaces. In fact, we will specify properties of real sur-
faces via direct comparison to those of the blackbody. For example, recall
that the blackbody, being the ideal emitter, emits the maximum radiation
possible. Therefore, we can define the emissivity of a body as the ratio of
radiation emitted by its surface to the radiation emitted by a blackbody
at the same temperature. The spectrum of radiation emitted by a real
surface is not necessarily the same as that specified by the Planck Distribu-
tion (Fig. 10.7) Also, the directional distribution may not qualify as diffuse.
Therefore, emissivity can assume various forms according to whether one
talks of emission at a given wavelength or emission in a given direction, or
some overall integrated value.
First, we define a quantity called the spectral directional emissivity,
ελ,θ (λ, θ, φ, T ) of a surface at temperature T as the ratio of the intensity
of the radiation emitted at wavelength λ in the (θ, φ) direction to the inten-
sity of the radiation emitted by a blackbody for the same values of λ and
T . This has the form
Iλ,e (λ, θ, φ, T )
(10.19) ελ,θ (λ, θ, φ, T ) = .
Iλ,b (λ, T )
Planck (blackbody)
real surface
spectral emissive power
Pl
an
ce
ck
rfa
su
(b
la
al
ck
re
b
od
y)
wavelength
Figure 10.7. Potential differences between the characteristics

of real versus idealized (blackbody) surfaces for spectral (left) and
directional (right) aspects of radiation emission.
Note that the blackbody term in the denominator appears correctly as hav-
ing no dependence upon direction since it is diffuse emitter. Likewise, the
total directional emissivity, εθ , is an average over the wavelength spectrum
and is defined as
Ie (θ, φ, T )
(10.20) εθ (θ, φ, T ) = .
Ib (T )
Note that these two quantities are actually defined in terms of radiation
intensity because of directional dependence.
Conversely, we can also defined an emissivity that has been averaged
over all directions, called the spectral hemispherical emissivity
Eλ (λ, T )
(10.21) ελ (λ, T ) = ,
Eλ,b (λ, T )
which is given in terms of emissive power. We recall from previous definitions
that these quantities represent emissive powers that have been integrated
over the hemispherical area of emission. How does this relate to the direc-
tional emissivity ελ,θ ? The spectral hemispherical emissivity can be written
according to the definitions
R 2π R π/2
Iλ,e (λ, θ, φ, T ) cos θ sin θ dθ dφ
(10.22) ελ (λ, T ) = 0 R 2π0 R π/2 .
0 0 I λ,b (λ, T ) cos θ sin θ dθ dφ
Using Eq. (10.19), substitute Iλ,e = ελ,θ Iλ,b into the top term, take note that
Iλ,b in both numerator and denominator do not depend upon direction (so
they can be taken outside their integral signs), and the relation simplifies to
R 2π R π/2
ελ,θ (λ, θ, φ, T ) cos θ sin θ dθ dφ
(10.23) ελ (λ, T ) = 0 0 R 2π R π/2 .
0 0 cos θ sin θ dθ dφ
If we make the further assumption that ε λ,θ is independent of φ (a rea-
sonable assumption for many surfaces), we can obtain (after evaluating the
denominator)
Z π/2
(10.24) ελ (λ, T ) = 2 ελ,θ (λ, θ, T ) cos θ sin θ dθ .
0
The total hemispherical emissivity, ε, accounts for all directions and all
wavelengths and is defined as
E(T ) E(T )
(10.25) ε(T ) = = .
Eb (T ) σ T4
R∞
Now, using our previously established definitions of E = 0 Eλ (λ) dλ and
Eλ (λ, T ) = ελ (λ, T )Eλ,b (λ, T ), we can see that
R∞
ελ (λ, T ) Eλ,b (λ, T ) dλ
(10.26) ε(T ) = 0 .
Eb (T )
If the emissivities of a surface are known, these definitions can be applied
directly to calculate all emission characteristics.
We note (again according to definition) that the directional emissivity
of a diffuse emitter is constant, i.e. independent of direction. For “real”
surfaces, it may be said in general that I&D Ex. 12.5
• emissivity of metallic surfaces is small pp 724
• oxide layers increase the emissivity of metallic surfaces I&D Ex. 12.6
• emissivity of non–conductors is large pp 727
• emissivity of conductors increases with T , but may either decrease
or increase for non–conductors
Moreover, emissivity can depend on other factors, for example local residual
stresses or chemical interactions. Özişik (1985) and Incropera and Dewitt
(2002) contain summary data, while more comprehensive data are available
in reference texts (e.g. Wood et al., 1964).
10.3.2. Absorptivity. It follows that, like for emission, we can quan-
tify the characteristics of irradiation in terms of surface properties. At the
beginning of this chapter, we introduced the idea of a semi–transparent
medium in which a portion of the irradiation is absorbed 10.10, a portion is
reflected10.11, and a portion is transmitted10.12. For the most part, we will
concentrate on situations where transmission vanishes, so that the problem
is one of only absorption and reflection. Recall that such a material is called
opaque.
A simple energy balance for opaque surfaces shows
(10.27) Gλ = Gλ,ref + Gλ,abs + Gλ,trans ,
where Gλ is the irradiation, Gλ,ref and Gλ,abs are the components reflected
and absorbed, respectively, and Gλ,trans is the transmitted component, which
10.10Manifested as heating at the surface.
10.11Thrown back off away from the surface.
10.12Allowed to pass through to lower layers of the material.
often vanishes. As we introduced various parameters that characterize emis-

sion, we likewise now introduce parameters that characterize absorption,
transmission, and reflection10.13. The spectral directional absorptivity de-
termines the fraction of irradiation absorbed by a surface
Iλ,i,abs (λ, θ, φ)
(10.28) αλ,θ (λ, θ, φ) = ,
Iλ,i (λ, θ, φ)
where these properties are not generally dependent upon surface tempera-
ture (as the corresponding parameters for emission were). The correspond-
ing parameter after directional averaging is the spectral hemispherical ab-
sorptivity which is defined as
Gλ,abs (λ)
(10.29) αλ (λ) = ,
Gλ (λ)
which, from the definition
Z 2π Z π/2
Gλ = Iλ,i (λ, θ, φ) cos θ sin θ dθ dφ
0 0
and Eq. (10.28), can be expressed directly as
R 2π R π/2
αλ,θ Iλ,i (λ, θ, φ) cos θ sin θ dθ dφ
(10.30) αλ (λ) = 0R 2π0 R π/2 .
0 0 Iλ,i (λ, θ, φ) cos θ sin θ dθ dφ
Note that if the incident radiation is diffuse, I λ,i (λ, θ, φ) → Iλ,i (λ) so this
comes outside the integrals and cancels on top and bottom. Also, if α λ,θ is
independent of φ we see that the expression becomes
Z π/2
(10.31) αλ (λ) = 2 αλ,θ (λ, θ) cos θ sin θ dθ .
0
in an analogous manner as Eq. (10.24) for the spectral emissivity.
The total hemispherical absorptivity is the integrated average over all
directions and the entire wavelength spectrum and is defined simply as the
fraction of the total irradiation absorbed by the surface
Gabs
(10.32) α = ,
G
which, from the previous definition
Z ∞
G = Gλ dλ
0
and Eq. (10.29) can be expressed directly as
R∞
αλ (λ) Gλ (λ) dλ
(10.33) α = 0 R∞ .
0 Gλ (λ) dλ
10.13Although again, we will largely stress absorption and reflection over transmission.
10.4. KIRCHHOFF’S LAW AND GRAY SURFACES 130
10.3.3. Reflectivity. Reflectivity determines the fraction of incident

radiation reflected by a surface. The complicating factor here is that the
phenomenon is inherently bidirectional. That is, it depends both upon the
direction of the incident radiation and the direction of the reflected radiation.
For now, we will avoid this problem by working with a form that represents
the integrated average over the hemisphere associated with the reflected part
of the radiation — thus, the directional dependence is simply associated with
the incident radiation. Accordingly, the spectral directional reflectivity is
defined as
Iλ,i,ref (λ, θ, φ)
(10.34) ρλ,θ (λ, θ, φ) = ,
Iλ,i (λ, θ, φ)
the spectral hemispherical reflectivity is defined as
Gλ,ref (λ)
(10.35) ρλ (λ) = ,
Gλ (λ)
and the total hemispherical reflectivity is defined as
Gref
(10.36) ρ = .
G
We can derive similar relations as we did for absorptivity, i.e.
R 2π R π/2
ρλ,θ Iλ,i (λ, θ, φ) cos θ sin θ dθ dφ
(10.37) ρλ (λ) = 0R 2π0 R π/2
0 0 Iλ,i (λ, θ, φ) cos θ sin θ dθ dφ
and
R∞
ρλ (λ) Gλ (λ) dλ
(10.38) ρ = 0 R∞ .
0 Gλ (λ) dλ
There are similar relationships for transmissivity τ .
10.3.4. Relationships. From the energy balance of irradiation we wrote
in Eq. (10.27), it can be shown by dividing by the irradiation that 10.14
(10.39) ρ λ + αλ = 1
and, if properties are averaged over the entire wavelength spectrum I&D Ex. 12.7
(10.40) ρ + α = 1. pp 735
Therefore, knowledge of one property implies the other.
10.4. Kirchhoff ’s Law and Gray Surfaces

Energy conservation applied to irradiation gives rise to a relationship
between absorbtivity and reflectivity, shown in Eqs. (10.39) and (10.40).
There is an additional set of relationships between absorbtivity and emis-
sion, known as Kirchhoff’s Law , which is derived as follows. Consider an
isothermal blackbody enclosure at temperature T s which contains a smaller
object, also at temperature Ts (Fig. 10.8). The two bodies are in thermal
10.14We assume here that the transmitted radiation vanishes, so that G
λ,trans = 0.
Ts
Ts
Figure 10.8. Radiative transfer between two bodies at thermal

equilibrium.
equilibrium. The small object experiences a diffuse irradiation flux which is

equal to the blackbody radiation flux emitted by the enclosure G = E b (Ts ).
The amount of radiation absorbed by the object is then obtained simply by
multiplying by its absorbtivity α and area A obj
(10.41) qabs = α Aobj G = α Aobj Eb (Ts ) .
However, the body must emit radiation equal to the rate it is absorbed in
order to satisfy thermal equilibrium, which implies q emit = qabs . Combining
this observation with Eq. (10.41), we can write
(10.42) qabs = α Aobj Eb (Ts ) = qemit .
From the emission standpoint, we recall that emissivity is defined as the
ratio of the actual emissive power to the total emissive power, as quantified
by Eq. (10.25). In particular, the emissivity of the small object is its actual
emissive power divided by the emissive power of a blackbody operating at the
same temperature. Converting these fluxes into absolute quantities based
on the area of the small body, we can thus write
(10.43) qemit = ε Aobj Eb (Ts ) .
Eqs. (10.42) and (10.43) clearly imply
(10.44) ε = α
because they are otherwise equivalent. Note that in Eq. (10.42), the black-
body emissive power Eb (Ts ) applies to the walls of the enclosure, while in
Eq. (10.43) the blackbody emissive power applies to our generic blackbody
object operating at temperature Ts . Therefore, the restriction of thermal
equilibrium is a primary requirement for the validity of Eq. (10.44). This
equation is one form of Kirchhoff’s Law.
By the same reasoning, both the spectral form of this law may be shown
to be true
(10.45) ε λ = αλ ,
and the spectral–directional version as well
(10.46) ελ,θ = αλ,θ .
Eq. (10.46) is applicable to all cases, because ε λ,θ and αλ,θ are indepen-
dent of their respective overall spectral and directional distributions. What
about the constraints on Eq. (10.45)? Are they as restrictive as those for
Eq. (10.44), i.e. that the irradiation is from a blackbody operating at the
same temperature?
Let us write Eq. (10.45) in terms of its respective definitions for ε λ in
Eq. (10.23) and αλ in Eq. (10.30), which yields
R 2π R π/2
0 0 ελ,θ (λ, θ, φ, T ) cos θ sin θ dθ dφ
R 2π R π/2 =
0 0 cos θ sin θ dθ dφ
R 2π R π/2
0 0 αλ,θ Iλ,i (λ, θ, φ) cos θ sin θ dθ dφ
R 2π R π/2 ·
0 0 I λ,i (λ, θ, φ) cos θ sin θ dθ dφ
This relation, and thus Eq. (10.45), will be satisfied for either of the following
two conditions
• If the irradiation Iλ,i (λ, θ, φ) is diffuse then Iλ,i (λ, θ, φ) → Iλ,i (λ) so
this comes outside the integrals and cancels, leaving the two sides
equal since ελ,θ is always equal to αλ,θ
• If the surface is diffuse, so that ελ,θ and αλ,θ are independent of
direction, then they come outside the integrals, in which case the
two sides are again equal
Are there any other conditions in which Eq. (10.44) will be satisfied?
As with the spectral components, let us recast Eq. (10.44) in terms of its
fundamental definitions for ε in Eq. (10.26) and α in Eq. (10.33). We find
R∞ R∞
0 ε λ (λ, T ) E λ,b (λ, T ) dλ αλ (λ) Gλ (λ) dλ
= 0 ·
Eb (T ) G
Assuming ελ = αλ , this expression is satisfied for either of the following two
conditions
• If the irradiation is the result of blackbody emission, then G λ (λ) =
Eλ,b (λ, T ) and G = Eb (T ), in which case the two sides are equal —
this was our original condition
• If the surface is gray, meaning that both ε λ and αλ are indepen-
dent of wavelength λ — they both come outside their respective
integrals, so that the expression is once again satisfied
We see now that there are a number of forms of Kirchhoff’s Law. For I&D Ex. 12.9
convenience, we summarize these forms and their restrictions in Table 10.1. pp 742
I&D Ex. 12.10
pp 744
Table 10.1. Forms of Kirchhoff’s Law
Form Restrictions
ελ,θ = αλ,θ none
ελ = α λ ελ,θ = αλ,θ and irradiation is diffuse or surface is diffuse
ε = α ελ = αλ and thermal equilibrium and black irradiation
or gray surface
CHAPTER 11
Radiation Exchange
In Chapter 10, we described some of the characteristics of thermal radia-

tion. Now we extend this treatment to consider radiation exchange between
two or more finite bodies.
11.1. The View Factor

Radiation exchange depends on surface geometries and their orienta-
tions, not to mention their radiative properties and temperatures. We will
still invoke the idealization that any two surfaces are separated by a non–
participating medium. That is, the medium has no affect upon the radiation
exchange, for example, as with a vacuum. We will initially focus on the geo-
metrical aspect by introducing the view factor concept 11.1. The view factor,
Fij , is defined simply as the fraction of radiation leaving surface i that is
intercepted by surface j. To derive F ij , we assume arbitrary surfaces Ai and
Aj , which have respective differential areas dA i and dAj (Fig. 11.1). The
θj
d Aj
R
θi Tj
Aj
d Ai Ti
Ai
Figure 11.1. Calculation of the view factor.
differential areas under consideration are connected by a radius of length R,

which forms polar angles θi and θj with the respective unit normal vectors.
In general, R, θi , and θj are not constant as we consider the entirety of these
two surfaces, i.e. the surfaces may have curvature, etc.
11.1The terms “configuration factor” and “shape factor” used in some texts are
synonymous.
134
11.1. THE VIEW FACTOR 135
From the form of Eq. (10.5) on pp. 120, we can deduce the rate at which
radiation leaves dAi and strikes dAj as
dqi→j = Ii dAi cos θi dωj,i ,
where Ii is the intensity of the radiation leaving dA i and dωj,i is the solid
angle subtended by dAj when viewed from dAi . Now recall the definition of
solid angle is dAnormal /R2 , which means that, taking the projection of dA j ,
gives
dAj cos θj
dωj,i = .
R2
Substituting, we find
cos θi cos θj
dqi→j = Ii dAi dAj .
R2
The total radiation coming from dAi is that which is emitted plus that which
is reflected, i.e. Ii should be written more exactly as Ii → Ii,e+r , where the
extra subscripts indicate emission and reflection.
Now we make an important assumption about radiation emanating from
dAi . We assume that this surface emits and reflects diffusely, so that the
total radiation is given by the radiosity relationship J i = πIi,e+r . We can
then cast dqi→j as
cos θi cos θj
(11.1) dqi→j = Ji dAi dAj .
πR2
We now make another critical assumption: J i is taken to be uniform over
surface i. Then, the total rate at which radiation leaves surface i and is
intercepted by surface j can be found simply by integrating Eq. (11.1) over
both surfaces, which gives
cos θi cos θj
Z Z
qi→j = Ji dAi dAj
Ai Aj πR2
cos θi cos θj
Z Z
= Ji dAi dAj .
Ai Aj πR2
Uniformity in Ji allows us to take it outside the integral.
Now, from the definition of view factor, the radiation which leaves sur-
face i and is intercepted by surface j is q i→j and the total radiation that
leaves surface i (emitted plus reflected) is A i Ji , therefore, by definition, the
view factor is
radiation from i that hits j qi→j
(11.2) Fij = = ·
total radiation from i Ai Ji
By inspection, we see that view factor can be calculated as I&D Ex. 13.1
1
Z Z
cos θi cos θj pp 798
(11.3) Fij = 2
dAi dAj .
A i Ai Aj πR
Eq. (11.3) is remarkable in the sense that, with the assumptions we have
made, the view factor is dependent strictly on geometry. We do not require
11.2. VIEW FACTOR ALGEBRA 136
any particular knowledge of the radiative quantities to calculate it. Again,

the important assumptions we have made in deriving Eq. (11.3) are
• the medium does not participate,
• surface i emits and reflects diffusely,
• radiosity of surface i is uniform
By a similar derivation, the complementary view factor F ji can be repre-
sented as
radiation from j that hits i qj→i
Fji = = ,
total radiation from j Aj Jj
from which we find
1 cos θi cos θj
Z Z
(11.4) Fji = dAi dAj .
A j Ai Aj πR2
11.2. View Factor Algebra

Based upon the definitions in Eqs. (11.3) and (11.4), certain identities
can be immediately deduced. For example, both F ij and Fji depend upon
the same integral, which implies
(11.5) Fij Ai = Fji Aj .
Eq. (11.5) is termed the reciprocity rule and is useful for finding one view
factor when the other is known. This relation is the first result in a system
of algebra we will construct in order to minimize the number of times that
the view factor integral actually has to be computed for a specific radiation
problem.
Let us now consider view factors from the standpoint of conservation
of energy. Imagine an enclosure involving a particular surface, which may,
in general, be concave, flat, or convex. This surface is denoted by cross–
hatching in Fig. 11.2. Without loss of generality, we label this surface as
concave
convex
flat
Figure 11.2. Enclosures for radiation exchange.
1, and the remaining surfaces in the enclosure 2, 3, . . . , N . Note that one

or more of the 2, 3, . . . , N surfaces may be “virtual” in the sense that they
represent an unbounded area.
The total radiation emanating from surface 1 is A 1 J1 . In a proper ac-
counting of how energy is conserved, all of this radiation must be intercepted
by the other surfaces. In cases where surface 1 is concave, some of the ema-
nating radiation will be self–intercepted because this surface can “see” itself.
In cases where the surface is flat or convex, the surface does not intercept
any of its own emanating radiation. With these observations, we can write
a simple conservation law for the radiation from surface 1 as
A1 J1 = q1→1 + q1→2 + q1→3 + · · · + q1→N .
Dividing by A1 J1 , we find
A1 J1 q1→1 q1→2 q1→3 q1→N
= 1 = + + + ··· + .
A1 J1 A1 J1 A1 J1 A1 J1 A1 J1
Of course, the right hand side simply represents the sum of all the view
factors, i.e.
1 = F11 + F12 + F13 + · · · + F1N .
Note according to the above discussion that F 11 = 0 for flat and convex
surfaces, but F11 6= 0 for concave surfaces. We can write this in compact
form for surface i = 1 as
N
X
(11.6) Fij = 1 ,
j=1
which is called the summation rule.

Once again, the summation rule is a simple statement of conservation of
radiative energy. We can write such an equation for each of the N surfaces
in the enclosure, i.e. 1 ≤ i ≤ N equations. For any of these equations, the
term Fii represents self–interception of radiation for that particular surface.
Generalizing the above observation, F ii = 0 for flat and convex surfaces, but
Fii 6= 0 for concave surfaces.
Based on the observation in Eq. (11.6) that there are N view factors
associated with each surface and there are N surfaces, the total number
of view factors that characterize a radiation problem is N 2 , which can be
expressed as the matrix
 
F11 F12 F13 · · · F1N
 F21 F22 F23 · · · F2N 
 
 F31 F32 F33 · · · F3N 
.
 .. .. .. .. 

..
 . . . . . 
FN 1 FN 2 FN 3 · · · F N N
Each of these view factors is defined by the view factor integral, as given
by Eq. (11.3). In general, the view factor integral is not trivial to evaluate.
However, Eqs. (11.5) and (11.6) clearly imply that not all of the factors has
to be calculated directly.
Using the summation rule, N view factors can be obtained from the N
equations derived from the summation rule applied to each surface of the
enclosure. Also, N (N − 1)/2 of the factors can be obtained by N (N − 1)/2
applications of the reciprocity relation. This number is derived as follows: as
applied to a given surface there are N − 1 reciprocity relations 11.2, and this
can be applied N times, thus giving N (N − 1) relationships. However, only
half of these are independent, giving N (N − 1)/2 relations 11.3. Therefore,
only N 2 − N − N (N − 1)/2 = N (N − 1)/2 view factors actually have to
be calculated directly. For example, for a 3–surface enclosure there are
32 = 9 view factors, only 3 · 2/2 = 3 of which have to be computed directly.
The other 6 factors can be calculated using the algebraic relationships in
Eqs. (11.5) and (11.6).
Example 11.1:
Consider a simple 2–surface enclosure of a sphere within a sphere
(Fig. 11.3). The smaller sphere has an outer surface area of A 1 , while the
Figure 11.3. Two concentric spheres exchanging radiation.
larger has an inner surface area of A 2 . Determine the view factors charac-
terizing this problem.
There are N 2 = 22 = 4 relevant view factors, only N (N −1)/2 = 2·1/2 =
1 of which has to actually be calculated directly, i.e. by evaluating the view
factor integral. The trick is to intelligently pick the view factor that must be
calculated! For example, all radiation leaving the inner sphere is intercepted
by the outer sphere, therefore by inspection, we conclude F 12 = 1. In this
case, we skipped having to evaluate Eq. (11.3).
The rest of the problem is trivial. From reciprocity, we see
A1 A1
F21 = F12 = .
A2 A2
From summation, we see
F11 + F12 = 1 → F11 = 0 .
11.2That is, there is a reciprocity relationship to every other surface besides the one
in question. There is no meaningful reciprocity relationship between a surface and itself.
11.3An even more direct proof is to realize that this problem is also quantified by the
number of combinations of N entities taken 2 at a time. From combinatorics, we find
N! N × (N − 1) × (N − 2)! N (N − 1)
CN,2 = = = ·
(N − 2)! 2! (N − 2)! × 2 × 1 2
Of course, this also could have been concluded by inspection, since the inner
sphere is completely convex. This would have been the other easy starting
point for the problem. Also from summation, we see
A1
F21 + F22 = 1 → F22 = 1 − .
A2
For more complex geometries, view factors may have to be calculated using
the double–integral definition. Solutions have been obtained for a variety of
configurations (e.g. Incropera and Dewitt, 2002, §13.1.2). ♦♦♦
Another important identity can be derived by considering the case where
a surface is broken up into its component parts (Fig. 11.4). First, it is evi-
2
1
Figure 11.4. Composite surfaces.
dent that radiation reaching a composite surface is the sum of the radiation
reaching each of its parts
n
X
(11.7) Fi(j) = Fik ,
k=1
where the parentheses around j indicate that it is a composite surface made

up of 1, 2, . . . , k, . . . , n components. The composite area is clearly
n
X
(11.8) A(j) = Ak .
k=1
For example, in Fig. 11.4, we would write F i(j) = Fi(1,2) = Fi1 + Fi2 and
A(j) = A(1,2) = A1 +A2 . Turning this concept around via reciprocity, we can
also determine a component–wise description of the surface that originates
the radiation. Multiplying Eq. (11.7) by A i , we obtain

Xn
Ai Fi(j) = Ai Fik
k=1

= Ai Fi1 + Fi2 + · · · + Fin
= Ai Fi1 + Ai Fi2 + · · · + Ai Fin
= A1 F1i + A2 F2i + · · · + An Fni (reciprocity)
Xn
= Ak Fki .
k=1
We can also write an “overall” reciprocity equation A i Fi(j) = A(j) F(j)i , so
that we finally conclude
n
1 X
(11.9) F(j)i = Ak Fki .
A(j)
k=1
Referring again to Fig. 11.4 as an example, we find I&D Ex. 13.2
A1 F1i + A2 F2i pp 799
F(1,2)i = .
A1 + A 2
Finally, other view factor relations may exist, depending upon the spe-
cific problem. For example, because we have already shown that the view
factor integral in Eq. (11.3) depends only upon geometry, any geometric sym-
metry also implies symmetry in the corresponding view factors. Fig. 11.5
shows two examples for N = 3. In the triangular duct F 12 = F13 . For
2 3
1
1
2
Figure 11.5. Radiation enclosures with N = 3 surfaces. Dot-

ted lines indicate planes of symmetry.
the cylinder, surface 1 is taken as the entire inside surface, so that again
F12 = F13 . Unlike Eqs. (11.5) and (11.6), such relations are problem specific
and will not necessarily exist for every configuration.
In retrospect, we see that N (N − 1)/2 is actually the upper bound for
the number of view factors that have to be calculated directly. Any extra
relations that we can deduce, e.g. by inspection of flat or convex surfaces
and by symmetry observations, decreases this number. Many problems are
such that the view factor integral never actually has to be evaluated.
11.4. EXCHANGE AMONG DIFFUSE GRAY SURFACES 141
11.3. Blackbody Radiation Exchange

In the general problem, radiation departs from a surface via both emis-
sion and reflection. At the target, it may be reflected and absorbed. How-
ever, for blackbodies the situation is greatly simplified since there is no
reflection. Energy only leaves a surface via emission and all incident radia-
tion is absorbed at the target surface. For radiation exchange between two
arbitrary blackbody surfaces, we can write
qi→j = Ji Ai Fij
directly from the view factor definition in Eq. (11.2). Now, invoking the
blackbody assumption, radiosity equals the emissive power for a blackbody,
so that
qi→j = Eb,i Ai Fij .
Similarly, we can write the radiation transfer in the opposite direction as
qj→i = Eb,j Aj Fji .
The net exchange between the two surfaces can be written simply as the
difference between these two quantities
qij = qi→j − qj→i
= Eb,i Ai Fij − Eb,j Aj Fji
= Eb,i Ai Fij − Eb,j Ai Fij (reciprocity)

= Ai Fij Eb,i − Eb,j
Ai Fij σ Ti4 − Tj4 .

(11.10) qij =
Eq. (11.10) is the net rate at which radiation leaves surface i (is lost from
surface i) as a result of interacting with surface j. Likewise, it is the net
rate at which surface j gains energy.
This concept can easily be extended to an enclosure of N blackbody
surfaces, all maintained at various temperatures. A similar treatment to the I&D Ex. 13.3
above yields pp 801
N
X N
X
Ti4 Tj4 Fij Ti4 − Tj4

(11.11) qi = Ai Fij σ − = σ Ai
j=1 j=1
as the net transfer of radiation from surface i as a result of interacting with

all of the other N − 1 surfaces of the enclosure.
11.4. Exchange Among Diffuse Gray Surfaces

Eq. (11.11) is interesting from a conceptual standpoint, but we have
already seen that real surfaces often reflect radiation, i.e. ρ > 0 and α < 1.
The reality of this situation is that “packets” of radiation will interact with
many surfaces consecutively, each which might absorb some portion and
reflect the remainder. Two surfaces will therefore exchange energy at many
“levels”, as the radiation bounces back and forth. This bouncing effect
complicates matters.
In this section, we will relax somewhat the assumption of a blackbody.
Specifically, we will model enclosures having the following properties for each
surface
• isothermal (uniform temperature)
• uniform radiosity and irradiation
• opaque (τ = 0), diffuse, and gray (ε = α)
and will then develop relations for the net rate at which radiation leaves a
surface i, defined as qi . In particular, we can write qi according to conser-
vation of energy at the radiative surface, either in terms of the difference
between radiosity and irradiation, or equivalently, between emission and
absorption (Fig. 11.6). We have
ρ *G*A G*A
J*A E*A
G*A
α*G*A
q q
Figure 11.6. Conservation of energy in two equivalent forms,

where q is the net loss from the surface.

(11.12) qi = A i Ji − G i

= A i Ei + ρ i Gi − G i

= A i Ei + G i ρi − 1

= A i Ei − G i 1 − ρ i

(11.13) qi = A i Ei − α i Gi .
Notice that the form containing radiosity implicitly includes absorption,

while the form based on emission implicitly accounts for reflection. Let us
focus on the radiosity Ji = Ei + ρi Gi . Using Ei = εi Ebi and ρi = 1 − αi =
1 − εi , we can write radiosity as Ji = εi Ebi + (1 − εi ) Gi . Solving for Gi , we
find
Ji − εi Ebi
Gi = ,
1 − εi
which, when substituted into qi above, yields

Ji − εi Ebi
qi = A i Ji −
1 − εi

1 − εi Ji − εi Ebi
= A i Ji −
1 − εi 1 − εi

Ji (1 − εi ) − Ji + εi Ebi
= Ai
1 − εi

εi Ebi − εi Ji
= Ai
1 − εi
Ai εi (Ebi − Ji )
=
1 − εi
Ebi − Ji
(11.14) qi =
(1 − εi ) / (Ai εi )
This expression is the net radiative energy from a surface. We readily notice
the analogy to circuits, i.e. Ebi − Ji represents driving potential, while q i
is heat flow (analogous to flow of electrical current) and (1 − ε i ) / (Ai εi )
is a radiative surface “resistance”. Fig. 11.7 illustrates this circuit analog.
Conceptually, we can think of Eq. (11.14) as quantifying the amount by
(1 − ε i ) / ( A i ε i )
Ei Ji
qi
Figure 11.7. Circuit analogy for radiative exchange at a sur-

face as characterized by Eq. (11.14).
which a real surface differs from the ideal blackbody radiator. In other
words, if we write Eq. (11.14) as
1 − εi
qi = Ebi − Ji ,
Ai εi
it is clear that εi → 1 implies Ji → Ebi . In other words, as the emissivity
goes to unity, the radiosity approaches the blackbody emissive power — the
surface is then equivalent to the ideal blackbody surface, for which ε i = αi =
1 (perfect absorptivity) and ρi = 0 (no reflection). For εi < 1, the surface is
not ideal, so that it is described by Eq. (11.14).
Unfortunately, the results expressed thus far are incomplete because
we do not know the value of the radiosity J i in Eq. (11.14). This must
be calculated based on interaction with the other N − 1 surfaces of the
problem. However, the radiosities for these surfaces then come into play as
well. Recalling the developments that led to the summation rule in Eq. (11.6)
were based on the conservation of energy of the radiation leaving surface i.
Let us now consider the conservation of energy for the irradiation of surface
i, that is, the total irradiation is the sum of all net radiation transfers to
surface i. We write this as
Ai Gi = q1→i + q2→i + · · · + qN →i
= A1 J1 F1i + A2 J2 F2i + · · · + AN JN FN i (apply Eq. (11.2))
= J1 A1 F1i + J2 A2 F2i + · · · + JN AN FN i
= J1 Ai Fi1 + J2 Ai Fi2 + · · · + JN Ai FiN (reciprocity)
= Ai (J1 Fi1 + J2 Fi2 + · · · + JN FiN ) .
Canceling Ai from both sides, we can write the irradiation as
N
X
Gi = Jj Fij .
j=1
In turn, we can substitute this into Eq. (11.12) for the net rate of radiation
leaving surface i, which yields
 
XN
qi = Ai Ji − Jj Fij 
j=1
 
N
X
= Ai Ji × 1 − Jj Fij  (multiply Ji by 1)
j=1
 
N
X N
X
= Ai Ji Fij − Jj Fij  (Eq. (11.6))
j=1 j=1
 
XN N
X
= Ai  Fij Ji − Jj Fij  (take Ji into Σ)
j=1 j=1
N
X
(11.15) qi = Ai Fij (Ji − Jj )
j=1
Eq. (11.15) represents the net radiation leaving surface i written in terms
of the interaction with radiosities of the other surfaces. We again notice a
clear analogy to linear circuits where J i − Jj is the potential, qi is flow,
and (Ai Fij )−1 is a resistance term11.4. Unlike the single resistor implied by
Eq. (11.14), this configuration represents a resistor for each surface interact-
ing with the given surface i (Fig. 11.8). Surface i thus comprises the node
for this circuit. Furthermore, it should be clear that in a given problem of N
11.4This quantity is usually referred to as the spatial resistance, geometric resistance,
or shape resistance.
11.5. THE GRAYBODY MATRIX PROBLEM 145
q
i −> 1
J1
i 1 ) −1
q
i −> 2
J2
i F
−1
(A
)
Fi2
q (A i J
3
i q
i −> 3
Ji
(A i Fi 3 ) −1
(A i
FiN
)
−1
JN
q
i −> N
Figure 11.8. Circuit analogy for radiative exchange at a sur-

face as characterized by Eq. (11.15).
surfaces, there will be N such circuits of the type shown in Fig. 11.8. That
is, there will be such a circuit for each node, where the nodes go from 1 to
N , each of which represents one surface of the problem.
11.5. The Graybody Matrix Problem

You have probably already deduced from the above discussion that the
problem must ultimately be cast as a matrix equation. Like Eq. (11.14),
there is insufficient information in Eq. (11.15) to solve the typical problem,
because we generally do not know the values for q i . However, we can readily
set Eq. (11.14) equal to Eq. (11.15) to obtain
N
Ebi − Ji X
= Ai Fij (Ji − Jj ) .
(1 − εi ) / (Ai εi )
j=1
Or, canceling Ai , we can write more simply

N
Ebi − Ji X
(11.16) = Fij (Ji − Jj ) .
(1 − εi ) /εi
j=1
Let us examine this equation closely. Eq. (11.16) is written for a specific
surface i, so in an actual problem, there would be N such equations for
i : 1 → N . This equation is simply the conservation of energy for each node
(surface) in the equation. In fact, it is really nothing more than Kirchhoff’s
Circuit Law , i.e. the sum of the currents (radiative transfers) for a node
must be zero. This same concept appears in Fig. 11.8.
11.5. THE GRAYBODY MATRIX PROBLEM 146
You have probably further deduced that the circuit analogy for this
problem becomes too unwieldy for large N . For example, the circuit rep-
resentation for N = 3 is shown in Fig. 11.9, but for higher values of N ,
such diagrams quickly become too difficult to reasonably draw. The matrix
Eb3
R
e3
J
R 3
s13
R
e1
R
s23
E J
b1 1
R s12
J2 R e2
Eb2
Figure 11.9. Circuit analogy for radiative exchange among 3

surfaces showing emissive powers, radiosity nodes, and all resis-
tances. Resistors with subscript “e” are emissive resistors, while
those with subscript “s” are shape resistors.
representation is rather more general. For the same N = 3 problem, we

would write a form of Eq. (11.16) for each surface as
ε1
Eb1 − J1 = F11 J1 − J1 + F12 J1 − J2 + F13 J1 − J3
1 − ε1
ε2
Eb2 − J2 = F21 J2 − J1 + F22 J2 − J2 + F23 J2 − J3
1 − ε2
ε3
Eb3 − J3 = F31 J3 − J1 + F32 J3 − J2 + F33 J3 − J3
1 − ε3
where the unknowns are J1 , J2 , and J3 . The emissive powers are considered
known because they can be calculated from the Stefan–Boltzmann Law in
Eq. (10.14) on pp. 124. Emissivities are also considered to be known, as are
the view factors. These equations can be written in matrix form as
    
F12 + F13 + ε̄1 − F12 − F13 J1 ε̄1 Eb1
 − F21 F21 + ε̄2 + F23 − F23   J2  =  ε̄2 Eb2 
− F31 − F32 F31 + F32 + ε̄3 J3 ε̄3 Eb3
where ε̄1 = ε1 / (1 − ε1 ), ε̄2 = ε2 / (1 − ε2 ), and ε̄3 = ε3 / (1 − ε3 ) are short-
hand notations. Of course, this particular system can be solved using any
suitable matrix iteration or inversion procedure (Kreyszig, 1988). Configu- I&D Ex. 13.4
rations having N > 3 lead to commensurately larger matrix problems. pp 808
One configuration where the circuit analogy is preferred is the case of
N = 2. This problem reduces to the simple series of resistors (Fig. 11.10).
11.6. ADDITIONAL CONFIGURATIONS 147
Here, we know the potential at the two termini from the Stefan–Boltzmann
(1 − ε 1 ) / (Α 1ε 1) (A 1 F1 2 ) −1 (1 − ε 2 ) / (Α 2ε 2)
q J J −q
1 E 1 2 E b2 2
b1
q
12
Figure 11.10. Series circuit for the N = 2 problem.
Law, and can thus write

σ T14 − T24

(11.17) q12 = ·
(1 − ε1 ) / (ε1 A1 ) + 1/ (A1 F12 ) + (1 − ε2 ) / (ε2 A2 )
Moreover, because there are only 2 surfaces, the net rate of transfer from
surface 1 to surface 2, q12 , must be equal to the net rate of transfer from
surface 1, q1 , and thus the net rate of transfer to surface 2, −q 2 .
11.6. Additional configurations

There are several interesting configurations related to radiation exchange.
For example, if we examine the form of the emissive resistance in Eq. (11.14)
1 − ε
, R =
Aε
it is clear that R will be large if the emissivity is small. This provides
a basis for using low emissivity materials as radiation shields to decrease
the heat transfer between two bodies 11.5. That is, a low–ε surface can be
placed between two bodies between which heat transfer is to be decreased.
This would result in additional high–resistance elements in Fig. 11.10 with a
commensurate decrease in q12 . For gray surfaces we have α + ρ = ε + ρ = 1,
so that low–ε implies that the surface is highly reflective of radiation.
Another interesting configuration is the re–radiating surface, which is
characterized by zero net radiation transfer, q i = 0. This is a reason-
able model when the surface is well–insulated on the opposite side. From
Eq. (11.14), we see that qi = 0 implies Ebi = Ji , so that the the radiosity at
node i will be known. How does this effectively change the problem?
Let us carefully examine the example of a 3 surface enclosure, N = 3,
where one of the surfaces is a re–radiating surface (Fig. 11.11). It is clear
that q13 = q23 , because there is no radiative transfer through resistor R e3 . In
other words, q3 = 0 because this is a re–radiating surface. Also, J 3 = Eb3 is
known. The circuit is thus reduced to a simpler parallel arrangement where
one of the legs is Rs12 and the other is Rs13 +Rs23 . The circuit can be solved I&D Ex. 13.6
pp 815
11.5For example, so–called low–ε glass is used in this capacity.
11.6. ADDITIONAL CONFIGURATIONS 148
R
e3
J = E b3
R 3
s13
R
e1
q
13 R
s23
E J
b1 1
q
23
R s12
J2 R e2
Eb2
Figure 11.11. Circuit for N = 3 enclosure with surface 3

acting as a re–radiating surface.
by formulating the overall resistance between E b1 and Eb2 . Using standard

circuit analysis, the total resistance R t is found to be
1
Rt = Re1 + 1 1 + Re2
Rs13 + Rs23 + Rs12
Rs12 (Rs13 + Rs23 )
= Re1 + + Re2 ,
Rs12 + Rs13 + Rs23
from which we can write the solution in the form
Eb1 − Eb2
q1 = − q 2 = .
Rt
Notice that the problem is independent of resistor R e3 , whose form is
1 − ε3
Re3 = ·
A3 ε3
In turn, this means that the problem is independent of the emissivity of
surface 3. This observation is true in general, i.e. the emissivity of a re–
radiating surface is irrelevant. Once q 1 is known, standard circuit theory can
again be applied to find the potentials J 1 and J2 . In a more general problem,
i.e. N > 3, we can apply Eq. (11.15) in the standard fashion to formulate
the matrix problem, except we simply set q i = 0 for any re–radiating surface
i.
APPENDIX A
Transient Conduction Example: The

Homogeneous Cooling Problem
The concept of transient conduction was examined in Chapter 4. While

certain idealizations were discussed, e.g. the lumped capacitance method,
there were appreciable limitations upon their applicability. Here, we intro-
duce a more generalized one–dimensional problem where spatial gradients do
not vanish and the Fourier number is such that the one term approximation
is not necessarily valid. We will focus on the case where boundary conditions
are of the first kind (Dirichlet boundary conditions). There are a number of
ways to approach this problem, however, we will use the well–known method
of Separation of Variables.
Eq. (4.1) describes the physics of this problem
∂T ∂2T
= α ,
∂t ∂x2
where T is only a function of x and t. Recall that this equation is a lin-
ear partial differential equation, so we suspect that we can treat specific
problems strictly with theoretical methods A.1. We take the problem domain
to be a finite region in Cartesian space 0 ≤ x ≤ L. Moreover, we assume
a temporal duration of t ≥ 0. To complete the description of a problem,
that is to make sure it is well–posed, we will require both boundary condi-
tions and initial conditions. We focus on homogeneous boundary conditions
of the first kind, i.e. T (0, t) = T (L, t) = 0. It is assumed that specific
non–homogeneous cases can be cast in homogeneous form using a suitable
transformation. Initial conditions are given by a function T (x, 0) = F (x).
A.1. Separation of Variables Method

The method begins with a conjecture that the solution can be written
in the form
(A.1) T (x, t) = Ψ(x) Γ(t) .
That is, we assume that the physical problem is such that the contribution
related to temporal response can be separated from the contribution related
to spatial variation (Carrier and Pearson, 1976). For example, consider the
imaginary case T = (x2 −1)t. This equation clearly represents a “separable”
√
problem, where Ψ = (x2 − 1) and Γ = t. Conversely, the case T = xt +
A.1As opposed to non–linear problems which often cannot be solved analytically.
149
A.1. SEPARATION OF VARIABLES METHOD 150
√
tanh(x2 t + x/t) is not readily separable. We therefore suspect that
p
even this method may be limited in the types of problems that may be
solved. Fortunately, it is known to work for the type of one–dimensional
configurations we are interested in here.
Under the conjecture of Eq. (A.1), partial derivatives have certain forms.
Using the Chain Rule of Calculus, we see
∂T ∂Ψ(x) ∂Γ(t)
(A.2) = Γ(t) + Ψ(x) = Ψ(x) Γ 0 (t) ,
∂t ∂t ∂t
where the prime symbol denotes the derivative of a univariate function.
Notice that the derivative of Ψ(x) with respect to t vanishes because Ψ is
only a function of x, not of t. Similarly, we can apply Chain Rule twice to
find
∂2T
(A.3) = Ψ 00 (x) Γ(t) .
∂x2
Proceeding, we now substitute Eqs. (A.2) and (A.3) into the conduction
equation to obtain
1 Γ 0 (t) Ψ 00 (x)
(A.4) = .
α Γ(t) Ψ(x)
We have now cast the problem in a separated form where the left hand side
is only a function of time t and the right hand side is only a function of
the spatial coordinate x. However, according to principle, Eq. (A.4) must
be valid for all x and t in the problem domain. It follows that each side
must be equal to a constant. Otherwise, either of x or t could be held fixed
while the other could be varied such that Eq. (A.4) would be contradicted.
Therefore,
1 Γ 0 (t)
(A.5) = C0
α Γ(t)
and
Ψ 00 (x)
(A.6) = C0 ,
Ψ(x)
where C0 is a constant. Positive values for C 0 lead to√exponentially increas-
ing behavior and imaginary values (involving i = −1 ) lead to periodic
behavior (Carrier and Pearson, 1976). These responses can be verified by
substitution. We are instead interested in the case where C 0 is negative.
This leads to behavior that decays exponentially in time, a phenomenon
compatible with the type of boundary and initial conditions we will employ.
Therefore, we define
(A.7) C0 = −λ2 ,
where λ > 0.
A.2. SOLUTION PROCEDURE 151
We have now developed two individual ordinary differential equations

from the single partial differential equation
(A.8) Γ 0 (t) + αλ2 Γ(t) = 0 t≥0
and
(A.9) Ψ 00 (x) + λ2 Ψ(x) = 0 0≤x≤L.
These taken separately are each easier to solve than the original partial
differential equation.
We must apply the same separation process to the boundary conditions.
Using Eq. (A.1), the two boundary conditions can be written as
Ψ(0) Γ(t) = Ψ(L) Γ(t) = 0 .
Now, Γ(t) cannot vanish for arbitrary values of t, otherwise the whole so-
lution would be trivial. Therefore we see that Ψ(0) and Ψ(L) must vanish
instead. The boundary conditions are therefore
(A.10) Ψ(0) = 0
and
(A.11) Ψ(L) = 0 .
We do not perform a similar reduction on the initial condition. This is
handled differently as discussed below. The problem in (x, t) has now been
completely separated into two simpler problems: a spatial problem in x and
a temporal problem in t.
A.2. Solution Procedure

We are now faced with solving the individual problems in x and t. As
mentioned above, we are focusing specifically on homogeneous boundary
conditions of the first kind. Cases involving other combinations of boundary
conditions are more complex and require a generalized treatment ( Özişik,
1980).
The first step is to realize that Eq. (A.8) has a solution of the form
2
(A.12) Γ(t) = e−αλ t .
This can be readily verified by substitution. Next, we see that the trigono-
metric functions can satisfy Eq. (A.9) as
(A.13) Ψ(x) = C1 sin λx + C2 cos λx ,
where C1 and C2 are constants of integration. If we apply Eq. (A.10), we
get
Ψ(0) = C1 sin 0 + C2 cos 0 ,
which implies that C2 = 0. Now apply Eq. (A.11) to obtain
Ψ(L) = C1 sin λL = 0 .
A.3. DETERMINING MODE COEFFICIENTS 152
It is clear that C1 6= 0, otherwise the entire solution would once again be

trivial. Therefore, it must instead be the case that sin λL = 0, so that
(A.14) λL = nπ ,
where n = 1, 2, 3, . . . That is, the problem is only satisfied for certain val-
ues of λ, called eigenvalues, i.e. λ = nπ/L. Thus, this is an eigenvalue
problemA.2.
Note, we will now change the nomenclature to reflect the fact that there
are many admissible eigen–related values, thus λ → λ n , C1 → Cn , and λn =
nπ/L. Also note that in this case “sin λ n x” is called the eigenfunction. Each
mode n yields an elementary solution to the problem. A general solution
is therefore obtained by the linear superposition of all modes n = 1 → ∞.
This can be written as
∞
2
X
(A.15) T (x, t) = Cn sin (λn x) e−αλn t .
n=1
Eq. (A.15) now satisfies both the governing equation and the boundary
conditions. However, we have yet to determine mode coefficients C n such
that the initial conditions are satisfied.
A.3. Determining Mode Coefficients

Directly plugging in initial conditions at t = 0 yields
X∞
(A.16) T (x, 0) = F (x) = Cn sin λn x ,
n=1
however, this appears to be a dead end at first glance since there are an
infinite number of unknown Cn values and only one equation. In actuality,
we see that this problem boils down to one of expanding an arbitrary func-
tion F (x) into a series. In this case, the series will be a Fourier “sin” series
(Boyce and DiPrima, 1977). Let us review how such a series is constructed.
First, we make use of the fact that the eigenfunctions are orthogonal,
that is they obey
Z L
sin λn x sin λm x dx = 0
0
for m 6= n and
Z L
sin λn x sin λm x dx = N (λn )
0
for m = n, where N (λn ) is the norm and is defined simply as
Z L
N (λn ) = sin2 λn x dx .
0
Özişik (1980) has an extensive discussion of orthogonality.

A.2Özişik (1980) discusses eigenvalue problems in detail.
A.3. DETERMINING MODE COEFFICIENTS 153
RL
Next, we operate on Eq. (A.16) with 0 sin λn x dx to obtain
Z L Z LX∞
(A.17) F (x) sin λn x dx = Cm sin λm x sin λn x dx .
0 0 m=1
Note, the mode number symbol is arbitrary and we changed n → m in the
summation to avoid ambiguity. We recall that the mode coefficients are
constants rather than functions of x, so Eq. (A.17) can be simplified to
Z L X∞ Z L
(A.18) F (x) sin λn x dx = Cm sin λm x sin λn x dx .
0 m=1 0
The summation sign and coefficients have been taken outside of the integral.
Now we make the observation that the right hand side vanishes except in
the case where m = n due to the orthogonality property. This can be better
visualized if we write out the terms explicitly. We obtain
Z L
C1 sin λ1 x sin λn x dx +
0
Z L
C2 sin λ2 x sin λn x dx +
0
Z L
C3 sin λ3 x sin λn x dx + · · · +
0
Z L
Cn sin λn x sin λn x dx +
0
Z L
Cn+1 sin λn+1 x sin λn x dx +
0
Z L
Cn+2 sin λn+2 x sin λn x dx + · · ·
0
as the explicit representation of the series. Clearly, only mode C n is non–
zero, as governed by orthogonality. The rest of the modes are trivial and
the summation itself vanishes. We can then simplify Eq. (A.18) to
Z L Z L
(A.19) F (x) sin λn x dx = Cn sin λn sin λn x dx ,
0 0
which can be written using the norm as
Z L
(A.20) F (x) sin λn x dx = Cn N (λn ) .
0
Solving Eq. (A.20), we find that the mode coefficients are given by
Z L
1
(A.21) Cn = F (x) sin λn x dx .
N (λn ) 0
It is left as an exercise to the reader to show that N (λ n ) for this case is
simply L/2.
A.4. EXAMPLE: THE UNIT INITIAL CONDITION 154
Eq. (A.15) satisfies the governing equation and the boundary conditions
while Eq. (A.21) determines coefficients such that the initial condition F (x)
is satisfied. If F (x) is very complicated, evaluation of Eq. (A.21) may not
be straightforward. Otherwise, it can usually be handled with the aid of
standard integral tables. These two equations give the final solution to the
problem specified above. Physically, this means that we know the temper-
ature distribution T (x, t) for the entire problem domain 0 ≤ x ≤ L and
t ≥ 0. We can therefore compute quantities of engineering interest such as
heat flux at any position and any time.
A.4. Example: The Unit Initial Condition

As an example of finishing the problem to obtain the exact solution, let
us consider the unit initial condition F (x) = 1, i.e. the temperature at any
x at time t = 0 is unity. We must now evaluate the mode coefficients using
Eq. (A.21) asA.3
Z L
1
Cn = F (x) sin λn x dx
N (λn ) 0
Z L Z L
2 2
= 1 sin λn x dx = sin λn x dx
L 0 L 0
2 L 2L nπx L
= − cos λn x = − cos

L λn Lnπ L 0

0

2 nπL
= − cos − cos 0
nπ L
2
= [1 − cos n π]
nπ
2
= [1 − (−1)n ]
nπ
The last result is obtained by noting that n varies as 1, 2, 3, . . . , for which the
cosine of nπ must correspondingly alternate between negative and positive
1. In fact, we can construct a table of how the term in brackets varies with
n (Table A.1). Clearly, all even–numbered modes drop out of the problem
Table A.1. Values of the alternating term: [1 − (−1) n ]
n 1 2 3 4 5 6 7 ···
value 2 0 2 0 2 0 2 · · ·
because their resulting mode coefficients are all 0. However, odd–numbered

modes remain. We can simplify Cn further by writing it only for the non–
trivial odd modes as
2 4
Cn = 2 = where n = 1, 3, 5, 7, . . .
nπ nπ
A.3Recall that you’ve already showed that N (λ ) = L/2.
n
A.4. EXAMPLE: THE UNIT INITIAL CONDITION 155
Finally, we can substitute Cn into the general solution in Eq. (A.15) to

obtain
∞
X 4 2
T (x, t) = sin (λn x) e−αλn t .
nπ
n=1,3,5,...
This expression can be used to evaluate temperature at any arbitrary (x, t),
e.g. by
• choose desired (x, t)
• set T (x, t) = 0
• set n = 1
• evaluate current mode n and add to running total of T (x, t)
• compare old value of T (x, t) to newly calculated value
• stop if converged, i.e. if change of T (x, t) is below user–specified
convergence criterion, otherwise go to next mode
APPENDIX B
Laminar Forced Convection in a Pipe
This appendix shows the long–hand derivation of a number of equations

relevant to laminar forced convection in fully developed pipe flow.
B.1. Volume Flow Rate

Eq. (7.4) gives the result for the volume flow rate in laminar pipe flow.
Here, we formally show this starting from the definition of volume flow rate
as the integral over the flow cross section
Z Z r0
Q = V · n̂ dA = u 2πr dr
A 0
and the exact solution given by Eq. (7.3). Substituting this expression we
find
Z r0 " 2 #
1 dP 2 r
Q = − r0 1 − 2πr dr
0 4 µ dx r0
Z r0 " 2 #
π r02 dP

r
= − 1− r dr
2µ dx 0 r 0
Z r0
π r02 dP r3

= − r − 2 dr
2µ dx 0 r0
2 2 4 r0
π r0 dP r r
= − − 2

2µ dx 2 4 r 0 0
π r02 dP r02

= −
2µ dx 4
4

π r0 dP
= − ,
8µ dx
which confirms Eq. (7.4).
B.2. Integration of Mean Temperature Equation

Eq. (7.23) gives the result for the expression A in the case of constant
wall heat flux. To show this, we must integrate Eq. (7.22), which has the
form Z r0
2
u θ r dr = 1 .
ū r02 0
156
B.2. INTEGRATION OF MEAN TEMPERATURE EQUATION 157
We substitute for u and θ to find

" 2 # " 4 2 #
2 A r02 r0 r 3 1 r 1 r
Z
− 2 2 ū 1 − + − r dr = 1 .
ū r0 0 r0 16 16 r0 4 r0
Writing out terms in long hand, we find
Z r0
7 r3 5 r5 1 r7

3
−4A r − + − dr = 1 .
0 16 16 r02 16 r04 16 r06
Carrying out term–by–term integration yields
r
7 r4 5 r6 1 r 8 0

3 2
−4A r − + − = 1
2 · 16 4 · 16 r02 6 · 16 r04 8 · 16 r06 0

36 − 42 + 20 − 3
−4A r02 =
24 · 16

11
−4A r02 =
24 · 16

11
−A r02 =
6 · 16
11
− A r02 = 1
96
Solving, we find
96
A r02 = − ,
11
which confirms Eq. (7.23).
APPENDIX C
Blackbody Radiation
C.1. Wien’s Displacement Law

Eq. (10.12) reports Wien’s Displacement Law as
λmax T = 2897.8 µm · K .
Here, we prove this assertion starting from the spectral blackbody emissive
power in Eq. (10.11)
C1
Eλ,b (λ, T ) = .
λ5 e C 2 /(λ T) − 1
We wish to determine the conditions for λ where E λ,b (λ, T ) is a maximum.
Differentiating with respect to λ and setting the result equal to zero, we find
dEλ,b (λ, T ) −1
= − 5 C1 λ−6 eC2 /(λ T ) − 1 −
dλ
−5

C2 /(λ T )
−2 C2 −2 C2 /(λ T )
C1 λ e − 1 − λ e
T
5 C1
= − · 5 C /(λ T ) +
λ λ e 2 − 1

C1 1 C2 C2 /(λ T )
· · e
λ5 eC2 /(λ T ) − 1 eC2 /(λ T ) − 1 T λ2
5 C2 eC2 /(λ T )
= − · Eλ,b + Eλ,b · 2
· C /(λ T )
λ Tλ e 2 − 1
= 0,
which can be recast as
5 C2 eC2 /(λ T )
− · Eλ,b + Eλ,b · · = 0.
λ T λ2 eC2 /(λ T ) − 1
Factoring out Eλ,b /λ, we can now simplify this expression as
C2 eC2 /(λ T )
· C /(λ T ) = 5.
Tλ e 2 − 1
Notice that the term C2 /(λ T ) appears as a unified variable. That is, we
can let x = C2 /(λ T ), and solve the equation
ex
x = 5.
ex − 1
158
C.2. RADIATION FUNCTIONS 159
With a little algebra we find

ex − 1 −x

x = 5 = 5 1 − e ,
ex
which is readily solved with a numerical procedure. For example, a few
iterations of the Bisection Method (Kreyszig, 1988) shows
C2
x = ≈ 4.965782 .
λT
Substituting C2 = 14, 390 µm · K (Incropera and Dewitt, 2002) and solving,
we find
C2 14390
λT ≈ = = 2897.833 ,
4.965782 4.965782
which confirms Wien’s Law in Eq. (10.12).
C.2. Radiation Functions

In §10.2, various blackbody radiation properties were discussed. Here,
we give a summary of these properties in terms of the combined variable λT
in units of µm · K. Units of the column containing the radiation intensity
are (µm · K · sr)−1 . This form, i.e. where temperature appears to the minus
fifth power, is convenient for tabulation owing to the fact the wavelength
appears as the negative fifth power in Planck’s Distribution, e.g. Eqs. (10.10)
and (10.11). See Dunkle (1954) and Siegel and Howell (2001) for more
comprehensive data.
References show some variations due to differences in which integration
is performed. For example, Incropera and Dewitt (2002, Table 12.1, pp. 716)
shows f0→λ = 0.701046 for λT = 5, 600, while Arpaci et al. (2000, Table 8.1,
pp. 411) indicates this value is f0→λ = 0.70102.
Table C.1. Blackbody radiation functions
Iλ,b / σT 5 Iλ,b / σT 5 f0→λ

λT f0→λ λT
200 3.7668e-28 0.0 5600 3.1624e-05 0.70102
400 4.9126e-14 0.0 5700 3.0396e-05 0.71077
500 2.1422e-11 0.0 5800 2.9217e-05 0.72013
600 1.0416e-09 0.0 5900 2.8086e-05 0.72914
700 1.4812e-08 0.0 6000 2.7000e-05 0.73779
800 9.9175e-08 0.000016 6100 2.5959e-05 0.74611
900 4.0592e-07 0.000087 6200 2.4961e-05 0.75411
1000 1.1855e-06 0.000321 6300 2.4005e-05 0.76181
continued on next page . . .

C.2. RADIATION FUNCTIONS 160
Iλ,b / σT 5 Iλ,b / σT 5

λT f0→λ λT f0→λ
1100 2.7224e-06 0.000911 6400 2.3088e-05 0.76920
1200 5.2404e-06 0.00213 6500 2.2210e-05 0.77632
1300 8.8324e-06 0.00432 6600 2.1369e-05 0.78317
1400 1.3442e-05 0.00779 6700 2.0563e-05 0.78976
1500 1.8889e-05 0.01285 6800 1.9792e-05 0.79610
1600 2.4913e-05 0.01972 6900 1.9053e-05 0.80230
1700 3.1227e-05 0.02853 7000 1.8345e-05 0.80808
1800 3.7554e-05 0.03934 7100 1.7667e-05 0.81373
1900 4.3654e-05 0.05211 7200 1.7018e-05 0.81918
2000 4.9337e-05 0.06673 7300 1.6396e-05 0.82443
2100 5.4466e-05 0.08305 7400 1.5800e-05 0.82949
2200 5.8955e-05 0.10089 7500 1.5229e-05 0.83437
2300 6.2760e-05 0.12003 7600 1.4682e-05 0.83910
2400 6.5873e-05 0.14026 7800 1.3656e-05 0.84801
2500 6.8312e-05 0.16136 8000 1.2713e-05 0.85625
2600 7.0113e-05 0.18312 8500 1.0672e-05 0.87457
2700 7.1325e-05 0.20536 9000 9.0103e-06 0.88999
2800 7.2005e-05 0.22789 9500 7.6496e-06 0.90304
2898 7.2212e-05 0.25011 10000 6.5296e-06 0.91416
2900 7.2212e-05 0.25056 10500 5.6024e-06 0.92367
3000 7.2005e-05 0.27323 11000 4.8308e-06 0.93185
3100 7.1441e-05 0.29578 11500 4.1852e-06 0.93892
3200 7.0576e-05 0.31810 12000 3.6421e-06 0.94505
3300 6.9458e-05 0.34011 12500 3.1831e-06 0.95401
3400 6.8132e-05 0.36173 13000 2.7932e-06 0.95509
3500 6.6640e-05 0.38291 13500 2.4605e-06 0.95921
3600 6.5017e-05 0.40360 14000 2.1753e-06 0.96285
3700 6.3295e-05 0.42376 14500 1.9299e-06 0.96607
3800 6.1501e-05 0.44337 15000 1.7178e-06 0.96893
3900 5.9657e-05 0.46241 15500 1.5338e-06 0.97149
4000 5.7785e-05 0.48087 16000 1.3736e-06 0.97377
4100 5.5901e-05 0.49873 16500 1.2336e-06 0.97581
4200 5.4019e-05 0.51600 17000 1.1109e-06 0.97765
4300 5.2151e-05 0.53268 18000 9.0777e-07 0.98081
4400 5.0306e-05 0.54878 19000 7.4879e-07 0.98341
4500 4.8492e-05 0.56430 20000 6.2298e-07 0.98555
4600 4.6716e-05 0.57926 25000 2.7633e-07 0.99217
4700 4.4982e-05 0.59367 30000 1.4039e-07 0.99529
4800 4.3293e-05 0.60754 35000 7.8621e-08 0.99695
4900 4.1654e-05 0.62089 40000 4.7364e-08 0.99792
5000 4.0065e-05 0.63373 45000 3.0200e-08 0.99852
5100 3.8528e-05 0.64608 50000 2.0150e-08 0.99890
5200 3.7043e-05 0.65795 55000 1.3952e-08 0.99917
5300 3.5610e-05 0.66936 75000 4.1837e-09 0.99971
5400 3.4230e-05 0.68034 100000 1.3568e-09 0.99991
5500 3.2902e-05 0.69088 ∞ 0 1.0
APPENDIX D
Document History
Version Year Remarks

1.0 1999 initial plain TEX implementation
2.0 2003 LATEX re–implementation using “amsbook”
class; implementation of figures, tables,
equation numbering, references, and index
2.1 2005 some minor corrections and additions; added
page references where appropriate for better
navigation
161
About the Author
Michael C. Wendl holds Bachelor’s and Master’s Degrees in Mechani-

cal Engineering and a Doctorate in Applied Science and Engineering from
Washington University. He is on the Faculties of Medicine (Genetics) and
Engineering (Mechanical and Aerospace Engineering) at Washington Uni-
versity. Dr. Wendl’s research activity primarily focuses on analytical work,
especially in pure and applied mathematics, incompressible flow, conduc-
tion heat transfer, random DNA processing, and other areas of theoretical
biology. His main teaching goal is to prepare and train students to think
critically not only about science/engineering, but also broader issues that
arise in life, e.g. decision–making, evaluating claims objectively, and logical
problem solving.
162
Creative Commons Public License
Attribution–NoDerivs–NonCommercial 1.0
THE WORK (AS DEFINED BELOW) IS PROVIDED UNDER THE TERMS OF THIS CRE-
ATIVE COMMONS PUBLIC LICENSE (“CCPL” OR “LICENSE”). THE WORK IS PROTECTED
BY COPYRIGHT AND/OR OTHER APPLICABLE LAW. ANY USE OF THE WORK OTHER THAN
AS AUTHORIZED UNDER THIS LICENSE IS PROHIBITED.
BY EXERCISING ANY RIGHTS TO THE WORK PROVIDED HERE, YOU ACCEPT AND
AGREE TO BE BOUND BY THE TERMS OF THIS LICENSE. THE LICENSOR GRANTS YOU
THE RIGHTS CONTAINED HERE IN CONSIDERATION OF YOUR ACCEPTANCE OF SUCH
TERMS AND CONDITIONS.
1. Definitions.
a. “Collective Work” means a work, such as a periodical issue, anthology or encyclopedia,
in which the Work in its entirety in unmodified form, along with a number of other con-
tributions, constituting separate and independent works in themselves, are assembled into
a collective whole. A work that constitutes a Collective Work will not be considered a
Derivative Work (as defined below) for the purposes of this License.
b. “Derivative Work” means a work based upon the Work or upon the Work and other pre-
existing works, such as a translation, musical arrangement, dramatization, fictionalization,
motion picture version, sound recording, art reproduction, abridgment, condensation, or
any other form in which the Work may be recast, transformed, or adapted, except that a
work that constitutes a Collective Work will not be considered a Derivative Work for the
purpose of this License.
c. “Licensor” means the individual or entity that offers the Work under the terms of this
License.
d. “Original Author” means the individual or entity who created the Work.
e. “Work” means the copyrightable work of authorship offered under the terms of this License.
f. “You” means an individual or entity exercising rights under this License who has not pre-
viously violated the terms of this License with respect to the Work, or who has received
express permission from the Licensor to exercise rights under this License despite a previous
violation.
2. Fair Use Rights. Nothing in this license is intended to reduce, limit, or restrict any rights
arising from fair use, first sale or other limitations on the exclusive rights of the copyright owner under
copyright law or other applicable laws.
3. License Grant. Subject to the terms and conditions of this License, Licensor hereby grants
You a worldwide, royalty-free, non-exclusive, perpetual (for the duration of the applicable copyright)
license to exercise the rights in the Work as stated below:
a. to reproduce the Work, to incorporate the Work into one or more Collective Works, and to
reproduce the Work as incorporated in the Collective Works;
b. to distribute copies or phonorecords of, display publicly, perform publicly, and perform
publicly by means of a digital audio transmission the Work including as incorporated in
Collective Works;
The above rights may be exercised in all media and formats whether now known or hereafter devised.
The above rights include the right to make such modifications as are technically necessary to exercise
the rights in other media and formats. All rights not expressly granted by Licensor are hereby reserved.
4. Restrictions. The license granted in Section 3 above is expressly made subject to and limited
by the following restrictions:
a. You may distribute, publicly display, publicly perform, or publicly digitally perform the
Work only under the terms of this License, and You must include a copy of, or the Uni-
form Resource Identifier for, this License with every copy or phonorecord of the Work You
distribute, publicly display, publicly perform, or publicly digitally perform. You may not
offer or impose any terms on the Work that alter or restrict the terms of this License or the
recipients’ exercise of the rights granted hereunder. You may not sublicense the Work. You
must keep intact all notices that refer to this License and to the disclaimer of warranties.
163
CREATIVE COMMONS PUBLIC LICENSE 164
You may not distribute, publicly display, publicly perform, or publicly digitally perform the
Work with any technological measures that control access or use of the Work in a manner
inconsistent with the terms of this License Agreement. The above applies to the Work as
incorporated in a Collective Work, but this does not require the Collective Work apart from
the Work itself to be made subject to the terms of this License. If You create a Collective
Work, upon notice from any Licensor You must, to the extent practicable, remove from the
Collective Work any reference to such Licensor or the Original Author, as requested.
b. You may not exercise any of the rights granted to You in Section 3 above in any manner
that is primarily intended for or directed toward commercial advantage or private monetary
compensation. The exchange of the Work for other copyrighted works by means of digital
file-sharing or otherwise shall not be considered to be intended for or directed toward com-
mercial advantage or private monetary compensation, provided there is no payment of any
monetary compensation in connection with the exchange of copyrighted works.
c. If you distribute, publicly display, publicly perform, or publicly digitally perform the Work
or any Collective Works, You must keep intact all copyright notices for the Work and give the
Original Author credit reasonable to the medium or means You are utilizing by conveying
the name (or pseudonym if applicable) of the Original Author if supplied; the title of the
Work if supplied. Such credit may be implemented in any reasonable manner; provided,
however, that in the case of a Collective Work, at a minimum such credit will appear where
any other comparable authorship credit appears and in a manner at least as prominent as
such other comparable authorship credit.
5. Representations, Warranties and Disclaimer.

a. By offering the Work for public release under this License, Licensor represents and warrants
that, to the best of Licensor’s knowledge after reasonable inquiry:
i. Licensor has secured all rights in the Work necessary to grant the license rights
hereunder and to permit the lawful exercise of the rights granted hereunder without
You having any obligation to pay any royalties, compulsory license fees, residuals or
any other payments;
ii. The Work does not infringe the copyright, trademark, publicity rights, common law
rights or any other right of any third party or constitute defamation, invasion of
privacy or other tortious injury to any third party.
b. EXCEPT AS EXPRESSLY STATED IN THIS LICENSE OR OTHERWISE AGREED IN
WRITING OR REQUIRED BY APPLICABLE LAW, THE WORK IS LICENSED ON
AN “AS IS” BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR
IMPLIED INCLUDING, WITHOUT LIMITATION, ANY WARRANTIES REGARDING
THE CONTENTS OR ACCURACY OF THE WORK.
6. Limitation on Liability. EXCEPT TO THE EXTENT REQUIRED BY APPLICABLE LAW,

AND EXCEPT FOR DAMAGES ARISING FROM LIABILITY TO A THIRD PARTY RESULTING
FROM BREACH OF THE WARRANTIES IN SECTION 5, IN NO EVENT WILL LICENSOR BE
LIABLE TO YOU ON ANY LEGAL THEORY FOR ANY SPECIAL, INCIDENTAL, CONSEQUEN-
TIAL, PUNITIVE OR EXEMPLARY DAMAGES ARISING OUT OF THIS LICENSE OR THE USE
OF THE WORK, EVEN IF LICENSOR HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH
DAMAGES.
7. Termination.
a. This License and the rights granted hereunder will terminate automatically upon any breach
by You of the terms of this License. Individuals or entities who have received Collective
Works from You under this License, however, will not have their licenses terminated provided
such individuals or entities remain in full compliance with those licenses. Sections 1, 2, 5,
6, 7, and 8 will survive any termination of this License.
b. Subject to the above terms and conditions, the license granted here is perpetual (for the
duration of the applicable copyright in the Work). Notwithstanding the above, Licensor
reserves the right to release the Work under different license terms or to stop distributing
the Work at any time; provided, however that any such election will not serve to withdraw
this License (or any other license that has been, or is required to be, granted under the terms
of this License), and this License will continue in full force and effect unless terminated as
stated above.
8. Miscellaneous.
a. Each time You distribute or publicly digitally perform the Work or a Collective Work, the
Licensor offers to the recipient a license to the Work on the same terms and conditions as
the license granted to You under this License.
b. If any provision of this License is invalid or unenforceable under applicable law, it shall
not affect the validity or enforceability of the remainder of the terms of this License, and
without further action by the parties to this agreement, such provision shall be reformed to
the minimum extent necessary to make such provision valid and enforceable.
c. No term or provision of this License shall be deemed waived and no breach consented to
unless such waiver or consent shall be in writing and signed by the party to be charged with
such waiver or consent.
CREATIVE COMMONS PUBLIC LICENSE 165
d. This License constitutes the entire agreement between the parties with respect to the Work
licensed here. There are no understandings, agreements or representations with respect to
the Work not specified here. Licensor shall not be bound by any additional provisions that
may appear in any communication from You. This License may not be modified without
the mutual written agreement of the Licensor and You.
Bibliography
Alexander, H. R. (2003). Hyperthermia and its modern use in cancer

treatment. Cancer , 98, 219–221.
Arpaci, V. S., Selamet, A., and Kao, S. H., Introduction to Heat

Transfer (Prentice Hall, Upper Saddle River NJ, 2000).
Bejan, A., Convection Heat Transfer (John Wiley & Sons, New York NY,
1984).
Bergmann, G., Graichen, F., Rohlmann, A., Verdonschot, N.,

and van Lenthe, G. H. (2001). Frictional heating of total hip implants:
Part 1: Measurements in patients. Journal of Biomechanics, 34, 421–428.
Boyce, W. E. and DiPrima, R. C., Elementary Differential Equations

and Boundary Value Problems (John Wiley & Sons, New York NY, 1977),
3rd edition.
Burmeister, L. C., Convective Heat Transfer (John Wiley & Sons, New
York NY, 1983).
Carrier, G. F. and Pearson, C. E., Partial Differential Equations:

Theory and Technique (Academic Press, New York NY, 1976).
Dunkle, R. V. (1954). Thermal radiation tables and applications. Trans.

ASME , 76, 549–549.
Gebhart, B., Jaluria, Y., Mahajan, R. L., and Sammakia, B.,

Buoyancy–Induced Flows and Transport (Hemisphere, Washington DC,
1988).
Gillman, L. and McDowell, R. H., Calculus (W. W. Norton and Co.,

New York NY, 1978), 2nd edition.
Incropera, F. P. and Dewitt, D. P., Fundamentals of Heat and Mass

Transfer (John Wiley & Sons, New York NY, 2002), 5th edition.
Kays, W. M. and Crawford, M. E., Convective Heat and Mass Transfer

(McGraw–Hill, New York NY, 1980), 2nd edition.
166
Bibliography 167
Kreyszig, E., Advanced Engineering Mathematics (John Wiley & Sons,

New York NY, 1988), 6th edition.
Lander, E. S., Linton, L. M., Birren, B., Nusbaum, C., Zody,

M. C., Baldwin, J., Devon, K., Dewar, K., Doyle, M., FitzHugh,
W., Funke, R., Gage, D., Harris, K., Heaford, A., Howland, J.,
Kann, L., Lehoczky, J., LeVine, R., McEwan, P., McKernan,
K., Meldrim, J., Mesirov, J. P., Miranda, C., Morris, W., Nay-
lor, J., Raymond, C., Rosetti, M., Santos, R., Sheridan, A.,
Sougnez, C., Stange-Thomann, N., Stojanovic, N., Subrama-
nian, A., Wyman, D., Rogers, J., Sulston, J., Ainscough, R.,
Beck, S., Bentley, D., Burton, J., Clee, C., Carter, N., Coul-
son, A., Deadman, R., Deloukas, P., Dunham, A., Dunham, I.,
Durbin, R., French, L., Grafham, D., Gregory, S., Hubbard, T.,
Humphray, S., Hunt, A., Jones, M., Lloyd, C., McMurray, A.,
Matthews, L., Mercer, S., Milne, S., Mullikin, J. C., Mungall,
A., Plumb, R., Ross, M., Shownkeen, R., Sims, S., Waterston,
R. H., Wilson, R. K., Hillier, L. W., McPherson, J. D., Marra,
M. A., Mardis, E. R., Fulton, L. A., Chinwalla, A. T., Pepin,
K. H., Gish, W. R., Chissoe, S. L., Wendl, M. C., Delehaunty,
K. D., Miner, T. L., Delehaunty, A., Kramer, J. B., Cook,
L. L., Fulton, R. S., Johnson, D. L., Minx, P. J., Clifton, S. W.,
Hawkins, T., Branscomb, E., Predki, P., Richardson, P., Wen-
ning, S., Slezak, T., Doggett, N., Cheng, J. F., Olsen, A., Lu-
cas, S., Elkin, C., Uberbacher, E., Frazier, M., Gibbs, R. A.,
Muzny, D. M., Scherer, S. E., Bouck, J. B., Sodergren, E. J.,
Worley, K. C., Rives, C. M., Gorrell, J. H., Metzker, M. L.,
Naylor, S. L., Kucherlapati, R. S., Nelson, D. L., Weinstock,
G. M., Sakaki, Y., Fujiyama, A., Hattori, M., Yada, T., Toyoda,
A., Itoh, T., Kawagoe, C., Watanabe, H., Totoki, Y., Taylor,
T., Weissenbach, J., Heilig, R., Saurin, W., Artiguenave, F.,
Brottier, P., Bruls, T., Pelletier, E., Robert, C., Wincker,
P., Rosenthal, A., Platzer, M., Nyakatura, G., Taudien, S.,
Rump, A., Yang, H. M., Yu, J., Wang, J., Huang, G. Y., Gu,
J., Hood, L., Rowen, L., Madan, A., Qin, S. Z., Davis, R. W.,
Federspiel, N. A., Abola, A. P., Proctor, M. J., Myers, R. M.,
Schmutz, J., Dickson, M., Grimwood, J., Cox, D. R., Olson,
M. V., Kaul, R., Raymond, C., Shimizu, N., Kawasaki, K., Mi-
noshima, S., Evans, G. A., Athanasiou, M., Schultz, R., Roe,
B. A., Chen, F., Pan, H. Q., Ramser, J., Lehrach, H., Reinhardt,
R., McCombie, W. R., de la Bastide, M., Dedhia, N., Blocker,
H., Hornischer, K., Nordsiek, G., Agarwala, R., Aravind, L.,
Bailey, J. A., Bateman, A., Batzoglou, S., Birney, E., Bork, P.,
Brown, D. G., Burge, C. B., Cerutti, L., Chen, H. C., Church,
D., Clamp, M., Copley, R. R., Doerks, T., Eddy, S. R., Eichler,
Bibliography 168
E. E., Furey, T. S., Galagan, J., Gilbert, J. G. R., Harmon,

C., Hayashizaki, Y., Haussler, D., Hermjakob, H., Hokamp, K.,
Jang, W. H., Johnson, L. S., Jones, T. A., Kasif, S., Kaspryzk,
A., Kennedy, S., Kent, W. J., Kitts, P., Koonin, E. V., Korf, I.,
Kulp, D., Lancet, D., Lowe, T. M., McLysaght, A., Mikkelsen,
T., Moran, J. V., Mulder, N., Pollara, V. J., Ponting, C. P.,
Schuler, G., Schultz, J. R., Slater, G., Smit, A. F. A., Stupka,
E., Szustakowki, J., Thierry-Mieg, D., Thierry-Mieg, J., Wag-
ner, L., Wallis, J., Wheeler, R., Williams, A., Wolf, Y. I.,
Wolfe, K. H., Yang, S. P., Yeh, R. F., Collins, F., Guyer, M. S.,
Peterson, J., Felsenfeld, A., Wetterstrand, K. A., Patrinos,
A., and Morgan, M. J. (2001). Initial sequencing and analysis of the
human genome. Nature, 409, 860–921.
Lee, T. K., Zhong, X. L., Gong, L., and Quinn, R. (2003). Hypersonic
aerodynamic heating prediction using weighted essentially nonoscillatory
schemes. Journal of Spacecraft and Rockets, 40, 294–298.
Mills, A. F., Heat Transfer (Prentice Hall, Upper Saddle River NJ, 1999),
2nd edition.
Odeh, S. D., Behnia, M., and Morrison, G. L. (2000). Hydrodynamic

analysis of direct steam generation solar collectors. Journal of Solar En-
ergy Engineering, 122, 14–22.
Özişik, M. N., Heat Conduction (John Wiley & Sons, New York NY, 1980).
Özişik, M. N., Heat Transfer: A Basic Approach (McGraw–Hill, New York

NY, 1985).
Panton, R. L., Incompressible Flow (John Wiley & Sons, New York NY,
1984).
Pastukhov, V. G., Maidanik, Y. F., Vershinin, C. V., and Ko-

rukov, M. A. (2003). Miniature loop heat pipes for electronics cooling.
Applied Thermal Engineering, 23, 1125–1135.
Pohlhausen, K. (1921). Zur näherungsweise Integration der differential

Gleichung der laminaren Reibungschicht. Zeitscrift für angewandte Math-
ematik und Mechanik , 1, 252–268.
Schlichting, H., Boundary Layer Theory (McGraw–Hill, New York NY,

1979), 7th edition.
Siegel, R. and Howell, J., Thermal Radiation Heat Transfer (Taylor

and Francis, London, 2001), 4th edition.
Bibliography 169
Sinha, S., Kumar, S., and Kumar, N. (1998). Energy conservation in

high–rise buildings: Changes in air conditioning load induced by verti-
cal temperature and humidity profile in Delhi. Energy Conversion and
Management, 39, 437–440.
Taymaz, I., Cakir, K., Gur, M., and Mimaroglu, A. (2003). Ex-
perimental investigation of heat losses in a ceramic coated diesel engine.
Surface and Coatings Technology, 169, 168–170.
TEMA, Standards of the Tubular Exchange Manufacturers Association

(TEMA, New York NY, 1978), 6th edition.
von Karman, T. (1921). Über laminare und turbulente Reibung. Zeitscrift

für angewandte Mathematik und Mechanik , 1, 233–252.
Wood, W. D., Deem, H. W., and Lucks, C. F., Thermal Radiative

Properties (Plenum Press, New York NY, 1964).
Index
Biot number, 40, 44 separable, 38

Bisection Method, 159 Dittus–Boelter correlation, 88
boundary conditions, 13 DNA, 23
adiabatic, 14
Dirichlet, 14, 149 Eckert number, 86
Neumann, 14 eigenvalues, 45, 152
no–slip, 55, 69, 71, 78, 85 emissive power, 120
Robbins, 14
fins, 27–36
boundary layer approximation, 57, 91
effectiveness, 34
boundary layer equations, 59, 92 efficiency, 35
Boussinesq approximation, 92 general equation, 29
buoyancy, 90 Fourier number, 44, 149
Fourier’s Law, 7, 17, 22, 28, 37, 50, 55,
Chain Rule of Calculus, 44, 49, 150
81
characteristic length scale, 41
Churchill–Chu correlation Gaussian Error Function, 50
horizontal cylinder, 98 gel electrophoresis, 23
vertical plate, 97 Grashof number, 93
circuit analogy, 17–23, 143
conduction, 2 heat exchanger, 99–100
equation, 9–12 –NTU analysis, 110–115
constant conductivity, 12 coefficient, 103
cylindrical coordinates, 21 concentric tube, 99
one–dimensional, 13 correction factors, 109
steady, 13 fouling factor, 104
one–dimensional, 16–17 heat capacity rate, 111
circuit analogy, 17–23 LMTD analysis, 104–109
continuum assumption, 4–6 shell–and–tube, 99
control surface, 3 thumbnail diagram, 100–102
control volume, 3 heat generation, 11, 23–27
convection, 2, 53–54 hydraulic diameter, 89
coefficient, 14, 18, 27, 53, 62 hyperbolic trig functions, 30
heat exchanger, overall, 103
Ideal Gas Law, 6, 93
infinite, 47
initial conditions, 13
external, 66–67
irradiation, 121
internal, 77
natural, 90 Kirchhoff’s Circuit Law, 145
Couette flow, 85
L’Hospitals rule, 108
differential equation Laplacian operator, 13
170
INDEX 171
lumped capacitance analysis, 38, 149 resistance, 17, 103

thermal expansion coefficient, 92
M–16 rifle, 2, 22 thumbnail diagram, 100–102
Navier–Stokes Equations, 57, 85 volumetric heat capacity, 8
Newton’s Law of Cooling, 18, 27, 29, 33,
38, 53, 81, 105
Newton’s Law of Viscosity, 55
Nusselt number, 40, 63
overall heat transfer coefficient, 103
Peclet number, 74
Prandtl number, 60, 86
product solution, 50
radiation, 2, 116–117
blackbody, 122
Planck’s Distribution, 123
Stefan–Boltzmann Law, 124
Wien’s Displacement Law, 124
diffuse, 121
emissivity, 126
gray body, 132
intensity, 119
Kirchhoff’s Law, 130
opaque body, 117, 128
re–radiating surface, 147
semi–transparent body, 117, 128
view factor, 134
reciprocity rule, 136
summation rule, 137
radiosity, 122
Rayleigh number, 96
resistors
in parallel, 19
in series, 19
Reynolds number, 56, 60, 78
Reynolds–Colburn analogy, 63
Sieder–Tate correlation, 88
similarity transform, 48, 67, 94
skin friction factor, 62, 64, 68, 73
solid angle, 117
stream function, 67, 95
Taylor series, 9, 28
temperature, 1
absolute and relative, 6
film, 66
gradient, 7–8
thermal
conductivity, 7, 8
contact resistance, 19
diffusivity, 8, 37

Heat Transfer: Fundamentals of

Uploaded by

Copyright:

Available Formats

Heat Transfer: Fundamentals of

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Heat Transfer: Fundamentals of

Uploaded by

Copyright:

Available Formats

Fundamentals Of

(Class Notes for ME 371)

Saint Louis, U.S.A.

Permission is granted to copy, distribute and/or display this document under

This document contains no color figures

Chapter 2. Elementary Heat Conduction 7

Chapter 3. One–Dimensional Steady Conduction 15

Chapter 4. Transient Conduction 37

Chapter 5. Introduction to Convection 53

Chapter 6. External Convection 66

6.1. Laminar Flow Over A Flat Plate 67

Chapter 7. Internal Convection 77

Chapter 8. Natural Convection 90

Chapter 9. Heat Exchangers 99

Chapter 10. Introduction to Radiation 116

Chapter 11. Radiation Exchange 134

Appendix A. Transient Conduction Example: The Homogeneous

Appendix B. Laminar Forced Convection in a Pipe 156

Appendix C. Blackbody Radiation 158

Heat transfer is fundamentally important in almost all aspects of en-

Heat Transfer can be described as the process of energy transmission due

• Solar power generation: design and optimize hardware for maximal

• HVAC1.2: size required systems for ultra–high–rise office towers and

• Biomechanics: analyze body temperature increases as a function of

• Electronics: increase effectiveness of laptop computer cooling sys-

1.1Temperature is a surprisingly difficult concept to define in terms of first principles.

• Medicine: analyze effectiveness of hyperthermia treatments for can-

• Aerodynamics: design fuselage to limit effects of aerodynamic heat-

1.1. Physical Mechanisms

• Convection relies on bulk motion of the medium. Example: cooling

• Radiation occurs by way of electromagnetic phenomena as described

convected air as heat is absorbed, causing a change in density and subse-

1.2. Concept of Conservation of Energy

Figure 1.2. Control volume and control surface schematic.

energy conducted to surface

Figure 1.3. Conservation of energy at an interface. Dotted

t is taken to be vanishingly small, so that the control surface encloses no

1.3. The Continuum Assumption

Figure 1.4. (a) Defining density as a point function according

located at coordinates (x0 , y0 , z0 ), where density is defined simply as mass

Since the (x0 , y0 , z0 ) location of point C is arbitrary, ρ at any point in

1.4. Absolute Versus Relative Temperature

that temperature can be interpreted in an absolute sense, that is, relative

1.7This is also referred to as centigrade in some countries.

∆T = Tk,2 − Tk,1 = (Tc,2 + 273.15) − (Tc,1 + 273.15) = Tc,2 − Tc,1 ,

Elementary Heat Conduction

Here we will introduce the basic concepts of conduction heat transfer.

2.1. Fourier’s Law

Figure 2.1. Heat is conducted along a negative gradient from

2.2. Thermal Properties of Matter

Property Notation Units

as well, including many of those associated with fluid mechanics problems

0.01 0.1 1 10 100 1000

Figure 2.2. Approximate ranges of thermal conductivity for

2.3. The Conduction Equation

Figure 2.3. Differential volume showing energy generation

if conduction is also steady (stationary), Eq. (2.20) reduces to

2.4. Boundary and Initial Conditions

where Ts is the temperature at the surface coinciding with the coordinate